Validation of an algorithm to identify antiretroviral-naïve status at time of entry into a large, observational cohort of HIV-infected patients.

ABSTRACT

PURPOSE: Large, observational HIV cohorts play an important role in answering questions which are difficult to study in randomized trials; however, they often lack detailed information regarding previous antiretroviral treatment (ART). Knowledge of ART treatment history is important when ascertaining the long-term impact of medications, co-morbidities, or adverse reactions on HIV outcomes. METHODS: We performed a retrospective study to validate a prediction algorithm for identifying ART-naïve patients using the Veterans Aging Cohort Study's Virtual Cohort-an observational cohort of 40 594 HIV-infected veterans nationwide. Medical records for 3070 HIV-infected patients were reviewed to determine history of combination ART treatment. An algorithm using Virtual Cohort laboratory data was used to predict ART treatment status and compared to medical record review. RESULTS: Among 3070 patients' medical records reviewed, 1223 were eligible for analysis. Of these, 990 (81%) were ART naïve at cohort entry based on medical record review. The prediction algorithm's sensitivity was 86%, specificity 47%, positive predictive value (PPV) 87%, and negative predictive value 45%, using a viral load threshold of <400 copies/ml. Sensitivity analysis revealed that PPV would be maximized by increasing the viral load threshold, whereas sensitivity would be maximized by lowering the viral load threshold. CONCLUSIONS: A prediction algorithm using available laboratory data can be used to accurately identify ART-naïve patients in large, observational HIV cohorts. Use of this algorithm will allow investigators to accurately limit analyses to ART-naïve patients when studying the contribution of ART to outcomes and adverse events.