This is the period over which drug coverage is assessed, and time-zero represents the point from which prediction of the subsequent outcome VL is made. One rationale for including only patients who had continuous records of prescription and undetectable

VLs was that there was evidence that such patients were actually picking up their prescribed drugs. Patients experiencing at least one DCVL episode were included in the analysis, with one or multiple DCVL episodes contributed by each patient. A DCVL episode was excluded from analyses if the drug coverage period met any of the following exclusion criteria: (i) there was a gap after the end of a prescription to the next prescription or to time-zero longer than 3 months (to exclude the possibility of missing data learn more or receipt of antiretroviral drugs from other sources), (ii) the duration of the prescription (i.e. amount of drug) was missing, unless this did not result in any gap in drug coverage, and (iii) time-zero was more than 2 weeks after the end of the last ever (at the time of the analysis) recorded prescription. Furthermore, only episodes with outcome VL up to 30 April 2008 and time-zero

between 1 January 2000 and 1 October 2007 were included. The analysis considered drug coverage measured in two different ways: as a continuous variable (per 10% increase) and categorized as ≤60, 60.1–80, 80.1–95, 95.1–99.9 and 100% Selleckchem 5 FU coverage. The endpoint in this analysis was viral rebound, defined by whether the outcome VL was >200 copies/mL or not. We chose a VL value of 200 copies/mL, because we were interested in predicting even relatively small rises in VL. A modified Poisson regression model, with robust error variances [43], was used to model the association between drug coverage

nearly and risk of viral rebound, after adjusting for other potential confounding variables. Adjustment was made for the following potential confounders: age (per 10-year increase), sex, ethnicity (white, black African and other), risk group (homo/bisexual, heterosexual and other), calendar year of start of HAART, continuous time with undetectable (i.e. ≤50 copies/mL) VL (per 1-year increase), previous virological failures (defined as VL >500 copies/mL while on HAART, classified as 0, 1 and 2 or more), current ART regimen [regimens containing nucleoside reverse transcriptase inhibitors (NRTIs) only, unboosted PIs, ritonavir-boosted PIs, NNRTIs and other], number of previous treatment interruptions (defined as discontinuation of all therapy for at least 2 weeks after having started ART while VL value >500 copies/mL and classified as 0, 1, 2, 3 or more), CD4 cell count at time-zero (<200, 200–350 and >350 cells/μL, and missing) and calendar year of time-zero.