Predicting virologic failure in an HIV clinic.

Abstract

We sought to use data captured in the electronic health record (EHR) to develop and validate a prediction rule for virologic failure among patients being treated for infection with human immunodeficiency virus (HIV). We used EHRs at 2 Boston tertiary care hospitals, Massachusetts General Hospital and Brigham and Women's Hospital, to identify HIV-infected patients who were virologically suppressed (HIV RNA level < or = 400 copies/mL) on antiretroviral therapy (ART) during the period from 1 January 2005 through 31 December 2006. We used a multivariable logistic model with data from Massachusetts General Hospital to derive a 1-year virologic failure prediction rule. The model was validated using data from Brigham and Women's Hospital.We then simplified the scoring scheme to develop a clinical prediction rule. The 1-year virologic failure prediction model, using data from 712 patients from Massachusetts General Hospital, demonstrated good discrimination (C statistic, 0.78) and calibration (chi(2)= 6.6; P= .58). The validation model, based on 362 patients from Brigham and Women's Hospital, also showed good discrimination (C statistic, 0.79) and calibration (chi(2)= 1.9; P= .93). The clinical prediction rule included 7 predictors (suboptimal adherence, CD4 cell count < 100 cells/microL, drug and/or alcohol abuse, highly ART experienced, missed > or = 1 appointment, prior virologic failure, and suppressed < or = 12 months) and appropriately stratified patients in the validation data set into low-, medium-, and high-risk groups, with 1-year virologic failure rates of 3.0%, 13.0%, and 28.6%, respectively. A risk score based on 7 variables available in the EHR predicts HIV virologic failure at 1 year and could be used for targeted interventions to improve outcomes in HIV infection.