Health-related quality of life in HIV1-infected patients on HAART: a five-years longitudinal analysis accounting for dropout in the APROCOCOPILOTE cohort (ANRS CO8

Health-related quality of life in HIV-1-infected patients onHAART: a five-years longitudinal analysis accounting fordropout in the APROCO-COPILOTE cohort (ANRS CO-8)

Camelia Protopopescu Æ Fabienne Marcellin Æ Bruno Spire Æ Marie Preau ÆRenaud Verdon Æ Dominique Peyramond Æ Francois Raffi Æ Genevieve Chene ÆCatherine Leport Æ Maria-Patrizia Carrieri

Received: 18 April 2006 / Accepted: 21 November 2006 / Published online: 1 February 2007� Springer Science+Business Media B.V. 2007

Abstract

Background The long-term efficacy of Highly Active

Antiretroviral Therapies (HAART) has enlightened

the crucial role of health-related quality of life

(HRQL) among HIV-infected patients. However, any

analysis of such extensive longitudinal data necessi-

tates a suitable handling of dropout which may corre-

late with patients’ health status.

Methods We analysed the HRQL evolution over

5 years for 1,000 patients initiating a protease inhibitor

(PI)-containing therapy, using MOS SF-36 physical

(PCS) and mental (MCS) scores. In parallel with a

classical separate random effects model, we used a

joint parameter-dependent selection model to account

for non-ignorable dropout.

Results HRQL evolved according to a two-phase

pattern, characterized by an initial improvement

during the year following HAART initiation and a

relative stabilization thereafter. Immunodepression

and self-reported side effects were found to be nega-

tive predictors of both PCS and MCS scores. Hepatitis

C virus coinfection and AIDS clinical stage were found

to affect physical HRQL. Results were not significantly

altered when accounting for dropout.

Conclusion Such results, obtained on a large sample

of HIV-infected patients with extensive follow-up,

underline the need for a regular monitoring of patients’

immunological status and for a better management of

their experience with hepatitis C and HAART.

Keywords HIV � HAART � Quality of life �Hepatitis C � Dropout

Introduction

The introduction of Highly Active Antiretroviral

Therapies (HAART) has led to a progressive decline

in HIV-related mortality [1]. With HIV infection

considered as a chronic illness, health-related quality

of life (HRQL) has become an important marker in

the follow-up of HIV-infected patients. In this context,

our study aimed at analysing HRQL long-term evo-

lution after initiation of a protease inhibitor (PI)-con-

taining therapy and the clinical and sociobehavioral

factors associated to HRQL changes. To that end, we

C. Protopopescu � F. Marcellin � B. Spire �M. Preau � M.-P. CarrieriHealth and Medical Research National Institute(INSERM), Research Unit 379, Social Sciences Applied toMedical Innovation, 23, rue Stanislas Torrents, 13006Marseilles, France

C. Protopopescu (&) � F. Marcellin � B. Spire �M. Preau � M.-P. CarrieriSoutheastern Health Regional Observatory (ORS-PACA),23, rue Stanislas Torrents, 13006 Marseilles, Francee-mail: [email protected]

R. VerdonCHU de la Cote de Nacre, Caen, France

D. PeyramondCroix-Rousse Hospital, Lyon, France

F. RaffiHotel-Dieu, Nantes, France

G. CheneINSERM Research Unit 330, Bordeaux, France

C. LeportFaculte Xavier Bichat, Paris, France

123

Qual Life Res (2007) 16:577–591

DOI 10.1007/s11136-006-9151-7

used data collected among HIV-infected patients from

a French cohort study (APROCO-COPILOTE -

ANRS CO8) during 5 years after enrolment. However,

such longitudinal data always present missing assess-

ments due to patients’ temporary (missed clinical visit

or non-response to a questionnaire) or permanent

withdrawal from the study. The probability of with-

drawal may depend on HRQL itself, as for instance,

patients with low HRQL may be more likely to leave

the study than the others. In this case, the underlying

dropout process is informative or non ignorable, fol-

lowing the terminology of Little et al. [2], and valid

results for longitudinal inference can be obtained only

if this process is explicitly modelled [3]. Different

methods have been developed to address this issue

[4–7]. Guo and Carlin proposed a joint modelling of

longitudinal and event time data using Bayesian

methods [8]. We adapted this modelling approach to

the analysis of our study data, then we compared the

obtained results with those of the standard approach,

which does not take into account the possibly

non-ignorable dropout.

Methods

Study population

The French APROCO-COPILOTE (ANRS CO-8)

multicenter cohort study, setup in 1997, aimed at

describing clinical, immunological, virological and

socio-behavioral characteristics of HIV-1-infected

patients who were beginning combination antiretrovi-

ral therapy (HAART) that included a PI [9]. Entry in

the cohort (Month 0, M0) corresponded to the begin-

ning of combination therapy. Along with clinical and

socio-demographic data collection, which have been

extensively described elsewhere [9, 10], follow-up of

patients included longitudinal assessment of HRQL

using the Medical Outcome Study 36-item Short Form

Health Survey (MOS SF-36) [11–13] validated in

French [14, 15]. This self-administered questionnaire is

designed to analyse eight health concepts, each related

to a physical or mental dimension of HRQL. The

corresponding scores, converted into a scale ranging

from 0 to 100, can be combined to obtain two aggre-

gate scores: the physical component summary (PCS)

and the mental component summary (MCS), with high

scores corresponding to better HRQL. Patients were

asked to complete the MOS SF-36 questionnaire at

baseline (M0), and a total of six other measures were

collected during the follow-up: at 12 months (M12), at

28 months (M28), then every 8 months thereafter until

Month 60 (M60).

Statistical methods

The longitudinal evolution between M0 and M60 of the

PCS and MCS scores was analysed using linear models.

Scores were log-transformed (–ln(100 – PCS) and

–ln(100 – MCS)) in order to satisfy normality and het-

eroskedasticity assumptions. Factors associated with

physical and mental HRQL evolution were explored

among patients’ sociodemographic, behavioral and

HIV-related characteristics: age, gender, geographic

origin, stability and comfort of housing conditions,

presence of a stable partner, daily alcohol consumption,

HIV transmission group (homosexuals, intravenous

drug users IDU, others), antiretroviral naivety, CD4

count (<200 cells/mm3), number of self-reported

symptoms of lipodystrophy, number of self-reported

treatments’ side-effects excluding lipodystrophy.

In a first approach (separate modelling), HRQL

scores evolution was analysed using a random effects

or mixed model, where all missing data are considered

missing at random (MAR), cf. Little et al. [2]. As

descriptive analyses clearly showed a two-phase evo-

lution in HRQL scores with a change in the slope

1 year after initiation of combination therapy (Fig. 1a

and 1b), a piecewise linear model with a change point

at M12 was chosen for this analysis. Each factor con-

tribution to the model was made of three covariates: a

first one representing the principal effect of the factor,

and two interaction-with-time variables (slopes) rep-

resenting respectively the changes in the short-time

effect of the factor (M0–M12), and the changes in the

effect of the factor during the maintenance phase of

the treatment (M12–M60). For example, a positive

principal effect followed by a negative first slope,

smaller in absolute value, and a non-significant second

slope means that the corresponding factor has a global

positive effect on the response variable during the

study period, but this effect decreases during the first

year, then does not change significatively after M12.

The effect of each factor on patients’ HRQL was first

tested in univariate analyses, and the factors which

achieved a significance of P < .2 were considered eli-

gible for the multivariate model. The two interactions

with time were included in the multivariate model if at

least one of them was significant in the univariate

analysis (P < .2). The final multivariate model was

determined using likelihood ratio tests to validate the

number of random effects used and backward analyses

with P < .2 threshold for the selection of factors.

