In February 2013, GlaxoSmithKline (GSK) announced a commitment to further clinical transparency through the public disclosure of GSK Clinical Study Reports (CSRs) on the GSK Clinical Study Register. The following guiding principles have been applied to the disclosure: Information will be excluded in order to protect the privacy of patients and all named persons associated with the study Patient data listings will be completely removed* to protect patient privacy. Anonymized data from each patient may be made available subject to an approved research proposal. For further information please see the Patient Level Data section of the GSK Clinical Study Register. Aggregate data will be included; with any direct reference to individual patients excluded *Complete removal of patient data listings may mean that page numbers are no longer consecutively numbered
184
Embed
*Complete removal of patient data listings may …...4 6.1.7 Change from baseline of SDI organ damage system subscores .....62 6.1.8 Frequency of increase from baseline of SDI organ
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
In February 2013, GlaxoSmithKline (GSK) announced a commitment to further clinical transparency through the public disclosure of GSK Clinical Study Reports (CSRs) on the GSK Clinical Study Register.
The following guiding principles have been applied to the disclosure: Information will be excluded in order to protect the privacy of patients and all named
persons associated with the study Patient data listings will be completely removed* to protect patient privacy. Anonymized
data from each patient may be made available subject to an approved research proposal. For further information please see the Patient Level Data section of the GSK Clinical Study Register.
Aggregate data will be included; with any direct reference to individual patients excluded
*Complete removal of patient data listings may mean that page numbers are no longer consecutively
skin, premature gonadal failure, diabetes and malignancy) summarized by year interval of
patients treated with belimumab or SoC, based on data from the US BLISS LTE trial
(BEL112233) and the TLC.
2.3 Exploratory Objectives
As a sensitivity analysis, the primary and secondary objectives above were retested using
pooled data from the BLISS LTE trials (BEL112233 and BEL112234) and the TLC.
3 Selection of the External SLE Cohort
A full report of the selection of the external SLE cohort is attached as Appendix A.2 Briefly, a
systematic literature review was performed to identify cohorts, registries or other databases
formed to support studies in SLE. The objective was to identify an SLE comparison cohort with
population characteristics similar to the BLISS trial population and with an adequate sample of
patients with complete clinical data and at least five years follow-up. Three hundred ninety-three
publications were identified referring to 92 cohorts. Twenty-one cohorts/databases of
approximately 400 or more patients were identified which had been studied in at least 3
19
publications. Data for each of these 21 cohorts were extracted from 317 publications to fill a
data extraction form of 53 items. Evaluation criteria included cohort size, ethnicity, age, duration
of SLE, severity of disease activity, extent of organ damage progression, duration of follow-up,
loss to follow-up, scope of data collection and data availability.
The review identified the Toronto Lupus Cohort (TLC) as the preferred source of SoC data for
this study based on the size of the cohort, the extent of organ damage seen in the patients and
severity of SLE disease activity, which was comparable to the BLISS LTE trial inclusion criteria.
The TLC collects over 500 data points at each visit with additional data collected on an annual
basis. Moreover, the scales for disease severity, organ damage progression and health-related
quality of life were compatible with those used in the BLISS trials (Table 1). A subset of the TLC
with patient baseline characteristics similar to the BLISS trials had previously been used in a
GlaxoSmithKline (GSK) study of mortality and damage progression in SLE.1
Table 1. Comparable instruments used for the BEL112233 LTE and TLC
BEL112233 LTE Toronto Lupus Cohort
SLE Disease Activity SELENA-SLEDAI SLEDAI-2K
SLE Organ Damage SDI SDI
Health-related Quality of Life SF-36 Version 2 SF-36 Version 1
Abbreviations: SDI, SLICC/ACR Damage Index; SELENA, Safety of Estrogens in Lupus Erythematosus National Assessment; SF-36, 36-Item Short Form Survey; SLE, systemic lupus erythematosus; SLEDAI, Systemic Lupus Erythematosus Disease Activity Index
4 Population
Eligibility criteria from the BLISS trials were applied to the patients from the TLC (Table 2).
Application of these criteria produced the raw sample sizes seen in Table 3 at baseline and 1, 2
and 5-years of follow-up.
20
Table 2. Eligibility criteria.
Inclusion Criteria
Diagnosis of systemic lupus erythematosus (ICD-9 710.0) using ≥ 4 of 11 American College of Rheumatology criteria (710.0)
Active severe lupus nephritis or central nervous system lupus
Receipt of B cell target therapy at any time
Abbreviations: SELENA, Safety of Estrogens in Lupus Erythematosus National Assessment; SLEDAI, Systemic Lupus Erythematosus Disease Activity Index
Table 3. Raw sample sizes in BEL112233, BEL112234 and the TLC
Sample Sizes
With ≥ 1 Year
Follow up
With ≥ 2 Years
Follow up
With ≥ 5 Years
Follow up
For Time to Event
Analyses
Toronto Lupus Cohort 940 817 546 940
Year 1-2 Year 2-3 Year 5-6 NA
BEL112233 (US BLISS LTE) 259 252 195 259
Pooled dataset
(BEL 112233, BEL112234) 949 871 592 949
Abbreviations: LTE, long term extension
The baseline date in BEL112233 and BEL112234 was set as the date of first exposure to
belimumab. For patients in the TLC, the baseline date was the first date the patient’s SLEDAI-
2K score reached or exceeded 6.
TLC patients were also excluded for the following reasons:
• Baseline date before 1990 (due to changes in care patterns prior to 1990)
• ≥15 or years follow-up
• No visit within 24 weeks of annual visit timing (due to irregularity of TLC annual visits)
With these exclusions, the sample used for the analyses below included 259 patients from
BEL112233, 949 patients from the pooled BEL 112233 and BEL112234 datasets and 592
patients from the TLC.
21
Table 4. Sample sizes in BEL112233, BEL112234 and the TLC for 5 year analyses and
time to event analyses.
Sample Sizes
With ≥ 1 Year
Follow up
With ≥ 2 Years
Follow up
With ≥ 5 Years
Follow up
For Time to Event
Analyses
Toronto Lupus Cohort 592 499 381 592
Year 1-2 Year 2-3 Year 5-6 NA
BEL112233 (US BLISS LTE) 259 252 195 259
Pooled dataset
(BEL 112233, BEL112234) 949 871 592 949
Abbreviations: LTE, long term extension
5 Methods
5.1 Choice of Propensity Score Matching Variables
Predictors of SLE organ damage were chosen for propensity score matching (PSM) variables. A
recent systematic literature review identifying factors influencing organ damage and damage
progression5 was used to identify publications which reported predictors of SLE organ damage
progression.6–9 These were augmented by an internal GSK study which studied the impact of
disease activity on mortality and organ damage progression.10 The predictors found in the
literature (Table 5) were then reviewed by clinical experts and limited to those for which data
was available in both BEL112233 and the TLC. One variable was available – disease activity
over time – but was not suitable as a PSM variable because it was not a baseline variable. This
process produced the list of 14 PSM variables seen in the first column of Table 6. All 14
variables (17 operationalized variables) were used in the PSM for the primary and secondary
analyses (checked in the second column of Table 6). The exploratory analysis on the pooled
(BEL112233, BEL112234) dataset had 13 PSM variables available (checked in the third column
of Table 6). The PSM variable smoker was excluded from the exploratory analyses on the
pooled dataset due to an inexplicably large difference in proportions between the pooled and
TLC datasets; 2% versus 24%, respectively. The PSM variables were operationalized as 17
variables in the BEL112233 dataset (checked in the fifth column of Table 6) and as 16 variables
in the pooled dataset (checked in the sixth column of Table 6). Definitions of these
operationalized variables from both the US LTE and the TLC cohorts are provided in Table 7.
22
Baseline SDI was operationalized as a categorical variable because there were so few patients
with baseline SDI > 2. The references for the operationalized Race/Ethnicity and Baseline SDI
variables were Caucasian and zero, respectively.
Table 5. Predictors of organ damage found in the literature
Predictors
Age6–8,10
Gender7,8,10
Race/Ethnicity7,10
Household income7
Educational attainment7
SLE duration7,9,10
History - hypertension7
History - dyslipidemia10
History - proteinuria7
History - lupus anticoagulant positivity7
History - anticardiolipin positivity7
History - anti-β2-glycoprotein I positivity7
History – anti-Ro positivity7
Current smoker10
Number of ACR criteria satisfied at diagnosis7
Baseline SLEDAI score8
Disease activity over time (i.e., time-weighted SLEDAI)6,8,9
Corticosteroid use/dose6,7,10
Hydroxychloroquine/other antimalarial drug use7,10
Cyclophosphamide/other immunosuppressive use7,10
Initial or prior SDI6,9
SF-20 physical functioning9
Abbreviations: ACR, American College of Rheumatology; SDI, SLICC/ACR Damage Index; SF-20, 20-Item Short Form Survey; SLE, systemic lupus erythematosus; SLEDAI, Systemic Lupus Erythematosus Disease Activity Index
23
Table 6. Propensity score baseline matching variables in the data and how they were operationalized.
PSM Variables Available in the Data
Primary/Secondary Analysis
Exploratory Analysis
Operationalized
Variables
Primary/Secondary Analysis
Exploratory Analysis
Age X X Age X X
Age squared X X
Gender X X Female X X
Race/Ethnicity X X Black * X X
Asian/Other Race * X X
SLE duration X X SLE duration X X
History - hypertension X X History - hypertension X X
History - dyslipidemia X X History - dyslipidemia X X
History - proteinuria X X History - proteinuria X X
Current smoker X Current smoker X
Number of ACR criteria satisfied at diagnosis
X X Number of ACR criteria satisfied at diagnosis
X X
Baseline SLEDAI score X X Baseline SLEDAI score X X
Corticosteroid use X X Corticosteroid use X X
Antimalarial use X X Antimalarial use X X
Immunosuppressive use X X Immunosuppressive use X X
Baseline SDI X X Baseline SDI = 1 ** X X
Baseline SDI = 2+ ** X X
Abbreviations: ACR, American College of Rheumatology; SDI, SLICC/ACR Damage Index; SLE, systemic lupus erythematosus; SLEDAI, Systemic Lupus Erythematosus Disease Activity Index * Caucasian is the reference ** SDI = 0 is the reference
24
Table 7. Definitions of variables
Operationalized
Variables Variable Type Long Term Extension Toronto Lupus Cohort
Age Continuous Calculated as the difference in years between date of birth and baseline date
Calculated as the difference in years between date of birth and baseline date
Age squared Continuous Calculated as the Age squared Calculated as the Age squared
Female True/False True if Female True if Female
Black * True/False True if subject’s race was Black True if subject’s race was Black
Asian/Other Race * True/False True if subject’s race was neither White nor Black
True if subject’s race was neither White nor Black
SLE duration Continuous Calculated as the difference in years between date of diagnosis and baseline date
Calculated as the difference in years between date of diagnosis and baseline date
Hypertension True/False True if SBP at >140 or DBP at >90 or history of antihypertensive therapy or history of adverse events related to hypertension
True if SBP>140 or DBP>90 or on antihypertensive therapy at baseline
Dyslipidemia True/False True if history of hyperlipidemia therapy or history of adverse events related to high cholesterol
True if total cholesterol at > 5.2 mmol/L or on hyperlipidemia therapy at baseline
Proteinuria True/False True if baseline proteinuria at > 500 mg/day True if baseline proteinuria at > 500 mg/day
Current smoker True/False True if tobacco user at baseline True if tobacco user at baseline
Number of ACR criteria satisfied at diagnosis
Integer Total ACR criteria at baseline Total ACR criteria at baseline
Baseline SLEDAI score Continuous Total SELENA-SLEDAI at baseline Total SLEDAI-2K at baseline
Corticosteroid use True/False True if taking any corticosteroid at baseline True if taking any corticosteroid at baseline
Antimalarial use True/False True if taking any antimalarial medication at baseline
True if taking any antimalarial medication at baseline
Immunosuppressive use True/False True if taking any immunosuppressive medication at baseline
True if taking any immunosuppressive medication at baseline
Baseline SDI = 1 ** True/False True if total SDI score at baseline is equal to 1 True if total SDI score at baseline is equal to 1
25
Baseline SDI = 2+ ** True/False True if total SDI score at baseline is greater
than or equal to 2 True if total SDI score at baseline is greater than or equal to 2
Abbreviations: ACR, American College of Rheumatology; DBP, diastolic blood pressure; mmol/L, millimoles per liter; SBP, systolic blood pressure; SDI, SLICC/ACR Damage Index; SELENA, Safety of Estrogens in Lupus Erythematosus National Assessment; SLE, systemic lupus erythematosus; SLEDAI, Systemic Lupus Erythematosus Disease Activity Index * Caucasian is the reference ** SDI = 0 is the reference
26
5.2 Propensity Score Matching Methods
Propensity scores were calculated using the logistic regression procedure in SAS Version 9.411
in the following manner:
• The model specification initially included all potential predictor variables in Table 6 as
independent variables (age squared was also added, as well as dummy variables for
race/ethnicity and baseline SDI strata) (Full Model)
• In a backward elimination step-wise fashion, the statistically least significant predictor
was dropped from the propensity score model, until all included predictors had a p-value
< 0.1 (Trimmed Model)
• The specific predictors of organ damage included as covariates in the trimmed model
was based on the model specification with the minimum Akaike information criterion
(AIC) value.
• Attention was devoted to assessing the adequacy of the match for baseline SDI score
(as likely the most important predictor of future organ damage), by comparing the
frequency distribution of baseline SDI scores for the belimumab and SoC samples.
The propensity score (PS) value for matching was defined as the estimated log-odds (i.e., the X
value) from the logistic regression, rather than the predicted probability, to enhance the range of
variation the PS distribution for matching. LTE patients were matched 1:1 to TLC patients
based on similar PS value (within a caliper1 value defined as 20% of the standard deviation for
the distribution of the PS variable in the full sample). Unmatched patients were excluded from
the analysis of the PS-matched patient sample. The matching process was implemented using
a commonly used SAS macro.12
Four sets of matches were performed. For the primary and secondary analysis, matching was
performed on the BEL112233 data for 1) the analyses requiring 5 years of follow-up and 2)
time-to-event analyses requiring ≥ 1 year of follow-up. Likewise, for the exploratory analysis,
matching was performed on the pooled data for 1) analyses requiring 5 years of follow-up and
2) time-to-event analyses requiring ≥ 1 year of follow-up.
PSM was performed twice, once with a PS value based on a model with all candidate covariates
(full model), and then with a PS value based on a model with selected covariates deleted
(trimmed model), as described above. Post-PSM balance in covariates using the full-model PS
1 the maximum permitted difference between matched subjects
27
was superior to balance using the trimmed model PS, so the former sample was selected for all
PS-matched sample analyses.
5.3 Assessment of Post PSM Covariate Balance
The measure of balance (bias) used was the standardized distance across the variables used to
determine the PS value for each patient. Standardized distance is defined as
d = (X̅T - X̅
C) /√[Var(XT) + Var(XC)]/2
for continuous variables and
d = (P̂T - P̂
C) /√[(P̂T(1 − P̂T) + (P̂C(1 − P̂C)]/2
for binary variables.
Generally, for adequate balance the standardized distance should be no larger than 10% for all
variables used to determine PS values. Ideally the standardized distance would be less than
5% for all variables.13 Variables with standardized distances larger than 10% were added as
covariates to each analysis.
5.4 Visit Selection
Longitudinal data was available from both datasets for estimation of change in SDI. Visit
intervals in the BEL112233 and pooled datasets were fixed. In BEL112233 the final visit of the
parent trial was at 76 weeks. Thereafter SDI was recorded every 48 weeks while SLEDAI was
recorded every 24 weeks. Visits at weeks 52 and 76 of the parent trials and every 48 weeks
thereafter were selected as SDI “annual” visits, with a short 24-week “second year” between
weeks 52 and 76 (SDI was not recorded at the week 100 visit.) In the pooled dataset the final
visits of the parent trials were at 52 or 76 weeks. Depending on the parent trial, “annual” visits
occurred as in BEL112233 (weeks 52 and 76) or at weeks 52 and 100. Thereafter, annual visits
occurred every 48 weeks. In both cases, if the patient was assigned to placebo in the parent
trial “annual” visits were every 48 weeks after the start of the LTE.
In the TLC data, visits were not performed at precise time intervals. For the purposes of the time
to event analyses, “annual” visits were defined based on the interval from the baseline date to
the visit closest to each 48-week interval and deviating no more than 24 weeks from that
interval.
28
For other SDI analyses, TLC “annual” visits were the visits that most closely matched the
intervals from baseline date of the “annual” visits of the LTE patients to whom they were
matched.
All SDI analyses were based only on “annual” visits, omitting other visits in each dataset. The
secondary analyses of mean SLEDAI and corticosteroid usage over 5 years include all visits
through the 5th “annual” visit in all three datasets. Since SLEDAI and corticosteroid usage was
recorded every 24 weeks in BEL112233, these 5 year analyses were through 240 weeks for
patients receiving placebo and 244 weeks for patients receiving belimumab in the parent study.
5.5 Primary and Secondary Endpoints
All inferential statistics were two-tail tests performed with an alpha of p=0.05.
5.5.1 Primary endpoint: difference in change of SDI from baseline to 5 years
Change of total SDI from baseline to 5 years was evaluated using linear regression with change
of total SDI from baseline as the dependent variable, and with a variable indicating treatment
group (belimumab or SoC). Unbalanced matching variable(s) determined via Section 5.3 above
were added as covariates. If statistically significant, the decade of entry into the study was also
a covariate. As a sensitivity analysis, the primary endpoint was also evaluated using inverse
propensity score weighting (IPSW), a PS method that uses the entire sample and the PS to
weight the observations, to confirm the robustness of results.
5.5.2 Difference in time to first SDI worsening
The time to the first worsening (increase) in total SDI score were analyzed using parametric
survival models with a binary indicator for treatment with belimumab as the covariate.
Unbalanced matching variable(s) determined via Section 5.3 above were added as covariates. If
statistically significant, the decade of entry into the study was also added as a covariate.
5.5.3 Change from baseline SDI score by year interval
Descriptive statistics of the change from baseline SDI score were estimated at the end of years
1 through 5 for both the belimumab and SoC groups. The counts and proportions of subjects in
each treatment arm, out to year 5, of the incremental changes from baseline of total SDI were
calculated.
Also, a continuation-ratio logit model was used to analyze the change from baseline for total SDI
score for each year. A binary indicator for treatment with belimumab was the only covariate. All
29
increases in total SDI ≥2 were combined into one category. Continuation-ratio logits were used
to model (1) the probability of any change (ΔSDI > 0) from baseline for a given year, and (2) the
conditional probability of a change greater than one given that there was a change from
baseline. The unconstrained continuation-ratio model allows for unequal odds ratios for the two
The probabilities for each of the three categories can be directly retrieved from the two logit
formulas. For subjects treated with belimumab the probabilities of each category were derived
as follows:
𝑃(ΔSDI ≥ 1) = 𝜋1 + 𝜋2 =exp [𝛼1 + 𝛽1]
1 + exp [𝛼1 + 𝛽1]
𝜋0 = 1 - (𝜋1 + 𝜋2) = 1 −exp [𝛼1+𝛽1]
1+exp [𝛼1+𝛽1]
𝜋+2 = exp [𝛼2 + 𝛽2]×𝜋1 = (exp [𝛼2+𝛽2]
1+exp [𝛼2+𝛽2]) (
exp [𝛼1+𝛽1]
1+exp [𝛼1+𝛽1])
𝜋1 = (exp [𝛼1 + 𝛽1]
1 + exp [𝛼1 + 𝛽1]) (
1
1 + exp [𝛼2 + 𝛽2])
For a given year the odds of any increase from baseline in total SDI score for subjects taking
belimumab versus those receiving SoC is given by exp(𝛽1). For subjects taking belimumab
versus those receiving SoC the odds of an increase greater than 1 given that there was any
increase is given by exp(𝛽2).
Using three categories to measure change in SDI leaves four degrees of freedom for each year
and thus the unconstrained model is also a saturated model. The fully constrained model
34
(𝛽1=𝛽2) specifies equality of odds ratios and is nested within the unconstrained model. The
deviance of the unconstrained model was used to test equality of odds ratios for the two
transitions. If the constrained model provided a poor fit the unconstrained model was fit and a
Wald test was used to test the significance of treatment effects.
5.6.4 Difference of change from baseline SDI by year interval
Change of SDI from baseline to end of years 1 through 5 were evaluated using linear regression
with change of SDI from baseline as the dependent variable, and with a variable indicating
treatment group (belimumab or SoC). Unbalanced matching variable(s) determined via Section
5.3 above were added as covariates. The decade of entry into the study was also a covariate.
5.6.5 Transition analysis of SDI from baseline over a 5-year interval
The SAP specified that multi-state Markov modelling transition analysis of SDI would be
performed independently for the belimumab and SoC groups using the Jackson et al. 2011
methodology.3,4 This methodology calculates transition probabilities between health states over
time, in this case health states defined by SDI strata.
Due to the number of empty cells in such a transition matrix, annual transition probabilities
instead were estimated based on the time to first SDI worsening analysis Section 5.6.2 above.
5.6.6 Change from baseline of SDI organ damage system subscores
Descriptive statistics of the change from baseline SDI organ system subscores were estimated
at the end of years 1 through 5 for the belimumab and SoC groups. The counts and proportions
of the incremental changes for each of the SDI organ system subscores were calculated.
Counts for any change from baseline were combined into one category and a two-sided Fisher’s
exact test was used to test for independence of no change from baseline in each SDI organ
system subscore based on treatment arm.
5.6.7 Frequency of increase from baseline of SDI organ damage system
subscores
The SAP specified that the frequency of increase of SDI organ system subscores from baseline
to censoring between patients treated with belimumab or SoC would be evaluated using logistic
regression with a variable indicating treatment group (belimumab or SoC) as the dependent
variable, and with the change of SDI organ system subscore from baseline and unbalanced
matching variable(s) determined via Section 5.3 above as covariates. The decade of entry into
the study would also be a covariate.
35
Because of the low frequencies of SDI organ damage system subscores and the non-
dichotomous nature of the values, changes from baseline for SDI organ system subscores were
instead analyzed using linear regression with the difference between the subject’s score in their
final year and their baseline score used as the response variable and an indicator variable for
treatment with belimumab as an independent variable. A categorical variable for the decade of
entry was included if statistically significant. Unbalanced matching variable(s) determined via
Section 5.3 above were added as covariates. The year from baseline was also included to
control for the length of time from baseline.
The data set for this analysis consisted of the matched patients from the propensity score
matching where subjects were not restricted to at least 5 years of follow-up. A second analysis
was performed using the smaller dataset of matched patients with 5 years follow-up. The results
from the second analysis were used to check the robustness of the results where subjects’
scores were recorded in different years.
5.6.8 Difference in mean SLEDAI score from baseline over a 5-year interval
The SAP specified that mean SLEDAI score from baseline through year 5 would be evaluated
using linear regression with mean SLEDAI score as the dependent variable and with a variable
indicating treatment group (belimumab or SoC). Unbalanced matching variable(s) determined
via 5.3 above would be added as covariates. The decade of entry into the study would also be
a covariate.
