A multi-state model of treatment states in an antiretroviral treatment programme cohort in Cape Town

Master Thesis


Permanent link to this Item
Journal Title
Link to Journal
Journal ISSN
Volume Title
Introduction A recent systematic review estimated that almost a quarter of patients in low- and middle-income countries are not retained on antiretroviral treatment (ART) beyond one year. Further, it is difficult to determine whether a patient who is not retained in care has interrupted their treatment, transferred to another treatment facility, or died. Previous studies have been deterministic in classifying loss to follow-up and treatment interruption. This study investigates treatment interruption and resumption rates when accounting for uncertainty in the occurrence of interruptions. The primary objective is to estimate the rate at which ART is interrupted and the rate at which ART is resumed after an interruption. Methods We fitted a multi-state model to data from the Khayelitsha cohort of the International Epidemiologic Databases to Evaluate AIDS. Between 2001 and 2012, 6796 adult patients starting ART were included. Potential treatment interruption periods were defined between contact points 3 or more months apart. To aid the model in determining if a patient truly interrupted treatment a CD4 count model was used. CD4 counts were modelled to drop to baseline by 3 months after the start of a treatment interruption. Bayesian estimation and Markov chain Monte Carlo were used to obtain posterior distributions of parameters. Several scenarios were used in sensitivity testing, including varying the threshold used to define potential treatment interruption periods, and either adjusting or excluding the data of those with CD4 counts that drop below baseline. Results The baseline annual rate of treatment interruption had a posterior mean of 0.060 (95% CI 0.038- 0.087) which is significantly lower than the prior distribution that had a mean of 0.145 (95% CI 0.080-0.229). The posterior distribution of the baseline annual rate of treatment resumption (mean 1.09; 95% CI 0.68-1.65) was consistent with the prior distribution (mean 1.46; 95% CI 0.21-3.90). The posterior distributions of the parameters related to treatment interruption and resumption did not change significantly in sensitivity testing. Conclusion This study indicates that treatment interruption rates may be significantly lower than previously estimated. The methodology of this study may be useful to those measuring retention within ART programmes. An important limitation was that the CD4 count model did not allow for CD4 counts to fall below baseline during periods of treatment interruption. This limits the generalisability of the posterior estimates of the parameters of the CD4 count model. Further research may require a more flexible CD4 count model.