We conducted a retrospective cohort study at Jaramogi Oginga Odinga teaching and referral hospital (JOOTRH): the largest referral hospital in western Kenya. The hospital is located in the southwest part of the country bordering Lake Victoria and serves an area with some of the worst health indicators in the country, including high prevalence of HIV infection (15.4%, which is greater than twice that of the national 7.1% prevalence) [18], [19]. Since 2003, the hospital provides comprehensive HIV care at no cost, as part of the national ART program through a joint effort with Columbia University (MTCT-plus program), Government of Kenya and the U.S. Centers for Disease Control (CDC).

Included in this analysis were non-pregnant adults of >15 years, who initiated first-line regimen between 1^st January 2009 and 31st January 2011, and had at least one follow-up visit record. During the study period, the WHO 2006 guidelines for adolescents and adults adopted by MOH-NASCOP were in use [20]. The first-line regimen consisted of the NRTI backbone zidovudine (AZT) or stavudine (d4T) or tenofovir (TDF), with lamivudine (3TC) and either Nevirapine (NVP) or Efavirenz (EFV). Patients were initiated on treatment when they either had CD4 counts of ≤200 cells/mm3 or when they had WHO stage IV disease. They would then be followed up for 2 weeks after initiating treatment, monthly if stable and six months thereafter. During the visits, clinicians would collect the patient’s demographics, clinical and pharmacological information in standardized optical character reader forms, which were then transcribed into the KEMRI/CDC HIV implementation science service program (HISS) electronic database, designed mainly for data management and program evaluation. Quality control for the stored data was done at regular intervals. At the time of registration, patients were given unique identifiers different from those in the patient support centers for concealment purposes.

The primary outcome in this analysis was time to first combined antiretroviral treatment (cART) modification, defined as the time from treatment initiation to change of one or more antiretroviral drugs used as part of the initial first-line cART. Reasons for treatment modification were based on those documented by the clinician, usually as, toxicity, treatment failure (defined as immunological failure, according to WHO 2006 guidelines as CD4 counts decrease of 50% from the on treatment peak value, or a persistent CD4 count lower than 100 cells or fall of CD4 counts to pre-therapy baseline or clinical failure defined as new or recurrent WHO stage IV condition), non-adherence, or others. In case the documented reason was recorded as “others”, further chart review at the patient support center clinic, was done to identify the exact documented reason.

Independent variables assessed were mainly demographic and clinical in nature and included age at treatment initiation, gender, baseline CD4 counts, baseline WHO clinical stage, type of NNRTI treatment in the regimen (NVP vs EFV) and the type of NRTI backbone (AZT or TDF or d4T). Baseline parameters were assessed at cART initiation, which was also the entry point for the participants in this study.

Baseline patient characteristics were described using percentages for categorical data and median and inter-quartile ranges for continuous data. Incidence rates were calculated as the number of events over the person years of follow-up and the confidence intervals obtained from Poisson distribution. Drug specific incidence rates were determined as rate per persons initiating the specific drug. Kaplan-Meier analyses were used to estimate the time to first cART modification. Patients were censored at the time of event or at their last clinical follow-up visit.

Cox proportional hazards models were used to determine factors associated with cART modification. Due to violation of proportionality of hazards (PH), piecewise Cox regression models were fitted in at ≤12 months and >12 months which were time periods corresponding to the time at which the hazards were proportional. Predictor variables assessed included gender, age at treatment initiation, baseline weight, CD4 counts (obtained at closest date to treatment initiation, usually taken 6 months prior or after cART initiation), WHO stage, and the patient’s cART regimen i.e. (NVP vs. EFV), (AZT vs. d4T/TDF), (TDF vs. AZT/d4T), (d4T vs. AZT/TDF). Information on baseline CD4 was missing for 178 patients (10.6% for those with cART modification and 20.5% for those who sustained treatment). The missing CD4 data was imputed by multiple imputation using chain equations (MICE) [21]. Prediction equation included WHO staging, baseline weight, age at treatment initiation, gender, time to treatment modification, treatment modification status and first-line regimens. Before imputation, continuous variables were normalized using square root transformation for age and log-transformation for baseline weight. A total of 10 imputed data sets were generated.