The BEL 112234 dataset did not contain longitudinal SLEDAI scores. Therefore, this analysis
could not be undertaken.
5.6.9 Difference in cumulative corticosteroid usage from baseline over a 5-year
interval
The SAP specified that cumulative use of corticosteroids from baseline through year 5 would be
evaluated using linear regression with cumulative corticosteroid use as the dependent variable
and with a variable indicating treatment group (belimumab or SoC). Unbalanced matching
variable(s) determined via 5.3 above would be added as covariates.
The BEL 112234 dataset did not contain enough longitudinal concomitant medication data for
this analysis to be feasible. Therefore, this analysis could not be undertaken.
5.7 Diagnostic Analyses
All inferential statistics were two-tail tests performed with an alpha of p=0.05.
36
5.7.1 Baseline characteristics of unmatched study arms
5.7.1.1 BEL112233 LTE and TLC
Comparisons of baseline characteristics were made using all subjects included in the
population. (See Section 4 above.) A separate analysis was performed for all subjects with at
least 5 years of follow-up as well as all subjects with at least 1 year of follow-up.
5.7.1.2 Pooled LTE and TLC
Comparisons of baseline characteristics was made using all subjects included in the population.
(See Section 4 above.) A separate analysis was also performed for all subjects with at least 5
years of follow-up as well as all subjects with at least 1 year of follow-up.
5.7.1.3 Comparison methods
Welch’s t-test was utilized to test for equality of means between study arm baseline
characteristics. The degrees of freedom for the test was approximated using the Welch-
Satterthwaite equation. The test statistic was calculated by the following formula:
t = �̅�𝑡−�̅�𝑐
√𝑠𝑡
2
𝑁𝑡+
𝑠𝑐2
𝑁𝑐
where:
• t (treatment) = belimumab
• c (control) = SoC
• �̅�𝑖 is the sample mean
• 𝑠𝑖2 is the sample variance
• 𝑁𝑖 is the number of subjects
The standardized difference is reported as the percent bias (%bias):
%bias =�̅�𝑡−�̅�𝑐
√(𝑠𝑡2+𝑠𝑐
2)/2
×100
5.7.2 Baseline characteristics of matched samples
5.7.2.1 BEL112233 LTE and TLC
Comparisons of baseline characteristics were made using all matched subjects. A separate
analysis was performed for matched subjects with at least 5 years of follow-up as well as all
subjects with at least 1 year of follow-up.
37
5.7.2.2 Pooled LTE and TLC
Comparisons of baseline characteristics were made using all matched subjects. A separate
analysis was performed for matched subjects with at least 5 years of follow-up as well as all
subjects with at least 1 year of follow-up.
5.7.2.3 Comparison methods
Study arms were tested for statistically significant differences in patient baseline characteristics
using Welch’s t-test, Section 5.7.1.3 above. The standardized mean difference was also
determined for each covariate.
5.7.3 Distribution of year 5 data point timing
5.7.3.1 US BLISS LTE and TLC
Patients in the TLC were not seen at specific intervals. Likewise, patients originating in
belimumab and SoC arms of the underlying belimumab trials had different intervals of
observation. Therefore the 5th year observation in both arms took place at time points not
strictly 5 years from baseline. The distributions of time from baseline to the 5th year observation
were reported.
5.7.3.2 Pooled BLISS LTE and TLC
Patients in the TLC were not seen at specific intervals. Likewise, patients originating in
belimumab and SoC arms of the underlying belimumab trials had different intervals of
observation. Therefore, the 5th year observation in both arms took place at time points not
strictly 5 years from baseline. The distributions of time from baseline to the 5th year observation
were reported.
5.7.3.3 Comparison methods
A paired t-test was used to test for differences between matched subjects in the length of
elapsed time from baseline.
5.7.4 Patients withdrawing from LTE and TLC cohorts
5.7.4.1 BEL112233 LTE and TLC
An analysis was conducted that included all study participants in both cohorts, i.e. subjects that
completed the full five years of follow-up as well as subjects that dropped out before study end.
The impact of the dropout rates were assessed by comparing those who completed the study
versus those who did not complete the study in terms of baseline and clinical characteristics.
38
5.7.4.2 Pooled LTE and TLC
An analysis was conducted that included all study participants in both cohorts, i.e. subjects that
completed the full five years of follow-up as well as subjects that dropped out before study end.
The impact of the dropout rates were assessed by comparing those who completed the study
versus those who did not complete the study in terms of baseline and clinical characteristics.
5.7.4.3 Comparison methods
Time to event analyses were used to test for differences in clinical outcomes. The time to event
analysis consisted of: (1) time to first SDI change; (2) time to mild/moderate flare; and, (3) time
to severe flare. The BEL112234 dataset did not contain longitudinal flare data. Therefore, the
time to mild/moderate and severe flare analysis could not be undertaken with the pooled
sample.
5.7.5 BLISS LTE subjects randomized to SoC in parent trial
5.7.5.1 BEL112233 LTE and TLC
Analysis was conducted to test for study effects associated with differences in quality of care
obtained within the setting of the randomized clinical trials. These analyses were performed by
comparing BLISS subjects randomized to SoC against TLC patients to determine whether
enrollment in the randomized clinical trial had a significant effect on clinical outcomes
associated with SoC. Propensity score matching were used to match BLISS SoC subjects to
TLC patients based on the BLISS subjects’ characteristics at core baseline, i.e., at
randomization to SoC + placebo in BLISS 76 (BEL110751).
5.7.5.2 Pooled LTE and TLC
Analysis was conducted to test for study effects associated with differences in quality of care
obtained within the setting of the randomized clinical trials. These analyses were performed by
comparing BLISS subjects randomized to SoC against TLC patients to determine whether
enrollment in the randomized clinical trial had a significant effect on clinical outcomes
associated with SoC. Propensity score matching were used to match BLISS SoC subjects to
TLC patients based on the BLISS subjects’ characteristics at core baseline, i.e., at
randomization to SoC + placebo.
5.7.5.3 Comparison Methods
Fisher’s exact test was used to test for differences in SDI worsening during the parent study
compared with clinical outcomes in the TLC over a follow-up period equal to the duration of the
parent study.
39
5.7.6 Belimumab baseline of BLISS LTE subjects
5.7.6.1 BEL112233 LTE and TLC
Analysis was conducted to test for potential biases introduced by whether the patient was
randomized to belimumab or placebo in the parent study. This analysis focused on tests of
equivalence among the subgroups of BLISS LTE subjects randomized to belimumab 10 mg/kg
in the parent study, BLISS LTE subjects randomized to belimumab 1 mg/kg in the parent study,
and BLISS LTE subjects randomized to placebo in the parent study.
5.7.6.2 Pooled LTE and TLC
Analysis was conducted to test for potential biases introduced by whether the patient was
randomized to belimumab or placebo in the parent study. This analysis focused on tests of
equivalence among the subgroups of BLISS LTE subjects randomized to belimumab 10 mg/kg
in the parent study, BLISS LTE subjects randomized to belimumab 1 mg/kg in the parent study,
and BLISS LTE subjects randomized to placebo in the parent study.
5.7.6.3 Comparison methods
Time to event analyses were used to test for differences in clinical outcomes during the follow-
up period of the comparative effectiveness analysis. The time to event analysis consisted of: (1)
time to first SDI change; (2) time to mild/moderate flare; and, (3) time to severe flare. The
BEL112234 dataset did not contain longitudinal flare data. Therefore, the time to mild/moderate
and severe flare analysis could not be undertaken with the pooled sample.
6 Results
6.1 Primary and Secondary Analyses
6.1.1 Propensity score matching
6.1.1.1 BEL112233 and TLC Patients with 5-years follow up
Table 8 and Table 9 show the results of the full propensity score logistic regression model over
the entire sample of 567 patients. The range of the PS distribution (Table 9) was -9.927 to
4.701. The range of common support (the range of “overlap” in the PS distributions) for the LTE
and TLC patient was -3.648 to 2.893, illustrated in Figure 1. With the caliper value of 0.53 (20%
of the standard deviation for the PS distribution), the range of support was -4.178 to 3.423. 95
TLC patients and the 11 LTE patients with PS values outside of the range of support (including
the caliper) could not be matched.
40
Using the PS values calculated from the full propensity score logistic regression model, 99 of
195 belimumab patients were matched 1:1 to 99 of the 372 TLC patients.
Table 8. Results of full propensity score logistic regression model, BEL112233 and TLC
dataset with 5 years follow-up (N=567)
Parameter Odds Ratio SE z p-value
Intercept 0.000 0.000 -5.45 <0.001
Age 1.332 0.085 4.51 <0.001
Age Squared 0.997 0.001 -4.11 <0.001
Female 0.968 0.437 -0.07 0.943
Black 0.907 0.296 -0.3 0.765
Asian/Other Race 0.302 0.116 -3.13 0.002
SLE Duration 0.986 0.018 -0.76 0.449
Smoker 0.049 0.026 -5.73 <0.001
Hypertension 4.382 1.248 5.19 <0.001
Dyslipidemia 0.142 0.043 -6.42 <0.001
Proteinuria 0.234 0.086 -3.96 <0.001
ACR Criteria 1.181 0.114 1.72 0.085
Baseline SLEDAI 0.946 0.030 -1.77 0.076
Corticosteroid Use 1.505 0.430 1.43 0.153
Antimalarial Use 2.931 0.800 3.94 <0.001
Immunosuppressive Use 2.771 0.762 3.71 <0.001
Baseline SDI = 1 2.928 0.969 3.25 0.001
Baseline SDI = 2+ 4.920 1.850 4.24 <0.001
Abbreviations: ACR, American College of Rheumatology; SDI, SLICC/ACR Damage Index; SE, standard error; SLE, Systemic lupus erythematosus; SLEDAI, Systemic Lupus Erythematosus Disease Activity Index; TLC, Toronto Lupus Cohort
Table 9. Summary statistics of PSM variable, BEL112233 and TLC dataset with 5 years
follow-up (N=567)
Statistic Value
Observations 567
Mean (SD) -1.365 (2.631)
Range -9.927, 4.701
Caliper (20% of SD) 0.53
Abbreviation: SD, standard deviation; TLC, Toronto Lupus Cohort
41
Figure 1. Common support in full model with all patients (n=567)
Prior to PSM, the LTE and TLC samples were not well balanced (Table 10). The percent bias is
larger than 10% for most of the variables (mean bias = 40%).
However, the PS-matched samples of 99 LTE and 99 TLC patients were well balanced (Table
11). Bias is less than 5% for nine of the seventeen variables, and less than 10% for all
variables (the mean bias is 4.6%).
-10
-50
5
Pre
dic
ted
PS
LTE TLC
42
Table 10. Bias prior to propensity score matching, BEL112233 and TLC dataset with 5
years follow-up (N=567)
Mean t-test
Variable Belimumab SoC % Bias t p>|t|
Age 42.769 37.303 45.5 5.01 <0.001
Age Squared 1947.4 1560.8 38.1 4.22 <0.001
Female 0.928 0.895 11.6 1.28 0.200
Black 0.231 0.153 19.7 2.29 0.022
Asian/Other Race 0.092 0.234 -39.0 -4.18 <0.001
SLE Duration 7.947 5.762 30.0 3.38 0.001
Smoker 0.036 0.237 -61.1 -6.27 <0.001
Hypertension 0.677 0.376 63.0 7.09 <0.001
Dyslipidemia 0.226 0.581 -77.5 -8.55 <0.001
Proteinuria 0.123 0.317 -48.1 -5.18 <0.001
ACR Criteria 5.923 5.651 19.8 2.22 0.027
Baseline SLEDAI 7.785 10.056 -48.4 -5.28 <0.001
Corticosteroid use 0.636 0.608 5.8 0.66 0.510
Antimalarial Use 0.738 0.519 46.6 5.17 <0.001
Immunosuppressive Use 0.538 0.315 46.4 5.31 <0.001
Baseline SDI = 1 0.272 0.148 30.7 3.60 <0.001
Baseline SDI = 2+ 0.287 0.108 46.2 5.55 <0.001
Abbreviations: ACR, American College of Rheumatology; PS, propensity score; SDI, SLICC/ACR Damage Index; SE, standard error; SLE, Systemic lupus erythematosus; SLEDAI, Systemic Lupus Erythematosus Disease Activity Index; TLC, Toronto Lupus Cohort
43
Table 11. Bias post PS matching, BEL112233 and TLC dataset with 5 years follow-up
(n=198)
Mean t-test
Variable Belimumab SoC % Bias t p>|t|
Age 39.980 38.993 8.4 0.59 0.557
Age Squared 1733.0 1661.7 7.2 0.51 0.611
Female 0.929 0.919 3.8 0.27 0.790
Black 0.212 0.232 -4.8 -0.34 0.734
Asian/Other Race 0.141 0.121 6.0 0.42 0.676
SLE Duration 7.368 7.569 -2.6 -0.19 0.853
Smoker 0.071 0.071 0.0 0.00 1.000
Hypertension 0.545 0.535 2.0 0.14 0.887
Dyslipidemia 0.283 0.313 -6.6 -0.46 0.643
Proteinuria 0.202 0.182 5.1 0.36 0.720
ACR Criteria 6.030 5.939 6.5 0.46 0.648
Baseline SLEDAI 8.455 8.546 -2.2 -0.16 0.875
Corticosteroid use 0.646 0.667 -4.2 -0.30 0.766
Antimalarial Use 0.697 0.687 2.2 0.15 0.878
Immunosuppressive Use 0.455 0.444 2.0 0.14 0.887
Baseline SDI = 1 0.242 0.273 -6.9 -0.49 0.628
Baseline SDI = 2+ 0.152 0.182 -8.1 -0.57 0.570
Abbreviations: ACR, American College of Rheumatology; SDI, SLICC/ACR Damage Index; SE, standard error; SLE, Systemic lupus erythematosus; SLEDAI, Systemic Lupus Erythematosus Disease Activity Index; TLC, Toronto Lupus Cohort
6.1.1.2 BEL112233 and TLC patients with ≥ 1-year follow up for time to event analyses
Table 12 and Table 13 show the results of the full propensity score logistic regression model
over the entire sample of 965 patients. The range of the PS distribution (Table 13) was -8.475 to
3.645. The range of common support (the range of “overlap” in the PS distributions) for the LTE
and TLC patient was -3.928 to 2.171, illustrated in Figure 2. With the caliper value of 0.400
(20% of the standard deviation for the PS distribution), the range of support was -4.328 to
2.571. Two hundred and forty-six TLC patients and the 13 LTE patients with PS values outside
of the range of support (including the caliper) cannot be matched.
Using the PS values calculated from the full propensity score logistic regression model, 179 of
259 belimumab patients were matched 1:1 to 179 of the 706 TLC patients.
44
Table 12. Results of full propensity score logistic regression model, BEL112233 and TLC
dataset with ≥ 1 year follow-up (N=965)
Parameter Odds Ratio SE z p-value
Intercept 0.000 0.000 -8.300 <.0001
Age 1.336 0.057 6.820 <.0001
Age Squared 0.997 0.000 -6.090 <.0001
Female 1.280 0.427 0.740 0.4580
Black 0.765 0.194 -1.050 0.2920
Asian/Other Race 0.260 0.075 -4.640 <.0001
SLE Duration 0.962 0.013 -2.840 0.0050
Smoker 0.069 0.028 -6.490 <.0001
Hypertension 1.709 0.361 2.540 0.0110
Dyslipidemia 0.421 0.093 -3.920 <.0001
Proteinuria 0.321 0.090 -4.070 <.0001
ACR Criteria 1.248 0.091 3.040 0.0020
Baseline SLEDAI 0.917 0.022 -3.540 <.0001
Corticosteroid use 1.375 0.277 1.580 0.1140
Antimalarial Use 2.118 0.420 3.780 <.0001
Immunosuppressive Use 2.530 0.510 4.610 <.0001
Baseline SDI = 1 3.186 0.788 4.690 <.0001
Baseline SDI = 2+ 4.618 1.264 5.590 <.0001
Abbreviations: ACR, American College of Rheumatology; SDI, SLICC/ACR Damage Index; SE, standard error; SLE, Systemic lupus erythematosus; SLEDAI, Systemic Lupus Erythematosus Disease Activity Index; TLC, Toronto Lupus Cohort
Table 13. Summary statistics of PSM variable, BEL112233 and TLC dataset with ≥ 1 year
follow-up (N=965)
Statistic Value
Observations 965
Mean (SD) -1.678 (2.027)
Range -8.475, 3.645
Caliper (20% of SD) 0.405
Abbreviation: SD, standard deviation; TLC, Toronto Lupus Cohort
45
Figure 2. Common support in full model with all patients (N=965)
Prior to PSM, the LTE and TLC samples are not well balanced (Table 14). The percent bias is
larger than 10% for most of the variables (mean bias = 35%).
However, the PS-matched samples of 179 LTE and 179 TLC patients are well balanced (Table
15). Bias is less than 5% for all but one variable, and less than 10% for all variables (the mean
bias is 2.2%).
46
Table 14. Bias prior to PS matching, BEL112233 and TLC dataset with ≥ 1 year follow-up
(N=965)
Mean t-test
Variable Belimumab SoC % Bias t p>|t|
Age 42.575 36.886 46.0 6.08 <0.001
Age Squared 1937.4 1541.0 37.6 5.01 <0.001
Female 0.934 0.888 16.3 2.13 0.033
Black 0.216 0.146 18.3 2.62 0.009
Asian/Other Race 0.093 0.280 -49.6 -6.26 <0.001
SLE Duration 7.746 6.208 21.5 2.91 0.004
Smoker 0.039 0.242 -61.2 -7.37 <0.001
Hypertension 0.533 0.380 31.1 4.31 <0.001
Dyslipidemia 0.228 0.347 -26.5 -3.55 <0.001
Proteinuria 0.135 0.330 -47.4 -6.10 <0.001
ACR Criteria 5.985 5.677 22.0 3.04 0.002
Baseline SLEDAI 7.857 10.030 -49.0 -6.37 <0.001
Corticosteroid use 0.649 0.625 5.0 0.68 0.494
Antimalarial Use 0.718 0.564 32.6 4.39 <0.001
Immunosuppressive Use 0.552 0.344 42.7 5.94 <0.001
Baseline SDI = 1 0.278 0.142 33.9 4.96 <0.001
Baseline SDI = 2+ 0.278 0.102 46.0 6.96 <0.001
Abbreviations: ACR, American College of Rheumatology; PS, propensity score; SDI, SLICC/ACR Damage Index; SE, standard error; SLE, Systemic lupus erythematosus; SLEDAI, Systemic Lupus Erythematosus Disease Activity Index; TLC, Toronto Lupus Cohort
47
Table 15. Bias post PS matching, BEL112233 and TLC dataset with ≥ 1 year follow-up
(N=358)
Mean t-test
Variable Belimumab SoC % Bias t p>|t|
Age 40.425 40.697 -2.4 -0.22 0.823
Age Squared 1763.4 1792.3 -3.0 -0.28 0.779
Female 0.916 0.916 0.0 0.00 1.000
Black 0.223 0.235 -2.7 -0.25 0.802
Asian/Other Race 0.128 0.128 0.0 0.00 1.000
SLE Duration 7.511 7.742 -3.2 -0.30 0.766
Smoker 0.056 0.067 -4.6 -0.44 0.661
Hypertension 0.458 0.458 0.0 0.00 1.000
Dyslipidemia 0.251 0.229 5.2 0.49 0.622
Proteinuria 0.168 0.179 -2.9 -0.28 0.781
ACR Criteria 5.955 5.927 1.9 0.18 0.855
Baseline SLEDAI 8.369 8.503 -3.7 -0.35 0.729
Corticosteroid use 0.682 0.693 -2.4 -0.23 0.820
Antimalarial Use 0.659 0.670 -2.4 -0.22 0.823
Immunosuppressive Use 0.458 0.464 -1.1 -0.11 0.916
Baseline SDI = 1 0.246 0.257 -2.6 -0.24 0.808
Baseline SDI = 2+ 0.168 0.168 0.0 0.00 1.000
Abbreviations: ACR, American College of Rheumatology; PS, propensity score; SDI, SLICC/ACR Damage Index; SE, standard error; SLE, Systemic lupus erythematosus; SLEDAI, Systemic Lupus Erythematosus Disease Activity Index; TLC, Toronto Lupus Cohort
6.1.2 Primary endpoint – Difference in change in SDI from baseline to 5 years
The total SDI score change from baseline to 5 years was evaluated using linear regression with
a binary indicator for treatment with belimumab as a covariate. Ninety-nine BEL112233 patients
were matched to 99 TLC patients using PSM. All PSM variables were balanced (Table 11) so
none were added as covariates. The baseline decade of entry also was not significant, so it was
not added as a covariate.
The difference in total SDI score change from baseline to the 5th year for PSM patients was
significantly lower (-0.4343, p<0.001) for subjects taking belimumab.
48
Table 16. Year 5 Total SDI difference of change from baseline
Variable
OLS Coefficient (SE)
[95% CI] P value
Robust SE Coefficient (SE)
[95% CI] P value
Intercept 0.7172 (0.0886)
95% CI: [0.5425 ; 0.8918] p<0.001
0.7172 (0.1106) 95% CI: [0.5004 ; 0.9339]
p<0.001
Belimumab -0.4343 (0.1252)
95% CI: [-0.6813 ; -0.1874] p<0.001
-0.4343 (0.1188) 95% CI: [-0.6673 ; -0.2014]
p<0.001
Abbreviations: CI, confidence interval; OLS, ordinary least squares; SDI, SLICC/ACR Damage Index; SE, standard error
As a sensitivity analysis, results were also produced using the entire sample of 567 patients and
IPSW. The difference in total SDI from baseline to the 5th year was very similar to the PSM
results (Table 17). However, weighted bias was statistically inadequate, with bias greater than
10% for 9 of the 17 propensity score variables (Table 18). An additional regression augmented
IPSW analysis was produced, adding variables with bias > 10% (Table 19) to the regression
model. All three propensity score methods estimated a significantly lower (p < 0.001) increase in
SDI from baseline to the 5th year of -0.4343 to -0.4499 (Table 20), when comparing belimumab
treatment with SoC.
