Theses Doctoral

Measuring HIV Exposure amongst Men who have Sex with Men in the USA: Implications for Risk Assessment in HIV Prevention Studies

Austin, Judith Florence

In the context of decreasing mortality and increasing prevalence, prevention of HIV-transmission represents a public health priority. In the United States, the majority of infections are sexually-acquired, with men who have sex with men and minorities disproportionately affected. Although a number of promising biomedical prevention approaches have emerged over the past decade, a further 20 years could be needed before a suitable product becomes widely available. Evidence from vaccine and microbicide trials has shown that success in one population may not be replicated in another. To understand surprising or unexpected results, investigators need chronologically concordant evidence of both study product adherence and viral exposure. Since exposure to HIV cannot yet be independently verified, in seeking to measure this variable, investigators target the sexual behaviors through which it takes place deriving data for these surrogate measures from study participants' voluntary self-reports.
Likely sources of reporting bias and efforts to minimize this phenomenon in the context of HIV-prevention research are critically reviewed in Chapter 1. Research describing the role of cognitive and affective functioning in the preparation of responses to potentially threatening questions is examined. Studies investigating techniques such as the use of colloquial language to facilitate comprehension, or variation in the length of the reference period to enhance recall are explored. Research comparing the effect of mode of administration on the amount of proscribed behavior reported - widely believed to correlate with validity - is reviewed. Contextual factors facilitating versus inhibiting disclosure of sensitive information are examined. Finally, risk-behavior measurement approaches used in selected HIV-prevention trials are inspected. Thereafter, the dissertation focuses on the properties of risk-assessment items, formulated specifically to elicit Global recall over six months, or Event-Specific (episodic) memory for selected recent episodes of limited duration, to capture sexual behaviors or temporally related activities. The capacity of the different questionnaire item formats to elicit responses with sufficient construct validity to serve as proxies for HIV-exposure is examined. Data for these studies are drawn from a large randomized controlled trial of a behavioral intervention to prevent HIV-transmission among men who have sex with men. Using a subset of 1295 cases and controls, models with good discriminant validity for HIV are derived separately for the Global and Event-Specific items. Thereafter, selected items from the two formats are combined to produce a single model with excellent discriminant validity, suggesting that these items can adequately represent true HIV-exposure.
Next, a preliminary investigation of the contribution of psychosocial items to the predictive model based on exposure measures is undertaken. Specifically, interaction with exposure measures and the increase in discriminant validity obtained using data derived from constructs of partner type/relationship status, substance use, depressive symptoms and perceived self-efficacy is examined first in stratified analysis and then in logistic regression analysis using the case-control data. Effect-modification is observed for perceived relationship status and non-injection drug use. Evaluation of psychosocial items continues in a cohort study with prospective analysis of follow-up data from all trial participants who returned for at least one follow-up visit. Informed by the case-control study, a series of items representing psychosocial constructs known for their association with HIV-infection are tested for main effects and effect-modification. Evidence of the interaction observed in stratified analyses and confirmed in ordinary logistic regression persists in separate, topic-specific GEE analyses with assorted exposure measures, but abates in repeated measures analyses drawing on all available psychosocial items. Lastly, a single lagged variable indicating primary relationship status of the most recent partner (with respect to the preceding study visit) provides a significant addition to the model. Significant main effects for all except depressive symptoms and perceived self-efficacy and the increase in discriminant validity obtained for the multivariable model versus the `exposure-only' model are sufficient to warrant continued use of these risk-assessment items.
Despite good predictive validity demonstrated for the sexual risk-behavior and psychosocial items, some inconsistent reporting between the Global and Event-specific formats is evident. Likely sources of this reporting bias are considered in light of the literature, and strengths and limitations of the overall study are discussed in the closing chapter.


  • thumnail for Austin_columbia_0054D_12604.pdf Austin_columbia_0054D_12604.pdf application/pdf 1.88 MB Download File

More About This Work

Academic Units
Thesis Advisors
Koblin, Beryl A.
Ph.D., Columbia University
Published Here
April 20, 2015