578 Qual Life Res (2007) 16:577–591

123

Patients drop out at a given visit if no HRQL

assessment is available from this visit onward (until

M60). This way, patients are defined as ‘‘drop out’’ for

HRQL if the last available measure occurs before M60

(censoring time). A discrete analogue of the continu-

ous proportional hazards (Cox) model [16] was used to

analyse separately the interval-censored time-to-drop-

out data. In this model, the effect of each factor can be

interpreted as in the classical Cox model.

In a second approach (joint modelling), HRQL

scores and time-to-dropout were simultaneously mod-

elled using the shared-parameter selection type-model

proposed by Henderson et al. [17] and Guo and Carlin

[8]. In this joint approach, the possible correlation

between the longitudinal and the time-to-dropout

models is modelled through two latent random vari-

ables and the common explanatory variables. Missing

data due to dropout are considered as non-ignorable,

while intermittent missing data are considered MAR.

All the analyses were performed using log-transformed

scores. In addition, each factor contribution to HRQL

between M0 and M12 was given for the untransformed

scores after translation back onto their original scale

(see Appendix). All the analyses were performed using

SAS software, version 9.1 for Windows [18]. A detailed

description of the statistical models and the SAS codes

are supplied in the Appendix.

Results

Description of dataset

Of the 1,281 patients who were enrolled in the AP-

ROCO-COPILOTE cohort (ANRS CO-8), 1,000 pa-

tients had completed the MOS SF-36 questionnaire at

M0, and were thus included in the analysis. There were

no significant differences in gender and viral load at

baseline between these patients and the 281 non-

respondents to the questionnaire. Respondents in-

cluded significantly higher proportions of young pa-

tients, patients infected through homosexual contacts,

patients not at AIDS-stage, and patients with low

baseline CD4 count (<200 cells/mm3) (data not shown).

Socio-demographic and clinical characteristics at

enrolment of the 1,000 study patients are presented in

Table 1.

n=1000 626 452 412 395 338 238

4949

.550

50.5

5151

.5

PC

S

0 12 28 36 44 52 60

Time (months)

Mean PCS Untransformed score SEs

Mean PCS Untransformed score SEs

n=1000 626 452 412 395 338 2384042

4446

48

MC

S

0 12 28 36 44 52 60

Time (months)

(a)

(b)

Fig. 1 Longitudinal evolution of physical and mental HRQLscores for HIV-infected patients included in the APROCO-COPILOTE cohort (ANRS CO-8) (a) Physical HRQL untrans-formed score. (b) Mental HRQL untransformed score

Table 1 Socio-demographic and clinical characteristics of APROCO-COPILOTE patients with available health related quality of lifedata at baseline (M0) (n = 1,000) (ANRS CO-8 cohort)

Characteristic No. (%)of patients

Male gender 773 (77.3)Age at enrolment—years (mean (SD)) 37.1 (0.3)Born in EU 746 (74.6)Stable partner 606 (60.6)Having children 348 (35.4)High-school education 322 (32.3)Full or part time employment 555 (55.6)Stable housing 803 (80.4)Comfortable housing 846 (85.9)High social support by partner,

family and friends140 (14.2)

HIV transmission groupHomosexual 430 (43.0)IDU 174 (17.4)Heterosexual/others 396 (39.6)

Time since first HIV-positive test—months (mean (SD))

58.6 (1.6)

Clinical stage of AIDS 183 (18.3)CD4+ cell count <200 cells/mm3 324 (32.4)Antiretroviral naivety 423 (42.3)Number of self-reported side-effects

(excluding lipodystrophy)(median [IQR])

1 [0–2]

Depression (CES-D score ‡16) 485 (50.4)Daily alcohol consumption 76 (7.6)HCV infection 215 (21.5)

IDU = intravenous drug use; CES-D = Center for Epidemio-logical Studies Depression Scale; SD = Standard deviation; IQR= interquartile range; HCV = hepatitis C virus

Qual Life Res (2007) 16:577–591 579

123

Most patients were male, natives of European

countries, mean age (SD) was 37.1 (0.3) years. There

were 174 patients (17.4%) infected through IDU, and

430 (43%) infected through homosexual contact. Mean

time (SD) since first HIV-positive test was 58.6 (1.6)

months. CD4 count was below 200 cells/mm3 in 32.4%

of patients, with a median [IQR] of 293 [154–440] cells/

mm3. As stated in the inclusion criteria, all patients

initiated an antiretroviral combination including a PI at

M0. Forty-two percent of patients had never received

antiretroviral therapy before (antiretroviral naive

patients). These patients had a significantly more

advanced HIV disease at baseline, characterized by a

CD4 count below 200 cells/mm3. They were also more

likely than the other patients to have unstable housing

conditions and to be infected through IDU. About one

fifth of all patients were coinfected with hepatitis C virus

(HCV). Between M0 and M60, a total of 78 patients died

and 46 patients signed a withdrawal request form.

We censored these 124 patients in the time to dropout

analysis, as we cannot consider that their quality of life

was observed at the time of dropout. The missing

HRQL data (intermittent and drop-out from the study)

for non-censored patients are due either to a missed

clinical visit, to a non-response to a self-administered

questionnaire or to a lost to follow-up. The proportion

of dropouts is quite high in this study: 37.4% of the 1,000

patients have dropped out after 1 year and 76.2% after

5 years. The distributions of the original aggregate PCS

and MCS scores and the number of individuals still in

the study at each of the seven dataset time points are

presented respectively in Fig. 1a, b.

Results using separate modelling

Multivariate random effects models estimated signifi-

cant positive slopes for both PCS and MCS scores from

M0 to M12 (‘‘Slope 1’’ in Tables 2a, 3a), thereby

underlining the increase in these two scores observed in

the first year after PI introduction (Fig. 1a, b). Slopes

estimated thereafter (‘‘Slope 2’’) were non-significantly

different from zero, underlining a relative stabilization

of the scores after M12, after accounting for the effect

of covariates.

Age, parenthood, experience of AIDS-defining

events and HCV coinfection were found to be inde-

pendent negative predictors of PCS score, while high

education level and homosexual HIV transmission

were associated with higher levels of physical HRQL.

CD4 count below 200 cells/mm3 and number of self-

reported treatment’ side effects excluding lipodystro-

phy were found to be negatively associated with both

physical and mental HRQL scores. However, the

impact of side effects on HRQL decreased over time,

Table 2 Longitudinal models for the evolution of log-transformed physical health-related quality of life score (PCS) over 5 years afterHAART initiation in HIV-infected adult patients (APROCO-COPILOTE ANRS CO-8 cohort)

Variable (a) Random effects model (n = 906) (b) Joint model (n = 906)

Coefficient [95% CI] P Coefficient [95% CI] P

Intercept –3.83 [–3.87, –3.80] <.0001 –3.84 [–3.87, –3.80] <.0001Slope 1 (· 10–3) 7.51 [5.65, 9.37] <.0001 7.52 [5.65, 9.40] <.0001Slope 2 (· 10–4) –2.41 [–8.97, 4.15] .47 –2.66 [–9.59, 3.55] .37Age at enrolment (years) (· 10–3) –2.75 [–3.60, –1.90] <.0001 –3.02 [–3.53, –1.80] <.0001Having children (· 10–2) –1.41 [–2.30, –.53] .0017 –1.41 [–2.30, –.52] .0018High-school education (· 10–2) 1.43 [.64, 2.22] .0004 1.38 [.58, 2.17] .0007Comfortable housing (· 10–2) 1.39 [.25, 2.54] .017 1.31 [.16, 2.47] .026Homosexual transmission group (· 10–3) 8.93 [.07, 17.79] .048 9.06 [.17, 17.94] .046IDU transmission group (· 10–3) –5.46 [–19.74, 8.82] .45 –5.46 [–19.82, 8.89] .45Clinical stage of AIDS (· 10–2) –5.60 [–7.03, –4.17] <.0001 –5.83[–7.26, –4.40] <.0001