Table 17. Year 5 Total SDI difference of change from baseline using inverse propensity
score weighting (N=567)
Variable
Coefficient (SE)
[95% CI] P value
Belimumab -0.4405 (0.1163)
95% CI: [-0.6685 ; -0.2216] p<0.001
Abbreviations: CI, confidence interval; SDI, SLICC/ACR Damage Index; SE, standard error
49
Table 18. Bias using inverse propensity score weighting (N=567)
Standardized differences (% Bias)
Variable Raw Weighted
Age 45.5% 15.2%
Age Squared 38.1% 12.2%
Female 11.6% 11.4%
Black 19.7% 18.1%
Asian/Other Race -39.0% -4.6%
SLE Duration 30.0% 7.9%
Smoker -61.1% -3.3%
Hypertension 63.0% 28.8%
Dyslipidemia -77.5% -13.5%
Proteinuria -48.1% -12.0%
ACR Criteria 19.8% 7.4%
Baseline SLEDAI -48.4% 0.6%
Corticosteroid use 5.8% 8.3%
Antimalarial Use 46.6% 25.6%
Immunosuppressive Use 46.4% 11.6%
Baseline SDI = 1 30.7% 8.8%
Baseline SDI = 2+ 46.2% 7.3%
Abbreviations: ACR, American College of Rheumatology; SDI, SLICC/ACR Damage Index; SLE, Systemic lupus erythematosus; SLEDAI, Systemic Lupus Erythematosus Disease Activity Index
Table 19. Year 5 Total SDI difference of change from baseline using regression
augmented IPSW (N=567)
Variable
Coefficient (SE)
[95% CI] P value
Belimumab -0.4499 (0.1155)
95% CI: [-0.6763 ; -0.2234] p<0.001
Abbreviations: CI, confidence interval; IPSW, inverse propensity score weighting; SDI, SLICC/ACR Damage Index; SE, standard error
50
Table 20. Year 5 Total SDI difference of change from baseline all methods (N=567)
Method
Coefficient (SE)
[95% CI] P value
Propensity Score Matched -0.4343 (0.1188)
95% CI: [-0.6673 ; -0.2014] p<0.001
IPSW* -0.4405 (0.1163)
95% CI: [-0.6685 ; -0.2216] p<0.001
Regression Augmented IPSW -0.4499 (0.1155)
95% CI: [-0.6763 ; -0.2234] p<0.001
Abbreviations: CI, confidence interval; IPSW, inverse propensity score weighting; SDI, SLICC/ACR Damage Index; SE, standard error * Bias statistically inadequate.
6.1.3 Difference in time to first SDI worsening
The time to the first worsening (increase) in total SDI score was analyzed using parametric
survival models with a binary indicator for treatment with belimumab as the covariate. All PSM
variables were balanced (Table 15) so none were added as covariates. The baseline decade of
entry also was not significant, so it was not added as a covariate. Results for exponential,
Weibull, Gompertz, log logistic, and log normal distributions were evaluated (Table 21 to Table
25). Results indicated that belimumab was associated with a significantly lower rate of organ
damage progression, p<0.001, regardless of the distribution used.
Table 21. Proportional hazards model of time to first change in total SDI score,
exponential distribution
Variable
Regression Coefficient
Estimate (SE)
[95% CI]
Hazard Rate/Ratio
Estimate (SE)
[95% CI] p value
Intercept -2.3967 (0.1209)
95% CI: [-2.6337 ; -2.1596]
0.0910 (0.0110)
95% CI: [0.0718 ; 0.1154]
<0.001
Belimumab -0.9389 (0.2230)
95% CI: [-1.3760 ; -0.5018]
0.3911 (0.0872)
95% CI: [0.2526 ; 0.6054]
<0.001
Abbreviations: CI, confidence interval; SDI, SLICC/ACR Damage Index; SE, standard error
51
Table 22. Proportional hazards model of time to first change in total SDI score, Gompertz
distribution.
Variable
Regression Coefficient
Estimate (SE)
[95% CI]
Hazard Rate/Ratio
Estimate (SE)
[95% CI] p value
Intercept -2.2616 (0.1593)
95% CI: [-2.5739 ; -1.9493]
0.1042 (0.0166)
95% CI: [0.0762 ; 0.1424]
<0.001
Belimumab -0.9651 (0.2205)
95% CI: [-1.3972 ; -0.5329]
0.3810 (0.0840)
95% CI: [0.2473 ; 0.5869]
<0.001
Abbreviations: CI, confidence interval; SDI, SLICC/ACR Damage Index; SE, standard error
Table 23. Proportional hazards model of time to first change in total SDI score, Weibull
distribution.
Variable
Regression Coefficient
Estimate (SE)
[95% CI]
Hazard Rate/Ratio
Estimate (SE)
[95% CI] p value
Intercept -2.5094 (0.1584)
95% CI: [-2.8199 ; -2.1990]
0.0813 (0.0129)
95% CI: [0.0596 ; 0.1109]
<0.001
Belimumab -0.9327 (0.2256)
95% CI: [-1.3749 ; -0.4906]
0.3935 (0.0888)
95% CI: [0.2529 ; 0.6123]
<0.001
p 1.0612 (0.06177)
95% CI: [0.9468 ; 1.1894]
NA 0.308
Abbreviations: CI, confidence interval; NA, not applicable; SDI, SLICC/ACR Damage Index; SE, standard error
Table 24. Accelerated failure time model of time to first change in total SDI score,
loglogistic distribution
Variable
Regression Coefficient
Estimate (SE)
[95% CI] p value
Intercept 1.9596 (0.1342)
95% CI: 1.6966; 2.2226]
<0.001
Belimumab 0.9310 (0.2140)
95% CI: 0.5515 ; 1.3504]
<0.001
gamma 0.7881 (0.0447)
95% CI: [0.7052 ; 0.8807]
<0.001
Abbreviations: CI, confidence interval; NR, not reported; SDI, SLICC/ACR Damage Index; SE, standard error
52
Table 25. Accelerated failure time model of time to first change in total SDI score,
lognormal distribution
Variable
Regression Coefficient
Estimate (SE)
[95% CI] p value
Intercept 1.9990 (0.1353)
95% CI: 1.7338; 2.2642]
<0.001
Belimumab 0.8930 (0.1960)
95% CI: [0.5087; 1.2772]
<0.001
sigma 1.3595 (0.0734)
95% CI: [1.2230 ; 1.5112]
<0.001
Abbreviations: CI, confidence interval; SDI, SLICC/ACR Damage Index; SE, standard error; NA, could not be calculated
The information criterion scores for the models are displayed in Table 26. The lognormal
distribution produced substantially better measures of fit (lower AIC and Bayesian information
criteria [BIC] scores) than the other distributions.
Table 26. Fit of regression models of time to first change in total SDI score
Distribution AIC BIC
Exponential 596.8 604.5
Gompertz 597.4 609.0
Loglogistic 591.5 603.1
Weibull 598.2 609.9
Lognormal 582.1 593.7
Abbreviations: AIC, Akaike information criterion; BIC, Bayesian information criterion
6.1.4 Change from baseline SDI score by year interval
The counts and proportions of subjects in each treatment arm, out to year 5, of the incremental
changes from baseline of total SDI score are displayed in Table 27. A substantial difference
occurs immediately. In the first year 17 (17.2%) SoC subjects had an increase in their total SDI,
with 7 (7.1%) subjects experiencing an increase of greater than 1. In contrast, only 3 (3.0%)
belimumab subjects had an increase and all were an increase of 1. By the fifth year only 5
(5.1%) belimumab subjects had an increase greater than 1 whereas 19 (19.2%) SoC subjects
had an increase greater than 1. Overall, by the fifth year 77 (77.8%) of belimumab subjects saw
no change, while 59 (59.6%) of SoC subjects saw no change.
53
Table 27. SDI change from baseline
Year 1 Year 2 Year 3 Year 4 Year 5
SDI
Change
SoC
N=99
Belim
N=99
SoC
N=99
Belim
N=99
SoC
N=99
Belim
N=99
SoC
N=99
Belim
N=99
SoC
N=99
Belim
N=99
0 [n (%)] 82
(82.8%) 96
(97.0%) 75
(75.8%) 87
(87.9%) 71
(71.7%) 79
(79.8%) 67
(67.7%) 78
(78.8%) 59
(59.6%) 77
(77.8%)
+1 [n (%)] 10
(10.1%) 3
(3.0%) 14
(14.1%) 11
(11.1%) 15
(15.2%) 16
(16.2%) 17
(17.2%) 17
(17.2%) 21
(21.2%) 17
(17.2%)
+2 [n (%)] 6
(6.1%) 0
8 (8.1%)
1 (1.0%)
8 (8.1%)
3 (3.0%)
9 (9.1%)
3 (3.0%)
12 (12.1%)
4 (4.0%)
+3 [n (%)] 0 0 0 0 2
(2.0%) 1
(1.0%) 2
(2.0%) 1
(1.0%) 3
(3.0%) 1
(1.0%)
+4 [n (%)] 1
(1.0%) 0
2 (2.0%)
0 3
(3.0%) 0
4 (4.0%)
0 3
(3.0%) 0
+5 [n (%)] 0 0 0 0 0 0 0 0 1
(1.0%) 0
Abbreviations: Belim, Belimumab; SDI, SLICC/ACR Damage Index; SoC, standard of care
For year 1, the constrained continuation ratio logit model provided an adequate fit (p=0.456);
allowing for the assumption of a constant odds ratio across SDI baseline change categories (0,
≥1; +1, ≥2). The type of treatment subjects received had a significant (p=0.002) effect on SDI
change from baseline (Table 28). The odds of an increase from baseline total SDI score and the
odds of having an increase greater than one given an increase were each 7.3728 (1 / 0.1356)
times greater for subjects receiving SoC versus subjects taking belimumab.
54
Table 28. SDI change from baseline constrained model year 1
Variable Coefficient (SE)
[95% CI] P value
Odds Ratio [95% CI]
Intercept:1 logit[P(ΔSDI>0|ΔSDI≥0)]
-1.5561 (0.2633) 95% CI: [-2.0722 ; -1.0401]
p<0.001 NA
Belimumab
-1.9978 (0.6387) 95% CI: [-3.2497 ; -0.7459]
p=0.002
0.1356 95% CI: [0.0388 ; 0.4743]
Intercept:2 logit[P(ΔSDI>1|ΔSDI≥1)]
-0.4168 (0.4812) 95% CI: [-1.3598 ; 0.5262]
p=0.386 NA
Degrees of freedom Deviance (P value)
1 0.55 (p=0.456)
Abbreviations: ΔSDI, SDI change from baseline; CI, confidence interval; NA, not applicable; SDI, SLICC/ACR Damage Index; SE, standard error
For year 2, the constrained continuation ratio logit model provided an adequate fit (p=0.262);
allowing for the assumption of a constant odds ratio across SDI baseline change categories (0,
≥1; +1, ≥2). The type of treatment subjects received had a significant (p=0.005) effect on SDI
change from baseline (Table 29). The odds of an increase from baseline total SDI score and the
odds of having an increase greater than one given an increase were each 2.7508 (1 / 0.3635)
times greater for subjects receiving SoC versus subjects taking belimumab.
55
Table 29. SDI change from baseline constrained model year 2
Variable Coefficient (SE)
[95% CI] P value
Odds Ratio [95% CI]
Intercept:1 logit[P(ΔSDI>0|ΔSDI≥0)]
-1.0791 (0.2245) 95% CI: [-1.5191 ; -0.6392]
p<0.001
NA
Belimumab
-1.0119 (0.3620) 95% CI: [-1.7214 ; -0.3025]
p=0.005
0.3635 95% CI: [0.1788 ; 0.7390]
Intercept:2 logit[P(ΔSDI>1|ΔSDI≥1)]
-0.5310 (0.3747) 95% CI: [-1.2654 ; 0.2034]
p=0.156
NA
Degrees of freedom Deviance (P value)
1 1.26 (p=0.262)
Abbreviations: ΔSDI, SDI change from baseline; CI, confidence interval; NA, not applicable; SDI, SLICC/ACR Damage Index; SE, standard error
For year 3, the constrained continuation ratio logit model provided an adequate fit (p=0.279);
allowing for the assumption of a constant odds ratio across SDI baseline change categories (0,
≥1; +1, ≥2). The type of treatment subjects received had a significant (p=0.041) effect on SDI
change from baseline (Table 30). The odds of an increase from baseline total SDI score and the
odds of having an increase greater than one given an increase were each 1.8449 (1 / 0.5420)
times greater for subjects receiving SoC versus subjects taking belimumab.
56
Table 30. SDI change from baseline constrained model year 3
Variable Coefficient (SE)
[95% CI] P value
Odds Ratio [95% CI]
Intercept:1 logit[P(ΔSDI>0|ΔSDI≥0)]
-0.8574 (0.2094) 95% CI: [-1.2679 ; -0.4469]
p<0.001 NA
Belimumab
-0.6124 (0.2995) 95% CI: [-1.1995 ; -0.0253]
p=0.041
0.5420 95% CI: [0.3014 ; 0.9750]
Intercept:2 logit[P(ΔSDI>1|ΔSDI≥1)]
-0.3595 (0.3206) 95% CI: [-0.9879 ; 0.2688]
p=0.262 NA
Degrees of freedom Deviance (P value)
1 1.17 (p=0.279)
Abbreviations: ΔSDI, SDI change from baseline; CI, confidence interval; NA, not applicable; SDI, SLICC/ACR Damage Index; SE, standard error
For year 4, the constrained continuation ratio logit model provided an adequate fit (p=0.298);
allowing for the assumption of a constant odds ratio across SDI baseline change categories (0,
≥1; +1, ≥2). The type of treatment subjects received had a significant (p=0.012) effect on SDI
change from baseline (Table 31). The odds of an increase from baseline total SDI score and the
odds of having an increase greater than one given an increase were each 2.0786 (1 / 0.4811)
times greater for subjects receiving SoC versus subjects taking belimumab.
57
Table 31. SDI change from baseline constrained model year 4
Variable Coefficient (SE)
[95% CI] P value
Odds Ratio [95% CI]
Intercept:1 logit[P(ΔSDI>0|ΔSDI≥0)]
-0.6718 (0.2026) 95% CI: [-1.0688 ; -0.2747]
p<0.001 NA
Belimumab
-0.7317 (0.2917) 95% CI: [-1.3033 ; -0.1601]
p=0.012
0.4811 95% CI: [0.2716 ; 0.8521]
Intercept:2 logit[P(ΔSDI>1|ΔSDI≥1)]
-0.3114 (0.3038) 95% CI: [-0.9068 ; 0.2840]
p=0.305 NA
Degrees of freedom Deviance (P value)
1 1.08 (p=0.298)
Abbreviations: ΔSDI, SDI change from baseline; CI, confidence interval; NA, not applicable; SDI, SLICC/ACR Damage Index; SE, standard error
For year 5, the constrained continuation ratio logit model provided an adequate fit (p=0.700);
allowing for the assumption of a constant odds ratio across SDI baseline change categories (0,
≥1; +1, ≥2). The type of treatment subjects received had a significant (p<0.001) effect on SDI
change from baseline (Table 32). The odds of an increase from baseline total SDI score and the
odds of having an increase greater than one given an increase were each 2.5148 (1 / 0.3976)
times greater for subjects receiving SoC versus subjects taking belimumab.
58
Table 32. SDI change from baseline constrained model year 5
Variable Coefficient (SE)
[95% CI] P value
Odds Ratio [95% CI]
Intercept:1 logit[P(ΔSDI>0|ΔSDI≥0)]
-0.3645 (0.1946) 95% CI: [-0.7459 ; 0.0168]
p=0.061
NA
Belimumab
-0.9222 (0.2799) 95% CI: [-1.4709 ; -0.3735]
p<0.001
0.3976 95% CI: [0.2297 ; 0.6883]
Intercept:2 logit[P(ΔSDI>1|ΔSDI≥1)]
-0.1580 (0.2778) 95% CI: [-0.7025 ; 0.3866]
p=0.570
NA
Degrees of freedom Deviance (P value)
1 0.15 (p=0.700)
Abbreviations: ΔSDI, SDI change from baseline; CI, confidence interval; NA, not applicable; SDI, SLICC/ACR Damage Index; SE, standard error
6.1.5 Difference of change from baseline SDI by year interval
The change of total SDI score from baseline to end of years 1 through 5 was analyzed using
linear regression with a binary indicator for treatment with belimumab as a covariate. All PSM
variables were balanced (Table 11) so none were added as covariates.
The baseline decade of entry was initially included as a covariate in the year 1 analysis, but it
was not statistically significant. In fact, baseline decade of entry was not a significant factor for
any of the five years from baseline. The results from the regression with baseline decade of
entry as a covariate were omitted for subsequent years.
When baseline decade of entry was included as a covariate the difference from baseline in the
first year total SDI score was significantly (p=0.002) lower for subjects taking belimumab. See
Table 33.
59
Table 33. Year 1 Total SDI difference of change from baseline controlled for entry decade
Variable
OLS Coefficient (SE)
[95% CI] P value
Robust SE Coefficient (SE)
[95% CI] P value
Intercept 0.2000 (0.0965)
95% CI: [0.0097 ; 0.3903] p=0.040
0.2000 (0.0992) 95% CI: [0.0055 ; 0.3945]
p=0.044
Belimumab -0.2801 (0.0755)
95% CI: [-0.4290 ; -0.1311] p<0.001
-0.2801 (0.0902) 95% CI: [-0.4569 ; -0.1032]
p=0.002
Entry Decade 2000 0.1203 (0.1133)
95% CI: [-0.1032 ; 0.3438] p=0.290
0.1203 (0.1352) 95% CI: [-0.1446 ; 0.3852]
p=0.374
Entry Decade 2010 -0.1253 (0.1589)
95% CI: [-0.4386 ; 0.1880] p=0.431
-0.1253 (0.1075) 95% CI: [-0.3360 ; 0.0854]
p=0.244
Abbreviations: CI, confidence interval; OLS, ordinary least squares; SDI, SLICC/ACR Damage Index; SE, standard error
Without controlling for baseline decade of entry the results were similar. The change in total SDI
score from baseline at the end of the first year was significantly (p<0.001) lower for subjects
taking belimumab. See Table 34. The average SDI change from baseline was lower by 0.2323
for subjects taking belimumab compared to those receiving SoC.
Table 34. Year 1 Total SDI difference of change from baseline
Variable
OLS Coefficient (SE)
[95% CI] P value
Robust SE Coefficient (SE)
[95% CI] P value
Intercept 0.2626 (0.0487)
95% CI: [0.1665 ; 0.3587] p<0.001
0.2626 (0.0669) 95% CI: [0.1315 ; 0.3937]
p<0.001
Belimumab -0.2323 (0.0689)
95% CI: [-0.3682 ; -0.0964] p<0.001
-0.2323 (0.0688) 95% CI: [-0.3671 ; -0.0975]
p<0.001
Abbreviations: CI, confidence interval; OLS, ordinary least squares; SDI, SLICC/ACR Damage Index; SE, standard error
For years 2 through 5 the change in total SDI score from baseline was always significantly lower
for belimumab. See Table 35 to Table 38. After two years the average SDI change from
60
baseline was lower by 0.2525 for subjects taking belimumab compared to those receiving SoC.
The magnitude of the difference decreased slightly in year 3 to 0.2424. This was followed by
subsequent substantial increases in the last two years. By the end of year 4 belimumab subjects
had an average difference that was lower by 0.3131. In the last year the belimumab average
change from baseline was 0.4343 less.
Table 35. Year 2 Total SDI difference of change from baseline
Variable
OLS Coefficient (SE)
[95% CI] P value
Robust SE Coefficient (SE)
[95% CI] P value
Intercept 0.3838 (0.0629)
95% CI: [0.2599 ; 0.5078] p<0.001
0.3838 (0.0811) 95% CI: [0.2250 ; 0.5427]
p<0.001
Belimumab -0.2525 (0.0889)
95% CI: [-0.4279 ; -0.0772] p=0.005
-0.2525 (0.0806) 95% CI: [-0.4105 ; -0.0946]
p=0.002
Abbreviations: CI, confidence interval; OLS, ordinary least squares; SDI, SLICC/ACR Damage Index; SE, standard error
Table 36. Year 3 Total SDI difference of change from baseline
Variable
OLS Coefficient (SE)
[95% CI] P value
Robust SE Coefficient (SE)
[95% CI] P value
Intercept 0.4949 (0.0785)
95% CI: [0.3402 ; 0.6497] p<0.001
0.4949 (0.0959) 95% CI: [0.3070 ; 0.6829]
p<0.001
Belimumab -0.2424 (0.1110)
95% CI: [-0.4612 ; -0.0236] p=0.030
-0.2424 (0.1078) 95% CI: [-0.4537 ; -0.0311]
p=0.025
Abbreviations: CI, confidence interval; OLS, ordinary least squares; SDI, SLICC/ACR Damage Index; SE, standard error
61
Table 37. Year 4 Total SDI difference of change from baseline
Variable
OLS Coefficient (SE)
[95% CI] P value
Robust SE Coefficient (SE)
[95% CI] P value
Intercept 0.5758 (0.0829)
95% CI: [0.4123 ; 0.7393] p<0.001
0.5758 (0.1029) 95% CI: [0.3741 ; 0.7774]
p<0.001
Belimumab -0.3131 (0.1172)
95% CI: [-0.5442 ; -0.0820] p=0.008
-0.3131 (0.1166) 95% CI: [-0.5417 ; -0.0845]
p=0.007
Abbreviations: CI, confidence interval; OLS, ordinary least squares; SDI, SLICC/ACR Damage Index; SE, standard error
Table 38. Year 5 Total SDI difference of change from baseline
Variable
OLS Coefficient (SE)
[95% CI] P value
Robust SE Coefficient (SE)
[95% CI] P value
Intercept 0.7172 (0.0886)
95% CI: [0.5425 ; 0.8918] p<0.001
0.7172 (0.1106) 95% CI: [0.5004 ; 0.9339]
p<0.001
Belimumab -0.4343 (0.1252)
95% CI: [-0.6813 ; -0.1874] p<0.001
-0.4343 (0.1188) 95% CI: [-0.6673 ; -0.2014]
p<0.001
Abbreviations: CI, confidence interval; OLS, ordinary least squares; SDI, SLICC/ACR Damage Index; SE, standard error
6.1.6 Transition analysis of SDI from baseline over a 5-year interval
The annual transition probability of a change in SDI was estimated by combining the results
from the time to first SDI worsening analysis (6.1.3) with the observed conditional probability
that the increase in SDI score was 1 point versus 2+ points. The resulting annual probabilities
are shown in
Table 39. A constant hazard was assumed and time to first SDI worsening was modeled using
an exponential distribution (Table 21). The conditional probabilities were derived separately
using the observed counts for the specific treatment arm.
62
Table 39. Annual transition probabilities
SoC Belimumab
No SDI change 0.9130 0.9650
SDI increase by 1 0.0604 0.0329
SDI increase by 2 0.0266 0.0021
Abbreviations: SoC, standard of care; SDI, SLICC/ACR Damage Index
6.1.7 Change from baseline of SDI organ damage system subscores
The counts and proportions of the incremental changes from baseline out to year 5 for each of
the SDI organ system subscores are displayed in the even numbered tables of Section 6.1.7.
Results of two-sided Fisher’s exact tests for each SDI organ system subscore are displayed in
the odd numbered tables for each year, also in Section 6.1.7.
The majority of the subscores showed no significant difference in the proportion of subjects with
a change from baseline at the end of year 5 or any prior year; the exceptions were
musculoskeletal and skin subscores.
At the end of the fifth year 11 (11.1%) SoC subjects had seen an increase from their baseline
ocular system subscore. See Table 40. In comparison, only 4 (4.0%) of belimumab subjects had
seen an increase. No subjects taking belimumab had an increase in the first two years.
