Background: The identification of promising interventions to reduce HIV transmission depends on the availability of reliable self-reports of sexual risk behaviors (SRB). Reliability of SRB data is influenced by characteristics of items (e.g. item type, response scale) and respondents (e.g. stigma, openness, outness, sexual communication). This study utilized data obtained from two different but equivalent item types (aggregate vs. encounter-specific) to evaluate SRB data reliability.
Methods: Data were obtained from 223 HIV+ MSM at the baseline observation of a behavioral intervention trial. An audio computer-assisted self-interviewing (ACASI) instrument was used to gather SRB in two ways. First, participants provided aggregate counts of specific behaviors for the prior 30 days. Second, participants provided information on 5 specific sexual encounters during the same period. Data from both item types was compared for the number of sexual encounters (NSEX), the number of sexual partners (NPART), and the number of encounters involving unprotected receptive/insertive anal intercourse (URAI, UIAI).
Results: The proportion of participants with consistent SRB (between aggregate and encounter-specific items) was highest for data on the number of sexual encounters (57%) and the number of sexual encounters (56%). Reporting consistency was lowest for UIAI (31%). Participants reported higher SRB frequencies on the encounter-specific items in all cases except NPART, where values were higher for the aggregate items. Higher HIV stigma scores and lower outness scores were related to inconsistencies in NSEX and NPART. Inconsistencies in reports of URAI/UIAI were not significantly related to any of the tested respondent characteristics.
Conclusions: Poor reliability in SRB data obtained from participants in trials of HIV prevention interventions could lead to poor decision-making regarding intervention effectiveness. Comparisons of data obtained from aggregate and encounter-specific items yielded potentially significant inconsistencies in self-reported SRB. More work is needed to understand how item and respondent characteristics influence SRB data reliability.