Introduction Methodology Simulations Application Conclusion Robust Methods for Health-related Quality-of-life Assessment Ian McCarthy Baylor Scott & White Health Center for Clinical Effectiveness Utah Health Services Research Conference April 30, 2014 This project was supported by grant number K99HS022431 from the Agency for Healthcare Research and Quality. The content is solely the responsibility of the author and does not necessarily represent the official views of the Agency for Healthcare Research and Quality. Robust Methods for Health-related Quality-of-life Assessment
24
Embed
Robust Methods for Health-related Quality-of-life Assessment
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
IntroductionMethodology
SimulationsApplicationConclusion
Robust Methods for Health-relatedQuality-of-life Assessment
Ian McCarthy
Baylor Scott & White HealthCenter for Clinical Effectiveness
Utah Health Services Research ConferenceApril 30, 2014
This project was supported by grant number K99HS022431 from the Agency for Healthcare Research andQuality. The content is solely the responsibility of the author and does not necessarily represent the official
views of the Agency for Healthcare Research and Quality.
Robust Methods for Health-related Quality-of-life Assessment
IntroductionMethodology
SimulationsApplicationConclusion
Background
Cost- and comparative-effectiveness studies becomingincreasingly important
Require assessment of health-related quality-of-life (HRQoL)outcomes and quality-adjusted life-years (QALYs)
Common approach first collapses the multi-dimensionalHRQoL profile into a one-dimensional QALY (Drummondet al., 2005; Brazier et al., 2002; Brazier & Ratcliffe, 2007)
EQ-5DSF-6DHUI
Robust Methods for Health-related Quality-of-life Assessment
IntroductionMethodology
SimulationsApplicationConclusion
Problem
Loss of information when reducing HRQoL profile into QALY, withpotentially biased and inconsistent marginal effects estimates(Mortimer & Segal, 2008; Devlin et al., 2010; Parkin et al., 2010;Gutacker et al., 2012):
1 Floor and ceiling effects not present in the underlying domainsbut imposed by the scoring algorithm.
2 Nonlinearities in the relationship between the outcome andindependent variables which are difficult to approximate usingthe summary score.
Robust Methods for Health-related Quality-of-life Assessment
IntroductionMethodology
SimulationsApplicationConclusion
Current Study
1 Monte Carlo study showing the bias of the estimatedcoefficients when relying solely on QALYs or some othersummary score based on several ordered outcome variables.
2 Propose new two-step methodology that first estimatescoefficients in each HRQoL domain and then transforms thecoefficients and marginal effects into the QALY domain basedon predicted values from the first-stage regressions.
Robust Methods for Health-related Quality-of-life Assessment
IntroductionMethodology
SimulationsApplicationConclusion
Estimating QALYsMarginal Effects: Standard ApproachMarginal Effects: Proposed Methodology
The SF-6D
Developed by John Brazier and other, the SF-6D is formed from asubset of questions from the SF-36 or SF-12 and is a commonHRQoL outcome intended to provide a general measure of apatient’s health status (Brazier et al., 2002; Brazier & Ratcliffe,2007).
Six dimensions/domains of health: (Physical functioning, rolelimitations, social functioning, pain, mental health, andvitality)
Each domain characterized numerically with a range ofintegers. Best value is 1, and worst value ranges from 4 to 6.