Table 40. SDI ocular system subscore change from baseline by year
SDI
Ocular Year 1 Year 2 Year 3 Year 4 Year 5
Change
from Baseline
SoC N=99
Belim N=99
SoC N=99
Belim N=99
SoC N=99
Belim N=99
SoC N=99
Belim N=99
SoC N=99
Belim N=99
0 [n (%)] 97
(98.0%) 99
(100.0%) 95
(96.0%) 99
(100.0%) 93
(93.9%) 96
(97.0%) 92
(92.9%) 96
(97.0%) 88
(88.9%) 95
(96.0%)
+1 [n (%)] 2
(2.0%) 0
3 (3.0%)
0 5
(5.1%) 3
(3.0%) 6
(6.1%) 3
(3.0%) 10
(10.1%) 4
(4.0%)
+2 [n (%)] 0 0 1
(1.0%) 0
1 (1.0%)
0 1
(1.0%) 0
1 (1.0%)
0
Abbreviations: Belim, belimumab; SDI, SLICC/ACR Damage Index; SoC, standard of care
For the ocular system subscore there was no significant difference in the proportion of subjects
with an increase from their baseline score by the end of the 5th year (p=0.104) or at the end of
any of the prior years. See Table 41.
63
Table 41. Fisher’s test for belimumab versus SoC SDI ocular subscore change from
baseline
YEAR Odds Ratio 95% CI P value
1 0 0 ; 5.316 0.497
2 0 0 ; 1.496 0.121
3 0.486 0.076 ; 2.356 0.498
4 0.412 0.067 ; 1.874 0.331
5 0.339 0.076 ; 1.196 0.104
Abbreviations: CI, confidence interval; SDI, SLICC/ACR Damage Index; SoC, standard of care
At the end of the fifth year 8 (8.1%) SoC subjects had seen an increase from their baseline
neuropsychiatric system subscore. See Table 42. For belimumab subjects, 7 (7.1%) had seen
an increase. The difference in the number of SoC subjects compared to belimumab subjects
experiencing any increase remained fairly constant throughout the five years.
Table 42. SDI neuropsychiatric system subscore change from baseline by year
SDI
Neuro Year 1 Year 2 Year 3 Year 4 Year 5
Change
From
Baseline
SoC N=99
Belim N=99
SoC N=99
Belim N=99
SoC N=99
Belim N=99
SoC N=99
Belim N=99
SoC N=99
Belim N=99
0 [n (%)] 97
(98.0%) 98
(99.0%) 96
(97.0%) 94
(94.9%) 93
(93.9%) 92
(92.9%) 93
(93.9%) 92
(92.9%) 91
(91.9%) 92
(92.9%)
+1 [n (%)] 2
(2.0%) 1
(1.0%) 3
(3.0%) 5
(5.1%) 6
(6.1%) 6
(6.1%) 6
(6.1%) 6
(6.1%) 8
(8.1%) 6
(6.1%)
+2 [n (%)] 0 0 0 0 0 1
(1.0%) 0
1 (1.0%)
0 1
(1.0%)
Abbreviations: Belim, belimumab; Nuero, neuropsychiatric; SDI, SLICC/ACR Damage Index; SoC, standard of care
For the neuropsychiatric system subscore there was no significant difference in the proportion of
subjects with an increase from their baseline score by the end of the 5th year (p=1.000) or at
the end of any of the prior years. See Table 43.
64
Table 43. Fisher’s test for belimumab versus SoC SDI neuropsychiatric system subscore
change from baseline
YEAR Odds Ratio 95% CI P value
1 0.497 0.008 ; 9.685 1.000
2 1.698 0.320 ; 11.241 0.721
3 1.178 0.325 ; 4.418 1.000
4 1.178 0.325 ; 4.418 1.000
5 0.866 0.256 ; 2.860 1.000
Abbreviations: CI, confidence interval; SDI, SLICC/ACR Damage Index; SoC, standard of care
At the end of the fifth year 2 (2.0%) SoC subjects had seen an increase from their baseline renal
system subscore, whereas, no belimumab subjects saw an increase. See Table 44. The
difference in the number of SoC subjects compared to belimumab subjects experiencing any
increase remained constant throughout the five years with the only SoC changes coming in the
first year.
Table 44. SDI renal system subscore change from baseline by year
SDI
Renal Year 1 Year 2 Year 3 Year 4 Year 5
Change
From
Baseline
SoC N=99
Belim N=99
SoC N=99
Belim N=99
SoC N=99
Belim N=99
SoC N=99
Belim N=99
SoC N=99
Belim N=99
0 [n (%)] 97
(98.0%) 99
(100.0%) 97
(98.0%) 99
(100.0%) 97
(98.0%) 99
(100.0%) 97
(98.0%) 99
(100.0%) 97
(98.0%) 99
(100.0%)
+1 [n (%)] 2
(2.0%) 0
2 (2.0%)
0 2
(2.0%) 0
2 (2.0%)
0 2
(2.0%) 0
Abbreviations: Belim, belimumab; SDI, SLICC/ACR Damage Index; SoC, standard of care
For the renal system subscore there was no significant difference in the proportion of subjects
with an increase from their baseline score by the end of the 5th year (p=0.497) or at the end of
any of the prior years. See Table 45.
65
Table 45. Fisher’s test for belimumab versus SoC SDI renal system subscore change
from baseline
YEAR Odds Ratio 95% CI P value
1 0 0 ; 5.316 0.497
2 0 0 ; 5.316 0.497
3 0 0 ; 5.316 0.497
4 0 0 ; 5.316 0.497
5 0 0 ; 5.316 0.497
Abbreviations: CI, confidence interval; SDI, SLICC/ACR Damage Index; SoC, standard of care
At the end of the fifth year 3 (3.0%) SoC subjects had seen an increase from their baseline
pulmonary system subscore, whereas, no belimumab subjects saw an increase. See Table 46.
One SoC subject saw an increase each year except in year 4.
Table 46. SDI pulmonary system subscore change from baseline by year
SDI
Pulmonary Year 1 Year 2 Year 3 Year 4 Year 5
Change
From
Baseline
SoC N=99
Belim N=99
SoC N=99
Belim N=99
SoC N=99
Belim N=99
SoC N=99
Belim N=99
SoC N=99
Belim N=99
0 [n (%)] 98
(99.0%) 99
(100.0%) 97
(98.0%) 99
(100.0%) 97
(98.0%) 99
(100.0%) 97
(98.0%) 99
(100.0%) 96
(97.0%) 99
(100.0%)
+1 [n (%)] 1
(1.0%) 0
2 (2.0%)
0 1
(1.0%) 0
1 (1.0%)
0 2
(2.0%) 0
+2 [n (%)] 0 0 0 0 1
(1.0%) 0
1 (1.0%)
0 1
(1.0%) 0
Abbreviations: Belim, belimumab; SDI, SLICC/ACR Damage Index; SoC, standard of care
For the pulmonary system subscore there was no significant difference in the proportion of
subjects with an increase from their baseline score by the end of the 5th year (p=0.246) or at
the end of any of the prior years. See Table 47.
66
Table 47. Fisher’s test for belimumab versus SoC SDI pulmonary system subscore
change from baseline
YEAR Odds Ratio 95% CI P value
1 0 0 ; 39.001 1.000
2 0 0 ; 5.316 0.497
3 0 0 ; 5.316 0.497
4 0 0 ; 5.316 0.497
5 0 0 ; 2.405 0.246
Abbreviations: CI, confidence interval; SDI, SLICC/ACR Damage Index; SoC, standard of care
At the end of the fifth year 6 (6.1%) SoC subjects had seen an increase from their baseline
cardiovascular system subscore. See Table 48. For belimumab subjects, 5 (5.1%) had seen an
increase. The difference in the number of SoC subjects compared to belimumab subjects
experiencing any increase remained fairly constant throughout the five years.
Table 48. SDI cardiovascular system subscore change from baseline by year
SDI
CV Year 1 Year 2 Year 3 Year 4 Year 5
Change
From
Baseline
SoC N=99
Belim N=99
SoC N=99
Belim N=99
SoC N=99
Belim N=99
SoC N=99
Belim N=99
SoC N=99
Belim N=99
0 [n (%)] 97
(98.0%) 99
(100.0%) 97
(98.0%) 97
(98.0%) 96
(97.0%) 95
(96.0%) 94
(94.9%) 95
(96.0%) 93
(93.9%) 94
(94.9%)
+1 [n (%)] 2
(2.0%) 0
2 (2.0%)
1 (1.0%)
2 (2.0%)
3 (3.0%)
4 (4.0%)
3 (3.0%)
5 (5.1%)
4 (4.0%)
+2 [n (%)] 0 0 0 1
(1.0%) 0
1 (1.0%)
0 1
(1.0%) 0
1 (1.0%)
+3 [n (%)] 0 0 0 0 1
(0.0%) 0
1 (0.0%)
0 1
(0.0%) 0
Abbreviations: Belim, Belimumab; CV, cardiovascular; SDI, SLICC/ACR Damage Index; SoC, standard of care
For the cardiovascular system subscore there was no significant difference in the proportion of
subjects with an increase from their baseline score by the end of the 5th year (p=1.000) or at
the end of any of the prior years. See Table 49.
67
Table 49. Fisher’s test for belimumab versus SoC SDI CV subscore change from baseline
YEAR Odds Ratio 95% CI P value
1 0 0 ; 5.316 0.497
2 1.000 0.071 ; 14.049 1.000
3 1.345 0.221 ; 9.429 1.000
4 0.793 0.152 ; 3.808 1.000
5 0.825 0.192 ; 3.371 1.000
Abbreviations: CI, confidence interval; CV, cardiovascular; SDI, SLICC/ACR Damage Index; SoC, standard of care
At the end of the fifth year 1 (1.0%) SoC subjects had seen an increase from their baseline
peripheral vascular system subscore. See Table 50. For belimumab subjects, 2 (2.0%) had
seen an increase. No subject saw an increase in the first year.
Table 50. SDI peripheral vascular system subscore change from baseline by year
SDI
PV Year 1 Year 2 Year 3 Year 4 Year 5
Change
From
Baseline
SoC N=99
Belim N=99
SoC N=99
Belim N=99
SoC N=99
Belim N=99
SoC N=99
Belim N=99
SoC N=99
Belim N=99
0 [n (%)] 99
(100.0%) 99
(100.0%) 99
(100.0%) 98
(99.0%) 99
(100.0%) 97
(98.0%) 98
(99.0%) 97
(98.0%) 98
(99.0%) 97
(98.0%)
+1 [n (%)] 0 0 0 1
(1.0%) 0
1 (1.0%)
1 (1.0%)
1 (1.0%)
1 (1.0%)
1 (1.0%)
+2 [n (%)] 0 0 0 0 0 1
(1.0%) 0
1 (1.0%)
0 1
(1.0%)
Abbreviations: Belim, Belimumab; PV, peripheral vascular; SDI, SLICC/ACR Damage Index; SoC, standard of care
For the peripheral vascular system subscore there was no significant difference in the
proportion of subjects with an increase from their baseline score by the end of the 5th year
(p=1.000) or at the end of any of the prior years. See Table 51.
68
Table 51. Fisher’s test for belimumab versus SoC SDI peripheral vascular system
subscore change from baseline
YEAR Odds Ratio 95% CI P value
1 0 0 ; Inf 1.000
2 Inf 0.026 ; Inf 1.000
3 Inf 0.188 ; Inf 0.497
4 2.014 0.103 ; 120.322 1.000
5 2.014 0.103 ; 120.322 1.000
Abbreviations: CI, confidence interval; Inf, infinity; SDI, SLICC/ACR Damage Index; SoC, standard of care
At the end of the fifth year 1 (1.0%) SoC subjects had seen an increase from their baseline
gastrointestinal system subscore. See Table 52. For belimumab subjects, 2 (2.0%) had seen an
increase. No changes occurred after the second year.
Table 52. SDI gastrointestinal system subscore change from baseline by year
SDI
GI Year 1 Year 2 Year 3 Year 4 Year 5
Change
From
Baseline
SoC N=99
Belim N=99
SoC N=99
Belim N=99
SoC N=99
Belim N=99
SoC N=99
Belim N=99
SoC N=99
Belim N=99
0 [n (%)] 99
(100.0%) 98
(99.0%) 98
(99.0%) 97
(98.0%) 98
(99.0%) 97
(98.0%) 98
(99.0%) 97
(98.0%) 98
(99.0%) 97
(98.0%)
+1 [n (%)] 0 1
(1.0%) 1
(1.0%) 2
(2.0%) 1
(1.0%) 2
(2.0%) 1
(1.0%) 2
(2.0%) 1
(1.0%) 2
(2.0%)
Abbreviations: Belim, belimumab; GI, gastrointestinal; SDI, SLICC/ACR Damage Index; SoC, standard of care
For the gastrointestinal system subscore there was no significant difference in the proportion of
subjects with an increase from their baseline score by the end of the 5th year (p=1.000) or at
the end of any of the prior years. See Table 53.
Table 53. Fisher’s test for belimumab versus SoC SDI gastrointestinal system subscore
change from baseline
YEAR Odds Ratio 95% CI P value
1 Inf 0.026 ; Inf 1.000
2 2.014 0.103 ; 120.322 1.000
3 2.014 0.103 ; 120.322 1.000
4 2.014 0.103 ; 120.322 1.000
5 2.014 0.103 ; 120.322 1.000
Abbreviations: CI, confidence interval; Inf, infinity; SDI, SLICC/ACR Damage Index; SoC, standard of care
69
At the end of the fifth year 16 (16.2%) SoC subjects had seen an increase from their baseline
musculoskeletal system subscore, with 7 of those being an increase greater than one. See
Table 54. In comparison, only 3 (3.0%) belimumab subjects had seen an increase with none
being greater than 1. The first year saw the largest number (10) of SoC subjects experiencing
an increase. The difference in the number of SoC subjects compared to belimumab subjects
experiencing any increase grew or remained the same from year to year.
Table 54. SDI musculoskeletal system subscore change from baseline by year
SDI
MS Year 1 Year 2 Year 3 Year 4 Year 5
Change
From
Baseline
SoC N=99
Belim N=99
SoC N=99
Belim N=99
SoC N=99
Belim N=99
SoC N=99
Belim N=99
SoC N=99
Belim N=99
0 [n (%)] 89
(89.9%) 98
(99.0%) 88
(88.9%) 97
(98.0%) 86
(86.9%) 96
(97.0%) 85
(85.9%) 96
(97.0%) 83
(83.8%) 96
(97.0%)
+1 [n (%)] 8
(8.1%) 1
(1.0%) 7
(7.1%) 2
(2.0%) 9
(9.1%) 3
(3.0%) 9
(9.1%) 3
(3.0%) 9
(9.1%) 3
(3.0%)
+2 [n (%)] 2
(2.0%) 0
4 (4.0%)
0 4
(4.0%) 0
4 (4.0%)
0 6
(6.1%) 0
+3 [n (%)] 0 0 0 0 0 0 1
(0.0%) 0
1 (0.0%)
0
Abbreviations: Belim, belimumab; MS, musculoskeletal; SDI, SLICC/ACR Damage Index; SoC, standard of care
For the musculoskeletal system subscore the odds of belimumab subjects experiencing an
increase from their baseline score by the end of the 5th year were significantly lower compared
to SoC subjects (p=0.003). In fact, for belimumab subjects, the odds were significantly less in
the first year (p=0.010) and continued to be significantly less for the intervening years. See
Table 55.
Table 55. Fisher’s test for belimumab versus SoC SDI musculoskeletal system subscore
change from baseline
YEAR Odds Ratio 95% CI P value
1 0.092 0.002 ; 0.667 0.010
2 0.166 0.017 ; 0.793 0.018
3 0.208 0.037 ; 0.793 0.016
4 0.191 0.034 ; 0.718 0.009
5 0.163 0.030 ; 0.599 0.003
Abbreviations: CI, confidence interval; SDI, SLICC/ACR Damage Index; SoC, standard of care
70
At the end of the fifth year 8 (8.1%) SoC subjects had seen an increase from their baseline skin
system subscore, with no increase for any belimumab subject. See Table 56. The first year saw
the largest number (4) of SoC subjects experiencing an increase. Except for the third year, SoC
subjects saw an increase each year.
Table 56. SDI skin system subscore change from baseline by year
SDI
Skin Year 1 Year 2 Year 3 Year 4 Year 5
Change
From
Baseline
SoC N=99
Belim N=99
SoC N=99
Belim N=99
SoC N=99
Belim N=99
SoC N=99
Belim N=99
SoC N=99
Belim N=99
0 [n (%)] 95
(96.0%) 99
(100.0%) 93
(93.9%) 99
(100.0%) 93
(93.9%) 99
(100.0%) 92
(92.9%) 99
(100.0%) 91
(91.9%) 99
(100.0%)
+1 [n (%)] 4
(4.0%) 0
6 (6.1%)
0 6
(6.1%) 0
7 (7.1%)
0 7
(7.1%) 0
+2 [n (%)] 0 0 0 0 0 0 0 0 1
(1.0%) 0
Abbreviations: Belim, belimumab; SDI, SLICC/ACR Damage Index; SoC, standard of care
For the skin system subscore the odds of belimumab subjects experiencing an increase from
their baseline score by the end of the 5th year were significantly lower compared to SoC
subjects (p=0.007). In fact, for belimumab subjects, the odds were significantly less for all but
the first year. See Table 57.
Table 57. Fisher’s test for belimumab versus SoC SDI skin subscore change from
baseline
YEAR Odds Ratio 95% CI P value
1 0 0 ; 1.496 0.121
2 0 0 ; 0.826 0.029
3 0 0 ; 0.826 0.029
4 0 0 ; 0.668 0.014
5 0 0 ; 0.559 0.007
Abbreviations: CI, confidence interval; SDI, SLICC/ACR Damage Index; SoC, Standard of Care
At the end of the fifth year 1 (1.0%) SoC subject had seen an increase from their baseline
premature gonadal failure subscore, whereas, no belimumab subjects saw an increase. See
Table 58. The one SoC subject who saw an increase did so in the first year.
71
Table 58. SDI premature gonadal failure subscore change from baseline by year
SDI
Gonadal Year 1 Year 2 Year 3 Year 4 Year 5
Change
From
Baseline
SoC N=99
Belim N=99
SoC N=99
Belim N=99
SoC N=99
Belim N=99
SoC N=99
Belim N=99
SoC N=99
Belim N=99
0 [n (%)] 98
(99.0%) 99
(100.0%) 98
(99.0%) 99
(100.0%) 98
(99.0%) 99
(100.0%) 98
(99.0%) 99
(100.0%) 98
(99.0%) 99
(100.0%)
+1 [n (%)] 1
(1.0%) 0
1 (1.0%)
0 1
(1.0%) 0
1 (1.0%)
0 1
(1.0%) 0
Abbreviations: Belim, belimumab; SDI, SLICC/ACR Damage Index; SoC, standard of care
For the premature gonadal failure subscore there was no significant difference in the proportion
of subjects with an increase from their baseline score by the end of the 5th year (p=1.000) or at
the end of any of the prior years. See Table 59.
Table 59. Fisher’s test for belimumab versus SoC SDI premature gonadal failure
subscore change from baseline
YEAR Odds Ratio 95% CI P value
1 0 0 ; 39.001 1.000
2 0 0 ; 39.001 1.000
3 0 0 ; 39.001 1.000
4 0 0 ; 39.001 1.000
5 0 0 ; 39.001 1.000
Abbreviations: CI, confidence interval; SDI, SLICC/ACR Damage Index; SoC, standard of care
At the end of the fifth year 1 (1.0%) SoC subject had seen an increase from their baseline
diabetes subscore. See Table 60. For belimumab subjects, 2 (2.0%) had seen an increase.
Table 60. SDI diabetes subscore change from baseline by year
SDI
Diabetes Year 1 Year 2 Year 3 Year 4 Year 5
Change
From
Baseline
SoC N=99
Belim N=99
SoC N=99
Belim N=99
SoC N=99
Belim N=99
SoC N=99
Belim N=99
SoC N=99
Belim N=99
0 [n (%)] 99
(100.0%) 99
(100.0%) 98
(99.0%) 99
(100.0%) 98
(99.0%) 98
(99.0%) 98
(99.0%) 97
(98.0%) 98
(99.0%) 97
(98.0%)
+1 [n (%)] 0 0 1
(1.0%) 0
1 (1.0%)
1 (1.0%)
1 (1.0%)
2 (2.0%)
1 (1.0%)
2 (2.0%)
Abbreviations: Belim, belimumab; SDI, SLICC/ACR Damage Index; SoC, standard of care
72
For the diabetes subscore there was no significant difference in the proportion of subjects with
an increase from their baseline score by the end of the 5th year (p=1.000) or at the end of any
of the prior years, See Table 61.
Table 61. Fisher’s test for belimumab versus SoC SDI diabetes subscore change from
baseline
YEAR Odds Ratio 95% CI P value
1 0 0 ; Inf 1.000
2 0 0 ; 39.001 1.000
3 1.000 0.013 ; 79.241 1.000
4 2.014 0.103 ; 120.322 1.000
5 2.014 0.103 ; 120.322 1.000
Abbreviations: CI, confidence interval; Inf, infinity; SDI, SLICC/ACR Damage Index; SoC, standard of care
No subject in either treatment arm saw an increase in their malignancy subscore.
Table 62. SDI malignancy subscore change from baseline by year
SDI
Malig Year 1 Year 2 Year 3 Year 4 Year 5
Change
From
Baseline
SoC N=99
Belim N=99
SoC N=99
Belim N=99
SoC N=99
Belim N=99
SoC N=99
Belim N=99
SoC N=99
Belim N=99
0 [n (%)] 99
(100.0%) 99
(100.0%) 99
(100.0%) 99
(100.0%) 99
(100.0%) 99
(100.0%) 99
(100.0%) 99
(100.0%) 99
(100.0%) 99
(100.0%)
Abbreviations: Belim, belimumab; Malig, malignancy; SDI, SLICC/ACR Damage Index; SoC, standard of care
6.1.8 Frequency of increase from baseline of SDI organ damage system
subscores
The total change from baseline for SDI subscores was analyzed using linear regression with the
difference between the subject’s score in their final year and their baseline score used as the
response variable. An indicator variable for treatment with belimumab along with a categorical
variable for the decade of entry were included as covariates. The year from baseline was also
included to control for the length of time from baseline. All PSM variables were balanced so
none were added as covariates.
The data set for this analysis consisted of the 358 PS matched patients where subjects were not
restricted to at least 5 years of follow-up. A second analysis was performed using the 5th year of
the primary 198 PS matched patients. The results from this analysis were used to check the
robustness of the results where subjects’ scores were recorded in different years.
73
The majority of the subscores showed no significant difference by treatment arm in the change
between the subject’s score in their final year and their baseline score; the exceptions were
musculoskeletal and skin subscores.
