Is Audio Computer-Assisted Self-Interview (ACASI) Useful in Risk Behaviour Assessment of Female and Male Sex Workers, Mombasa, Kenya?

Background Audio computer-assisted self-interview (ACASI) may elicit more frequent reporting of socially sensitive behaviours than face-to-face (FtF)-interview. However, no study compared responses to both methods in female and male sex workers (FSW; MSW) in Africa. Methodology/Principal Findings We sequentially enrolled adults recruited for an HIV-1 intervention trial into a comparative study of ACASI and FtF-interview, in a clinic near Mombasa, Kenya. Feasibility and acceptability of ACASI, and a comparative analysis of enrolment responses between ACASI and FtF on an identical risk assessment questionnaire were evaluated. In total, 139 women and 259 men, 81% of eligible cohort participants, completed both interviews. ACASI captured a higher median number of regular (2 vs. 1, p<0.001, both genders) and casual partners in the last week (3 vs. 2, p = 0.04 in women; 2 vs. 1, p<0.001 in men). Group sex (21.6 vs. 13.5%, p<0.001, in men), intravenous drug use (IDU; 10.8 vs. 2.3%, p<0.001 in men; 4.4 vs. 0%, p = 0.03 in women), and rape (8.9 vs. 3.9%, p = 0.002, in men) were reported more frequently in ACASI. A surprisingly high number of women reported in ACASI that they had paid for sex (49.3 vs. 5.8%, p<0.001). Behaviours for recruitment (i.e. anal sex, sex work, sex between males) were reported less frequently in ACASI. The majority of women (79.2%) and men (69.7%) felt that answers given in ACASI were more honest. Volunteers who were not able to take ACASI (84 men, and 37 women) mostly lacked reading skills. Conclusions/Significance About 1 in 5 cohort participants was not able to complete ACASI, mostly for lack of reading skills. Participants who completed ACASI were more likely to report IDU, rape, group sex, and payment for sex by women than when asked in FtF interview. ACASI appears to be a useful tool for high risk behaviour assessments in the African context.


Introduction
Audio computer-assisted self-interview (ACASI), which is known to capture some sensitive behaviours more reliably in high risk populations in developed nations [1], has not been evaluated in populations recruited for HIV-1 prevention trials in Africa. As such populations are purposefully selected for presumed 'high-risk' behaviour, interviewer attitudes may impact socially desirable responses of the interviewee, especially because socially stigmatised behaviours such as male same sex behaviour, or anal sex practice, are rarely assessed in Africa [2,3].
Research on computerized interviewing in the USA has shown that replacing the interviewer with a computer can provide conditions, including privacy and the perception of anonymity, that facilitate reliable and frank reporting, thereby increasing reports of sensitive behaviour in surveys of the general population [4,5], adolescents [1,6], and injection drug users (IDU) [7,8]. In addition to reducing social desirability bias, ACASI has several advantages over face-to-face (FtF)-interview: it standardises data collection by using recorded audio question tracks and captures data directly into a database, allowing for automated skip patterns and built in logic checks. The use of ACASI has been recommended for socio-behavioural research in developing countries [1,9].
Perhaps due to its cost and technological complexity, the use of ACASI in African research studies has been low. Comparison studies of ACASI and FtF-interview in the general population in Zimbabwe and in 4 other developing countries suggested that participants are at ease with completing ACASI, and preferred a computer for answering sensitive questions, although low literacy may pose problems [10,11]. Among adolescents in Kenya, substantial differences between interview methods were reported for questions related to premarital sex, although not always in the expected direction [12,13]. Ongoing studies targeting adolescents in rural Malawi aim to evaluate which method of interview elicits increased reports of sexual risk behaviour, including an assessment of the impact of order of the interview mode on responses [14].
