Objective: To develop and validate a claims signature model that estimates proportions of HIV-infected patients in administrative claims databases who switched combination antiretroviral therapy (cART) regimens because of virologic failure.
Methods: The authors used an HIV-specific registry (development data set) to develop logistic regression models to estimate odds of virologic failure among patients who switched cART regimens. Models were validated in a sample of administrative claims with laboratory values (validation data set). The final model was applied to an application data set as a worked example.
Results: There were 1691, 1073, and 3954 eligible patients with cART switches in the development, validation, and application data sets, respectively. In the development data set, virologic failure before a switch was observed 21.8% of the time. Failure more likely caused the regimen switch among patients who were treatment experienced, had been receiving their baseline regimen for > 180 days, had ≥ 2 or more physician visits within 90 days, had > 1 HIV RNA or CD4 cell count test within 30 days, had any resistance test within 180 days, or had a change in regimen type. The final model had good discriminatory ability (C = 0.885) and fit (Hosmer-Lemeshow P = 0.8692). Failure was estimated to occur in 18.9% (v. 18.6% observed) of switches in the validation data set and 13.8% in the application data set.
Conclusions: This claims signature model allows payers to use claims data to estimate virologic failure rates in their patient populations, thereby better understanding plan costs of failure.