Comparing models of HIV transmission risk on concurrent HIV/AIDS diagnosis using multiple imputation and other methods for handling missing risk data, New York City 2008

Background: In New York City (NYC), ~25% of persons newly diagnosed with HIV are diagnosed concurrently (≤31 days) with AIDS, of whom >30% have missing transmission risk information. To identify factors related to late HIV diagnosis, we evaluated the impact of different methods for handling missing risk data on the association between transmission risk and concurrent HIV/AIDS diagnosis. Methods: Conventional methods for handling missing data, case deletion and indicator, were compared with the Centers for Disease Control and Prevention's (CDC) multiple imputation (MI) method for risk among adults reported with HIV diagnosed during 2008 (reported through 9/2009). Transmission risk was classified using CDC-defined categories. Odds ratios (OR) and 95% confidence intervals (CI) were estimated using logistic regression, stratified by sex and age quartiles and adjusted for race, country-of-birth, borough of residence, incarceration history, and poverty. Results: Among 3,801 new diagnoses, 24.7% had concurrent HIV/AIDS. In multivariable models, transmission risk was associated with concurrent diagnosis using the MI approach only. Compared to heterosexuals, male injection drug users (IDU) ages 25-33 and 43+ years were more likely to be concurrently diagnosed (OR=3.5, 95%CI=1.3-9.3; OR=2.2, 95%CI=1.4-3.3; respectively). Similarly, female IDUs were at increased odds among all age quartiles, most strikingly among ages 13-24 (OR=7.0, 95%CI=1.4-34.7) and 25-33 (OR=5.3, 95%CI=1.9-14.4). Conclusions: MI enabled detection of an association between transmission risk and concurrent diagnosis, a finding absent from analyses using conventional missing data methods for transmission risk. Application of MI can identify groups at-risk for concurrent diagnosis, and could therefore be useful for intervention/prevention planning.