To our knowledge, no study has compared responses by ACASI with FtF-interview in high risk populations recruited for HIV-1 prevention trials in Africa. In Mombasa, Kenya, male and female sex workers (MSW, FSW) have been invited to join an ongoing cohort study to monitor HIV-1 seroincidence in preparation for studies of new HIV-1 prevention interventions [15,16]. Previously, at enrolment to the cohort study, HIV risk behaviours have been established through FtF-interview. For this study, we invited cohort participants to also take ACASI at their enrolment visit, and compared risk behaviour in volunteers who had undergone both ACASI and FtF-interview. Our study was exploratory and aimed at hypothesis generation and assessment of contextual acceptability and feasibility of ACASI in FSW and a large group of MSW who have sex with men. Both sex work and homosexual sex are illegal in Kenya. We assumed ACASI to be helpful in detecting behaviours that may be under-reported in FtFinterview due to social desirability bias. We also assumed that the order of the interview mode might influence responses given in either method. Hence we crossed over the order of interview methods. All interviews were taken on the same day and prior to HIV-1 testing.
Socio-demographic characteristics and laboratory confirmed STI for participants and non-participants are shown in table 1. HIV-1 prevalence did not differ between participating and nonparticipating men and women. Not surprisingly, both nonparticipating genders had a similar low education attainment. Non-participating men presented more often with non-specific urethritis at enrolment, and a larger proportion was unemployed. There were no differences in key risk behaviour comparisons established in FtF interview (i.e. gender of sex partner, anal sex, been paid, or having paid for sex, partner counts, group sex, rape, and IDU) between participating and non-participating men. Nonparticipating women presented more often with syphilis at enrolment, and a larger proportion had no prior knowledge of their HIV-1 status. Non-participating women more frequently reported anal sex in the past 3 months (62.2 vs. 16.8%, p = 0.05), and recalled a higher number of recent casual partners (median 4 vs. 2, p = 0.003).
A total of 259 men and 139 women completed both interview methods, and were similar in age, education attainment, and unemployment status, although more men had formal employment (table 1). The majority (73.0%) of men did not know their HIV-1 status while 66.9% of women had taken an HIV test before.

Acceptability of ACASI
An almost equal proportion, 33.3% in women, and 30.8% in men, had used a computer previously (table 2). Over 90% in both groups found the questions easy to understand and were comfortable in answering. Overall, almost 80% of women and 70% of men felt that answering questions in ACASI would give more honest answers, 11.7% of women, and 16.0% of men felt the methods were about the same, and 9.2% of women, and 14.3% of men felt answers in ACASI would be less honest than in a FtFinterview, p = 0.16. These proportions were similar regardless of the order of methods, with the exception that no woman who took ACASI after FtF believed that ACASI would provide less honest answers (data not shown). Mean duration to complete ACASI was almost double the duration of the FtF-interview for both women and men (31.2 vs. 15.9 minutes for men; 35.7 vs. 18.9 minutes for women, p = 0.001). Women took longer than men to complete both FtF-and ACASI-interviews, by 3 and 5 minutes, respectively. Men and women experienced the duration of ACASI as just right (67.1 vs. 62.3%) or too short (11.9 vs. 17.5%), while 20% of both groups felt that ACASI took too long.
Comparisons of sexual risk behaviour, substance misuse, and violence ACASI captured a significantly higher median number of casual sex partners in women (3 vs. 2, table 3) and men (2 vs. 1, table 4), with excellent agreement between both interview methods in women, and good agreement in men. Agreement between methods for reported regular sex partners for women and men was poor. Bland Altman plots (figure 1) show the average of the reported number of partners (regular or casual) by both methods versus the difference between reported number of partners by ACASI minus FtF-interview for women and men. For both men and women, ACASI was more sensitive in capturing a higher regular partner number (2 or more in the last week), while FtFinterview captured a slightly higher report of regular partner numbers when lower numbers (,2) were reported. Bland-Altman plots demonstrate that for each increase in average partner number participants reported more regular partners by ACASI (figure 1, A and B). For casual partners, there appears to be no such increase between methods (figure 1, C and D).
Reported partner gender preference for women and men was similar in ACASI and FtF-interview. Fewer women reported anal sex practice (39.6 vs. 43.9, p = 0.15), and fewer women reported having received cash for sex (87.1 vs. 95.7%, p = 0.01) in ACASI than in FtF-interviews; these behaviours were recruitment criteria for cohort enrolment. Almost half of the women reported having paid for sex themselves in ACASI, while few admitted to this in FtF-interview (49.3 ACASI vs. 5.6% FtF, p,0.001).
