Ethics Statement

All patients enrolled in the Hospital Nacional Cayetano Heredia (HNCH) HIV cohort sign an informed consent allowing the use of their data in clinical research. We used de-identified data previously recorded in the HNCH HIV cohort; as such patients were not contacted for this particular study. Institutional review boards at HNCH and the University of Alabama at Birmingham (UAB) approved this study.

Study Setting and Design

The HNCH is part of the Peruvian National HIV program and includes a cohort of over 1600 patients receiving HIV/AIDS care. This prospective cohort collects detailed socio-demographic, psychosocial, and clinical information on all patients enrolled in care. All HIV care is provided on-site, and is coordinated by a team of infectious disease specialists. For patients prescribed ART during the observation period from January 2006 through December 2010 (CD4<200 cells/ml or symptomatic), patients are scheduled for visits in three concurrent facets of the system: medication pick-up, laboratory appointments and clinic visits. Medications are dispensed free of charge, directly to patients by a team of nurses. Initially, however, a 1-week supply of ART supply is dispensed for the first 2 weeks of therapy and monthly supplies are dispensed thereafter if there are no complications. Patients are scheduled for separate laboratory monitoring (CD4 count and viral load) visits every six months (HIV genotyping is not routinely obtained). Patients also attend clinic visits for evaluation by an infectious disease provider at least once every six months. Both laboratory testing and clinic visit frequency take place as recommended by the Peruvian Ministry of Health guidelines [12]. Due to operational constraints such as the lack of coordination across environments and staffing limitations, medication pick-up, laboratory and clinic appointments often does not occur on the same date. All laboratory and clinical information is stored in an electronic database on site at the HNCH.

Eligibility Criteria

All treatment-naive patients 18 years or older and initiating ART between January 2, 2006 and December 30, 2010 were included. Treatment initiation was defined as the start of a first-line regimen (2 nucleoside reverse transcriptase inhibitors (NRTI) and 1 non nucleoside reverse transcriptase inhibitor (NNRTI)) [13] and continuation for ≥ 14 days.

Patients who discontinue ART during the first two weeks primarily do so for adverse side effects and were excluded from the analysis. Also excluded were pregnant women (As this patients normally switch to PIs containing ART regimens due to pregnancy and not due to failure), patients whose initial regimens included a protease inhibitor (PI; therefore non-first-line regimen per Peruvian National Guidelines) and patients purchasing their medications privately (hence not receiving standard issue regimens dispensed via the national treatment program).

Study variables

Socio-demographic, psychosocial, and clinical variables were retrieved by direct chart abstraction from clinic notes (J.L.S., J.L.A., J.P., F.M., D.C.G., D.C.), medication pick up details were retrieved from standardized ART dispensing documents while laboratory data and clinic visit dates were extracted via Microsoft Access queries to the HNCH database where all arrived laboratory visit dates are recorded. A number of independent patient-level variables included age, gender, baseline weight (kg), HIV acquisition risk factor (heterosexual, men who have sex with men [MSM]) and standardized body mass index (BMI): weight/(height in meters)². Socio-economic variables included travel distance from home district to hospital (in miles, using Google Maps: maps.google.com consulted on February 2011), and income (obtained from social worker assessments prior to ART initiation where poverty was stratified as earning < minimum wage and extreme poverty: < half the minimum wage) [14]. Clinical data included medical history of opportunistic infections, tuberculosis and sexually transmitted diseases (STD) prior to ART initiation. Laboratory variables included baseline and subsequent values for the following tests: HIV RNA (copies/ml), CD4 value (cells/ml), and hematocrit (%). Antiretroviral regimen variables were collected for both first-line regimens (defined as 2 NRTIs and an NNRTI) and second-line regimens (NRTIs in combination with a PI), if prescribed. We also constructed 2 clinical markers, CD4 and BMI progression, comparing baseline values with values at the beginning of second-line regimens: (1) CD4 progression: A positive progression was dichotomously defined as either achieving a CD4 count above 200 cells/ml in those starting with a CD4 < 200 or having a CD4 value higher than baseline at the end of the first-line in those starting with CD4 ≥ 200 (2) BMI progression: was dichotomously stratified as positive if there was any increase of BMI noted, otherwise was recorded as negative.

Three OPs, as previously validated, were constructed by the study statistician using pharmacy pick-up, laboratory completion and clinic arrived visit dates; OPs were measured only during the first-line period. They were:

1. 1. Medication possession ratio (MPR): days with antiretrovirals dispensed/days on first-line therapy [15];
2. 2. Laboratory monitory constancy (LMC): proportion of 6 months intervals in follow-up with ≥ 1 viral load (VL) or CD4 reported [16];
3. 3. Clinic visit constancy (CVC): proportion of 6 months intervals in follow-up with ≥ 1 clinic visit [4,17].

