Comparative Effectiveness of First-Line Antiretroviral Therapy

Results From a Large Real-World Cohort After the Implementation of Dolutegravir

Mariana Veloso Meireles; Ana Roberta P. Pascom; Elisabeth C. Duarte; Willi McFarland


AIDS. 2019;33(10):1663-1668. 

In This Article

Materials and Methods

Data Sources

Data originate from three BMoH information systems: the antiretroviral drug dispensing system (SICLOM), which registers every ART prescription; the HIV laboratory examination system (SISCEL), which registers CD4+ cell count and viral load tests performed within the country's public health system (SUS); and the notifiable diseases information system (SINAN), which contains HIV/AIDS case reports. SICLOM and SISCEL capture information at provider level, with real-time data available to the BMoH. SICLOM covers virtually 100% of patients on treatment, as antiretroviral drugs are not sold privately in Brazil. SICLOM and SISCEL share a common registration database with a single unique identifier. SINAN de-duplicates cases once a year at the national level, followed by a probabilistic linkage method to link cases in SINAN to SICLOM/SISCEL.[16,17] Underreporting in SINAN is estimated at 30%.[16] In this study, SINAN case reports were available through 30 June 2017; ART prescriptions, viral load and CD4+ cell counts were available from SICLOM/SISCEL through 31 August 2018.

Inclusion/Exclusion Criteria

We included antiretroviral-naive adults (age 15–80 years) who started ART between January 2014 and June 2017 and had a viral load in SISCEL at 365 (±90) days after treatment initiation. We excluded patients who had undetectable (<50 copies/ml) viral load at baseline due to uncertainty as to whether they were ART-naive.


Our outcome variable was viral suppression (<50 copies/ml) at 12 months (365 ± 90 days) from treatment initiation. If an individual had more than one viral load in the period, the first after 365 days was used. If no viral load was recorded after 365 days, then the one closest to 365 days was used.

We tested hypotheses for the effectiveness of different ART regimens in achieving suppression compared with TLE as the standard. Patients were categorized using their first prescription in SICLOM, analogous to 'intent-to-treat'. The most common regimens (≥1000 individuals) were reported individually; the remaining were grouped as 'other regimens'. Potential confounders were year of ART initiation, sex/exposure category, age, education, race ('color' in the Brazilian context), region, baseline CD4+ cell count and viral load, and ART adherence. Baseline CD4+ cell count was defined as the count recorded within -100 to +20 days; baseline viral load was the count within -100 to 0 days from treatment initiation.

Adherence was calculated from pharmacy refill data. SICLOM registers the duration the prescription covers, typically 30–90 days. An adherence indicator was calculated for each individual, based on the dates and durations of prescriptions and date of the outcome:

Whenever the duration of the last prescription exceeded the interval until the viral load measurement, that duration was recategorized to the specific number of days in the interval, avoiding overestimation of adherence. Adherences above 100% were observed because patients often collected their drugs before the expected date.

Associations for the likelihood of achieving suppression at 12 months were assessed using univariable and multivariable logistic regression. We present proportions within groups, crude [odds ratio (OR)] and adjusted ORs (aORs) with 95% confidence intervals (95% CIs). Variables with P value less than 0.20 in univariable analyses were included in the multivariable analysis. We used stepwise backward selection with P value less than 0.05 for retention and the cutoff for significance.

The primary analysis used an intention-to-treat-analogous approach, ignoring discontinuations or regimen changes. Secondary analysis did a per-protocol analogous approach, including only patients who did not switch regimens and who had high adherence, defined as 95–105% (adherences above 105% appeared more subject to data input errors). To assess potential indication bias resulting from changing guidelines, we conducted sensitivity analyses excluding patients who started TLE in 2017 (TLE was recommended for patients with nonsevere tuberculosis), female patients who started TLE in 2016 and 2017 (TLE was recommended for pregnant women), and patients with unknown baseline CD4+ cell count, and combinations of the previous criteria. We also assessed potential bias resulting from some regimens being used more frequently in certain subgroups.