OBJECTIVES: When studying HIV infection, administrative claims databases can provide information on treatment and cost for large numbers of patients but most do not contain laboratory test results, making it impossible to know when a change in antiretroviral treatment (ART) was due to virologic failure. Our objective was to develop and validate a model for identifying patients in a claims database who switched ART due to virologic failure. METHODS: We identified three databases of adult HIV-positive patients who switched ART regimens between January 1, 2003 and March 31, 2008.  The HIV Insight clinical registry was used to develop a logistic regression model incorporating demographics, regimen characteristics, and other independent variables to estimate the odds of virologic failure.  Next, the subset of the Ingenix i3 LabRx health insurance claims database with HIV viral load test results (claims/lab database) was used to validate the model.  The model was then used to estimate the proportion of patients with virologic failure in the full Ingenix i3 LabRx claims database (claims database). RESULTS: There were 1,691 patients with ART switches in HIV Insight; 1,073 in the claims/lab database, and 3,954 in the claims database. The base model (main effects only) had good discriminatory ability (c=0.875); but poor overall model fit (Hosmer Lemeshow test, p<0.001). Adding three significant two-way interaction terms improved fit (p=0.8692) and discriminatory ability (c=0.885).  When the final model was applied to the claims/lab database, it predicted 18.9% of patients would have virologic failure; the actual proportion was 18.6%. CONCLUSIONS: We developed and validated a model that could be used in administrative claims to predict the proportion of ART switches due to virologic failure. Health plans may use this model to identify treatments with rising rates of virologic failure and to examine costs related to such failure.