Slope 1 (· 10–3) 3.44 [2.14, 4.75] <.0001 3.56 [2.26, 4.87] <.0001Slope 2 ( · 10–4) 2.78 [–.63, 6.20] .11 2.72 [–.69, 6.13] .12

HCV infection (· 10–2) –1.24 [–2.47, –.02] .047 –1.24 [–2.47, –.02] .047No. of self-reported side effects (excluding

lipodystrophy)a (· 10–2)–3.12 [–4.11, –2.14] <.0001 –3.17 [–4.16, –2.17] <.0001

Slope 1 (· 10–3) 1.04 [.22, 1.87] .013 1.07 [.23, 1.90] .012Slope 2 (· 10–4) 1.45 [.64, 2.27] .0005 1.46 [.64, 2.27] .0005

CD4 count <200 cells/mm3 a (· 10–2) –2.47 [–3.55, –1.39] <.0001 –2.29 [–3.37, –1.21] <.0001Slope 1 ( · 10–3) 1.62 [.21, 3.03] .024 1.48 [.07, 2.88] .040Slope 2 ( · 10–4) –.52 [–6.12, 5.07] .85 –.83 [–6.43,4.76] .77

IDU = intravenous drug use; a time-dependent covariates; HCV = hepatitis C virus

580 Qual Life Res (2007) 16:577–591

123

as shown by the significant positive associated slopes.

In the same way, the impact of low CD4 count on

physical HRQL progressively decreased during the

first year of the follow-up, remaining constant there-

after. Having comfortable housing conditions was the

only positive predictor of both physical and mental

HRQL, but its impact on mental HRQL also de-

creased between M0 and M12. Additional predictors

of better mental HRQL were male gender and pres-

ence of a stable partner. HIV transmission through

IDU and number of self-reported symptoms of lip-

odystrophy1 were negatively associated with mental

HRQL score. The impact of antiretroviral naivety on

mental HRQL had a varying pattern over time: neg-

ative at baseline, positive at M12 and re-becoming

negative during the maintenance phase (M12–M60).

Depressive symptomatology, defined as a score ‡16 in

the French version of the Center for Epidemiological

Studies Depression Scale CES-D [19], was collinear

with poor mental HRQL, as 64% of patients with

CES-D score over 16 had mental HRQL below the

first quartile. Thus, depression could be considered to

be more a proxy than an explanatory variable of the

patients’ mental health status.

Results from the separate analysis for the dropout

process are presented in Table 4a.

Socio-demographic and behavioral characteristics

that were found independently associated with a higher

risk of dropout were: female gender, younger age,

difficult housing conditions (unstable or uncomfortable

housing), non-European origin and daily consumption

of alcohol (average number of alcohol units per day ‡1

[20]). Concerning HIV-related characteristics, risk

factors for a dropout included: homosexual and IDU

transmission, antiretroviral naivety and CD4 count

below 200 cells/mm3.

Results using joint modelling

Results using joint modelling of HRQL scores and

dropout process are presented in Tables 2b, 3b and 4b.

The coefficients associated with significant predictors

of HRQL estimated with the joint model were similar

to those obtained with separate models. Changes did

not exceed 10% in magnitude, except for the coeffi-

cient associated with the number of self-reported

symptoms of lipodystrophy, whose absolute value

changed by a rate of 11% between separate and joint

modelling. However, the 95% confidence intervals

obtained with the two methods for this coefficient

overlapped greatly. Some of the dropout model coef-

ficients exhibited marked changes, but remained in the

Table 3 Longitudinal models for the evolution of log-transformed mental health-related quality of life score (MCS) over 5 years afterHAART initiation in HIV-infected adult patients (APROCO-COPILOTE ANRS CO-8 cohort)

Variable (a) Random effects model (n = 925) (b) Joint model (n = 924)

Coefficient [95% CI] P Coefficient [95% CI] P

Intercept –4.08 [–4.11, –4.06] <.0001 –4.08 [–4.11, –4.05] < .0001Slope 1 ( · 10–3) 8.99 [6.47, 11.50] <.0001 8.66 [6.10, 11.23] <.0001Slope 2 ( · 10–4) 3.34 [–4.78, 11.47] .42 2.75 [–5.47, 10.97] .51Male gender ( · 10–2) 3.51 [2.16, 4.86] <.0001 3.56 [2.20, 4.92] <.0001Comfortable housing ( · 10–2) 4.19 [2.41, 5.96] <.0001 4.18 [2.40, 5.97] <.0001

Slope 1 ( · 10–3) –2.59 [–4.44, –.75] .0059 –2.60 [–4.47, –.72] .007Slope 2 ( · 10–4) .98 [–5.05, 7.01] .75 .93 [–5.17, 7.02] .77

Stable partner ( · 10–2) 1.24 [.23, 2.25] .016 1.22 [.21, 2.24] .018Homosexual transmission group ( · 10–3) –9.06 [–21.18, 3.07] .14 –9.19 [–21.41, 3.03] .14IDU transmission group ( · 10–2) –3.30 [–4.77, –1.84] <.0001 –3.31 [–4.78, –1.83] <.0001Antiretroviral naivety ( · 10–2) –2.16 [–3.40, –.92] .0007 –2.14 [–3.39, –.90] .0007

Slope 1 ( · 10–3) 2.76 [1.59, 3.93] <.0001 2.72 [1.55, 3.90] <.0001Slope 2 ( · 10–4) –4.21 [–7.67, –.75] .017 –4.23 [–7.70, –.76] .017

No. of self-reported side effects (excludinglipodystrophy) a ( · 10–2)

–4.43 [–5.59, –3.27] <.0001 –4.40 [–5.57, –3.23] <.0001

Slope 1 ( · 10–3) 2.27 [1.29, 3.25] <.0001 2.26 [1.27, 3.24] <.0001Slope 2 ( · 10–4) .20 [–.92, 1.32] .72 .20 [–.92, 1.33] .72

No. of self-reported symptoms of lipodystrophya

Slope 1 ( · 10–3) –.41 [–.75, –.06] .021 –.36 [–.71, –.015] .041Slope 2 ( · 10–4) .43 [–1.07, 1.92] .57 .33 [–1.17, 1.83] .67

CD4 count <200 cells/mm3 a ( · 10–2) –1.10 [–2.03, –.16] .021 –1.14 [–2.07, –.20] .018

IDU = intravenous drug use; a time-dependent covariates

1 Note that there is no principal effect for the ‘‘Number of self-reported symptoms of lipodystrophy’’ in Table 3, as lipodystro-phy symptoms were absent at M0. The coefficient of the firstslope is used in this case for the interpretation.

Qual Life Res (2007) 16:577–591 581

123

same order of magnitude. Moreover, random effects’

coefficients in the joint models (U0 and U1 in Table 4b,

see Appendix) were non significantly different from

zero. Consequently, given the parametric assumptions of

the model, no significant effect of dropout on the lon-

gitudinal HRQL estimates was detected in this study.