Behaviours that were recruitment criteria for cohort enrolment, including anal sex practice, anal receptive role-taking, and having received cash for sex in the last 3 months, were reported significantly less frequently in ACASI than in FtF interviews with men. However, more men in ACASI than in FtF-interview reported participating in group sex (21.6 vs. 13.5%, p,0.001) and having paid for sex themselves (45.2 vs. 39.5%, p = 0.09).
ACASI and FtF-interview had excellent agreement for daily cigarette and marijuana use in women and men, good agreement for daily alcohol use, but poor agreement for being drunk daily. A lower proportion of women and men said in ACASI that they always used alcohol before sex (14.0% vs. 31.8%, p,0.001, for women; 14.2 vs. 40.0%, p,0.001, for men, data not shown). IV drug use in the last three months was reported by six women in ACASI, but not at all in FtF-interview (4.4 vs. 0%, p = 0.03). Similarly, a higher proportion of men reported IV drug use in ACASI than FtF (10.8 vs. 2.3%, p,0.001). Both men and women reported having been raped (in the last three months) more often in ACASI compared to FtF (women: 6.6 vs. 4.4%, p = 0.4, men 8.9 v 3.9%, p = 0.002).

Order of interview mode
The order of interview mode for categorical variables was assessed for three risk behaviours that were significantly different between methods in women (i.e. received cash for sex, paid for sex, and IV drug use, table 3) and eight in men (i.e. sex partner gender, anal sex and role taking, received cash for sex, paid for sex, group sex, IV drug use and rape, table 4).
Of 95 women who took ACASI first, 16 (16.8%) denied receiving cash for sex in ACASI, but 15 of these 16 women subsequently reported this activity in FtF-interview. Of the 44 women who took ACASI after FtF-interview, only 2 (4.5%) denied transactional sex in ACASI, while one of the two women had reported this during the preceding FtF-interview (16.8 vs. 4.5%, p = 0.04). An equal proportion of women admitted to having paid another person for sex independent of when ACASI was taken (48.8 vs. 49.5%, p = 0.9). No differences were found for IV drug use by order of interview mode in women (data not shown).
In men, the order of the interview method mattered only for sex partner gender in men who had self-identified as men who have sex with men (MSM) in the FtF-interview (91.5% of all men). Of 140 MSM (by FtF-standard) who took ACASI before the FtFinterview, 14 (10.0%) admitted to have sex with women only to the computer, while none said so in the FtF-interview that they did after ACASI. Of 91 MSM who took ACASI after the FtFinterview (all reported to be either homosexual, or bisexual to the counsellor), 2 (2.2%) changed their sex partner preference in ACASI, and reported sex with women only (10.0 vs.2.2%, p = 0.02).

Discussion
This is the first study comparing ACASI with FtF-interview in African high risk populations. We compared responses to an identical risk assessment questionnaire in 81% of the consecutively enrolled cohort participants (139 women, 259 men). Almost one out of five newly enrolled cohort participants (84 men and 37 women) were not able to take ACASI, mostly for lack of reading skills. Risk behaviour assessed in FtF-interview between non-ACASI and ACASI participants was not different in men; but women unable to use ACASI more frequently reported anal sex in the previous 3 months, and a higher number of recent casual partners.
The majority of women and men felt that answers given in ACASI were more honest. Over 90% of women and men were comfortable with ACASI, and, -although ACASI was on average twice as long as FtF-interview-, the majority of the volunteers had no objection to the longer interview time. Noteworthy is a recent assessment in ACASI of intentional provision of misinformation in FtF-interview by female microbicide trial participants in South Africa. Almost 80% had done this at least once, for reasons including politeness, to avoid criticism or seek praise, and embarrassment [17].
We explored responses to an identical questionnaire offered in ACASI and in FtF-interview and assessed the impact of the order of the interview mode. Among participants who took both interview methods, ACASI and FtF-interview had excellent agreement for less sensitive behaviours in both men and women, as demonstrated by reports on daily cigarette or marijuana use. Group sex, IDU, (in men), and rape (in men) were significantly more frequently reported in ACASI. A similar trend towards higher responses in ACASI for group sex and rape was seen in women. Thus, ACASI led to more frequent reporting of sensitive behaviours that were not linked to recruitment criteria. In contrast, potentially stigmatising behaviours that were also recruitment criteria were reported less frequently in ACASI compared to FtF-interviews. Transactional sex in the 3 months prior to cohort enrolment in men and women were less likely to be reported in ACASI. Moreover, 38 (14.7%) of men who reported anal sex in FtF-interviews did not report anal sex in ACASI, and MSM who took ACASI before FtF-interview were more likely to deny same sex behaviour than men who took ACASI after FtFinterview.