Scoring algorithm developed in Brazier et al. (2002) andBrazier & Ratcliffe (2007) for calculating a population-basedindex score from the SF-6D questionnaire
Robust Methods for Health-related Quality-of-life Assessment
IntroductionMethodology
SimulationsApplicationConclusion
Estimating QALYsMarginal Effects: Standard ApproachMarginal Effects: Proposed Methodology
Scoring the SF-6D
Physical Functioning (PF)PF=2 or PF=3 -0.035PF=4 -0.044PF=5 -0.056PF=6 -0.117Role Limitations (RL)RL=2 or RL=3 or RL=4 -0.053Social Functioning (SF)SF=2 -0.057SF=3 -0.059SF=4 -0.072SF=5 -0.087Pain (P)P=2 or P=3 -0.042P=4 -0.065P=5 -0.102P=6 -0.171Mental Health (MH)MH=2 or MH=3 -0.042MH=4 -0.100MH=5 -0.118Vitality (V)V=2 or V=3 or V=4 -0.071V=5 -0.092Combination of Domains“Most Severe” -0.061
Robust Methods for Health-related Quality-of-life Assessment
IntroductionMethodology
SimulationsApplicationConclusion
Estimating QALYsMarginal Effects: Standard ApproachMarginal Effects: Proposed Methodology
Focus on QALYs
By far the most common methodology for estimatingcoefficients and ultimately marginal effects is to first reducethe multi-dimensional health profile to a one-dimensionalQALY (Austin et al., 2000; Austin, 2002; Richardson &Manca, 2004; Manca et al., 2005; Basu & Manca, 2012)
Recent literature on how best to accommodate distributionalfeatures somewhat specific to QALYs (Austin, 2002; Basu &Manca, 2012), including a censored least absolute deviationmodel and a Beta MLE approach
Robust Methods for Health-related Quality-of-life Assessment
IntroductionMethodology
SimulationsApplicationConclusion
Estimating QALYsMarginal Effects: Standard ApproachMarginal Effects: Proposed Methodology
First Stage Regression
1 Estimate an ordered probit model separately for each domain,d = 1, ..., 6, with the follow-up HRQoL response (yid ,t1)modeled as a function of person-specific variables (xi ),baseline HRQoL response (yid ,t0), and treatment status (Ti ).
2 Form predicted probabilities of every possible response, j , ineach domain, d , denoted p̂d
j .
The regression results provide a predicted (marginal) probability foreach of 31 possible outcomes for each person.
Robust Methods for Health-related Quality-of-life Assessment
IntroductionMethodology
SimulationsApplicationConclusion
Estimating QALYsMarginal Effects: Standard ApproachMarginal Effects: Proposed Methodology
“Most Severe” Category
1 Defined as any one of the following (Brazier et al., 2002): 4or more in the physical functioning, social functioning, mentalhealth, or vitality domains; 3 or more in the role limitationdomain; or 5 or more in the pain domain
2 Since the probabilities, Pdij , are potentially correlated across
domains, the probability of a “most severe” health status canbe calculated following the principle of inclusion and exclusionfor probability:
P (A1 ∪ A2 ∪ ... ∪ AN) = P (A1) + ...+ P (AN) +
N∑n=2
(−1)n+1P (∩ n events) .
Robust Methods for Health-related Quality-of-life Assessment
IntroductionMethodology
SimulationsApplicationConclusion
Estimating QALYsMarginal Effects: Standard ApproachMarginal Effects: Proposed Methodology
Estimate QALYs
Q̂ALY i = 1− 0.035×(
P̂PFi2 + P̂PF
i3
)− 0.044× P̂PF
i4 − 0.056× P̂PFi5 − 0.117× P̂PF
i6
− 0.053×(
P̂RLi2 + P̂RL
i3 + P̂RLi4
)− 0.057× P̂SF
i2 − 0.059× P̂SFi3 − 0.072× P̂SF
i4 − 0.087× P̂SFi5
− 0.042×(
P̂Paini2 + P̂Pain
i3
)− 0.065× P̂Pain
i4 − 0.102× P̂Paini5 − 0.171× P̂Pain
i6
− 0.042×(
P̂MHi2 + P̂MH
i3
)− 0.100× P̂MH
i4 − 0.118× P̂MHi5
− 0.071×(
P̂Vi2 + P̂V
i3 + P̂Vi4
)− 0.092× P̂V
i5
− 0.061× P̂ (Most Severe) .
Robust Methods for Health-related Quality-of-life Assessment
IntroductionMethodology
SimulationsApplicationConclusion
Marginal Effects on QALYsTreatment Effects with Selection
Data Generating Processes
The D × 1 vector of latent HRQoL values, y∗i , is simulated asfollows:
y∗i = γ + βx ′i + εi , where
ε ∼ N (0D×1, ID×D) ,
x ∼ U[0, 1],
γ = ID×1, and
β = 1.5× ID×1.
Discrete HRQoL values are generated based on the value of thelatent value, y∗id , relative to the Jd × 1 vector of threshold values ineach domain.