Results for untransformed scores

To facilitate the results interpretation, we calculated

factors’ effects on the untransformed HRQL scores, as

obtained from each modelling approach, using a back

translation onto the original scores scale. Table 5

shows the contribution of each factor to the untrans-

formed PCS and MCS scores evolution between M0

and M12, which corresponds to the phase exhibiting

the more significant changes in HRQL. After adjusting

for the contribution of all factors, mean increase in

HRQL during the first year after HAART initiation

was estimated at about 4 score units for PCS and about

6 score units for MCS.

Discussion

In this study, we analysed HRQL data collected during

the 5 years after HAART initiation in a cohort of

French HIV-infected patients (APROCO-COPIL-

OTE, ANRS CO-8). Considering the distribution of

HRQL scores at each scheduled visit, we hypothesized

that HRQL evolution over time after initiation of such

a therapy followed a two-phase pattern, with a change

in slope after 1 year of treatment.

Indeed, when HRQL trajectories were estimated

using a multivariate mixed effects model, the change in

PCS and MCS scores over the first 12 months was

positive, but these scores showed a relatively stable

trend in the maintenance phase of the treatment, from

M12 to M60. The significant increase in HRQL scores

during the first year of HAART confirmed the results

obtained for the same patients using a cross-sectional

analysis at M12 [9]. Previous clinical trials have already

pointed out the improvement of quality of life after

6 months to 1 year of various treatment regimens

including PI [21–23]. The long-term effect of PI regi-

mens has also been addressed in several studies, leading

to various conclusions. Nieuwkerk et al. [24] reported a

significant improvement in quality of life over 96 weeks

among patients treated with a triple combination ther-

apy. Other studies, in patients treated with different

combinations including PI, reported either no overall

change in HRQL [25, 26] or deterioration in emotional

HRQL [27] after several years of treatment. These

somewhat conflicting conclusions could be attributed to

the relatively small sample sizes used in the analyses

(<100 patients) and to the heterogeneity of scales used

to assess HRQL. One advantage of the APROCO-

COPILOTE cohort is that it provided us with an

extended dataset of 1,000 HIV patients followed-up

during the 5 years after they began HAART.

Despite a global trend for a two-phase pattern, we also

observed some variability between patients’ individual

responses to HAART in terms of quality of life. In a

multivariate analysis, we tried to determine which factors

could explain these differences.

We found that low CD4 count had detrimental

effects on both physical and mental quality of life

[28, 29], which may be due to the faster disease

progression observed among patients beginning

HAART at an advanced immunodepression stage [30].

However, the impact on physical HRQL tended to

decrease over time, underlining the benefits of HA-

ART for these patients.

Table 4 Modelling of thetime-to-dropout of patientsincluded in the APROCO-COPILOTE cohort (ANRSCO-8)

IDU = intravenous drug use;a time-dependent covariatesb the random effects thatappear in the linear randomeffects model

Variable (a) Separateanalysis

(b) Joint analysis

With PCS With MCS

Coefficient P Coefficient P Coefficient P

Male gender –.15 .007 –.16 .006 –.14 .02Age at enrolment (years) –.03 <.0001 –.03 <.0001 –.03 <.0001Stable housing –.11 .03 –.10 .09 –.11 .04Comfortable housing –.26 <.0001 –.24 .0001 –.26 <.0001Born in EU –.22 <.0001 –.23 <.0001 –.22 <.0001Homosexual transmission group .11 .04 .11 .04 .11 .05IDU transmission group .20 .0008 .23 .0002 .21 .0005Antiretroviral naivety .24 <.0001 .24 <.0001 .24 <.0001CD4+ cell count < 200 cells/mm3 a .12 .05 .15 .02 .14 .02Daily alcohol consumption a

.29 <.0001 .32 <.0001 .31 <.0001U 0

b – – .66 .28 .10 .79U 1

b – – .99 .94 16.32 .29

582 Qual Life Res (2007) 16:577–591

123

We identified self-perceived treatment’ side effects

as another factor impairing both aspects of HRQL. In

fact, the adverse effects of HAART [31, 32] interfere

with patients’ daily living activities. Moreover, the

number of self-reported side effects has been shown to

be linked to the occurrence of non-adherence behav-

iours [33, 34], which in turn compromise the virological

response to treatment. The observed time trend toward

a lesser impact on HRQL may be explained by the

patients’ adaptation and adjustment to the drawbacks

of treatments as described by Bury [35].

Having comfortable housing conditions was the lone

factor identified as positively associated with both

physical and mental HRQL. However, the influence of

a comfortable housing on patients’ mental status ten-

ded to decrease over time, probably denoting that the

psychological weight of material living conditions

compete with multiple other factors such as clinical or

emotional status.

Among socio-demographic factors, older age was

found to be associated with an impaired physical

HRQL, while high education level was associated with

a better physical HRQL [28]. Results also showed a

negative impact of parenthood on physical status, which

may be explained by the additional care required when

bringing up children. The negative predictors of phys-

ical HRQL related to patients’ clinical status were

clinical stage of AIDS and coinfection with HCV.

Indeed, patients who presented an advanced stage of

the HIV disease were more likely to be physically

affected by their infection, and HCV coinfection has

been long associated with an impairment of HRQL [36,

37]. Finally, patients infected through homosexual

contact were more likely to present a better physical

HRQL than patients infected through heterosexual

contact or IDU, as already shown [38].

Supporting the results obtained in other studies,

we found that male gender was associated with a

better mental quality of life [39, 40]. Patients having

a stable partner reported better mental status than

others, which emphasizes the role of the emotional

support in patients’ well-being [26]. As substance use

is closely related to psychological distress, patients

contaminated through drug injection reported worse

mental status than others. The number of self-

reported symptoms of lipodystrophy was another

negative predictor of mental HRQL, confirming the

emotional impact of these physical and metabolic

symptoms [41–43]. Finally, antiretroviral naivety was

also negatively associated with mental HRQL. This

may be explained by the fact that naive patients in

APROCO-COPILOTE were more likely to present

an advanced stage of the HIV disease or to meet

social difficulties than the other patients. The

temporary positive impact at M12 of antiretroviral

naivety on mental HRQL may be explained by a

short-term positive effect of HAART initiation in

patients who had never experienced antiretroviral

treatments before. Towards the end of the study

period, this effect seems to be gradually counterbal-

anced perhaps by the awareness of the chronic nature

of the HIV disease.

The dropout process shared most of the predictors

found for HRQL, as patients reporting low HRQL

scores may also feel a lack of motivation and be

tempted to withdraw from the study.

Table 5 Factors’contributions to the evolutionof untransformed health-related quality of life scoresduring the first year afterHAART initiation in HIV-infected adult patients(APROCO-COPILOTEANRS CO-8 cohort)

IDU = intravenous drug use;HCV = hepatitis C virus;a time-dependent covariates

Variable (a) PCS (b) MCS

Randomeffects model(n = 906)

Joint model(n = 906)

Randomeffects model(n = 925)

Joint model(n = 924)

Intercept 3.99 4.00 6.06 5.85Male gender 1.84 1.87Age at enrolment (years) –.12 –.01Having children –.60 –.60High-school education .60 .58Comfortable housing .59 .55 .57 .57Stable partner .66 .65Homosexual transmission group .38 .38 –.48 –.49IDU transmission group –.23 –.23 –1.79 –1.80Antiretroviral naivety .61 .60Clinical stage of AIDS –.63 –.66HCV infection –.53 –.53No. of self-reported side effects

(excluding lipodystrophy) a–.80 –.81 –.92 –.91

No. of self-reported symptoms oflipodystrophy a

.01 .00

CD4 count <200 cells/mm3 a –0.22 –0.22 –.59 –.61

Qual Life Res (2007) 16:577–591 583

123

The method that we chose for the analyses taking

into account the dropout process (use of a shared

parameter selection model) enabled us to estimate

in a single model the effects of each patients’ char-

acteristics on the evolution of their HRQL and the

impact of the time that they spent on study. This is

of particular interest in the context of a longitudinal

cohort study, in which study dropouts are relatively

frequent. Indeed, patients who withdrew before the

end of the study may have a particular clinical

profile, covering a large number of cases from the

patient presenting a satisfying global health status

who do not feel the need of maintaining regular

clinical visits, to the one who is lost of follow-up

because of a worsening health status. The proposed

joint model, derived from the one presented by Guo

and Carlin [8], can be easily fitted using standard

packages such as the NLMIXED procedure of SAS

(see Appendix).

