Modeling Changes in CD4-positive T-Lymphocyte Counts after the Start of Highly Active Antiretroviral Therapy and the Relation with Risk of Opportunistic Infections The Aquitaine Cohort, 1996–1997

Abstract

After initiation of a treatment for human immunodeficiency virus type 1 infection containing a protease inhibitor, immune restoration associated with increases in CD4-positive (CD4+) T lymphocyte count may be delayed. In a sample of patients who had been prescribed protease inhibitors for the first time, the authors tested to see whether there was a minimal duration of CD4+ cell count increase before the increase had an impact on the occurrence of opportunistic infections. The evolution (difference between time t and baseline) of CD4+ cell count was modeled using a mixed effects linear model. Changes in CD4+ count estimated by this model were then included as time-dependent covariates in a proportional hazards model. Finally, the authors tested for the existence of a CD4+ change x time interaction. The authors used a sample of 553 French patients first prescribed protease inhibitors in 1996 and followed for a median of 16 months. During the first 120 days, there was no association between CD4+ change and the rate of opportunistic infections. After 120 days, each 50-cell/mm3 increase in CD4+ count was associated with a 60% (95% confidence interval: 45, 72) reduction in the incidence of opportunistic infections. These results, based on modeling of CD4+ cell response, at least indirectly reinforce the concept of a delayed but possible immune recovery with the use of protease inhibitors. The findings support the potential for interruption of certain types of prophylaxis against opportunistic infections under reasonable conditions of duration of antiretroviral therapy and sustained CD4+ cell response.

Prognosis of HIV-1-infected patients up to 5 years after initiation of HAART: collaborative analysis of prospective studies

“…Twelve cohorts contributed data: the French Hospital Database on HIV (FHDH) ANRS CO4 [6] and the Aquitaine Cohort ANRS CO3 (France) [7]; the AIDS Therapy Evaluation Project Netherlands (ATHENA) [8]; the Italian Cohort of Antiretroviral-Naive Patients (ICONA) [9]; the Swiss HIV Cohort Study (SHCS) [10]; the Frankfurt HIV Cohort [11] and Köln/Bonn Cohort (Germany) [12]; the EuroSIDA study (20 countries in Europe and Argentina) [13]; the Collaborations in HIV Outcomes Research US (CHORUS; USA) [14]; the Royal Free Hospital Cohort (UK) [15]; and the British Columbia Centre for Excellence in HIV/AIDS [16] and the South Alberta Clinic (Canada) [17]. At all sites, institutional review boards had approved the collection of data.…”

Penalized likelihood approach to estimate a smooth mean curve on longitudinal data

“…Estimates of the variance parameters were 2 w =1:2; =0:08 and 2 =0:09. Thus, without doing a priori hypotheses on the form of the curve, this analysis led to similar conclusion compared to the parametric ones using high degree polynomials [10] or a piecewise linear function [11].…”

“…Moreover, the combination of a non-parametric function for the mean curve and of stochastic Gaussian processes for subject-speciÿc deviation allows exible modelling of individual evolutions as well as the population evolution. Thus, using empirical Bayes, the proposed method allows estimation of an individual prediction in order to include it as time-dependent explanatory variable in a proportional hazards model, as it was done using a parametric longitudinal model, to study the relation between progression of CD4+ count and risk of an AIDS event [10]. ; is a squared matrix of size (K + 2 + p) with elements equal to zero except the ÿrst (K + 2) × (K + 2) block which is 1 .…”

“…The sample consisted of subjects included in a prospective hospital-based cohort of HIV-1 infected patients, the Aquitaine cohort, who were prescribed for the ÿrst time a treatment regimen containing at least one protease inhibitor, between March and 31 December 1996 and had at least one measurement of CD4+ and CD8+ cell counts at treatment initiation and one measurement during the follow-up [10]. The follow-up extended to 31 December 1997 and the median time of follow-up was 16 months (interquartile range [12][13][14][15][16][17][18].…”

“…This data set was previously analysed using either a high degree polynomial function of time [10] or a piecewise linear function [11] to approximate the non-linear evolution of CD4 lymphocyte counts after the initiation of the treatment. The ÿrst approach raises numerical problems and is not convenient for individual prediction outside of the range of the data [12].…”