Standardization of Rates Using Logistic Versus Hierarchical Modeling

• Activity Number: 136
• Type: Contributed
• Date/Time: Monday, August 10, 2015 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Jun Zhang* and Patricia Dietz and Michelle Van Handel and Samah Hayek
• Companies: CDC and CDC and CDC and CDC
• Keywords: HIV testing ; Rate ; statistical models

Abstract

In program evaluation, a frequent challenge is the comparison of rates from different populations to assess performance. Different methods can be used to standardize the rate. Hierarchical model has been advocated for analysis when data have a clustered structure. We compared two different methods of standardizing estimated state-level HIV testing percentages using 2011 Behavioral Risk Factor Surveillance System data. Both conventional and hierarchical logistic regression models were fit to estimate the percentage of persons who have ever been tested for HIV adjusting for sex and age by state. Based on conventional logistic regression, HIV testing percentages ranged from 26.4% (Utah) to 74.1% (Washington, D.C). From hierarchical logistic regression, HIV testing percentages ranged from 25.3% (Utah) to 73.1% (D.C.). The average difference in the HIV testing percentages between the two methods was 0.15%, the maximum difference was 1.1% (Utah) and 1.0% (D.C.) and the minimum difference was 0.002% (New Mexico). Our analysis found minimal differences between the two methods for standardizing most clusters. However, state ranking for HIV testing differed for each method.