Two limitations of our study must be underlined.

First, the use of the MOS SF-36 generic psychometric

instrument may not allow to take into account all the

specificities of HIV infection. Although this instru-

ment showed satisfying validity and reliability prop-

erties in a recent study invoving HIV-infected patients

[44], it neglects some important dimensions of quality

of life, especially sexual affective and cognitive func-

tioning or sleep problems. Second, the analysis ex-

cluded patients who did not complete the MOS SF-36

questionnaire at enrolment. A descriptive analysis

showed that these patients had a worse HIV status at

baseline than respondents.

To conclude, our study brought to light the early

effects of HAART on HIV patients’ quality of life,

characterized by a significant increase in both physical

and mental HRQL scores occurring during the first

year of treatment. Data collected over 5 years of

HAART showed no overall change in HRQL scores

after this initial increase. In addition, our study iso-

lated both immunodepression and coinfection with

HCV as being two negative predictors of patients’

quality of life, along with the number of self-reported

side-effects, and underlined the role of lipodystrophy

symptoms on patients’ mental status. These results,

obtained on a large sample of HIV-infected patients

followed up during an extended period of time, con-

firm those of previous studies conducted on fewer

subjects with a cross-sectional design. They underline

the necessity to investigate and manage the patients’

perception of treatments’ adverse effects throughout

their clinical follow-up. Besides, they confirm the need

for a careful monitoring of viral comorbidities and for

a prompt management of therapeutic failure to ensure

patients’ well-being.

Acknowledgements The authors would like to thank all pa-tients, nurses and physicians in clinical sites. The APROCO-COPILOTE Study Group is composed of the follow-ing:Steering Committee: Principal Investigators: C. Leport, F.Raffi; Methodology: G. Chene, R. Salamon; Social Sciences:J-P. Moatti, J. Pierret , B. Spire; Virology: F. Brun-Vezinet,H. Fleury, B. Masquelier; Pharmacology: G. Peytavin, R.Garraffo.Scientific Committee: Members of Steering Commit-tee and other members: D. Costagliola, P. Dellamonica, C.Katlama, L. Meyer, M. Morin, D. Salmon, A. Sobel; Projectcoordination: F. Collin; Events Validation Committee: L.Cuzin, M. Dupon, X. Duval, V. Le Moing, B. Marchou, T.May, P. Morlat, C. Rabaud, A. Waldner-Combernoux; Clini-cal Research Group : V. Le Moing, C. Lewden.Clinical cen-ters: Amiens (Pr Schmit), Angers (Dr Chennebault), Belfort(Dr Faller), Besancon (Dr Estavoyer, Pr Laurent, Pr Vuit-ton), Bordeaux (Pr Beylot, Pr Lacut, Pr Le Bras, Pr Rag-naud), Bourg-en-Bresse (Dr Granier), Brest (Pr Garre), Caen(Pr Bazin), Compiegne (Dr Veyssier), Corbeil-Essonne (DrDevidas), Creteil (Pr Sobel), Dijon (Pr Portier), Garches (PrPerronne), Lagny (Dr Lagarde), Libourne (Dr Ceccaldi),Lyon (Pr Peyramond), Meaux (Dr Allard), Montpellier (PrReynes), Nancy (Pr Canton), Nantes (Pr Raffi), Nice (PrCassuto, Pr Dellamonica), Orleans (Dr Arsac), Paris (PrBricaire, Pr Caulin, Pr Frottier, Pr Herson, Pr Imbert, DrMalkin, Pr Rozenbaum, Pr Sicard, Pr Vachon, Pr Vilde),Poitiers (Pr Becq-Giraudon), Reims (Pr Remy), Rennes (PrCartier), Saint-Etienne (Pr Lucht), Saint-Mande (Pr Roue),Strasbourg (Pr Lang), Toulon (Dr Jaureguiberry), Toulouse(Pr Massip), Tours (Pr Choutet).Data monitoring and statis-tical analysis: C. Alfaro, F. Alkaied, C. Barennes, S. Bouc-herit, AD. Bouhnik, C. Brunet-Francois, MP. Carrieri, JL.Ecobichon, V. El Fouikar, V. Journot, R. Lassalle, JP. Le-grand, M. Francois, E. Pereira, M. Preau, V. Villes, C. Pro-topopescu, H. Zouari, F. Marcellin.Promotion: AgenceNationale de Recherches sur le Sida (ANRS, CoordinatingAction n�7.) Other support: College des Universitaires deMaladies Infectieuses et Tropicales (CMIT ex APPIT), Si-daction Ensemble contre le Sida and associated pharmaceu-tical companies: Abbott, Boehringer-Ingelheim, Bristol-MyersSquibb, Glaxo- SmithKline, Merck Sharp et Dohme, Roche.

Appendix

The linear random effects model

Let yij be the transformed score (MCS or PCS) mea-

sured for patient i at time sj, i ¼ 1; . . . ;N; j ¼ 1; . . . ; 7.

We used a piecewise linear random effects model with

a change in the slope at M12, written as:

yij¼a0þx01i tj

� �b0þU0iþ a1þx02i tj

� �b1þU1i

� �min tj;12

� �

þ a2þx02i tj

� �b2þU2i

� �tj�12� �þþeij

Here x1iðtÞ and x2iðtÞ are vectors including fixed and

time-varying explanatory variables, and tj � 12� �þ

584 Qual Life Res (2007) 16:577–591

123

¼ tj � 12 if tj � 12 , and 0 otherwise. U0i is the subject-

specific intercept, U1i and U2i are the random slopes,

modelling the true individual level trajectories after

they have been adjusted for the overall mean trajectory

and the other fixed effects, such that U0i;U1i;U2ið Þ0�N 0;Gð Þ, and eij � N 0; r2

� �are mutually independent

measurement errors. The first term a0 þ x01i tj

� �b0 þU0i

includes principal effects of the factors, the second

term a1 þ x02i tj� �

b1 þU1i

� �represents the changes in

the short term evolution (M0–M12), and the third term

a2 þ x02i tj

� �b2 þU2i

� �represents the changes in the

long-term evolution (M12–M60).

We used the random effects covariance matrix G to

perform a likelihood ratio test for choosing the number

of random effects in the model. We tested whether we

needed a model with q = 3 random effects (i.e. two

random slopes: U1i;U2i) or whether a simpler model,

with only two random effects, would be more ade-

quate. We fitted both models with the same set of fixed

effects and recorded the deviance (–2 Log Likelihood)

for each of them. The distribution of the difference in

deviances is a mixture of a v2 with q degrees of free-

dom and a v2 with q–1 degrees of freedom [45].

For the MCS (resp. PCS) transformed score we

found an increase in the deviance of 16.2 (resp. 23.6) for

q = 3 degrees of freedom, so the test rejected the

specification with two random effects in both cases.