Robust Methods for Health-related Quality-of-life Assessment
IntroductionMethodology
SimulationsApplicationConclusion
Marginal Effects on QALYsTreatment Effects with Selection
Simulated QALY Distributions
010
2030
4050
Fre
quen
cy
.4 .6 .8 1SF-6D Index Score
010
2030
40F
requ
ency
.2 .4 .6 .8 1SF-6D Index Score
010
2030
4050
Fre
quen
cy
.4 .6 .8 1SF-6D Index Score
020
4060
80F
requ
ency
.3 .4 .5 .6 .7 .8SF-6D Index Score
050
100
150
200
Fre
quen
cy
.4 .6 .8 1SF-6D Index Score
Robust Methods for Health-related Quality-of-life Assessment
IntroductionMethodology
SimulationsApplicationConclusion
Marginal Effects on QALYsTreatment Effects with Selection
Monte Carlo Results
Model Incremental Effect St. Dev. Mean % Bias Lower % Bias Upper % Bias RMSE
ATE on QALY 0.033*** 0.038*** 0.032*** 0.029***(0.011) (0.011) (0.011) (0.010)
RMSE 0.098 0.111 0.098 0.097
Robust Methods for Health-related Quality-of-life Assessment
IntroductionMethodology
SimulationsApplicationConclusion
Intuition
Collapsing multi-dimensional profile into a single summarymeasure introduces floor/ceiling effects and nonlinearities thatare difficult to accommodate in a single equation framework.
With selection into treatment (whether on observables orunobservables), standard methods relying only on QALYsprovide biased estimates of true treatment effect.
An alternative approach is to estimate coefficients based onthe full health profile and then re-interpret effects in theQALY domain based on predicted probabilities in thefirst-stage regressions.
Robust Methods for Health-related Quality-of-life Assessment
IntroductionMethodology
SimulationsApplicationConclusion
Thank You
Robust Methods for Health-related Quality-of-life Assessment
IntroductionMethodology
SimulationsApplicationConclusion
Bibliography I
Austin, P.C. 2002. A comparison of methods for analyzing health-related quality-of-life measures. Value in Health,5(4), 329–337.
Austin, P.C., Escobar, M., & Kopec, J.A. 2000. The use of the Tobit model for analyzing measures of healthstatus. Quality of Life Research, 9(8), 901–910.
Basu, A., & Manca, A. 2012. Regression Estimators for Generic Health-Related Quality of Life andQuality-Adjusted Life Years. Medical Decision Making, 32(1), 56–69.
Brazier, J., & Ratcliffe, J. 2007. Measuring and valuing health benefits for economic evaluation. Oxford UniversityPress, USA.
Brazier, J., Roberts, J., & Deverill, M. 2002. The estimation of a preference-based measure of health from theSF-36. Journal of health economics, 21(2), 271–292.
Devlin, N.J., Parkin, D., & Browne, J. 2010. Patient-reported outcome measures in the NHS: new methods foranalysing and reporting EQ-5D data. Health economics, 19(8), 886–905.
Drummond, M.F., Sculpher, M.J., & Torrance, G.W. 2005. Methods for the economic evaluation of health careprogrammes. Oxford University Press, USA.
Gutacker, N., Bojke, C., Daidone, S., Devlin, N., & Street, A. 2012. Analysing Hospital Variation in HealthOutcome at the Level of EQ-5D Dimensions.
Manca, A., Hawkins, N., & Sculpher, M.J. 2005. Estimating mean QALYs in trial-based cost-effectiveness analysis:the importance of controlling for baseline utility. Health economics, 14(5), 487–496.
Mortimer, D., & Segal, L. 2008. Comparing the incomparable? A systematic review of competing techniques forconverting descriptive measures of health status into QALY-weights. Medical decision making, 28(1), 66.
Parkin, D., Rice, N., & Devlin, N. 2010. Statistical analysis of EQ-5D profiles: does the use of value sets biasinference? Medical Decision Making, 30(5), 556–565.
Richardson, G., & Manca, A. 2004. Calculation of quality adjusted life years in the published literature: a review ofmethodology and transparency. Health economics, 13(12), 1203–1210.
Robust Methods for Health-related Quality-of-life Assessment