Relative risk for HIV in India –An estimate using conditional auto-regressive models with Bayesian approach.

Chandrasekaran, K.; Kaushik, G.

Spatial and Spatio-temporal Epidemiology; 2017; 20: 27–34.

Abstract: Indian states are currently classified into HIV-risk categories based on the observed prevalence counts, percentage of infected attendees in antenatal clinics, and percentage of in-fected high-risk individuals. This method, however, does not account for the spatial depen-dence among the states nor does it provide any measure of statistical uncertainty.

We provide an alternative model-based approach to address these issues. Our method uses Poisson log-normal models having various conditional autoregressive structures with neighborhood-based and distance-based weight matrices and incorporates all available co-variate information. We use R and WinBugs software to fit these models to the 2011 HIV data. Based on the Deviance Information Criterion, the convolution model using distance-based weight matrix and covariate information on female sex workers, literacy rate and intravenous drug users is found to have the best fit.

The relative risk of HIV for the various states is estimated using the best model and the states are then classified into the risk categories based on these estimated values. An HIV risk map of India is constructed based on these results. The choice of the final model suggests that an HIV control strategy which focuses on the female sex workers, intravenous drug users and literacy rate would be most effective.

Keywords: HIV; Relative risk; Disease mapping; Spatial model; CAR models; Bayesian approach

Back to List of publications / Home