Nevertheless, the estimated G matrix for the model

with three random effects was almost singular, which

suggested that the variability in the last slope was very

small after considering heterogeneity which was

accounted for by the fixed factors. Moreover, the esti-

mated parameters and their P-values were very similar

for the two models, so we finally decided to include only

two random effects in the longitudinal model, which

allowed a faster estimation of the joint model.

The time-to-dropout model

Patients had scheduled follow-up visits at fixed times

(M0, M12, M28, M36, M44, M52, M60). The proposed

model allowed the investigation of the relationship

between times-to-dropout and a set of possibly

time-dependent explanatory variables. When dropout

time is discrete with tied events without underlying

ordering, this model is equivalent to a logistic model

(‘‘discrete-time logit model’’) for a data set with pa-

tient-time as the unit of analysis, which is the same data

set used for the mixed longitudinal analysis [46]. For

each of these observations the response variable w is

dichotomous, corresponding to 1 = ‘‘dropout’’ and

0 = ‘‘still in the study’’ in that time interval. Dropouts

due to death or withdrawal from the study, as well as

all observations corresponding to patients followed

until the end of the study (M60) were censored. A

censored observation can be modelled by having a

record of all zeros until the end of the follow-up.

The time scale of the study follow-up is partitioned

into six disjoint intervals, say ½th; thþ1Þ , h ¼ 1; . . . ; 6 . Let

Ti be the discrete time-to-dropout variable of the patient

i. Then the hazard of dropout in the interval ½th; thþ1) for

the patient i given covariates zi is the conditional prob-

ability that its dropout is at time th+1, given that it was still

in the study at the beginning of the interval:

kiðthÞ ¼ P Ti ¼ thþ1 Ti[th; zi thð Þjð Þ

Using elementary properties of conditional

probabilities, it can be shown that

P Ti ¼ thþ1ð Þ ¼ ki thð ÞYh�1

j¼1

1� ki tj� �� �

and

P Ti[thþ1ð Þ ¼Yh

j¼1

1� ki tj

� �� �

Suppose that ½si;�siÞ is the observed interval of dropout

of the ith subject, and di = 1 if Ti ¼ �si and 0 otherwise.

Let h ið Þ be such that si ¼ th ið Þ. The likelihood for the

discrete time data is given by

L ¼YN

i¼1

P Ti ¼ �sið Þ½ �di P Ti[�sið Þ½ �1�di

¼YN

i¼1

P Ti ¼ th ið Þþ1

� �

P Ti[th ið Þþ1

� �

" #di

P Ti[th ið Þþ1

� �� �

¼YN

i¼1

Yh ið Þ

j¼1

1� ki tj

� �� � ki tj� �

1� ki th ið Þ� �

" #wij

where wij ¼ 1 if j ¼ hðiÞ þ 1 and 0 otherwise. Prentice

and Gloeckler [16] have shown that

kiðthÞ ¼ 1� exp � exp k0h þ z0i thð Þc� �� �

where k0h is the logarithm of the integral of the

baseline hazard on the relevant interval ½th; thþ1Þ. This

can be written equivalently,

ln �ln 1� kiðthÞð Þ½ � ¼ k0h þ z0i thð Þc

Given this complementary log–log transformation, the

parameter c is interpreted as the effect of covariates in

Qual Life Res (2007) 16:577–591 585

123

zi on the hazard rate of dropout, assuming the hazard

rate to be constant on each interval of the study.

As noted by Vonesh et al. [4], one advantage when

using such a model is that it allows the inclusion of

time-dependent covariates, defined at the start

of each interval ½th; thþ1Þ. In this way, the inclusion of

time-dependent covariates is not affected by the fact

that an actual measurement may not be available

for them at the time-of-dropout. Moreover, the

specification of the time-to-dropout model gives a

closed-form expression for the marginal and joint

likelihood, allowing us to use existing standard soft-

ware like the SAS procedure NLMIXED to estimate

both the separate and joint model (see also Guo and

Carlin [8]).

The log-likelihood function for the sample of indi-

viduals used in this study can then be given by

ln L ¼XN

i¼1

Xh ið Þ

j¼1

wij ln 1� exp � exp k0j þ z0i tj

� �c

� �� �� ��

þ 1� wij

� �� exp k0j þ z0i tj

� �c

� �� �g

This takes the form of a ‘‘sequential binary response’’

model with data in the ‘‘vertical’’ form with the ith

subject contributing h(i) observations. Allison [47

] and Jenkins [48–50] have shown that, for such

suitably organized data, the log-likelihood function is

the same as the log-likelihood for a generalized linear

model of the binomial family with complementary log–

log link function.

The joint model

In the joint approach, the linear random effects model

and the time-to-dropout model are estimated simulta-

neously (see Guo and Carlin [8]). The discrete-time

proportional hazard model is now written as follows:

ln � ln 1� kiðthÞð Þ½ � ¼ c0U0i þ c1U1i þ k0h þ z0iðthÞc;

where U0 and U1 are the random effects that appear in

the linear random effects model. In this way, the HRQL

model and the time-to-dropout model are connected

through the stochastic dependence on unobserved fac-

tors represented by these random variables. The

parameters c0 and c1 in the time-to-dropout model

measure the association between the two submodels

induced through these two common latent variables. In

the absence of association between the two models, the

parameters c0 and c1 are not statistically significant and

there is no gain from the joint analysis.

As in Guo and Carlin [8], the binary explanatory

variables were coded with 1 for the active modality and

–1 for the reference modality.

The SAS codes used to estimate the separate and

joint models for the MCS transformed score are given

below.

Determination of factors’ contribution to the

untransformed HRQL scores changes between

M0 and M12

In order to have an idea of the contribution of each

factor on the HRQL scores changes between M0 and

M12 on their original scale (varying between 0 and

100), we adopted the following approach. For each

factor, the corresponding univariate two–phase ran-

dom effects model on transformed HRQL scores was

translated back onto the original scale using the inverse

transformation of -ln (100-score). Then, the factor’s

contribution to the HRQL scores evolution during the

first year of HAART was calculated as the difference

between the score values estimated by the back

translated model at M12 and M0. The same approach

was used to determine the intercepts’ contribution,

from the two-phase random effects model without

covariates.

Random slopes and variance components

parametrization for the mixed model:

Here obstime1 and obstime2 are the two time

variables defined in the ‘‘Statistical methods’’ section

(i.e. minðtj; 12Þ and tj � 12� �þ

respectively). The legend

for the rest of the covariates is given at the end of this

Appendix. Multivariate influence analysis was per-

formed using the INFLUENCE option of the MODEL

proc mixed data = mydata covtest method=ml;

class patient;

model lnmcs = obstime1 obstime2 sexe_r naif_tr

naif_tr_1 naif_tr_2

part_pr confort confort_1 confort_2 homo

toxico cd4t_200 nbs30 nbs30_1

nbs30_2 nbtm_1 nbtm_2 / influence

(iter = 5 effect = patient est) s cl;

random obstime1 intercept /

subject = patient type = fa0(2);

ods output covparms = cp;

ods output solutionf = solf;

ods output Influence = inf;

run;

586 Qual Life Res (2007) 16:577–591

123

statement, in order to test the stability of the linear

mixed model to perturbations of the data. According to

the likelihood distance and Cook’s D measure [51], two

patients identified as influential observations were

dropped from the basis for each HRQL score.

As in Guo and Carlin [8], we used the FA0(q)

structure (No Diagonal Factor Analytic) for the G

matrix in the RANDOM statement of SAS PROC

MIXED, where q is the number of random effects.