As African research sites are often located in settings underserved by formal health care [18], and prospective trial populations may misrepresent or over-report behaviour to ensure access to health care benefits through participation in research programmes, it seems probable that there is some misclassification at cohort enrolment for both MSM and transactional sex work in our cohort. Our study evaluating the effect of the order of interview mode on highly stigmatised and sensitive behaviours among 398 high-risk volunteers (most of whom were sex workers), suggested that ACASI is better conducted prior to FtF-assessment. Failure to disclose risk, or the reverse (i.e.overstating risk, as we suspect was the case in our study) would have significant implications for participant selection and behaviour tracking during intervention trials [17,19]. Spearman or kappa statistic. 2 Included those active in past week: 126 women in ACASI; 125 women in FtF. 3 Included those not active in past week: 13 women in ACASI; 14 in FtF. 4 Variable missing for 30 women who completed the study before this question added. 5 Missing values in 1 woman. doi:10.1371/journal.pone.0005340.t003 Spearman correlation coefficient or kappa statistic. 2 Included those active in past week: 198 men in ACASI; 206 in FtF. 3 Included those not active in past week: 61 men in ACASI; 53 in FtF. 4 Variable not asked in 23 subjects in ACASI. doi:10.1371/journal.pone.0005340.t004 That almost half of the women admitted to having paid for sex in ACASI was a surprise, and initially doubted. A focus group discussion conducted in 2008 with 13 women who admitted to payment for sex in ACASI suggested that this behaviour was indeed common, and gave some women a sense of control over the sex they purchased. Furthermore, a prospective sexual behaviour study conducted among 81 male sex workers recruited from the same cohort population, revealed that 68% of 238 female partners had paid for sex [20]. The finding that FSW purchase sex themselves has not been previously reported to our knowledge.
The crux to establishing high risk cohorts for HIV-1 prevention studies is a reliable and accurate assessment of volunteers' risk behaviour, that is sensitive to marginalised and stigmatised behaviours, and is conducive to truthful reporting. Surprisingly, ACASI has not been thoroughly evaluated among sex workers. The few ACASI comparison studies in Africa involved adolescents or general population, and firm conclusions of its usage have not been drawn, either because sample sizes were too small, or study findings were contradicting expectations [10,11,13]. A randomised crossover study of ACASI and computer-assisted personal interview (CAPI) in 445 adult volunteers from China, India, Peru, Russia, and Zimbabwe, found few differences in responses between methods, except for China, where volunteers gave a significantly higher response to some sensitive questions (i.e. ever had sex, number of partners, and unprotected acts) in CAPI [11]. Similar to our study, ACASI took twice as long on average in 4 of the 5 countries, and volunteers preferred a computer to an interviewer for answering sensitive questions [11]. ACASI's longer duration to complete suggest that volunteers take more time to consider their responses, which makes more honest reporting probable. ACASI enabled women, more than men, to report a higher median number of regular and casual sex partners.
Why then is ACASI not more widely used in research settings in Africa? While start-up costs or lack of systems capacity may pose some challenges, trial sites are often able to access financial and technical support from internal and external sources. That ACASI is merely used experimentally and is not more established in African settings is partly due to unfamiliarity with the interview method, but more likely based on mixed and sometimes contradicting findings. Indeed, in a recent prospective, randomised, cross-over design of 655 women enrolled in a study on hormonal contraceptive use with ACASI and FtF-interview in Zimbabwe, ACASI yielded higher reports on several reproductive health behaviours but discrepancies between self-reports and clinical data highlighted persistent measurement challenges [21]. Minnes et al. suggest that epidemiological studies should use multiple data sources, where possible, to estimate the range and direction of potential bias, and minimize misclassification [22].