If baseline decade was not a significant factor in the SDI subscore change from baseline the
results were omitted for models with baseline decade as an explanatory variable.
For the cardiovascular SDI system subscore there was no significant difference (p=0.502)
between belimumab and SoC in the change from baseline score. See Table 63.
Table 63. SDI cardiovascular system subscore change from baseline
Variable
OLS Coefficient (SE)
[95% CI] P value
Robust SE Coefficient (SE)
[95% CI] P value
PS matched Year 5 Coefficient (SE)
[95% CI] P value
Intercept -0.0501 (0.0392)
95% CI: [-0.1272 ; 0.0270] p=0.202
-0.0501 (0.0317) 95% CI: [-0.1122 ; 0.0120]
p=0.114
0.0808 (0.0329) 95% CI: [0.0159 ; 0.1457]
p=0.015
Belimumab -0.0249 (0.0370)
95% CI: [-0.0977 ; 0.0479] p=0.502
-0.0249 (0.0299) 95% CI: [-0.0835 ; 0.0337]
p=0.405
-0.0202 (0.0465) 95% CI: [-0.1119 ; 0.0715]
p=0.665
Final Year 0.0201 (0.0037)
95% CI: [0.0128 ; 0.0274] p<0.001
0.0201 (0.0064) 95% CI: [0.0076 ; 0.0326]
p=0.002
NA
Abbreviations: CI, confidence interval; NA, not applicable; OLS, ordinary least squares; PS, propensity score; SDI, SLICC/ACR Damage Index; SE, standard error
When controlling for decade of entry, for the SDI diabetes subscore there was no significant
difference (p=0.138) between belimumab and SoC in the change from baseline score. See
Table 64.
74
Table 64. SDI diabetes subscore change from baseline
Variable
OLS Coefficient (SE)
[95% CI] P value
Robust SE Coefficient (SE)
[95% CI] P value
PS matched Year 5 Coefficient (SE)
[95% CI] P value
Intercept 0.0394 (0.0295)
95% CI: [-0.0186 ; 0.0974] p=0.182
0.0394 (0.0348) 95% CI: [-0.0288 ; 0.1077]
p=0.257
0.0000 (0.0246) 95% CI: [-0.0486 ; 0.0486]
p=1.000
Belimumab 0.0303 (0.0204)
95% CI: [-0.0098 ; 0.0704] p=0.138
0.0303 (0.0138) 95% CI: [0.0032 ; 0.0574]
p=0.029
0.0048 (0.0193) 95% CI: [-0.0332 ; 0.0428]
p=0.803
Final Year 0.0032 (0.0018)
95% CI: [-0.0004 ; 0.0068] p=0.081
0.0032 (0.0023) 95% CI: [-0.0013 ; 0.0077]
p=0.165 NA
Entry Decade
2000
-0.0659 (0.0269) 95% CI: [-0.1188 ; -0.0130]
p=0.015
-0.0659 (0.0346) 95% CI: [-0.1338 ; 0.0020]
p=0.057
0.0161 (0.0289) 95% CI: [-0.0410 ; 0.0732]
p=0.579
Entry Decade
2010
-0.0533 (0.0311) 95% CI: [-0.1145 ; 0.0079]
p=0.088
-0.0533 (0.0332) 95% CI: [-0.1183 ; 0.0117]
p=0.108
-0.0013 (0.0406) 95% CI: [-0.0813 ; 0.0787]
p=0.975
Abbreviations: CI, confidence interval; NA, not applicable; OLS, ordinary least squares; PS, propensity score; SDI, SLICC/ACR Damage Index; SE, standard error
For the SDI gastrointestinal system subscore there was no significant difference (p=0.675)
between belimumab and SoC in the change from baseline score, See Table 65.
Table 65. SDI gastrointestinal system subscore change from baseline
Variable
OLS Coefficient (SE)
[95% CI] P value
Robust SE Coefficient (SE)
[95% CI] P value
PS matched Year 5 Coefficient (SE)
[95% CI] P value
Intercept -0.0031 (0.0148)
95% CI: [-0.0322 ; 0.0259] p=0.832
-0.0031 (0.0129) 95% CI: [-0.0284 ; 0.0222]
p=0.808
0.0101 (0.0123) 95% CI: [-0.0142 ; 0.0344]
p=0.414
Belimumab 0.0059 (0.0140)
95% CI: [-0.0216 ; 0.0333] p=0.675
0.0059 (0.0127) 95% CI: [-0.0190 ; 0.0308]
p=0.644
0.0101 (0.0174) 95% CI: [-0.0243 ; 0.0445]
p=0.563
Final Year 0.0025 (0.0014)
95% CI: [-0.0003 ; 0.0052] p=0.078
0.0025 (0.0020) 95% CI: [-0.0015 ; 0.0064]
p=0.222 NA
Abbreviations: CI, confidence interval; NA, not applicable; OLS, ordinary least squares; PS, propensity score; SDI, SLICC/ACR Damage Index; SE, standard error
75
When controlling for decade of entry, for the SDI premature gonadal failure subscore there was
no significant difference (p=0.318) between belimumab and SoC in the change from baseline
score. See Table 66.
Table 66. SDI premature gonadal failure subscore change from baseline
Variable
OLS Coefficient (SE)
[95% CI] P value
Robust SE Coefficient (SE)
[95% CI] P value
PS matched Year 5 Coefficient (SE)
[95% CI] P value
Intercept 0.0015 (0.0107)
95% CI: [-0.0195 ; 0.0225] p=0.890
0.0015 (0.0017) 95% CI: [-0.0018 ; 0.0048]
p=0.383
0.0000 (0.0143) 95% CI: [-0.0281 ; 0.0281]
p=1.000
Belimumab -0.0133 (0.0074)
95% CI: [-0.0278 ; 0.0012] p=0.073
-0.0133 (0.0133) 95% CI: [-0.0393 ; 0.0128]
p=0.318
-0.0147 (0.0112) 95% CI: [-0.0367 ; 0.0073]
p=0.188
Final Year -0.0001 (0.0007)
95% CI: [-0.0014 ; 0.0012] p=0.849
-0.0001 (0.0001) 95% CI: [-0.0004 ; 0.0002]
p=0.383 NA
Entry Decade
2000
0.0129 (0.0097) 95% CI: [-0.0062 ; 0.0321]
p=0.185
0.0129 (0.0129) 95% CI: [-0.0125 ; 0.0383]
p=0.318
0.0152 (0.0167) 95% CI: [-0.0178 ; 0.0482]
p=0.365
Entry Decade
2010
0.0002 (0.0113) 95% CI: [-0.0219 ; 0.0223]
p=0.985
0.0002 (0.0008) 95% CI: [-0.0013 ; 0.0018]
p=0.780
0.0039 (0.0235) 95% CI: [-0.0423 ; 0.0502]
p=0.867
Abbreviations: CI, confidence interval; NA, not applicable; OLS, ordinary least squares; PS, propensity score; SDI, SLICC/ACR Damage Index; SE, standard error
For the SDI malignancy subscore there was no significant difference (p=0.776) between
belimumab and SoC in the change from baseline score. See Table 67.
76
Table 67. SDI malignancy subscore change from baseline
Variable
OLS Coefficient (SE)
[95% CI] P value
Robust SE Coefficient (SE)
[95% CI] P value
Intercept -0.0102 (0.0120)
95% CI: [-0.0338 ; 0.0134] p=0.397
-0.0102 (0.0096) 95% CI: [-0.0290 ; 0.0087]
p=0.290
Belimumab -0.0032 (0.0113)
95% CI: [-0.0255 ; 0.0191] p=0.776
-0.0032 (0.0074) 95% CI: [-0.0178 ; 0.0113]
p=0.664
Final Year 0.0033 (0.0011)
95% CI: [0.0011 ; 0.0056] p=0.003
0.0033 (0.0022) 95% CI: [-0.0009 ; 0.0076]
p=0.124
Abbreviations: CI, confidence interval; OLS, ordinary least squares; PS, propensity score; SDI, SLICC/ACR Damage Index; SE, standard error
When controlling for decade of entry, for the SDI musculoskeletal system subscore belimumab
subjects had a significantly smaller (p=0.006) increase from baseline score compared to SoC
subjects. See Table 68. The treatment coefficient (-0.2424) for the 5th year change from
baseline is within the 95% confidence interval (-0.3748 ; -0.0627) of the treatment coefficient
determined when using a subject’s last visit.
77
Table 68. SDI musculoskeletal system subscore change from baseline
Variable
OLS Coefficient (SE)
[95% CI] P value
Robust SE Coefficient (SE)
[95% CI] P value
PS matched Year 5 Coefficient (SE)
[95% CI] P value
Intercept 0.3645 (0.0964)
95% CI: [0.1750 ; 0.5541] p<0.001
0.3645 (0.1384) 95% CI: [0.0932 ; 0.6359]
p=0.008
0.2000 (0.0894) 95% CI: [0.0237 ; 0.3763]
p=0.026
Belimumab -0.2188 (0.0666)
95% CI: [-0.3498 ; -0.0877] p=0.001
-0.2188 (0.0796) 95% CI: [-0.3748 ; -0.0627]
p=0.006
-0.2424 (0.0700) 95% CI: [-0.3804 ; -0.1044]
p<0.001
Final Year 0.0148 (0.0060)
95% CI: [0.0031 ; 0.0265] p=0.013
0.0148 (0.0097) 95% CI: [-0.0043 ; 0.0338]
p=0.128 NA
Entry Decade
2000
-0.1967 (0.0879) 95% CI: [-0.3696 ; -0.0238]
p=0.026
-0.1967 (0.1387) 95% CI: [-0.4685 ; 0.0751]
p=0.156
0.0787 (0.1050) 95% CI: [-0.1284 ; 0.2857]
p=0.455
Entry Decade
2010
-0.3470 (0.1017) 95% CI: [-0.5470 ; -0.1470]
p<0.001
-0.3470 (0.1265) 95% CI: [-0.5949 ; -0.0991]
p=0.006
-0.0687 (0.1471) 95% CI: [-0.3589 ; 0.2215]
p=0.641
Abbreviations: CI, confidence interval; NA, not applicable; OLS, ordinary least squares; PS, propensity score; SDI, SLICC/ACR Damage Index; SE, standard error
For the SDI neuropsychiatric system subscore there was no significant difference (p=0.494)
between belimumab and SoC in the change from baseline score. See Table 69.
Table 69. SDI neuropsychiatric system subscore change from baseline
Variable
OLS Coefficient (SE)
[95% CI] P value
Robust SE Coefficient (SE)
[95% CI] P value
PS matched Year 5 Coefficient (SE)
[95% CI] P value
Intercept -0.0028 (0.0367)
95% CI: [-0.0750 ; 0.0694] p=0.939
-0.0028 (0.0351) 95% CI: [-0.0716 ; 0.0661]
p=0.937
0.0808 (0.0293) 95% CI: [0.0229 ; 0.1387]
p=0.006
Belimumab -0.0237 (0.0346)
95% CI: [-0.0919 ; 0.0444] p=0.494
-0.0237 (0.0295) 95% CI: [-0.0815 ; 0.0341]
p=0.421
0.0000 (0.0415) 95% CI: [-0.0818 ; 0.0818]
p=1.000
Final Year 0.0135 (0.0035)
95% CI: [0.0067 ; 0.0204] p<0.001
0.0135 (0.0057) 95% CI: [0.0024 ; 0.0247]
p=0.017
NA
Abbreviations: CI, confidence interval; NA, not applicable; OLS, ordinary least squares; PS, propensity score; SDI, SLICC/ACR Damage Index; SE, standard error
78
For the SDI ocular system subscore there was no significant difference (p=0.339) between
belimumab and SoC in the change from baseline score. See Table 70.
Table 70. SDI ocular system subscore change from baseline
Variable
OLS Coefficient (SE)
[95% CI] P value
Robust SE Coefficient (SE)
[95% CI] P value
PS matched Year 5 Coefficient (SE)
[95% CI] P value
Intercept -0.0217 (0.0306)
95% CI: [-0.0820 ; 0.0386] p=0.479
-0.0217 (0.0275) 95% CI: [-0.0755 ; 0.0322]
p=0.430
0.1212 (0.0291) 95% CI: [0.0639 ; 0.1785]
p<0.001
Belimumab -0.0277 (0.0289)
95% CI: [-0.0846 ; 0.0292] p=0.339
-0.0277 (0.0252) 95% CI: [-0.0771 ; 0.0217]
p=0.272
-0.0808 (0.0411) 95% CI: [-0.1618 ; 0.0002]
p=0.051
Final Year 0.0166 (0.0029)
95% CI: [0.0109 ; 0.0223] p<0.001
0.0166 (0.0045) 95% CI: [0.0078 ; 0.0253]
p<0.001
NA
Abbreviations: CI, confidence interval; NA, not applicable; OLS, ordinary least squares; PS, propensity score; SDI, SLICC/ACR Damage Index; SE, standard error
When controlling for decade of entry, for the SDI peripheral vascular system subscore there was
no significant difference (p=0.692) between belimumab and SoC in the change from baseline
score. See Table 71.
79
Table 71. SDI peripheral vascular system subscore change from baseline
Variable
OLS Coefficient (SE)
[95% CI] P value
Robust SE Coefficient (SE)
[95% CI] P value
PS matched Year 5 Coefficient (SE)
[95% CI] P value
Intercept 0.1223 (0.0432)
95% CI: [0.0373 ; 0.2073] p=0.005
0.1223 (0.0747) 95% CI: [-0.0240 ; 0.2686]
p=0.101
0.0000 (0.0349) 95% CI: [-0.0687 ; 0.0687]
p=1.000
Belimumab -0.0119 (0.0299)
95% CI: [-0.0706 ; 0.0469] p=0.692
-0.0119 (0.0248) 95% CI: [-0.0605 ; 0.0368]
p=0.633
0.0146 (0.0273) 95% CI: [-0.0392 ; 0.0684]
p=0.594
Final Year -0.0006 (0.0027)
95% CI: [-0.0058 ; 0.0047] p=0.827
-0.0006 (0.0037) 95% CI: [-0.0078 ; 0.0066]
p=0.874 NA
Entry Decade
2000
-0.0893 (0.0394) 95% CI: [-0.1669 ; -0.0118]
p=0.024
-0.0893 (0.0632) 95% CI: [-0.2133 ; 0.0346]
p=0.158
0.0166 (0.0409) 95% CI: [-0.0642 ; 0.0973]
p=0.686
Entry Decade
2010
-0.1191 (0.0456) 95% CI: [-0.2088 ; -0.0295]
p=0.009
-0.1191 (0.0671) 95% CI: [-0.2507 ; 0.0124]
p=0.076
-0.0039 (0.0574) 95% CI: [-0.1170 ; 0.1093]
p=0.946
Abbreviations: CI, confidence interval; NA, not applicable; OLS, ordinary least squares; PS, propensity score; SDI, SLICC/ACR Damage Index; SE, standard error
For the SDI pulmonary system subscore there was no significant difference (p=0.114) between
belimumab and SoC in the change from baseline score. See Table 72.
Table 72. SDI pulmonary system subscore change from baseline
Variable
OLS Coefficient (SE)
[95% CI] P value
Robust SE Coefficient (SE)
[95% CI] P value
PS matched Year 5 Coefficient (SE)
[95% CI] P value
Intercept 0.0219 (0.0190)
95% CI: [-0.0155 ; 0.0594] p=0.250
0.0219 (0.0231) 95% CI: [-0.0233 ; 0.0672]
p=0.343
0.0404 (0.0174) 95% CI: [0.0062 ; 0.0746]
p=0.021
Belimumab -0.0285 (0.0180)
95% CI: [-0.0638 ; 0.0069] p=0.114
-0.0285 (0.0167) 95% CI: [-0.0612 ; 0.0043]
p=0.088
-0.0404 (0.0245) 95% CI: [-0.0888 ; 0.0080]
p=0.101
Final Year 0.0021 (0.0018)
95% CI: [-0.0014 ; 0.0057] p=0.236
0.0021 (0.0027) 95% CI: [-0.0032 ; 0.0075]
p=0.437
NA
Abbreviations: CI, confidence interval; NA, not applicable; OLS, ordinary least squares; PS, propensity score; SDI, SLICC/ACR Damage Index; SE, standard error
80
For the SDI renal system subscore there was a marginally significant difference (p=0.084)
between belimumab and SoC in the change from baseline score. See Table 73. Subjects taking
belimumab tended to see a smaller increase from their baseline score. The treatment coefficient
(-0.0202) for the 5th year change from baseline is within the 95% confidence interval (-0.0720 ;
0.0046) of the treatment coefficient determined when using a subject’s last visit.
Table 73. SDI renal system subscore change from baseline
Variable
OLS Coefficient (SE)
[95% CI] P value
Robust SE Coefficient (SE)
[95% CI] P value
PS matched Year 5 Coefficient (SE)
[95% CI] P value
Intercept 0.0131 (0.0206)
95% CI: [-0.0274 ; 0.0537] p=0.525
0.0131 (0.0278) 95% CI: [-0.0413 ; 0.0675]
p=0.636
0.0202 (0.0101) 95% CI: [0.0004 ; 0.0400]
p=0.046
Belimumab -0.0337 (0.0195)
95% CI: [-0.0720 ; 0.0046] p=0.084
-0.0337 (0.0189) 95% CI: [-0.0708 ; 0.0033]
p=0.074
-0.0202 (0.0142) 95% CI: [-0.0482 ; 0.0078]
p=0.157
Final Year 0.0046 (0.0019)
95% CI: [0.0008 ; 0.0084] p=0.018
0.0046 (0.0032) 95% CI: [-0.0017 ; 0.0109]
p=0.150
NA
Abbreviations: CI, confidence interval; NA, not applicable; OLS, ordinary least squares; PS, propensity score; SDI, SLICC/ACR Damage Index; SE, standard error
For the SDI skin system subscore belimumab subjects had a significantly smaller (p=0.029)
increase from baseline score compared to SoC subjects. See Table 74. The treatment
coefficient (-0.0909) for the 5th year change from baseline is within the 95% confidence interval
(-0.0921 ; -0.0051) of the treatment coefficient determined when using a subject’s last visit.
81
Table 74. SDI skin system subscore change from baseline
Variable
OLS Coefficient (SE)
[95% CI] P value
Robust SE Coefficient (SE)
[95% CI] P value
PS matched Year 5 Coefficient (SE)
[95% CI] P value
Intercept 0.0045 (0.0234)
95% CI: [-0.0416 ; 0.0506] p=0.847
0.0045 (0.0352) 95% CI: [-0.0644 ; 0.0735]
p=0.898
0.0909 (0.0229) 95% CI: [0.0457 ; 0.1361]
p<0.001
Belimumab -0.0486 (0.0221)
95% CI: [-0.0921 ; -0.0051] p=0.029
-0.0486 (0.0188) 95% CI: [-0.0855 ; -0.0117]
p=0.010
-0.0909 (0.0324) 95% CI: [-0.1548 ; -0.0270]
p=0.006
Final Year 0.0078 (0.0022)
95% CI: [0.0034 ; 0.0121] p<0.001
0.0078 (0.0050) 95% CI: [-0.0021 ; 0.0176]
p=0.122
NA
Abbreviations: CI, confidence interval; NA, not applicable; OLS, ordinary least squares; PS, propensity score; SDI, SLICC/ACR Damage Index; SE, standard error
6.1.9 Difference in mean SLEDAI score from baseline over a 5-year interval
The 5-year adjusted mean SLEDAI (AMS) was analyzed using linear regression with a binary
indicator for treatment with belimumab as a covariate. All PSM variables were balanced (Table
11) so none were added as covariates. The decade of entry was initially included as a covariate
(Table 75), but was not statistically significant (p=0.118 and p= 0.066 for 2000 and 2010,
respectively).
Controlling for decade of entry there was no significant difference (p=0.892) in 5-year AMS
between belimumab and SoC. See Table 75. After removing decade of entry from the analysis
there still was not a statistically significant difference (p=0.715) in AMS over 5 years between
belimumab and SoC. See Table 76.
82
Table 75. Regression model AMS through year 5 with decade of baseline
Variable
OLS Coefficient (SE)
[95% CI] P value
Robust SE Coefficient (SE)
[95% CI] P value
Intercept 5.4179 (0.5943)
95% CI: [4.2451 ; 6.5907] p<0.001
5.4179 (0.5844) 95% CI: [4.2726 ; 6.5632]
p<0.001
Belimumab 0.0640 (0.4750)
95% CI: [-0.8734 ; 1.0013] p=0.893
0.0640 (0.4695) 95% CI: [-0.8563 ; 0.9842]
p=0.892
Entry Decade 2000 -1.0174 (0.7021)
95% CI: [-2.4028 ; 0.3680] p=0.149
-1.0174 (0.6516) 95% CI: [-2.2944 ; 0.2596]
p=0.118
Entry Decade 2010 -1.7400 (1.0132)
95% CI: [-3.7395 ; 0.2594] p=0.088
-1.7400 (0.9450) 95% CI: [-3.5921 ; 0.1120]
p=0.066
Abbreviations: AMS, adjusted mean SLEDAI; CI, confidence interval; OLS, ordinary least squares; SE, standard error; SLEDAI, Systemic Lupus Erythematosus Disease Activity Index
Table 76. Regression model AMS through year 5 without decade of baseline
Variable
OLS Coefficient (SE)
[95% CI] P value
Robust SE Coefficient (SE)
[95% CI] P value
Intercept 4.5974 (0.3063)
95% CI: [3.9930 ; 5.2018] p<0.001
4.5974 (0.2636) 95% CI: [4.0806 ; 5.1141]
p<0.001
Belimumab -0.1647 (0.4332)
95% CI: [-1.0195 ; 0.6901] p=0.704
-0.1647 (0.4516) 95% CI: [-1.0498 ; 0.7205]
p=0.715
Abbreviations: AMS, adjusted mean SLEDAI; CI, confidence interval; OLS, ordinary least squares; SE, standard error; SLEDAI, Systemic Lupus Erythematosus Disease Activity Index
6.1.10 Difference in cumulative corticosteroid usage from baseline over a 5-year
interval
The 5-year cumulative average daily corticosteroid usage was analyzed using linear regression
with a binary indicator for treatment with belimumab as a covariate. All PSM variables were
balanced (Table 11) so none were added as covariates. The decade of entry was initially
included as a covariate (Table 77), but was not statistically significant (p=0.407 and p=0.376,
83
respectively). Controlling for decade of entry belimumab was associated with a significantly
lower (p=0.011) 5-year mean daily steroid dose. See Table 77. For subjects who entered their
respective studies in the same decade belimumab subjects had a lower mean daily dose of over
2 mg/kg. After removing decade of entry the conclusion was unchanged; belimumab subjects
had a statistically significant (p=0.002) 5-year mean daily corticosteroid usage 2.35 mg/kg lower
than patients receiving SoC. See Table 78.