This parametrization is equivalent to specifying a

Cholesky root for the G matrix, to constrain it to be

non-negative definite. The lower triangular Choleski

root of G is a lower triangular matrix C, with non-

negative diagonal elements that is a ‘square root’ of G

in the sense that CC¢=G.

The time-to-dropout model

Here w is the dropout indicator variable, and time1

to time6 are the indicator variables for the six intervals

between the visits. The legend for the rest of the

covariates is given at the end of this Appendix.

The NOINT option was specified to prevent the

redundancy in the estimation of the coefficients of

time1–time6 indicator variables. The Newton–Raphson

algorithm with Gauss–Hermite quadrature was used

for the maximum-likelihood estimation of the

parameters.

The joint model

The following statements were used to estimate the

joint model with two random effects (u0 and u1,

corresponding to the intercept and the first slope), and

the same list of predictors used in separate analyses.

This is a modified version of the model XI in Guo and

Carlin [8] (see also their SAS code supplied at http://

www.biostat.umn.edu/~brad/software.html).proc logistic data = mydata;

model w(event=¢1¢)= time1–time6 homo

toxico ue_bis cd4t_200 naif_tr

confort sexe_r ageinc logem alc / noint

link = cloglog technique = newton;

run;

Qual Life Res (2007) 16:577–591 587

123

588 Qual Life Res (2007) 16:577–591

123

The estimate of the longitudinal and dropout sepa-

rate models were combined in a table named Sepa-

rateEstimates. This data set was used to provide

starting values for the joint model, cf. PARAMETERS

statement.

Unlike Guo and Carlin’s [8] program, we used a

discrete-time model, rather than a continuous one.

This way, the computation of the log likelihood

contribution of the dropout data is not made when the

last observation of a subject is reached, but instead for

each observation in the vertical dataset of follow-up

visits.

Variance and covariance of the random effects

were obtained by the delta method, using ESTI-

MATE statement. Approximate 95% prediction

intervals could then be obtained by assuming asymp-

totic normality.

Legend for the covariates used in the MCS mixed model:

Variable Label

Intercept interceptSlope 1 obstime1Slope 2 obstime2Male gender sexe_rComfortable housing confort

Slope 1 confort_1Slope 2 confort_2

Stable partner part_prHomosexual transmission group homoIDU transmission group toxicoAntiretroviral naivety naif_tr

Slope 1 naif_tr_1Slope 2 naif_tr_2

Number of self-reportedside-effects(excludinglipodystrophy)

nbs30

Qual Life Res (2007) 16:577–591 589

123

References

1. Palella, F. J., Jr., Delaney, K. M., Moorman, A. C., Loveless,M. O., Fuhrer, J., Satten, G. A., et al. (1998). Decliningmorbidity and mortality among patients with advanced hu-man immunodeficiency virus infection. HIV OutpatientStudy Investigators. New England Journal of Medicine, 338,853–860.

2. Little, R., & Rubin, D. (2002). Statistical analysis withmissing data.

3. Laird, N. M. (1988). Missing data in longitudinal studies.Statistics in Medicine, 7, 305–315.

4. Vonesh, E. F., Greene, T., & Schluchter, M. D. (2006). Sharedparameter models for the joint analysis of longitudinal dataand event times. Statistics in Medicine, 15, 143–163.

5. Hogan, J. W., Lin, X., & Herman, B. (2004). Mixtures ofvarying coefficient models for longitudinal data with discreteor continuous nonignorable dropout. Biometrics, 60, 854–864.

6. Lin, D. Y., & Ying, Z. (2003). Semiparametric regressionanalysis of longitudinal data with informative drop-outs.Biostatistics, 4, 385–398.

7. Hu, C., & Sale, M. E. (2003). A joint model for nonlinearlongitudinal data with informative dropout. Journal ofPharmacokinetics and Pharmacodynamics, 30, 83–103.

8. Guo, X., & Carlin, B. P. (2004). Separate and joint modelingof longitudinal and event time data using standard computerpackages. The American Statistician, 58, 1–9.

9. Carrieri, P., Spire, B., Duran, S., Katlama, C., Peyramond,D., Francois, C., et al. (2003). Health-related quality of lifeafter 1 year of highly active antiretroviral therapy. Journal ofAcquired Immune Deficiency Syndromes, 32, 38–47.

10. Carrieri, M. P., Leport, C., Protopopescu, C., Cassuto, J. P.,Bouvet, E., Peyramond, D., et al. (2006), Factors associatedwith nonadherence to highly active antiretroviral therapy: A5-year follow-up analysis with correction for the bias inducedby missing data in the treatment maintenance phase. Journalof Acquired Immune Deficiency Syndromes, 41, 477–485.

11. Ware, J. E., Jr., & Sherbourne, C. D. (1992). The MOS 36-item short-form health survey (SF-36). I. Conceptualframework and item selection. Medical Care, 30, 473–483.

12. McHorney, C. A., Ware, J. E., Jr., & Raczek, A. E. (1993).The MOS 36-Item Short-Form Health Survey (SF-36): II.Psychometric and clinical tests of validity in measuringphysical and mental health constructs. Medical Care, 31,247–263.

13. McHorney, C. A., Ware, J. E., Jr., Lu, J. F., & Sherbourne, C. D.(1994). The MOS 36-item Short-Form Health Survey (SF-36):

III Tests of data quality, scaling assumptions, and reliabilityacross diverse patient groups. Medical Care, 32, 40–66.

14. Ware, J. E., Jr., Keller, S. D., Gandek, B., Brazier, J. E., &Sullivan, M. (1995). Evaluating translations of health statusquestionnaires. Methods from the IQOLA project. Interna-tional Quality of Life Assessment. International Journal ofTechnology Assessment in Health Care, 11, 525–551.

15. Leplege, A., Mesbah, M., & Marquis, P. (1995). [Preliminaryanalysis of the psychometric properties of the French versionof an international questionnaire measuring the quality oflife: The MOS SF-36 (version 1.1)]. Revue d Epidemiologie etde Sante Publique, 43, 371–379.

16. Prentice, R. L., & Gloeckler, L. A. (1978). Regressionanalysis of grouped survival data with application to breastcancer data. Biometrics, 34, 57–67.

17. Henderson, R., Diggle, P., & Dobson, A. (2000). Jointmodelling of longitudinal measurements and event timedata. Biostatistics, 1, 465–480.

18. SAS/STAT 9.1 (2004). User’s Guide. Cary, NC, SAS Insti-tute Inc: SAS Publishing.

19. Furher, R., & Rouillon, F. (1989). La version francaise del’echelle CES-D. Description and translation of theauto-evaluation (in French). Psychiatrie et Psychobiologie, 4,163–166.

20. Bissell, D., Paton, A., & Ritson, B. (1982). ABC of alcohol.Help: Referral. British Medical Journal (Clinical ResearchEd), 284, 495–497.

21. Nieuwkerk, P. T., Gisolf, E. H., Colebunders, R., Wu, A. W.,Danner, S. A., & Sprangers, M. A. (2000). Quality of life inasymptomatic- and symptomatic HIV infected patients in atrial of ritonavir/saquinavir therapy. The Prometheus StudyGroup. Acquired Immune Deficiency Syndromes, 14, 181–187.

22. Cohen, C., Revicki, D. A., Nabulsi, A., Sarocco, P. W., &Jiang, P. (1998). A randomized trial of the effect of ritonavirin maintaining quality of life in advanced HIV disease. Ad-vanced HIV Disease Ritonavir Study Group. Acquired Im-mune Deficiency Syndromes, 12, 1495–1502.

