Evaluations of behavioural interventions designed to prevent transmission of HIV/sexually transmitted infections frequently rely on self-reported condom use data to measure success, yet such data are often acknowledged to be potentially inaccurate, either owing to faulty recall or social desirability bias.^1,2 Although research techniques to improve the validity of self-reported data on sexual behaviour have been recommended,³ their ability to reduce information bias has not been shown. Until recently, the objective assessment of research methods designed to improve the validity of self-reports of sex and condom use has not been possible. However, verifying recent unprotected sex is now feasible by testing women for prostate-specific antigen (PSA), a protein found in high concentrations in semen and detectable in vaginal secretions from women up to 48 h after unprotected sex.^4–8

We conducted a randomised controlled trial to determine whether the process of informing participants about PSA testing would reduce reporting bias. We compared women who knew that their vaginal specimens were to be tested for PSA before they completed the questionnaire on sex and condom use with women who did not have this knowledge in advance. We hypothesised that those with advance knowledge of the PSA testing would report more unprotected sex and would have better agreement between the PSA test results and their self-reported data than those without advance test knowledge.

METHODS

The population consisted of 210 female sex workers associated with a peer education programme designed to reduce transmission of HIV/sexually transmitted infections in Mombasa, Kenya. The trial took place during the baseline measurement of a larger study examining the effects of introducing the female condom in a peer education programme with sex workers.⁹ The institutional review boards of the sponsoring institution and Kenyatta National Hospital, Nairobi, Kenya, approved the study.

Sex workers who fulfilled the eligibility criteria were randomised into two groups of equal sizes using a previously created randomisation list. Both the randomisation process and the group assignments were unblinded. The first group provided written informed consent for both the questionnaire and the PSA test before we administered the sex and condom use questionnaire and took vaginal specimens for PSA testing. By contrast, the second group was first asked for informed consent for the questionnaire and were told that they would be asked to participate in a second study only after they had completed the questionnaire. At that point, we requested informed consent for the PSA test. The informed consent for the PSA testing advised participants that if they had sex without a male or female condom or if they used the condom incorrectly, semen could be detected for 1–2 days afterwards. Although we informed participants that we would test for this, we did not assess the participants’ comprehension of the study procedures or objectives. For both groups, a trained nurse used pictures to instruct participants on sample collection and then left the women to self-swab in private. Specimens were air dried and sent by air to a laboratory in the US for analysis. We identified PSA using a microparticle enzyme immunoassay (IMx, Abbott Laboratories, Abbott Park, Illinois, USA).

We examined two hypotheses: (1) whether the group with advance knowledge of the PSA test would report more unprotected sex for the past 48 h than the group without advance knowledge and (2) whether the group with advance knowledge would have less disagreement between their PSA results and self-reports of unprotected sex. We measured only one type of disagreement: reporting no sex or protected sex for the past 48 h, yet testing positive for PSA. We did not assess disagreement resulting from reporting exposure yet testing negative for PSA because other factors could explain the absence of PSA, such as the effect of practices such as douching or the marker’s short clearance time. Although PSA can be detected with high sensitivity (98–100%) immediately after semen exposure, the antigen clears rapidly from the vagina.^4,5 Macaluso et al⁴ found PSA in only 29% of specimens collected from women 24 h after clinical insemination with 1 ml of their partner’s semen and in 3% of specimens at 48 h. We tested the hypotheses with one-sided χ² tests.

RESULTS

All of the 210 eligible women who were recruited to the study agreed to participate. Two participants in the group without the advance knowledge of PSA testing declined to self-swab at baseline; however, both contributed a swab for PSA testing at the follow-up visit. Owing to missing test results, we have PSA data for 194 participants.

We found no difference between the advance knowledge (n = 105) and no advance knowledge groups (n = 105) in the proportion of women who reported exposure to unprotected sex for the past 48 h (14.3% and 11.4%, respectively; p = 0.27). A greater proportion of those who knew about the PSA test in advance (15.8%; n = 15) reported no exposure for the past 48 h and yet tested positive for PSA than those without the advance knowledge (9.1%; n = 9). However, the odds of having this type of discordant self-reported and PSA data did not differ by randomisation group (p = 0.08; OR 1.9; 95% CI 0.8 to 4.5).

DISCUSSION

Women who knew about the PSA test in advance did not report more exposure than those without advance knowledge. Surprisingly, more women with the advance test knowledge had discordant self-report and PSA results than the comparison group; however, this difference was not statistically significant. Whether respondents who know that their responses can be verified would be less likely to misrepresent their sex and condom use to researchers is unknown. However, these results do not support the assumption that such information necessarily would motivate individuals to provide more accurate responses. Other psychosocial and cultural factors—besides the desire to present oneself in a positive light—could motivate a person to respond inaccurately. For example, participants could become confused while answering questionnaires about what they are “supposed to do”,¹⁰ or they could lie on purpose to “test” the system. Owing to the small sample size, we recommend that similar studies using PSA be conducted to determine the directionality of the effects (if any) on accuracy of self-report of sex and condom use derived from using research methods (eg, advance knowledge of biological testing or audio computer-assisted self-interviewing) that are believed to reduce social desirability bias.