Table 77. Regression model cumulative corticosteroid usage through year 5 controlled
for decade of entry
Variable
OLS Coefficient (SE)
[95% CI] P value
Robust SE Coefficient (SE)
[95% CI] P value
Intercept 8.3490 (1.1466)
95% CI: [6.0875 ; 10.6105] p<0.001
8.3490 (1.3309) 95% CI: [5.7404 ; 10.9575]
p<0.001
Belimumab -2.0445 (0.9031)
95% CI: [-3.8258 ; -0.2632] p=0.025
-2.0445 (0.8061) 95% CI: [-3.6245 ; -0.4646]
p=0.011
Entry Decade 2000 -1.2248 (1.3496)
95% CI: [-3.8866 ; 1.4372] p=0.365
-1.2248 (1.4781) 95% CI: [-4.1218 ; 1.6723]
p=0.407
Entry Decade 2010 -1.7979 (1.8878)
95% CI: [-5.5214 ; 1.9256] p=0.342
-1.7979 (2.0297) 95% CI: [-5.7761 ; 2.1803]
p=0.376
Abbreviations: CI, confidence interval; OLS, ordinary least squares; SE, standard error
Table 78. Regression model cumulative corticosteroid usage through year 5 without
decade of entry covariates
Variable
OLS Coefficient (SE)
[95% CI] P value
Robust SE Coefficient (SE)
[95% CI] P value
Intercept 7.4633 (0.5860)
95% CI: [6.3074 ; 8.6192] p<0.001
7.4633 (0.5888) 95% CI: [6.3092 ; 8.6174]
p<0.001
Belimumab -2.3504 (0.8287)
95% CI: [-3.9850 ; -0.7158] p=0.005
-2.3504 (0.7724) 95% CI: [-3.8643 ; -0.8365]
p=0.002
Abbreviations: CI, confidence interval; OLS, ordinary least squares; SE, standard error
84
6.2 Exploratory Analyses
6.2.1 Propensity score matching
6.2.1.1 Pooled LTE and TLC Patients with 5-years follow up
Table 79 and Table 80 show the results of the full propensity score logistic regression model
over the entire sample of 973 patients. The range of the PS distribution (Table 80) was -5.730 to
4.866. The range of common support (the range of “overlap” in the PS distributions) for the LTE
and TLC patient was -3.125 to 4.522, illustrated in Figure 3. With the caliper value of 0.402
(20% of the standard deviation for the PS distribution), the range of support was -3.527 to
4.924. Thirty TLC patients and zero LTE patients with PS values outside of the range of support
(including the caliper) cannot be matched.
Using the PS values calculated from the full propensity score logistic regression model, 181 of
592 belimumab patients were matched 1:1 to 181 of the 381 TLC patients.
85
Table 79. Results of full propensity score logistic regression model, pooled LTE and TLC
dataset with 5 years follow-up (N=973)
Parameter Odds Ratio SE z p-value
Intercept 0.001 0.001 -6.76 <0.001
Age 1.265 0.051 5.80 <0.001
Age Squared 0.997 0.000 -5.24 <0.001
Female 1.325 0.402 0.93 0.354
Black 0.766 0.215 -0.95 0.342
Asian/Other Race 2.807 0.575 5.04 <0.001
SLE Duration 0.987 0.015 -0.84 0.402
Hypertension 2.047 0.444 3.30 0.001
Dyslipidemia 0.086 0.018 -11.50 <0.001
Proteinuria 0.450 0.101 -3.57 <0.001
ACR Criteria 1.173 0.082 2.30 0.022
Baseline SLEDAI 0.922 0.020 -3.78 <0.001
Corticosteroid Use 3.858 0.837 6.22 <0.001
Antimalarial Use 1.945 0.343 3.77 <0.001
Immunosuppressive Use 2.027 0.390 3.67 <0.001
Baseline SDI = 1 2.026 0.496 2.88 0.004
Baseline SDI = 2+ 3.730 1.173 4.19 <0.001
Abbreviations: ACR, American College of Rheumatology; LTE, long term extension; SDI, SLICC/ACR Damage Index; SE, standard error; SLE, systemic lupus erythematosus; SLEDAI, Systemic Lupus Erythematosus Disease Activity Index; TLC, Toronto Lupus Cohort
Table 80. Summary statistics of PSM variable, pooled LTE and TLC dataset with 5 years
follow-up (N=973)
Statistic Value
Observations 973
Mean (SD) 0.619 (2.001)
Range -5.730, 4.866
Caliper (20% of SD) 0.402
Abbreviation: LTE, long term extension trial; SD, standard deviation; TLC, Toronto Lupus Cohort
86
Figure 3. Common support in full model with all patients (n=973)
Prior to PSM, the LTE and TLC samples are not well balanced (Table 81). The percent bias is
larger than 10% for all of the variables (mean bias = 31.5%).
However, the PS-matched samples of 181 LTE and 181 TLC patients are well balanced (Table
82). Bias is less than 5% for all variables except antimalarial (14.9%) and immunosuppressive
(7.8%). The mean bias is 3.7%.
-6-4
-20
24
Pre
dic
ted
PS
Va
lue
(X
B)
LTE_OUS LTE_US TLC
87
Table 81. Bias prior to PS matching, pooled LTE and TLC dataset with 5 years follow-up
(N=973)
Mean t-test
Variable Belimumab SoC % Bias t p>|t|
Age 39.674 37.336 19.3 3.00 0.003
Age Squared 1693.9 1565.8 12.8 1.99 0.047
Female 0.927 0.895 11.4 1.76 0.078
Black 0.091 0.150 -18.0 -2.80 0.005
Asian/Other Race 0.471 0.234 51.3 7.68 <0.001
SLE Duration 6.683 5.738 13.9 2.16 0.031
Hypertension 0.426 0.370 11.4 1.73 0.085
Dyslipidemia 0.132 0.570 -103.1 -16.36 <0.001
Proteinuria 0.167 0.312 -34.4 -5.37 <0.001
ACR Criteria 5.932 5.646 20.8 3.18 0.002
Baseline SLEDAI 8.027 10.016 -42.5 -6.64 <0.001
Corticosteroid use 0.843 0.606 54.9 8.63 <0.001
Antimalarial Use 0.698 0.522 36.5 5.61 <0.001
Immunosuppressive Use 0.458 0.310 30.8 4.65 <0.001
Baseline SDI = 1 0.233 0.150 21.3 3.19 0.001
Baseline SDI = 2+ 0.181 0.105 21.8 3.23 0.001
Abbreviations: ACR, American College of Rheumatology; LTE, long term extension; PS, propensity score; SDI, SLICC/ACR Damage Index; SE, standard error; SLE, systemic lupus erythematosus; SLEDAI, Systemic Lupus Erythematosus Disease Activity Index; TLC, Toronto Lupus Cohort
88
Table 82. Bias post PS matching, pooled LTE and TLC dataset with 5 years follow-up
(n=362)
Mean t-test
Variable Belimumab SoC % Bias t p>|t|
Age 39.337 39.105 1.9 0.18 0.858
Age Squared 1691.8 1685.5 0.6 0.06 0.953
Female 0.895 0.906 -3.7 -0.35 0.726
Black 0.116 0.133 -5.0 -0.48 0.634
Asian/Other Race 0.315 0.331 -3.5 -0.34 0.737
SLE Duration 7.044 6.946 1.3 0.12 0.901
Hypertension 0.403 0.409 -1.1 -0.11 0.915
Dyslipidemia 0.326 0.309 3.6 0.34 0.736
Proteinuria 0.210 0.204 1.4 0.13 0.897
ACR Criteria 5.856 5.823 2.5 0.23 0.815
Baseline SLEDAI 9.094 8.912 4.4 0.42 0.675
Corticosteroid use 0.707 0.713 -1.2 -0.12 0.908
Antimalarial Use 0.669 0.597 14.9 1.42 0.157
Immunosuppressive Use 0.431 0.392 7.8 0.75 0.456
Baseline SDI = 1 0.204 0.193 2.8 0.26 0.793
Baseline SDI = 2+ 0.177 0.160 4.4 0.42 0.675
Abbreviations: ACR, American College of Rheumatology; LTE, long term extension; PS, propensity score; SDI, SLICC/ACR Damage Index; SE, standard error; SLE, systemic lupus erythematosus; SLEDAI, Systemic Lupus Erythematosus Disease Activity Index; TLC, Toronto Lupus Cohort
6.2.1.2 Pooled LTE and TLC Patients with ≥ 1-year follow up for time to event analyses
Table 83 and Table 84 show the results of the full propensity score logistic regression model
over the entire sample of 1,541 patients. The range of the PS distribution (Table 84) was -5.403
to 4.428. The range of common support (the range of “overlap” in the PS distributions) for the
LTE and TLC patient was -3.192 to 4.214, illustrated in Figure 4. With the caliper value of 0.343
(20% of the standard deviation for the PS distribution), the range of support was -3.535 to
4.558. The twenty TLC patients and the seven LTE patients with PS values outside of the range
of support (including the caliper) cannot be matched.
Using the PS values calculated from the full propensity score logistic regression model, 323 of
949 belimumab patients were matched 1:1 to 323 of the 592 TLC patients.
89
Table 83. Results of full propensity score logistic regression model, pooled LTE and TLC
dataset with ≥ 1 year follow-up (N=1541)
Parameter Odds Ratio SE z p-value
Intercept 0.002 0.001 -9.02 <0.001
Age 1.250 0.037 7.62 <0.001
Age Squared 0.998 0.000 -6.87 <0.001
Female 2.074 0.488 3.10 0.002
Black 0.493 0.116 -3.02 0.003
Asian/Other Race 1.793 0.269 3.90 <0.001
SLE Duration 0.960 0.011 -3.70 <0.001
Hypertension 1.607 0.254 3.01 0.003
Dyslipidemia 0.129 0.021 -12.44 <0.001
Proteinuria 0.386 0.066 -5.60 <0.001
ACR Criteria 1.238 0.063 4.17 <0.001
Baseline SLEDAI 0.920 0.015 -5.27 <0.001
Corticosteroid Use 4.845 0.807 9.48 <0.001
Antimalarial Use 1.296 0.176 1.91 0.057
Immunosuppressive Use 1.577 0.224 3.22 0.001
Baseline SDI = 1 2.242 0.423 4.28 <0.001
Baseline SDI = 2+ 3.713 0.853 5.71 <0.001
Abbreviations: ACR, American College of Rheumatology; LTE, long term extension; PS, propensity score; SDI, SLICC/ACR Damage Index; SE, standard error; SLE, systemic lupus erythematosus; SLEDAI, Systemic Lupus Erythematosus Disease Activity Index; TLC, Toronto Lupus Cohort
Table 84. Summary statistics of PSM variable, pooled LTE and TLC dataset with ≥ 1 year
follow-up (N=1541)
Statistic Value
Observations 1541
Mean (SD) 0.639 (1.717)
Range -5.403, 4.428
Caliper (20% of SD) 0.343
Abbreviation: LTE, long term extension; SD, standard deviation; TLC, Toronto Lupus Cohort
90
Figure 4. Common support in full model with all patients (N=1541)
Prior to PSM, the LTE and TLC samples are not well balanced (Table 85). The percent bias is
larger than 10% for most of the variables (mean bias = 26.7%).
However, the PS-matched samples of 323 LTE and 323 TLC patients are well balanced (Table
86). Bias is less than less than 10% for all variables and less than 5% for all but five variables,
with the largest bias belonging to the baseline SDI greater than or equal to two variable (9.8%).
The mean bias is 3.7%.
-6-4
-20
24
Pre
dic
ted
PS
Va
lue
(X
B)
LTE_OUS LTE_US TLC
91
Table 85. Bias prior to PS matching, pooled LTE and TLC dataset with ≥ 1 year follow-up
(N=1541)
Mean t-test
Variable Belimumab SoC % Bias t p>|t|
Age 38.782 36.735 16.2 3.16 0.002
Age Squared 1634.0 1538.5 9.0 1.77 0.078
Female 0.942 0.885 20.4 4.04 <0.001
Black 0.075 0.150 -24.1 -4.77 <0.001
Asian/Other Race 0.457 0.301 32.7 6.19 <0.001
SLE Duration 6.737 6.358 5.5 1.07 0.283
Hypertension 0.400 0.383 3.3 0.63 0.528
Dyslipidemia 0.128 0.471 -80.7 -16.14 <0.001
Proteinuria 0.169 0.346 -41.3 -8.11 <0.001
ACR Criteria 5.971 5.674 21.6 4.12 <0.001
Baseline SLEDAI 8.273 10.100 -39.8 -7.76 <0.001
Corticosteroid use 0.859 0.639 52.5 10.40 <0.001
Antimalarial Use 0.669 0.593 15.8 3.04 0.002
Immunosuppressive Use 0.472 0.372 20.4 3.89 <0.001
Baseline SDI = 1 0.241 0.150 22.9 4.29 <0.001
Baseline SDI = 2+ 0.177 0.103 21.3 3.97 <0.001
Abbreviations: ACR, American College of Rheumatology; LTE, long term extension; PS, propensity score; SDI, SLICC/ACR Damage Index; SE, standard error; SLE, systemic lupus erythematosus; SLEDAI, Systemic Lupus Erythematosus Disease Activity Index; TLC, Toronto Lupus Cohort
92
Table 86. Bias post PS matching, pooled LTE and TLC dataset with ≥ 1 year follow-up
(n=646)
Mean t-test
Variable Belimumab SoC % Bias t p>|t|
Age 38.108 37.416 5.4 0.69 0.492
Age Squared 1611.1 1566.9 4.2 0.53 0.598
Female 0.926 0.913 4.5 0.58 0.564
Black 0.115 0.108 2.0 0.25 0.803
Asian/Other Race 0.337 0.362 -5.2 -0.66 0.510
SLE Duration 7.061 6.803 3.6 0.46 0.648
Hypertension 0.412 0.402 1.9 0.24 0.810
Dyslipidemia 0.279 0.282 -0.7 -0.09 0.930
Proteinuria 0.245 0.263 -4.3 -0.54 0.588
ACR Criteria 5.901 5.892 0.7 0.09 0.931
Baseline SLEDAI 9.105 9.046 1.4 0.18 0.856
Corticosteroid use 0.715 0.743 -6.3 -0.80 0.426
Antimalarial Use 0.656 0.628 5.8 0.74 0.461
Immunosuppressive Use 0.399 0.409 -1.9 -0.24 0.810
Baseline SDI = 1 0.235 0.195 9.8 1.24 0.214
Baseline SDI = 2+ 0.133 0.139 -1.8 -0.23 0.819
Abbreviations: ACR, American College of Rheumatology; LTE, long term extension; PS, propensity score; SDI, SLICC/ACR Damage Index; SE, standard error; SLE, systemic lupus erythematosus; SLEDAI, Systemic Lupus Erythematosus Disease Activity Index; TLC, Toronto Lupus Cohort
6.2.2 Difference in change in SDI from baseline to 5 years
The total SDI score change from baseline to 5 years was evaluated using linear regression with
a binary indicator for treatment with belimumab as a covariate. 262 pooled LTE patients were
matched to 262 TLC patients using PSM. All but one of the PSM variables (antimalarial) were
balanced (Table 82). Therefore, an indicator variable for antimalarial use at baseline was
included as a covariate. The baseline decade of entry was also included as a covariate, with
1990 as the reference level.
When controlling for antimalarial use and decade of entry, the difference from baseline in the
fifth year total SDI score was significantly (p<0.001) lower for subjects taking belimumab. See
Table 87. None of the other covariates were statistically significant.
93
Without controlling for antimalarial use and decade of entry, the difference from baseline in the
fifth year total SDI score increased slightly but was still significantly (p<0.001) lower for subjects
taking belimumab. See Table 88.
Table 87. Year 5 Total SDI difference of change from baseline controlled for entry decade
and antimalarial use
Variable
OLS Coefficient (SE)
[95% CI] P value
Robust SE Coefficient (SE)
[95% CI] P value
Intercept 0.7183 (0.1489)
95% CI: [0.4254 ; 1.0112] p<0.001
0.7183 (0.1774) 95% CI: [0.3706 ; 1.0659]
p<0.001
Belimumab -0.4913 (0.1075)
95% CI: [-0.7026 ; -0.2800] p<0.001
-0.4913 (0.1210) 95% CI: [-0.7284 ; -0.2542]
p<0.001
Antimalarial use -0.0443 (0.1018)
95% CI: [-0.2446 ; 0.1560] p=0.664
-0.0443 (0.1011) 95% CI: [-0.2424 ; 0.1538]
p=0.661
Entry Decade 2000 0.0713 (0.1648)
95% CI: [-0.2528 ; 0.3954] p=0.665
0.0713 (0.1954) 95% CI: [-0.3117 ; 0.4543]
p=0.715
Entry Decade 2010 -0.2453 (0.2610)
95% CI: [-0.7586 ; 0.2680] p=0.348
-0.2453 (0.2266) 95% CI: [-0.6894 ; 0.1989]
p=0.279
Abbreviations: CI, confidence interval; OLS, ordinary least squares; SDI, SLICC/ACR Damage Index; SE, standard error
94
Table 88. Year 5 Total SDI difference of change from baseline
Variable
OLS Coefficient (SE)
[95% CI] P value
Robust SE Coefficient (SE)
[95% CI] P value
Intercept 0.7182 (0.0687)
95% CI: [0.5830 ; 0.8534] p<0.001
0.7182 (0.0871) 95% CI: [0.5475 ; 0.8890]
p<0.001
Belimumab -0.4530 (0.0972)
95% CI: [-0.6442 ; -0.2619] p<0.001
-0.4530 (0.0984) 95% CI: [-0.6460 ; -0.2601]
p<0.001
Abbreviations: CI, confidence interval; OLS, ordinary least squares; SDI, SLICC/ACR Damage Index; SE, standard error
6.2.3 Difference in time to first SDI worsening
The time to the first worsening (increase) in total SDI score was analyzed using parametric
survival models with a binary indicator for treatment with belimumab as the covariate. All PSM
variables had a bias less than 10%, however, the bias for the baseline SDI score of greater than
or equal to two variable was 9.8%. (Table 86). Therefore, baseline SDI score was included as a
covariate with the same levels used in the propensity score matching. The baseline decade of
entry also was included as a covariate.
Models with exponential, Weibull, Gompertz, log logistic, and log normal distributions were
evaluated (Table 89 to Table 94). Neither the baseline SDI score covariate nor the decade of
entry covariate was statistically significant for any of the distributions. In fact, there was only a
minor change in the coefficient estimate for belimumab when both covariates were excluded.
Since their impact was negligible, the results with all covariates are shown only for the
exponential distribution (Table 89).
Results indicated that belimumab was associated with a significantly lower rate of organ
damage progression, p<0.001, regardless of the distribution used.
95
Table 89. Proportional hazards model of time to first change in total SDI score controlling
for baseline SDI score and decade of study entry, exponential distribution
Copyright 2017 the GlaxoSmithKline group of companies. All rights reserved. Unauthorized copying or use of this information is prohibited.
HEALTH OUTCOMES STUDY PROTOCOL
UNIQUE IDENTIFIER HO-16-16611/206347
FULL TITLE A propensity score-matched study of systemic lupus erythematosus related organ damage in the BLISS long term extension trials (BEL112233 and BEL112234) and the Toronto Lupus Cohort
ABBREVIATED TITLE A propensity score-matched study of the BLISS long term extension trials vs. the Toronto Lupus Cohort
FINAL PROTOCOL APPROVED
SPONSORSHIP Sponsored
DIVISION Pharma
BUSINESS UNIT Research & Development
DEPARTMENT GHO / Medical Decision Modeling Inc.
STUDY ACCOUNTABLE PERSON
CONTRIBUTING AUTHORS Medical Decision Modeling Inc.
ASSET ID GSK1550188
GSK ASSET Belimumab (Benlysta®)
INDICATION Systemic lupus erythematosus
REVISION CHRONOLOGY:
Version Date Document Type Change(s) since last version
15-JUN-2016 Original n/a
15-NOV-2017 Amendment 01 Updated sensitivity and exploratory analyses to replace use of interim pooled US/OUS LTE dataset (201223) with new pooled dataset of US LTE (BEL112233) and OUS LTE (BEL112234) to be constructed de novo
Description: A propensity score-matched study of the BLISS long term extension trials vs. the Toronto Lupus Cohort
PPD
SPONSOR SIGNATORY
z+Drt"
Senior Line ManagerVP, IIVEO
tl .zot/
2
PPD
PPD
3
PROTOCOL SYNOPSIS
Unique Identifier
Abbreviated Title A propensity score-matched study of the BLISS long term extension trials vs. the Toronto Lupus Cohort
GSK Product Belimumab (Benlysta®)
Rationale To demonstrate the reduction in organ damage of long term treatment with belimumab plus standard of care (SoC) versus SoC alone for systemic lupus erythematosus.
Objectives (Primary, Secondary)
To compare the mean change in SDI scores from baseline to year 5 between patients treated with belimumab or SoC, based on data from the US BLISS LTE trial (BEL112233) and the TLC. To compare the time to first SDI worsening between patients treated with belimumab or SoC, based on data from the US BLISS LTE trial (BEL112233) and the TLC. To compare the total SDI score at yearly intervals between patients treated with belimumab or SoC, based on data from the US BLISS LTE trial (BEL112233) and the TLC. To perform MSM transition analyses of SDI independently for the belimumab and SoC groups using the Jackson et al. 2011 methodology based on data from the US BLISS LTE trial (BEL112233) and the TLC. To describe the change from baseline in SDI organ damage system (ocular, neuropsychiatric, renal, pulmonary, cardiovascular, peripheral vascular, gastrointestinal [GI], musculoskeletal, skin, premature gonadal failure, diabetes and malignancy) summarized by year interval of patients treated with belimumab or SoC, based on data from the US BLISS LTE trial (BEL112233) and the TLC. To compare the difference in mean SLEDAI score from baseline over a 5-year period between patients treated with belimumab or SoC, based on data from the US BLISS LTE trial (BEL112233) and the TLC. To compare the difference in cumulative corticosteroid usage from baseline over a 5-year period between patients treated with belimumab or SoC, based on data from the US BLISS LTE trial (BEL112233) and the TLC.
4
As a sensitivity analysis, the primary and secondary objectives above will be retested using a pooled BLISS LTE dataset to be constructed de novo from the BLISS LTE trials (BEL112233 and BEL112234) and the TLC.
Study Design A longitudinal propensity score-matched study comparing individual patients of the BLISS LTE trial(s) to clinically and demographically similar patients in the TLC
Study Population and Sampling Methods
Inclusion criteria:
• Diagnosis of systemic lupus erythematosus (ICD-9 710.0) using ≥ 4 of 11 American College of Rheumatology (ACR) criteria (710.0)
• Active severe lupus nephritis or central nervous system lupus
• Receipt of B cell target therapy at any time
• For TLC patients, previous use of belimumab
Data Source BLISS long term extension trials and Toronto Lupus Cohort
Data Analysis Methods Primary endpoint:
• The difference in change in SDI from baseline to year 5 interval between patients treated with belimumab or SoC, based on the data of the US BLISS LTE trial (BEL112233) and the TLC.
Secondary endpoints: The analysis of all secondary endpoints will use data from the US BLISS LTE trial (BEL112233) for patients treated with belimumab and data from the TLC for patients treated with SoC.