Dependent variables

Two primary study outcomes were decided a priori and included:

1. 1. First-line regimen persistence was measured as the time from first-line regimen initiation to switch to a second-line regimen (NRTIs and a PI) or to discontinuation of medications completely. Patients who had not discontinued their first-line regimen were censored at date of death, date lost to follow-up (30 days past expected medication run out date per drug pick-up data, as recommended by the Peruvian national guidelines), date of transfer of care or administratively at January 31^st, 2011 (end of observation period).
2. 2. Time to virologic failure of a second-line regimen, measured from second-line initiation (defined as start of a PI containing regimen) to the first viral load ≥ 400 copies/ml after 90 days of therapy. Patients who started a second-line regimen and had at least 1 follow-up viral load more than 90 days after the start of a second-line regimen were included. Patients who did not experience virologic failure were censored at the date of the last available viral load.

The lack of third-line drugs in Peru at the time these data were captured made the measurement of second-line regimen persistence impossible. We chose time to virologic failure as a surrogate of second-line persistence since it would be a reasonable indication to the time to switch to third-line regimens if these were available. We used 90 days instead of the more conventional 180 days as the cut-off for virologic failure as most patients lacked VL measured after 180 days, this definition has been used in other studies in the literature [18,19].

Statistical analysis

Descriptive statistics were employed to evaluate overall patient and regimen-level characteristics to ensure distributional assumptions for statistical tests were met. Continuous variables are reported as median [first quartile-third quartile] and categorical variables are reported as frequencies and percentages. Univariate analyses (UV) were performed to identify factors affecting initial regimen longevity and time to second-line regimen failure. Next, three multi-variable proportional hazard (PH) models (one per each OP) were fit for each of our study outcomes:

1. 1. Time to second-line ART initiation;
2. 2. Time to virologic failure of second-line ART.

Each model was adjusted for socio-demographic, clinical and laboratory variables. OPs were treated as time-varying covariates for first-line persistence assessments but fixed for second-line regimen time to virologic failure. We carried out sensitivity analyses using MPR as a continuous and as a categorical variable (using 85% as the cut off), both were time-varying measures over the preceding 180 days; thus, Hazard Ratios (HR) for these two analyses can be interpreted at any given time during the first-line period (≥180 days). Analyses were performed using SAS V9.1.3 software (SAS Institute, Cary, NC).

First-line persistence analyses

We included 856 patients in first-line regimen persistence analyses. Among these, age was 35.6 years [interquartile range (IQR) 29.4-42.9]; most were male (n=624, 73%) and baseline CD4 was 95 cells/ml (IQR 42-200) (Table 1). In UV analysis, MPR (per 10% increase) (HR=0.69; 95% confidence interval=0.60-0.70) and LMC (per 10% increase) (HR=0.88; 0.70-0.90) were associated with prolonged first-line persistence while CVC was not (HR=0.94; 0.84-1.04). Baseline CD4 count (per 50 cells/ml) was the only laboratory variable value associated with a decreased hazard of first-line regimen discontinuation (HR=0.87; 0.77-0.97). All other demographic and laboratory variables did not meet statistical significance.

For first-line regimen persistence we fit three multivariable proportional hazard models, one for each OP (Table 2). The first model, included MPR, showed that 10% increments in MPR (HR=0.66; 0.61-0.71) were associated with decreased hazard of switch to a second-line regimens while demographic, clinical or laboratory characteristics failed to show an association with prolonged time on first-line regimens. We then performed sensitivity analyses using MPR. When measured as a continuous variable over the preceding 180 days, 10% increases in MPR were associated with a decreased risk of switching to a second-line regimen (HR=0.71; 0.66-0.77) and when used as a categorical variable, A MPR < 85% in the previous 180 days had an increased risk of switching to a PI based regimen (HR=22.8; 11.02-47.16).

In the second model, fit for LMC, only this OP (HR=0.83; 0.71-0.96) was associated with prolonged first-line regimen persistence while other demographic, clinical and laboratory characteristics were not associated with increased first-line longevity in this model. In the third model, CVC showed a trend towards longer first-line regimens (HR=0.91; 0.79-1.05) but failed to reach statistical significance. No other demographic, clinical or laboratory variables in this model were associated with this outcome.

Analysis of time to second-line virologic failure

We included 79 individuals switching to a PI including regimen in the analyses of factors associated with second-line regimens virologic failure (See table 1 for baseline characteristics). In UV analyses, only MPR (HR=0.89; 0.80-0.99) reached a statistically significant association with prolonged time to virologic failure in these regimens, other two variables: history of opportunistic infections (HR=0.35; 0.15-0.83) and STDs (HR=0.35; 0.13-0.92) were also associated with prolonged time to virologic failure of the second-line regimen, but we suspected these were not clinically significant and we did not include them in subsequent analyses.

We again fit three separate multivariable Cox PH models, one including each OP (Table 3). In the model that included MPR, higher MPR values (per 10%) were associated with a decreased hazard of second-line regimen virologic failure (HR=0.88; 0.77-0.99) while no demographic or clinical variables were associated with time to second-line virologic failure. The models fit for LMC and CVC failed to show any statistically significant associations with this outcome (Table 3).