23. Carr, A., Chuah, J., Hudson, J., French, M., Hoy, J., Law, M.,et al. (2000). A randomised, open-label comparison of threehighly active antiretroviral therapy regimens including twonucleoside analogues and indinavir for previously untreatedHIV-1 infection: The OzCombo1 study. Acquired ImmuneDeficiency Syndromes, 14, 1171–1180.

24. Nieuwkerk, P. T., Gisolf, E. H., Reijers, M. H., Lange, J. M.,Danner, S. A., & Sprangers, M. A. (2001). Long-term qualityof life outcomes in three antiretroviral treatment strategiesfor HIV-1 infection. Acquired Immune Deficiency Syn-dromes, 15, 1985–1991.

25. Burgoyne, R. W., Rourke, S. B., Behrens, D. M., & Salit, I.E. (2004). Long-term quality-of-life outcomes among adultsliving with HIV in the HAART era: The interplay of changesin clinical factors and symptom profile. Acquired ImmuneDeficiency Syndromes Behavior, 8, 151–163.

26. Burgoyne, R., & Renwick, R. (2004). Social support andquality of life over time among adults living with HIV in theHAART era. Social Science and Medicine, 58, 1353–1366.

27. Eriksson, L. E., Bratt, G. A., Sandstrom, E., & Nordstrom,G. (2005). The two-year impact of first generation proteaseinhibitor based antiretroviral therapy (PI-ART) on health-related quality of life. Health Quality of Life Outcomes, 3, 32.

28. Campsmith, M. L., Nakashima, A. K., & Davidson, A. J.(2003). Self-reported health-related quality of life in personswith HIV infection: Results from a multi-site interviewproject. Health Quality of Life Outcomes, 1, 12.

29. Jia, H., Uphold, C. R., Wu, S., Chen, G. J., & Duncan, P. W.(2005). Predictors of changes in health-related quality of life

continued

Variable Label

Slope 1 nbs30_1Slope 2 nbs30_2

Number of self-reportedsymptoms oflipodystrophySlope 1 nbtm_1Slope 2 nbtm_2

CD4+ cell count <200 cells/mm3 cd4t_200Born in EU ue_bisAge at enrolment ageincStable housing logemDaily alcohol consumption alc

590 Qual Life Res (2007) 16:577–591

123

among men with HIV infection in the HAART era. Ac-quired Immune Deficiency Syndromes Patient Care STDSPatient Care STDS, 19, 395–405.

30. Hogg, R. S., Yip, B., Chan, K. J., Wood, E., Craib, K. J.,O’Shaughnessy, M. V., & Montaner, J. S. (2001). Rates ofdisease progression by baseline CD4 cell count and viral loadafter initiating triple-drug therapy. JAMA, 286, 2568–2577.

31. Bonfanti, P., Valsecchi, L., Parazzini, F., Carradori, S., Pus-terla, L., Fortuna, P., et al. (2000). Incidence of adversereactions in HIV patients treated with protease inhibitors: acohort study Coordinamento Italiano Studio Allergia e In-fezione da HIV (CISAI) Group. Journal of Acquired Im-mune Deficiency Syndromes, 23, 236–245.

32. Fellay, J., Boubaker, K., Ledergerber, B., Bernasconi, E.,Furrer, H., Battegay, M., et al. (2001). Prevalence of adverseevents associated with potent antiretroviral treatment: SwissHIV Cohort Study. Lancet, 358, 1322–1327.

33. Ammassari, A., Murri, R., Pezzotti, P., Trotta, M. P.,Ravasio, L., De Longis, P., et al. (2001). Self-reportedsymptoms and medication side effects influence adherence tohighly active antiretroviral therapy in persons with HIVinfection. Journal of Acquired Immune Deficiency Syn-dromes, 28, 445–449.

34. Duran, S., Spire, B., Raffi, F., Walter, V., Bouhour, D.,Journot, V., et al. (2001). Self-reported symptoms after ini-tiation of a protease inhibitor in HIV-infected patients andtheir impact on adherence to HAART. HIV Clinical Trials,2, 38–45.

35. Bury, M. (1991). The sociology of chronic illness: A review ofresearch and prospects. Sociology of Health and Illness, 13,451–468.

36. Foster, G. R., Goldin, R. D., & Thomas, H. C. (1998).Chronic hepatitis C virus infection causes a significantreduction in quality of life in the absence of cirrhosis. Hep-atology, 27, 209–212.

37. Rodger, A. J., Jolley, D., Thompson, S. C., Lanigan, A., &Crofts, N. (1999). The impact of diagnosis of hepatitis C viruson quality of life. Hepatology, 30, 1299–1301.

38. Preau, M., Leport, C., Salmon-Ceron, D., Carrieri, P., Por-tier, H., Chene, G., et al. (2004). Health-related quality oflife and patient–provider relationships in HIV-infected pa-tients during the first three years after starting PI-containing

antiretroviral treatment. Acquired Immune Deficiency Syn-dromes Care, 16, 649–661.

39. Cederfjall, C., Langius-Eklof, A., Lidman, K., & Wredling, R.(2001). Gender differences in perceived health-related qual-ity of life among patients with HIV infection. Acquired Im-mune Deficiency Syndromes Patient Care STDS, 15, 31–39.

40. Mrus, J. M., Williams, P. L., Tsevat, J., Cohn, S. E., & Wu, A.W. (2005). Gender differences in health-related quality oflife in patients with HIV/AIDS. Quality of Life Research, 14,479–491.

41. Nicholas, P. K., Kirksey, K. M., Corless, I. B., & Kemppainen,J. (2005). Lipodystrophy and quality of life in HIV: Symptommanagement issues. Applied Nursing Research, 18, 55–58.

42. Collins, E., Wagner, C., & Walmsley, S. (2000). Psychosocialimpact of the lipodystrophy syndrome in HIV infection.Acquired Immune Deficiency Syndromes Read, 10, 546–550.

43. Preau, M., Bouhnik, A., Spire, B., Leport, C., Saves, M.,Pıcard, O., et al. (2006). Health related quality of life andlipodystrophy syndrome among HIV-infected patients.Encephale, 32, 713–719.

44. Hsiung, P. C., Fang, C. T., Chang, Y. Y., Chen, M. Y., Wang,J. D. (2005). Comparison of WHOQOLbREF and SF-36 inpatients with HIV infection. Quality of Life Research, 14,141–150.

45. Monnete, G., Shao, Q., & Kwan, E. (2002). A first look atmultilevel models. In S.C.S. Institute for Social Research.York University: Institute for Social Research. StatisticalConsulting Service, York University.

46. Allison, P. D. (1995). Survival analysis using the SAS system:A practical guide. The SAS Institute, Cary, NC.

47. Allison, P. D. (1982). Discrete-time methods for the analysisof event histories. Sociological Methodology, 15, 61–98.

48. Jenkins, S. P. (1995). Easy estimation methods for discrete-time duration models. Oxford Bulletin of Economics andStatistics, 57, 129–138.

49. Jenkins, S. P. (1997). Estimation of discrete time (groupedduration data) proportional hazards models: pgmhaz. StataTechnical Bulletin Reprints, STB 17, 1–12.

50. Jenkins, S. P. (2005). Survival analysis. In Institute for Socialand Economic Research, University of Essex.

51. Zewotir, T., & Galpin, J. S. (2005). Influence diagnostics forlinear mixed models. Journal of Data Science, 3, 153–177.

Qual Life Res (2007) 16:577–591 591

123