• The difference in time to first SDI worsening between patients treated with belimumab or SoC.
• The change from baseline SDI score by year interval for patients treated with belimumab or SoC.
• The difference in change from baseline SDI score by year interval between patients treated with belimumab or SoC.
• Transition analysis of SDI from baseline over a 5-year period for patients treated with belimumab or SoC.
• Change from baseline in SDI organ damage system (ocular, neuropsychiatric, renal,pulmonary, cardiovascular, peripheral vascular, gastrointestinal [GI], musculoskeletal,skin, premature gonadal failure, diabetes
5
and malignancy) summarized by year interval for patients treated with belimumab or SoC.
• The frequency of increase from baseline in SDI organ damage system subscores between patients treated with belimumab or SoC.
• The difference in mean SLEDAI over the 5-year period.
• The difference in cumulative corticosteroid usage over the 5-year period.
Exploratory endpoints: The analysis of all exploratory endpoints will use the pooled BLISS LTE dataset to be constructed from the BLISS LTE trials (BEL112233 and BEL112234) for patients treated with belimumab and data from the TLC for patients treated with SoC.
• The difference in time to first SDI worsening between patients treated with belimumab or SoC.
• The change from baseline SDI score by year interval for patients treated with belimumab or SoC.
• The difference in change from baseline SDI score by year interval between patients treated with belimumab or SoC.
• Transition analysis of SDI from baseline over a 5 year period for patients treated with belimumab or SoC.
• Change from baseline in SDI organ damage system (ocular, neuropsychiatric, renal, pulmonary, cardiovascular, peripheral vascular, gastrointestina [GI], musculoskeletal, skin, premature gonadal failure, diabetes and malignancy) summarized by year interval for patients treated with belimumab or SoC.
Sample Size and Power Sample size:
LTE TLC
Primary and secondary endpoints
US LTE Sample (≥ 5 years Tx duration) with 2:1 Match
192 384
Exploratory endpoints
Pooled LTE Sample (≥5 years Tx duration) with 1:1 Match*
530 530
Power:
80%, =0.05 to detect a difference of
• Primary endpoint - 0.155 (34%)
• First exploratory endpoint -- 0.113 (27%)
Limitations The primary limitations are the numbers of patients in the BLISS LTE trials, the number of patients in the TLC, and the number of
6
patients with matching characteristics. This limits the power to reach statistically significant conclusions.
Other limitations include:
• Dissimilar data collection cycles, i.e. data in the BLISS LTEs were collected on a pre-defined schedule, while data in the TLC are collected when patients schedule visits.
• The controls will not have been randomly selected from the same population as the BLISS LTE patients. Instead propensity score matching will be used to match belimumab patients with controls based on a number of patient characteristics.
• BLISS patients may have received care from different types of health care systems than external cohort controls.
• SELENA-SLEDAI scores are available for BLISS LTE patients while SLEDAI-2K scores are available for TLC patients. While the components of SELENA-SLEDAI remain the same as the SLEDAI-2K, the definitions of several components are slightly different. The length of assessment was also different in the BLISS trials than in the TLC; up to 10 days prior to a visit in the BLISS trials while up to 30 days prior to a visit in the TLC. A study has indicated that there is minimal difference between 10 and 30 days assessment. The SELENA-SLEDAI is yet to be rigorously validated.
• Patients were free to withdraw from the BLISS studies and from the TLC. This may introduce bias.
• The sensitivity analysis will utilize pooled patient-level data from both BLISS LTE trials (BEL112233 and BEL112234), treating the data as through from one, rather than two, trials. In fact the populations and health systems through which care was given may have been significantly different. The preferred method of aggregating data from multiple trials is meta-analysis. However, using meta-analysis for just two trials would result in loss of power. The pooled dataset will be constructed de novo from the datasets of the BLISS LTE trials (BEL112233 and BEL112234) as part of this analysis.
• The BLISS trials used 48 weeks as a year.
• The PS matching approach to be used only accounts for sample selection bias (aka confounding by indication) for the observed confounders (predictors of organ damage also potentially affecting treatment assignment) included in the PS model. To the extent additional clinically important confounders exist in the data but cannot be observed,
7
sample selection bias cannot be fully addressed using PS adjustment methods.
• The TLC is treated at a single Canadian clinical site while the BLISS BEL112233 LTE trial was conducted at multiple US sites. Differences in outcomes may be confounded by differences in the national health systems. Outcomes may also be confounded by treatment practices specific to the TLC clinical site.
8
TABLE OF CONTENTS 1 INTRODUCTION/BACKGROUND................................................................................................ 11
3.5 Sample Size / Power Calculations ....................................................................................... 19
3.5.1 Available sample sizes for BLISS LTE and TLC data after propensity score matching ....................................................................................................................... 19
3.5.2 Sample size requirements for “Mean Change from Baseline SDI” endpoint: Alternative treatment effect size assumptions (SDLTE=0.6, SDTLC=0.7, power=80%,
4.2 Determination of patient characteristics to be used to propensity score matching .............. 21
4.3 Assessment of suitability of final propensity score model for matching ............................... 22
4.4 Identify matches between BLISS LTE (treatment) patients and TLC (comparison) patients, using propensity score with a 2:1 match ratio (for US LTE data used for study) . 22
4.5 Assess degree of post-matching balance in predictors used in propensity score model across treatment and comparison groups ............................................................................ 23
4.6 Index date for patients in the TLC ........................................................................................ 23
4.7 Withdrawals from BLISS LTE trials ...................................................................................... 23
4.8 Analysis of propensity score matching variables ................................................................. 23
4.9 Rationale for inclusion of decade of study entry as a covariate ........................................... 23
SDI Systemic Lupus International Collaborating Clinics/American College of Rheumatology (ACR) Damage Index
SLE Systemic lupus erythematosus
SLEDAI Systemic Lupus Erythematosus Disease Activity Index
SoC Standard of care
TLC Toronto Lupus Cohort
Tx Treatment
11
1 INTRODUCTION/BACKGROUND
Two Phase 3 randomized controlled trials of intravenously (IV) administered belimumab have established the clinical effectiveness of belimumab plus standard of care (SoC) versus SoC alone at 52 (BLISS 52) and 76 (BLISS 76) weeks. The long term extensions (LTE) (BEL112233 and BEL112234) of these trials, however, did not have comparison SoC arms. Thus the question of the long-term relative efficacy of belimumab with SoC versus SoC alone remains unanswered. The purpose of this study is to provide a long-term comparative analysis between belimumab plus SoC versus SoC alone in the treatment of systemic lupus erythematosus (SLE). It plans to do so by comparing BLISS LTE patients to propensity score-matched SLE patients with similar baseline characteristics taken from an external SLE cohort. A systematic review of the literature was previously performed to identify research cohorts of SLE patients (attached as Appendix A).1 The review identified the Toronto Lupus Cohort (TLC) as the preferred source of SoC data for this study based on the size of the cohort, the extent of organ damage seen in the patients and severity of SLE disease activity. A subset of the TLC with patient baseline characteristics similar to the BLISS trials has previously been used in a GSK study of mortality and damage progression in SLE. The use of a similar subset of the TLC is envisioned in this study. This will be the first analysis of long term efficacy of belimumab plus SoC versus SoC alone. The primary analysis will take place on data from the US LTE trial (BEL112233) versus the TLC patient data. As an exploratory sensitivity analysis the same analysis will be performed on the pooled BLISS LTE dataset to be constructed from the two BLISS LTE trials (BEL112233 and BEL112234) versus the TLC. Throughout the remainder of this document “belimumab treatment” refers to treatment with belimumab supplemented by SoC while SoC refers to SoC alone. Similarly, “TLC” throughout the remainder of this document refers to a subset of the TLC with patient characteristics similar to the patient baseline characteristics in the BLISS trials. Observations in both the LTE and TLC data may not fall on annual intervals, thus, for instance, a 5th year observation will be the first observation to take place ≥ 5 years but before 6 years from the index date.
2 OBJECTIVES
2.1 Primary To compare the mean change in SDI scores from baseline to year 5 between patients treated with belimumab or SoC, based on data from the US BLISS LTE trial (BEL112233) and the TLC.
2.2 Secondary To compare the time to first SDI worsening between patients treated with belimumab or SoC, based on data from the US BLISS LTE trial (BEL112233) and the TLC.
12
To compare the total SDI score at yearly intervals between patients treated with belimumab or SoC, based on data from the US BLISS LTE trial (BEL112233) and the TLC. To perform MSM transition analyses of SDI independently for the belimumab and SoC groups using the Jackson et al. 2011 methodology based on data from the US BLISS LTE trial (BEL112233) and the TLC.2,3,4 To describe the change from baseline in SDI organ damage system (ocular, neuropsychiatric, renal, pulmonary, cardiovascular, peripheral vascular, gastrointestinal [GI], musculoskeletal, skin, premature gonadal failure, diabetes and malignancy) summarized by year interval of patients treated with belimumab or SoC, based on data from the US BLISS LTE trial (BEL112233) and the TLC.
2.3 Exploratory As a sensitivity analysis, the primary and secondary objectives above will be retested using the pooled BLISS LTE dataset to be constructed from the BLISS LTE trials (BEL112233 and BEL112234) and the TLC.
3 RESEARCH METHODOLOGY
3.1 Study Design This is a longitudinal propensity score-matched study comparing individual patients of the BLISS LTE trial(s) to clinically and demographically similar patients in the TLC.
3.2 Study Population
3.2.1 Eligibility Criteria
3.2.1.1 Inclusion Criteria
• Diagnosis of systemic lupus erythematosus (ICD-9 710.0) using ≥ 4 of 11 American College of Rheumatology (ACR) criteria (710.0)
• Active severe lupus nephritis or central nervous system lupus
• Receipt of B cell target therapy at any time
• For TLC patients, previous use of belimumab
13
3.2.2 Sampling
All patients in the BLISS LTE trial(s) will be propensity score-matched to patients in the TLC meeting the inclusion and exclusion criteria above. The numbers of patients available from each source appear in the table below.
Sample Sizes Study start 1 year 2 year 5 year
TLC with SLEDAI ≥ 6 5 649 649 649 536
Study start Year 1-2 Year 2-3 Year 5-6
US BLISS LTE (BEL112233)6 268 259 244 192
LTE pooled dataset MITT7 998 955 861 531
3.2.3 Matching Procedure
For purposes of propensity score matching, the baseline for BLISS LTE subjects is defined as the date of first exposure to belimumab in the BLISS trials. The figure below illustrates the definition of the belimumab baseline for subjects enrolled in the US BLISS LTE (BEL112233).
• For subjects randomized to belimumab 10 mg/kg in BLISS 76 (BEL110751), the belimumab baseline is the core baseline (baseline at randomization) in BLISS 76 (BEL110751).
• For subjects randomized to belimumab 1 mg/kg in BLISS 76 (BEL110751), the belimumab baseline is the core baseline (baseline at randomization) in BLISS 76 (BEL110751).
• For subjects randomized to placebo in BLISS 76 (BEL110751), the belimumab baseline is the extension baseline (baseline at start of LTE) in US BLISS LTE (BEL112233).
Definition of belimumab baseline for BLISS US LTE (BEL112233) subjects
The matching procedure is illustrated in the following figure. Collectively, the subjects enrolled in the BLISS US LTE (BEL112233) comprise the belimumab cohort of the primary analysis. Propensity score matching is performed by comparing the characteristics of BLISS subjects at their respective belimumab baselines with the characteristics of TLC patients at their respective index dates. Data analysis considerations for defining the index dates for TLC patients are discussed in section 4.6.
Core
Baseline
Extension
Baseline
Belimumab Baseline
BLISS 76 Open label extension (BEL112233)
PBO BEL 10mg ≤5 years of follow-upPlacebo
BEL 1mg BEL 10mg ≤6.5 years of follow-upBEL 1mg/kg
BEL 10mg ≤6.5 years of follow-upBEL 10mg/kg
14
Propensity score matching procedure for primary analyses
The procedure outlined above ensures that the propensity score matching is performed against the point in time at which patients in the different cohorts were assigned to different treatments. However, it introduces two potential sources of biases:
• For BLISS LTE subjects randomized to belimumab 1 mg/kg or belimumab 10 mg/kg in the parent study, the follow-up period includes the duration of the parent study. However, for BLISS subjects randomized to placebo in the core study and for the TLC cohort, the follow-up period does not contain any period of enrollment in a randomized clinical trial. Differences in quality of care obtained within the setting of the randomized clinical trial, versus the quality of care during the LTE or in the TLC, may lead to differences in outcomes during the follow-up period for BLISS LTE subjects randomized to belimumab in the parent study versus BLISS LTE subjects randomized to placebo in the parent study and the TLC cohort.
• For BLISS LTE subjects randomized to belimumab 1 mg/kg in the parent study, the follow-up period includes the period of exposure to belimumab 1 mg/kg. However, for BLISS subjects randomized to belimumab 10 mg/kg or placebo in the parent study, the subjects were exposed to belimumab 10 mg/kg for the entire follow-up period. A dose response effect may lead to differences in outcomes for BLISS LTE subjects randomized to belimumab 1 mg/kg in the parent study versus BLISS LTE subjects randomized to belimumab 10 mg/kg or placebo in the parent study.
Diagnostics to test for these potential effects are discussed in sections 4.12.6 and 4.12.7.
BenlystaBLISS
TLC SoC
Belimumab
Baseline
PBO BEL 10mgPlacebo
BEL 1mg BEL 10mgBEL 1mg/kg
BEL 10mgBEL 10mg/kg
BLISS 76 Open label extension (BEL112233)
BLISS 76 Open label extension (BEL112233)
BLISS 76 Open label extension (BEL112233)
PS matching
conducted at
belimumab Baseline
Collectively, these three groups make up the ‘BLISS arm’ of the comparison
Up to 6.5 years of exposure
15
3.2.4 Matching Criteria
The set of criteria to be considered for inclusion in the propensity score model will be based on patient level characteristic predictors of organ damage. Potential criteria were identified by reviewing the literature for factors predicting organ damage. Only baseline patient characteristics available in both the BLISS LTE data and the TLC data will be evaluated for inclusion (see 3.2.3). Final predictors of organ damage to be used in the propensity score model will be determined using statistical evaluation criteria (see 3.2.4.2) subject to validation by two experts in the treatment and research of SLE. Matching criteria will be restricted to baseline characteristics because post-baseline variables could be confounded by treatment effects. As discussed in section 3.2.3, the matching criteria for BLISS LTE subjects are based on the first date of exposure to belimumab.
• Randomized to belimumab 10 mg/kg in BLISS 76: Matching criteria are based on characteristics at BLISS 76 baseline.
• Randomized to belimumab 1 mg/kg in BLISS 76: Matching criteria are based on characteristics at BLISS 76 baseline.
• Randomized to SoC in BLISS 76: Matching criteria are based on characteristics at first exposure to belimumab in the BLISS LTE (BEL112233) (extension baseline).
• If the data recorded for characteristics at extension baseline are identical to the characteristics recorded at BLISS 76 baseline, then the data will be used as-is.
• If there is no distinct data field available for a characteristic at extension baseline, then the value of the characteristic at BLISS 76 baseline will be used as a proxy.
3.2.4.1 Predictors of Organ Damage Progression Reported in the Literature, Availability in the TLC and BLISS LTE Data, and Candidate Variables to be used or Propensity Score Adjustment
Predictor TLC BLISS LTE
(US / Pooled) Potential PS
matching factor
Age5,8,9 Yes Yes / Yes Yes
Gender8,9 Yes Yes / Yes Yes
Race, Ethnicity8,9 Yes Yes / Yes Yes
Household income8 SES data No / No
Geocode proxy?
Educational attainment8 SES data No / No
Geocode proxy?
Disease duration8,9 Yes Yes / Yes Yes
Current Smoker5 Yes Probably / Probably
Probably
History-hypertension8 Yes§ Yes / Yes Yes
History-dyslipidemia Yes‡‡ Proxy†† No
16
History-proteinuria8 Yes‡‡ Yes / Yes Yes
History-lupus anticoag positivity8
Yes‡‡ No / No No
Baseline # ACR criteria satisfied8
At entry† No§§ / No§§ No
Baseline SLEDAI score5 Yes Yes / Yes‡ Yes
Disease activity over time (i.e., time-weighted SLEDAI)9,10,11 Yes Yes / No
No (not a baseline variable)
Corticosteroid use/dose5,8,9 Yes§ Yes / Yes†† Yes
Hydroxychloroquine/other antimalarial drug use5,8,9*
Baseline SDI9,12,13 Yes Yes / Yes Yes *TLC uses “antimalarial” drug variable but Petri et al. (2012) uses ‘hydroxychloroquine’ (specific antimalarial).
**TLC uses “immunosuppressive” drug variable but Petri et al. (2012) uses ‘cyclophosphamide’ (specific immunosuppressive).
† The SLE Diagnostic Check List is completed on entry to the cohort.
‡ SLEDAI score at baseline and at last visit of parent trial (BLISS 52/76) should be available.
†† Concomitant medications are comprehensively reported in the BEL112233 clinical study report. Statin use is a potential proxy for presence of dyslipidemia; however, it is inexact due to the use of statins for other medical conditions.
§ Every visit ‡‡ Annual test §§ Email from February 3, 2016.
3.2.4.2 Approach to be used to select predictors of organ damage to be included in final propensity score model
• The specific functional form used for the propensity score model does not have a profound impact on propensity score model performance, but following convention a logistic regression approach will be used.
• The model specification initially will include as independent variables all plausible (and available) predictors as listed in the table in section 3.2.4.1 as well as predictors recommended by external experts. In some cases, proxy measures may be used for unavailable predictors (e.g., geocode-level measures of population income or educational attainment)
• Drop the least statistically significant predictor from inclusion in propensity score model, then drop the next least statistically significant predictor, etc. until all included predictors have a p-value <0.1 (backward elimination)
• The specific predictors of organ damage to be included as covariates in the final model will be based on the model specification with the minimum Akaike information criterion (AIC) value, subject to modification as needed based on recommendations of external experts.
• Particular attention will be devoted to assessing the adequacy of the match for baseline SDI score (as likely the most important predictor of future organ damage), by comparing
PPD
17
the frequency distribution of baseline SDI scores for the treatment and comparison samples.
3.3 Data Source / Data Collection
3.3.1 TLC Data
The TLC is composed of patients seen at the who have agreed to participate in the cohort. Participants agree to a protocol that includes visits every 2 to 6 months regardless of disease activity.14 To insure standardization of assessments, training sessions are held for each new set of clinic providers.14
3.3.1.1 Instruments Completed on Entry to Cohort
The SLE Diagnostic Check List and 1000 Faces Cultural Background instruments are completed at entry to the cohort.15
3.3.1.2 Data Collected at Every Visit
The Lupus Protocol Version 6.0.1, a 560 question instrument, is completed at each visit. Questions are included on demographics, vitals, infections, cancer, lifestyle, organ systems (head and neck, retina, mucous membranes, respiratory, cardiac, vascular, gastrointestinal, reticuloendothelial, renal, fertility, skin, muculoskeletal, neurophychiatric, and endocrine), and therapies (NSAIDS, steroids, antimalarials, immunosuppressives, biologics, anticoagulants and antiplatelet).16 At each visit comprehensive sets of standard and lupus-related blood work are performed.
3.3.1.3 Data Collected Annually
Annually the SLICC Damage Index, SF36, Fatigue Severity Scale, and Family History instruments are completed. 15 Additional lupus-related bloodwork is performed on an annual basis.
3.3.1.4 Synchronization of Observations between TLC and LTE data
Per protocol, visits in the TLC are every 3 to 6 months with SDI, SF36, fatigue and family history collected on an annual basis. The timing of visits in the LTE trials varied between the parent study and extension phase. As shown in the figures of section 3.2.3, the baseline of the LTE study varies by treatment in the parent study. The baseline for patients receiving belimumab is the baseline of the parent study, while the baseline for patients receiving SoC is the baseline of the extension phase. The frequency of assessment in the parent study was generally every 4 weeks, except for SDI, which was assessed at 52 weeks and last visit. In the extension phase it was every 24 weeks, except for SDI and QoL, which were assessed every 48 weeks.
PPD
18
Data points will be synchronized at 6-month and annual intervals using a conservative algorithm to adjust for differences in assessment frequency. We recognize that bias can be introduced into time-to-event analyses by the frequency of the observations.
3.4 Endpoints
3.4.1 Primary Endpoint
The difference in change in SDI from baseline to year 5 interval between patients treated with belimumab or SoC, based on the data of the US BLISS LTE trial (BEL112233) and the TLC.
3.4.2 Secondary Endpoint(s)
The analysis of all secondary endpoints will use data from the US BLISS LTE trial (BEL112233) for patients treated with belimumab and data from the TLC for patients treated with SoC. The difference in time to first SDI worsening between patients treated with belimumab or SoC. The change from baseline SDI score by year interval for patients treated with belimumab or SoC. The difference in change from baseline SDI score by year interval between patients treated with belimumab or SoC. Transition analysis of SDI from baseline over a 5-year period for patients treated with belimumab or SoC. Change from baseline in SDI organ damage system (ocular, neuropsychiatric, renal, pulmonary, cardiovascular, peripheral vascular, gastrointestinal [GI], musculoskeletal, skin, premature gonadal failure, diabetes and malignancy) summarized by year interval for patients treated with belimumab or SoC. The frequency of increase from baseline in SDI organ damage system subscores between patients treated with belimumab or SoC. The difference in mean SLEDAI score from baseline over a 5-year period. The difference in cumulative corticosteroid usage from baseline over a 5-year period.
3.4.3 Exploratory Endpoint(s)
The analysis of all exploratory endpoints will use the pooled BLISS LTE dataset to be constructed from the BLISS LTE trials (BEL112233 and BEL112234) for patients treated with belimumab and data from the TLC for patients treated with SoC.
19
The difference in change in SDI from baseline to year 5 interval between patients treated with belimumab or SoC The difference in time to first SDI worsening between patients treated with belimumab or SoC. The change from baseline SDI score by year interval for patients treated with belimumab or SoC. The difference in change from baseline SDI score by year interval between patients treated with belimumab or SoC. Transition analysis of SDI from baseline over a 5 year period for patients treated with belimumab or SoC. Change from baseline in SDI organ damage system (ocular, neuropsychiatric, renal, pulmonary, cardiovascular, peripheral vascular, gastrointestinal [GI], musculoskeletal, skin, premature gonadal failure, diabetes and malignancy) summarized by year interval for patients treated with belimumab or SoC. The difference in cumulative corticosteroid usage from baseline over a 5-year period.
3.5 Sample Size / Power Calculations
3.5.1 Available sample sizes for BLISS LTE and TLC data after propensity score matching
Belimumab TLC Comparison
US LTE Sample (≥ 5 years Tx duration) with 2:1 Match
192 384
Pooled LTE Sample (≥5 years Tx duration) with 1:1 Match*
530 530
Tx = treatment. *TLC sample size is not sufficient for 2:1 match
• As reported in the TLC mortality and damage progression study, the maximum number of TLC patients available for matching who meet inclusion criteria and who have at least 5 years of post-index date data is 536, which is not sufficient for 2:1 matching for the pooled LTE sample.
