Adherence to antiretroviral therapy, virological response, and time to resistance in the Dakar cohort

Abstract

In 1998, with the launch of the Senegalese Initiative for Antiretroviral Access (ISAARV), Senegal became one of the first African countries to propose an antiretroviral access program. Our objective in this paper is to study the time to any first drug resistance, as well as predictors of the time to resistance. We propose a joint model to study the effect of adherence to the HAART therapy, and virological response on the time to resistance mutations. A logistic mixed model is used to model the time-dependent adherence process; and a Markov model is used to study the virological response. Given the presence of missing data in the adherence process and in the virological response, the latent adherence and virological states are then included in the linear predictor of the time to resistance model. The proposed time to resistance model takes into account interval-censored data as well as null hazard periods, during which the viral replication is very low. A Bayesian approach is used for accommodating with missing data and for prediction. We also propose model checking tools to study model adequacy.

Figure B1

Posterior predictive distribution of test quantities T₁ (mean of the proportion of time a patient is not adherent), T₂ (variance of the proportion of time a patient is not adherent), T₃ (mean of the proportion of adherence state changes per patient), and T₄ (variance of the proportion of adherence state changes per patient). Vertical lines represents the observed value of each test quantity.

Figure C1

Posterior predictive distribution of test quantities (number of transitions from state m= 1,2,3 (rows) to state n= 1,2,3 (columns)). Vertical lines represent the observed values of each test quantity.

Figure D1

Posterior predicted survival curves (black curves) and Turnbull survival estimation (gray curve).

Figure 1

Observed virological response Y, observed adherence pattern S, and observed number of non-adherence states among the three last months A for two patients. An absence of a point (at biannual times for Y and A, and every months for S) indicates missing data. Patient A (left panel) has a last negative resistance test at time t = 35 months and has a PI in its therapy during all the follow-up. Patient B (right panel) has a positive resistance test at time t = 24 months and has no PI in its therapy during all the follow-up.

Figure 2

QQ-plot of the distribution of the posterior means of the current adherence state random effects, previous adherence state random effects, and penultimate adherence state random effects against the normal distribution.

Figure 3

Posterior mean (dots) and 95 per cent credible intervals (segments) of transition probabilities from virological state m= 1,2,3 to state n= 1,2,3, according to the presence of a protease inhibitor in the initial therapy (PI= 1 if a protease inhibitor is included in the initial therapy and 0 otherwise), time since inclusion in the cohort (T = 1 if the patient has been followed for more than 2 years), and adherence to antiretroviral therapy. Each graph is divided into three subgraphs (according to a dotted line); the lower subgraph corresponds to perfect adherence during the three months preceding the transition, the middle subgraph to 1 or 2 non-adherence states during the three months preceding the transition, and the upper subgraph to non-adherence during the three months preceding the transition.

Figure 4

Each subfigure ((a) to (h)) presents predictions for eight new patients: the predictive distribution of the number of non-adherence states during the three last months preceding each biannual visit (bottom graph), of the virological states (middle graph), and of the resistance status (upper graph) for a new patient. At each biannual visit, the number of non-adherence states during the three last months can be 0, 1, 2, 3 (from light to dark gray); the virological state can be 1,2,3 (from light to dark grey); and the resistance status can be 0—non-resistant (light gray) or 1—resistant (dark gray). Patients (a)–(d) were assumed to have no PI in their therapy, and to have no resistance mutation at their last visit. Patients (e) to (h) were assumed to have no resistance mutation at their last visit. Patients (e) and (f) were assumed to have no PI in the therapy, unlike patients (g) and (h). (a) Perfectly adherent for 12 months, and in virological state 1 at 12 months. (b) Never adherent for 12 months, and in virological state 1 at 12 months. (c) Perfectly adherent for 30 months, and in virological state 3 at 30 months. (d) Never adherent for 30 months, and in virological state 3 at 30 months. (e) Perfectly adherent for 12 months, and in virological state 2 at 12 months. (f) Never adherent for 12 months, and in virological state 2 at 12 months. (g) Perfectly adherent for 12 months, and in virological state 2 at 12 months. (h) Never adherent for 12 months, and in virological state 2 at 12 months.

Figure 4

Each subfigure ((a) to (h)) presents predictions for eight new patients: the predictive distribution of the number of non-adherence states during the three last months preceding each biannual visit (bottom graph), of the virological states (middle graph), and of the resistance status (upper graph) for a new patient. At each biannual visit, the number of non-adherence states during the three last months can be 0, 1, 2, 3 (from light to dark gray); the virological state can be 1,2,3 (from light to dark grey); and the resistance status can be 0—non-resistant (light gray) or 1—resistant (dark gray). Patients (a)–(d) were assumed to have no PI in their therapy, and to have no resistance mutation at their last visit. Patients (e) to (h) were assumed to have no resistance mutation at their last visit. Patients (e) and (f) were assumed to have no PI in the therapy, unlike patients (g) and (h). (a) Perfectly adherent for 12 months, and in virological state 1 at 12 months. (b) Never adherent for 12 months, and in virological state 1 at 12 months. (c) Perfectly adherent for 30 months, and in virological state 3 at 30 months. (d) Never adherent for 30 months, and in virological state 3 at 30 months. (e) Perfectly adherent for 12 months, and in virological state 2 at 12 months. (f) Never adherent for 12 months, and in virological state 2 at 12 months. (g) Perfectly adherent for 12 months, and in virological state 2 at 12 months. (h) Never adherent for 12 months, and in virological state 2 at 12 months.