October 21, 2015

To improve public health, we must first determine what puts people at risk for disease and what promotes good health. This begins with accurately classifying who is most at risk for certain diseases and which behaviors are most likely to put someone at risk. A recent article, To Be or Not to Be: Bayesian Correction for Misclassification of Self-reported Sexual Behaviors Among Men Who Have Sex with Men, by Neal Goldstein, PhD’15, Seth Welles, PhD, ScD, and Igor Burstyn, PhD has gained national attention in the field of epidemiology because the authors were able to quantify misclassification resulting from common approaches used to estimate HIV risk. The article was published in the September issue of Epidemiology.

A common method for assessing risk is to simply ask people to self-identify and report personal characteristics or behaviors (for example when a physician asks a patient how often they exercise). However, in the case of assessing risk for infection with HIV, this can be challenging. Stigma and discrimination regarding certain sexual preferences and behaviors can make it difficult to determine true risk through self-reports. Since most new infections in the United States are found among men who have sex with men (MSM) this limitation in self-reports can have a major impact on understanding the distribution and level of risk.

To improve risk estimates among this population, Goldstein et al reanalyzed survey data from two previously published studies and compared three different proxy measures of sexual behavior: self-reported sexual identity (heterosexual/straight versus homosexual/gay), the gender of sexual partners, and specific sexual acts. This analysis showed that a person’s responses to these three measures can provide discordant estimates of risk; and asking a person’s sexual identity was in fact a poor indication of sexual risk when compared to asking about sexual partners or sexual acts among this population. For example, in one study of MSM, 43.8 percent reported not having male sex partners, but subsequently reported specific sexual acts such as oral and/or anal sex with other men that might put them at risk for infection. The authors used Bayesian modeling that takes estimates of the extent of misclassification of sexual orientation from one study where multiple measures of sexual behaviors were collected, and mathematically applies these to other studies that utilized imperfect measures of sexual orientation. After correcting for the misclassification, estimates of HIV/STI risk associated with sexual orientation, for example, are readjusted using the corrected measures of sexual orientation. The resulting estimates of risk often differ markedly from the initially reported estimates. 

Improving the accuracy of risk estimates, particularly in areas where self-report is known to be challenging, improves the accuracy of planning public health interventions that are based on risk. Better targeting of interventions can save valuable resources and improve health outcomes. Conversely, risk estimates that are biased or inaccurate make planning difficult, resources are wasted, and prevention can be compromised.

The approach suggested by Goldstein et al can significantly impact HIV risk assessment and planning. In addition, the methods from this study can be applied to correcting risk estimates in a multitude of other areas that face similar challenges with accurate self-reporting, such as dietary habits, mental health issues, or drug use. Making the modeling methods proposed by Goldstein et al accessible to agencies and public health departments can improve prevention planning and resource allocation.