• For the comparative effectiveness analysis of the US LTE sample only, 2:1 matching may be possible, but at most 192 belimumab patients and 384 propensity score-matched TLC patients would be available for use in the comparative effectiveness for the primary study endpoint.
20
3.5.2 Sample size requirements for “Mean Change from Baseline SDI” endpoint: Alternative
• Using the US-only LTE data for analysis of the 5-year Mean Change from Baseline SDI, the available sample size would be at most 192 belimumab patients and 384 propensity score-matched TLC patients (see 3.5.1), which would provide 80% power to detect a treatment effect size of a 0.155 unit difference (or 34% change) for a two-tailed test. [For a one-tailed test, a treatment effect size of a 0.138 unit difference (or 32%) could be detected.]
• Using the pooled LTE data for analysis of the 5-year Mean Change from Baseline SDI, under a “best case scenario” the available sample size would be 530 belimumab patients and 530 propensity score-matched TLC patients (see 3.5.1), which would provide 80% power to detect a treatment effect size of a 0.113 unit difference (or 27% change) for a two-tailed test. [For a one-tailed test, a treatment effect size of a 0.099 unit difference (or 25% change) could be detected.]
• However, given that there are only 536 potential comparison matches in the TLC data (see 3.2.2), it is likely that a suitable match for some of the 531 pooled LTE patients will not be found, especially given differences in patient demographics (e.g., race) across the geographies represented in the pooled LTE sample. If, for example, 395 of the pooled LTE sample (approximately 75%) are matched to 395 TLC patients, a treatment effect size of a 0.13 unit difference (or 30% change) could be detected with 80% power (two-tailed test). If only 297 of the pooled LTE sample (approximately 60%) can be matched, a treatment effect size of a 0.15 unit difference (33% change) could be detected with 80% power.
• Power for the pooled analysis will also be diminished to some degree by the need to include some form of statistical adjustment for differences in health system characteristics across the countries represented in the pooled LTE sample.
3.6 Hypotheses
21
• The change in SDI from baseline to year 5 will be the same for patients receiving belimumab as for patients receiving SoC.
• The time to first SDI worsening will be the same for patients receiving belimumab as for patients receiving SoC.
• The total SDI score at each year interval will be the same for patients receiving belimumab as for patients receiving SoC.
• The frequency of increase in SDI organ damage subscores will be the same for patients receiving belimumab as for patients receiving SoC.
4 DATA ANALYSIS CONSIDERATIONS
Subgroup analysis is not planned due to the limited power to detect statistical significance within subgroups.
4.1 Clinical Trial Datasets Patient characteristics for the belimumab patients will be extracted from the following analysis datasets from the BLISS LTEs (BEL112233 and BEL112234):
4.2 Determination of patient characteristics to be used to propensity score matching
• Patient characteristics shown to be potentially predictive of organ damage in the literature that are available in both the BLISS LTE and TLC data will considered for inclusion in the propensity score logistic regression model to be used for matching (see Table under 3.2.4.1)
• An initial model including all candidate predictors will be modified by sequentially dropping the least statistically significant predictor from inclusion in propensity score model, until all included predictors have a p-value <0.1 (backward elimination)
• The final model to be used for propensity score matching will be based on the model variant with the minimum Akaike information criterion (AIC) value.
22
• Final predictors of organ damage will be subject to validation by two experts in the treatment and research of SLE.
4.3 Assessment of suitability of final propensity score model for matching
• Determine range of common support for the logit index values (i.e., “X”) across belimumab and SoC patients
o If range of common support is too narrow, the propensity score model will be re-evaluated to determine if an alternative specification would improve common support
o Recommend refraining from pre-trimming patients off of the common support, given planned use of caliper matching, to preserve sample size.
• Examine propensity score model statistics (e.g., Hosmer–Lemeshow statistic, c-statistic, McFadden’s R2) to assure model has adequate predictive qualities.
4.4 Identify matches between BLISS LTE (treatment) patients and TLC (comparison) patients, using propensity score with a 2:1 match ratio (for US LTE data used for study)
• Matching will be based on the propensity score defined as the untransformed logit
index value (e.g., “X” for each individual from the logistic regression sample), rather than predicted probability, to increase variance at extremes of the propensity score distribution.
• The initial matching approach is to select for each treatment patient the “nearest neighbor” match among patients in the TLC comparison sample.
Matching for each treatment patient from the pool of TLC comparison patients proceeds (without replacement) until all treatment patients are matched
The process is then repeated to obtain the second comparison match for each treatment patient.
• The matching effort initially will employ a “rule of thumb” caliper (equal to 25% of the
standard deviation of the X distribution from the propensity score model) to try to eliminate the potential for excessively dissimilar “nearest” matches. As a result, some treatment patients may be dropped from the sample due to lack of adequate match.
• To focus on preserving sample size, the matching effort will be repeated with the caliper relaxed as needed to assure matches are attained for all treatment patients. This will likely decrease the extent of improvement in the balance for predictor variables resulting from propensity score matching. However, the caliper will not be relaxed to an extent that results in unacceptable post-match balance (see 4.5).
• Also will consider using inverse propensity score weighting approach as an alternative to matching to preserve sample size if needed (e.g., if achieving adequate balance via propensity score matching reduces the usable sample size to a degree that would jeopardize study power).
23
4.5 Assess degree of post-matching balance in predictors used in propensity score model across treatment and comparison groups
• The post-match “standardized bias” (SB) for each predictor included in the propensity
score model, calculated as
SB = 100 [(X̅T X̅C) / (0.5 {Var(XT) + Var(XC)}½], will be examined to assure that SB < 5% for all or most of the included predictors of organ damage.
• The trade-off between greater bias reduction (lower values of SB), associated with
fewer but more exact matches for treatment patients to comparison patients, versus
retention of sample size (associated with larger caliper value for matching) will be
assessed with the matching approach modified as needed.
• Given limited sample size, trade-off assessments will emphasize sample preservation
over precision in post-match balance, while maintaining adequate balance (e.g., the
SB for predictors of organ damage included in the final propensity score model must
be < 5% for all or most of the included predictors).
4.6 Index date for patients in the TLC
The index date for TLC patients will be the first point in their clinical record that their SLEDAI score reaches the SLEDAI inclusion criteria of the BLISS trials ( ≥ 6.0). An analysis will be made of the variability of SLEDAI scores in TLC and LTE patients to determine whether a single SLEDAI score ≥ 6 is sufficient to set the index date or that a sustained period ≥ 6 should be required. This analysis as well as clinical input will guide the final establishment of the index date criteria for TLC patients.
4.7 Withdrawals from BLISS LTE trials The population for time to event endpoints will include the MITT population of the BLISS LTE trial(s). The population for other endpoints will include only the populations available for analysis at that time point i.e., no imputation will be done to include patients who have withdrawn from the trial(s) by that point.
4.8 Analysis of propensity score matching variables
A subset of variables enumerated in 3.2.4.1 will be used to match patients from the BLISS LTE trial(s) to those in the TLC as described above. After matching has taken place, a comparison of each matching variable for belimumab patients to SoC patients will be performed. If a substantial difference is detected between treatment groups (belimumab and SoC) that variable will be added as a covariate to the statistical analysis of endpoints described below.
4.9 Rationale for inclusion of decade of study entry as a covariate
24
The decade of entry into the TLC is specified as a covariate in 4.11.1, 4.11.2.1, 4.11.2.3, 4.11.2.6, 4.11.2.7 and 4.11.2.8 below. The reason for doing so is that while the treatment of SLE has not changed greatly over the past decades and treatment in Canada and the US are comparable, the use of corticosteroids has abated in recent years as their role as a risk factor for organ damage has become more apparent. While it would be desirable to use the decade of entry as a propensity score matching variable, all LTE patients would be in a single decade, severely limiting the LTC patients that could be matched.
4.10 Descriptive Statistics Descriptive statistics will be performed on baseline demographic and clinical characteristics for both the belimumab-treated and SoC arms. This analysis will include a comparison of the two arms with p values.
4.11 Statistical analysis of endpoints All inferential statistics will be two-tail tests performed with an alpha of p=.05.
4.11.1 Primary endpoint
Change of SDI from baseline to censoring will be evaluated using linear regression with change of SDI from baseline as the dependent variable, and with a variable indicating treatment group (belimumab or SoC) and matching variable(s) determined in 4.8 above as covariates. The decade of entry into the study will also be a covariate.
4.11.2 Secondary endpoints
4.11.2.1 Difference in time to first SDI worsening
Cox proportional hazards regression will be used to estimate the hazard ratio between belimumab and SoC and its statistical significance. If substantial differences are found among matching variables in 4.8 above, they will be added as covariate(s). The decade of entry into the study will also be a covariate.
4.11.2.2 Change from baseline SDI score by year interval
Descriptive statistics of the change from baseline SDI score will be estimated at the end of years 1 through 5 for both the belimumab and SoC groups.
4.11.2.3 Difference of change from baseline SDI by year interval
Change of SDI from baseline to end of years 1 through 5 will be evaluated using linear regression with change of SDI from baseline as the dependent variable, and with a variable indicating treatment group (belimumab or SoC) and matching variable(s) determined in 4.8 above as covariates. The decade of entry into the study will also be a covariate.
25
4.11.2.4 Transition analysis of SDI
Multi-state Markov modelling transition analysis of SDI will be performed independently for the belimumab and SoC groups using the Jackson et al. 2011 methodology.2,3,4 This methodology calculates transition probabilities between health states over time, in this case health states defined by SDI strata.
4.11.2.5 Change from baseline of SDI organ damage system subscores
Descriptive statistics of the change from baseline SDI organ system subscores will be estimated at the end of years 1 through 5 for the belimumab and SoC groups.
4.11.2.6 Frequency of increase from baseline of SDI organ damage system subscores
Frequency of increase of SDI organ system subscores from baseline to censoring between patients treated with belimumab or SoC will be evaluated using logistic regression with a variable indicating treatment group (belimumab or SoC) as the dependent variable, and with the change of SDI organ system subscore from baseline and matching variables determined in 4.8 above as covariates. The decade of entry into the study will also be a covariate.
4.11.2.7 Mean SLEDAI score
Mean SLEDAI score from baseline through year 5 will be evaluated using linear regression with mean SLEDAI score as the dependent variable and with a variable indicating treatment group (belimumab or SoC) and matching variable(s) determined in 4.8 above as covariates. The decade of entry into the study will also be a covariate.
4.11.2.8 Cumulative corticosteroid usage
Cumulative use of corticosteroids from baseline through year 5 will be evaluated using linear regression with cumulative corticosteroid use as the dependent variable and with a variable indicating treatment group (belimumab or SoC) and matching variable(s) determined in 4.8 above as covariates.
4.11.3 Exploratory endpoints
The methods described above will also be used for the corresponding exploratory endpoints. The following exploratory analyses will be conducted separately for the LTE and TLC cohorts: study completers will be compared to non-completers for a) baseline subject characteristics and b) SDI scores by follow-up year (ie., SDI in year 1 for the subjects that drop out in year one, subjects that complete year 1 but drop out later and subjects that complete the full 5 years; the same comparison for year 2 and so on). The following exploratory analyses will be conducted separately for the LTE and TLC cohorts and for the two cohorts combined: a time-to-event analysis, such a Cox proportional hazards model, will be employed to estimate the relative risk of a series of endpoints: time to SDI
26
increase, varying from 1 point to 3 points, as well as time to a specific absolute SDI score (scores to be determined), with baseline SDI being a key covariate.
4.12 Diagnostics
4.12.1 Baseline characteristics of study arms
Study arms will be tested for statistically significant differences in patient baseline characteristics using t tests and Fisher’s exact tests, as appropriate.
4.12.2 Baseline characteristics of sample versus population
Matched samples will be tested for statistically significant differences in patient baseline characteristics using t tests and Fisher’s exact tests, as appropriate.
4.12.3 Distribution of year 5 data point timing
Patients in the TLC are not seen at specific intervals. Likewise, patients originating in belimumab and SoC arms of the underlying belimumab trials have different intervals of observation. Therefore the 5th year observation in both arms will take place at time points not strictly 5 years from baseline. The distributions of time from baseline of the 5th year observation will be reported.
4.12.4 Patients withdrawing for LTE and TLC Cohorts
Analysis will be conducted that includes all study participants in both cohorts, i.e. subjects that complete the full five years of follow-up as well as subjects that drop out before study end. The impact of the dropout rates will be assessed by comparing those who completed the study versus those who did not complete the study in terms of baseline and clinical characteristics. Time to event analyses will be used to test for differences in clinical outcomes including SRI response, flares, and increase in SLICC Damage Index.
4.12.5 BLISS subjects that did not enroll in LTE
Analysis will be conducted to detect selection bias among the BLISS subjects who enrolled in an LTE study versus BLISS subjects who did not proceed to enroll in an LTE. This analysis will compare LTE subjects with subjects in the corresponding parent study who either discontinued the parent study before completion, or completed the parent study but did not enroll in the LTE. The potential presence of selection bias will be assessed in terms of baseline and clinical characteristics, and time to event analyses will be used to test for differences in clinical outcomes during the parent study.
27
4.12.6 BLISS LTE subjects randomized to SoC
Analysis will be conducted to test for study effects associated with differences in quality of care obtained within the setting of the randomized clinical trials (see section 3.2.3). These analyses will be performed by comparing BLISS subjects randomized to SoC against TLC patients to determine whether enrollment in the randomized clinical trial had a significant effect on clinical outcomes associated with SoC. Propensity score matching will be used to match BLISS SoC subjects to TLC patients based on the BLISS subjects’ characteristics at core baseline, e.g., at randomization to SoC + placebo in BLISS 76 (BEL110751). Time to event analyses will be used to test for differences in clinical outcomes during the parent study compared with clinical outcomes in the TLC over a follow-up period equal to the duration of the parent study. This analysis will be conducted as early in the process as possible and assessed for statistical significance. If it is statistically significant, it will be assessed for clinical significance. If both statistically and clinically significant, methods will be adopted to mitigate this bias in further analyses , e.g., by incorporating fixed effects or stratification into the statistical models, or by performing subgroup analyses of BLISS LTE subjects by randomized study medication in the parent study.
4.12.7 Belimumab baseline of BLISS LTE subjects
Analysis will be conducted to test for potential biases introduced by the method used to define belimumab baseline for the propensity score matching procedure (section 3.2.3). This analysis will focus on tests of equivalence among the subgroups of BLISS LTE subjects randomized to belimumab 10 mg/kg in the parent study, BLISS LTE subjects randomized to belimumab 1 mg/kg in the parent study, and BLISS LTE subjects randomized to placebo in the parent study. Time to event analyses will be used to test for differences in clinical outcomes during the follow-up period of the comparative effectiveness analysis.
5 LIMITATIONS
The primary limitations are the numbers of patients in the BLISS LTE trials, the number of patients in the TLC, and the number of patients with matching characteristics. This limits the power to reach statistically significant conclusions. Other limitations include:
• Dissimilar data collection cycles, i.e. data in the BLISS LTEs were collected on a pre-defined schedule, while data in the TLC are collected when patients schedule visits.
• The controls will not have been randomly selected from the same population as the BLISS LTE patients. Instead propensity score matching will be used to match belimumab patients with controls based on a number of patient characteristics.
• BLISS patients may have received care from different types of health care systems than external cohort controls.
• SELENA-SLEDAI scores are available for BLISS LTE patients while SLEDAI-2K scores are available for TLC patients. While the components of SELENA-SLEDAI remain the same as the SLEDAI-2K, the definitions of several components are slightly different.17
28
The length of assessment was also different in the BLISS trials than in the TLC; up to 10 days prior to a visit in the BLISS trials while up to 30 days prior to a visit in the TLC.18,19 A study has indicated that there is minimal difference between 10 and 30 days assessment.20 The SELENA-SLEDAI is yet to be rigorously validated.21
• Patients were free to withdraw from the BLISS studies and from the TLC. This may introduce bias.
• The sensitivity analysis will utilize pooled patient-level data from both BLISS LTE trials (BEL112233 and BEL112234), treating the data as through from one, rather than two, trials. In fact the populations and health systems through which care was given may have been significantly different. The preferred method of aggregating data from multiple trials is meta-analysis. However, using meta-analysis for just two trials would result in loss of power. The pooled dataset will be constructed de novo from the datasets of the BLISS LTE trials (BEL112233 and BEL112234) as part of this analysis.
• The BLISS trials used 48 weeks as a year.
• The PS matching approach to be used only accounts for sample selection bias (aka confounding by indication) for the observed confounders (predictors of organ damage also potentially affecting treatment assignment) included in the PS model. To the extent additional clinically important confounders exist in the data but cannot be observed, sample selection bias cannot be fully addressed using PS adjustment methods. However, it is likely that in this study that differential access to belimumab across the TLC and LTE cohorts is likely to be the primary factor driving differences in treatment received for clinically similar patients, not unobserved confounders.
• The TLC is treated at a single Canadian clinical site while the BLISS BEL112233 LTE trial was conducted at multiple US sites. Differences in outcomes may be confounded by differences in the national health systems. Outcomes may also be confounded by treatment practices specific to the TLC clinical site.
6 STUDY CONDUCT, MANAGEMENT & ETHICS
6.1 Ethics Committee/IRB Approval IRB approval will be sought from the University of Toronto IRB.
6.2 Informed Consent Consents have previously been obtained for use of the data. No new data will be collected.
6.3 Data Protection Data have been de-identified according to HIPAA standards.
6.4 Personally Identifiable Information (PII) Data have been de-identified according to HIPAA standards.
29
6.5 Adverse Event (AE), Pregnancy Exposure, and Incident Reporting
Not applicable.
7 EXTERNAL INVOLVEMENT
7.1 Third Party Supplier Data will be analyzed by: Medical Decision Modeling Inc. 201 N. Illinois St., Ste. 1730 Indianapolis IN 46204
Data will be provided by:
Dr.
7.2 External Expert/Health Care Professionals (Consultants & Research PIs)
Dr. Consultant Dr. Consultant
PPD
PPD
PPD
PPD
PPD
PPD
PPD
PPD
PPD
PPD
30
REFERENCES 1. Medical Decision Modeling Inc. Research of Lupus Cohorts as Comparator Arm in
Belimumab Subcutaneous Cost Effectiveness Model - Internal Document. January 2016.
2. Bruce IN, O’Keeffe AG, Farewell V, et al. Factors associated with damage accrual in patients with systemic lupus erythematosus: results from the Systemic Lupus International Collaborating Clinics (SLICC) Inception Cohort. Ann Rheum Dis. 2015;74(9):1706-1713. doi:10.1136/annrheumdis-2013-205171.
3. Jackson C. Multi-state models for panel data: the msm package for R. J Stat Softw. 2011;38(8). https://www.jstatsoft.org/article/view/v038i08. Accessed February 24, 2016.
4. Jackson C. Multi-state modelling with R: the msm package. November 2015. https://cran.r-project.org/web/packages/msm/vignettes/msm-manual.pdf. Accessed February 24, 2016.
5. Impact of disease activity on mortality and damage progression in SLE patients with active disease despite standard of care using data from the Toronto Lupus Cohort - Preliminary Report. September 2014.
6. et al. A Multi-Center, Continuation Trial of Belimumab (HGS1006, LymphoStat-B), a Fully Human Monoclonal Anti-BLyS Antibody, in Subjects with Systemic Lupus Erythematosus (SLE) Who Completed the Phase 3 Protocol HGS1006-C1056 in the United States. December 2015.
7. Results Report for 201223, a Pooled Analysis of BEL112233 and BEL112234, Two Long-term, Extension Studies of BEL110751 and BEL110752, to Investigate Long-term Safety and Organ Damage Accrual with Belimumab Treatment in Systemic Lupus Erythematosus - GSK Internal Report. November 2015.
8. Petri M, Purvey S, Fang H, Magder LS. Predictors of organ damage in systemic lupus erythematosus: the Hopkins Lupus Cohort. Arthritis Rheum. 2012;64(12):4021-4028. doi:10.1002/art.34672.
9. Sutton EJ, Davidson JE, Bruce IN. The systemic lupus international collaborating clinics (SLICC) damage index: a systematic literature review. Semin Arthritis Rheum. 2013;43(3):352-361. doi:10.1016/j.semarthrit.2013.05.003.
10. Becker-Merok A, Nossent HC. Damage accumulation in systemic lupus erythematosus and its relation to disease activity and mortality. J Rheumatol. 2006;33(8):1570-1577.
11. Stoll T, Sutcliffe N, Mach J, Klaghofer R, Isenberg DA. Analysis of the relationship between disease activity and damage in patients with systemic lupus erythematosus--a 5-yr prospective study. Rheumatol Oxf Engl. 2004;43(8):1039-1044. doi:10.1093/rheumatology/keh238.
PPD
PPD
PPD
31
12. Cardoso CRL, Signorelli FV, Papi J a. S, Salles GF. Initial and accrued damage as predictors of mortality in Brazilian patients with systemic lupus erythematosus: a cohort study. Lupus. 2008;17(11):1042-1048. doi:10.1177/0961203308093829.
13. Alarcón GS, Friedman AW, Straaton KV, et al. Systemic lupus erythematosus in three ethnic groups: III. A comparison of characteristics early in the natural history of the LUMINA cohort. LUpus in MInority populations: NAture vs. Nurture. Lupus. 1999;8(3):197-209.
14. Urowitz MB, Gladman DD. Contributions of observational cohort studies in systemic lupus erythematosus: the University of Toronto Lupus Clinic experience. Rheum Dis Clin North Am. 2005;31(2):211-221, v. doi:10.1016/j.rdc.2005.01.008.
15. Toronto Lupus Cohort. Lupus Database Research Program: Data collection forms.
16. The Lupus Clinic, Centre for Prognosis Studies in the Rheumatic Diseases. Lupus Protocol Version 6.0.1. 2012.
17. Touma Z. Derivation of an Appropriate Outcome Measure in Lupus. 2012. https://www.researchgate.net/publication/256443973_Derivation_of_an_Appropriate_Outcome_Measure_in_Lupus. Accessed March 11, 2016.
18. GlaxoSmithKline. BENLYSTA (belimumab). GSKsource. https://www.gsksource.com/pharma/content/gsk/source/us/en/brands/benlysta/pi/primary-endpoint.html. Accessed March 11, 2016.
19. SLEDAI-2K length of assessment in Toronto Lupus Cohort - personal communication. March 2016.
20. Touma Z, Urowitz MB, Gladman DD. SLEDAI-2K for a 30-day window. Lupus. 2010;19(1):49-51. doi:10.1177/0961203309346505.
21. Mikdashi J, Nived O. Measuring disease activity in adults with systemic lupus erythematosus: the challenges of administrative burden and responsiveness to patient concerns in clinical research. Arthritis Res Ther. 2015;17:183. doi:10.1186/s13075-015-0702-6.