Variables significant at univariate analysis (P<0.10) were included in the multivariate models. Estimates of hazard coefficients were derived through averaging of the 10 iterations and appropriate standard errors calculated using the Rubin’s rules [21], [22].

We also assessed factors associated with specific reasons of treatment modification grouped as toxicity and contraindication (TB treatment and other drug contraindications) for which there was sufficient data to conduct the sub-analysis. All analysis was done in Stata version11 (StataCorp, College Station, Texas).

This study was approved by the ethics review committees of Kenya Medical Research Institute and Makerere University School of Medicine and the Institutional Review Board of JOOTRH. Since this was a retrospective study of already collected anonymous data, consent waiver was sought and obtained from the above Ethics reviews committees.

A total of 1140 participants aged 15 years and above who initiated treatment between 1^st January 2009 and 31^st January 2011 were enrolled in this study. Of these 185 had no follow-up visit and were excluded. Subsequently 955 participants who met the inclusion criteria were enrolled; of these 66.5% were female. At cART initiation, median patient age was 31 years (inter-quartile range IQR 26–38), median CD4 counts (available for 777 patients) was 257 (IQR 164–358) and median weight 60kg (IQR 53–67); 53.1% of the patients started cART at WHO stage III/IV. A majority of the patients initiated a d4T containing first-line regimen (59.7%), as well as a nevirapine-containing regimen (89.1%) (Table 1). The baseline CD4 of 309 (39.8%) participants was collected post-CART at a median period of 2.1 months (IQR 1.2–4.1).

The median follow-up time from cART initiation was 10.7 months during which a total of 178 individuals modified regimen. This represented an overall incidence rate of 18.64 per 100 person years [95% CI 16.09–21.59] over 946 person years of follow-up. The rate of modification was higher in the first year post-cART (IR; 44.08 95% CI: 36.69–52.97) compared to second (IR; 11.24 95% CI: 8.67–14.58) and the third year (IR; 3.88 95% CI: 1.85–8.12).

Table 2 shows the reasons for cART modification as reported by the clinicians. The most commonly cited reason for modification was toxicity (66.3%, IR 12.47; 95% CI: 10.41–14.94) followed by drug contraindication (12.4%, IR; 2.33 95% CI: 1.53–3.53) while treatment failure accounted for only 2.81%, (IR; 0.53 95% CI: 0.22–1.27). A further 18.5% were recorded as either others or non-adherence (2.23%). Information on adverse events was available for 34 of 118 persons who modified regimen due to toxicity. Of these d4T related peripheral neuropathy (38.2%) and lipodystrophy (26.5%) were the most common documented drug toxicities. On the other hand modification due to contraindication was mainly of NVP to EFV substitutions (68%) as a result of rifampicin-NVP contraindication with TB patients. Figure 1 further illustrates the time to cART modification; overall and stratified by key reasons for cART modification i.e. toxicity and drug contraindications. There was a steady increase in cART modification for both overall as well as by toxicity, throughout the follow-up time. The graph for toxicity closely mimicked that for overall cART modification and this was because toxicity accounted for up to 66.3% of all modifications. On the other hand the proportion of cART modification due to drug contraindication remained steadily low at less than 5% throughout the follow-up period.

A majority of cART modifications were single drug substitutions (n = 157, 88.2%), the drugs changed were d4T (n = 92), NVP (n = 48), AZT (n = 9), EFV (n = 9), TDF (n = 2). Treatment switch from first to second-line drugs accounted for 11.8% (n = 21) of all cART modifications. Overall rates for treatment modification was highest among persons initiating d4T (IR 18.83, 95% CI 15.56–22.78) based regimen as compared to either AZT (IR 4.03, 95% CI 2.17–7.49) or TDF (IR 1.43, 95% CI 0.36–5.71). This was equally the same when the rate of modifications in this NRTI’s was assessed by toxicity, treatment failure, contraindication and other reasons. Between the NNRTI’s the overall rate of cART modification was higher with EFV (IR 9.80, 95% CI 5.28–18.22) as compared to NVP (IR 7.17, 95% CI 5.58–9.21). The rate of modifications due to toxicity, treatment failure and drug contraindications was however higher with NVP as compared to EFV (Table 3).

Following the identification of violation of proportionality of hazard assumption for the variables baseline weight, d4T vs AZT/TDF, AZT vs. d4T/TDF and age, a piecewise Cox-regression model was fitted for two time periods ≤12 and >12 months. This time period coincided to that which the PH assumption had been met for all the variables.

In the first 12 months post cART, baseline WHO stage (III/IV vs. I/II, aHR 1.82, 95% CI 1.25–2.66), presence of d4T in regimen (aHR 2.21, 95% CI 1.49–3.30) and a yearly increase in age (aHR 1.02, 95% CI 1.0–1.04) were significantly associated with increased risk for cART modifications. On the other hand, use of either AZT or TDF was associated with reduction in risk of cART modification (aHR 0.60 95% CI: 0.38–0.96 and aHR 0.51 95% CI: 0.29–0.91 respectively) (Table 4).

After 12 months post cART, a yearly increase in age (aHR 1.05 95% CI 1.02–1.07) baseline CD4 count ≤350 vs. >350 (aHR 2.45 95% CI 1.14–5.26), presence of d4T in regimen (aHR 2.75 95% CI 1.25–6.05) and baseline weight (>60 kg vs. ≤60 kg) (aHR 2.69 95% CI 1.58–4.59) were significantly associated with an increased hazard for cART modification (Table 4).

Table 5 describes the factors associated with cART modification due to drug toxicity and contraindications. Patients, who initiated a d4T containing regimen and those who were older, were significant more likely to modify regimen due to toxicity within the first year of treatment (aHR 1.93, 95% CI: 1.18–3.15 and aHR 1.03, 95% CI 1.01–1.05 respectively). After the first year of cART initiation, patients who started treatment with low CD4 counts of ≤350 vs. >350, (aHR 2.75, 95% CI: 1.05–7.21), a high baseline weight (>60 kg vs. ≤60 kg) (aHR 3.89, 95% CI 2.01–7.54), those with d4T in their regimen (aHR 3.84, 95% CI 1.37–10.75) and those who were older (aHR 1.06, 95% CI: 1.05–1.09) were more likely to modify regimen due to toxicity.

Similarly patients who initiated treatment at low CD4 counts of ≤200 vs. >200 (aHR 5.98, 95% CI: 1.78–20.14), those who had WHO clinical stage III/IV vs I/II (aHR 5.92, 1.70–20.57) those who had a d4T containing regimen (aHR 4.10, 95% CI: 1.17–14.41) were more likely to modify treatment due to drug contraindications within the first year after cART initiation. Beyond the first year of cART only patients initiating a TDF containing regimen were more likely to modify treatment due to drug contraindication (aHR 10.08, 95% CI 1.41–72.15).

Of the 1140 participants who initiated treatment during the study period, 185 (16%) did not have any follow-up visit and thus they were probably lost to follow-up (ltfu) and were excluded in this analysis. Baseline characteristics of the participants lost to follow-up and excluded were similar to those who enrolled apart from disease stage at cART initiation, with those ltfu having advanced disease stage (p = 0.003) and were also likely to have missing CD4 counts (p = 0.001) (Table 1).

A further 178 (18.6%) participants had missing data on CD4 counts. These participants had similar characteristics to those whose baseline CD4 counts was available and differed only in WHO stage III/IV (57.9% vs. 40.4% p<0.001). Subsequently imputation was done for the missing CD4 values. There was no difference in the determination of predictors of cART modification when the analysis was done without the imputation, with only slight adjustments in the hazard ratios (i.e. at time periods >12 months, age (aHR 1.04 95% CI 1.02–1.07), CD4 counts ≤350 vs. >350 (aHR 2.64, 95% CI: 1.25–5.59), d4T (aHR 2.52 95% CI 1.14–5.55 and baseline weight (>60 kg vs. ≤60 kg) (aHR 2.23 95% CI 1.31–3.79). However there were differences in the predictors of cART modification due to toxicity and contraindications when analysis was done without imputation of missing CD4 counts. Low CD4 counts ≤350 vs. >350 (aHR 2.87 95% CI 1.11–7.42) was a significant predictor of cART modification due to toxicity at >12 months, in addition to age, d4T and high baseline weight. On the other hand, low CD4 counts ≤200 vs. >200 was no longer associated with cART modifications due to contraindication at ≤12 months post treatment initiation.