We agree that multiple data sources should be used, and prefer ACASI as a screening tool to help identify high risk populations as it elicits sensitive behaviours (ie. IDU and rape) that would otherwise go unreported. We also believe that ACASI provides an environment more conducive to truthful reporting, that may help prospective volunteers avoid overstating their risk (e.g. anal sex, transactional sex).
ACASI has a number of other benefits over FtF-interview, including consistency and standardisation of data collection, and elimination of the need for further data transcription. Admittedly, comparing ACASI-, or FtF-interview, to sexual diary studies kept over a corresponding recall period would have provided a better opportunity to assess accuracy and reliability [22]. Our study also was not designed to test for noninferiority and superiority of ACASI and did not use biomarkers as some others have done. [23], but analysis is ongoing of data collected for a 3-way comparison between ACASI, FtF-interview, and sex partner diaries prospectively kept by 59 MSM cohort participants. A formal evaluation of ACASI as a screening tool to determine eligibility in settings as ours is justified, as planned intervention studies will include Pre-Exposure Prophylaxis that may be especially appealing to prospective volunteers to overstate their sexual risk behaviour and ensure enrolment.
This study has a number of additional limitations. First, our study was not able to distinguish honesty from accuracy; peer mobilization prior to cohort enrolment may have encouraged prospective volunteers to over-report 'high risk' behaviour in FTFinterviews. Second, the questionnaire had intrinsic weaknesses such as the assumption that regular and casual partners are clearly distinguishable. It also seems reasonable to assume that some questions were explained by counsellors differently during FtFinterview than in ACASI, and that participant had the opportunity to clarify uncertainties, but we have no documentation of this. Third, the FtF-interviews were performed by a number of different counsellors, and we were not able to gender-match counsellors to clients. It may be that some volunteers were less comfortable with counsellors of the opposite gender or with the interview style of a specific counsellor. Fourth, the study excluded volunteers who were not able to read and may have had higher sexual risk, given that more sexually transmitted infections were diagnosed in these volunteers. A more user friendly version of ACASI could have facilitated use irrespective of education level [24]. Lastly, as in other ACASI comparison studies [14], some discrepant responses (for example, a case of proctitis in a woman who denied anal sex in ACASI, and would on that basis not have been examined) were not fully understood.
Despite these limitations, we feel that ACASI could have an important role in risk behaviour assessments among high-risk populations in Africa. The majority of women (79.2%) and men (69.7%) felt that answers given in ACASI were more honest. Over 90% of women and men were comfortable with ACASI, and,although ACASI was on average twice as long as FtF-interview-, the majority (80%) of the volunteers had no objection to the longer interview time. ACASI revealed some misclassification of same-sex behaviour, and an over-report of anal sex at cohort enrolment. Largely, ACASI elicited reports of risk behaviours in the same range as FtF-interview, and helped us to capture behaviours hitherto not recognised in our cohort populations. These behaviours were IDU in men and women, rape and group sex in men, and payment for sex by women. Such risk behaviour needs specific risk reduction counselling and further investigation.

Recruitment of study population
Since July 2005, populations in and around Mombasa at higher risk for HIV infection, including men and women who admitted to transactional sex in the past 3 months, have been targeted for an HIV-1 prevention cohort. Identification and recruitment of prospective study participants has been described elsewhere [16]. In short, peer educators identified volunteers and accompanied them to a Drop-in Centre adjacent to the research clinic. Prospective cohort participants were shown a 25-minute video describing cohort procedures, including a short demonstration of ACASI. Cohort eligibility was assessed by a pre-enrolment counsellor who discussed sexual and other risk behaviour. Cohort inclusion criteria were verified by an enrolment counsellor and were either a self-report of transactional sex, anal sex in the past three months, or same sex behaviour for men. Upon cohort enrolment, participants committed to an HIV-1 test, a risk assessment by standardized FtF-interview, and a medical examination with screening for sexually transmitted infections (STI). Details on cohort procedures, including case definitions used for HIV-1 and STI diagnosis, are described elsewhere [16,25]. Participants who tested HIV-1 positive at screening were offered follow up into a parallel HIV-1 positive cohort [16].
All study participants provided written informed consent. This study received approval from the National Ethical Review Board.

Eligibility for ACASI and allocation to interview mode
Newly enrolled cohort participants in the period March 2006-May 2008, who had basic reading skills were invited to undergo both interview methods (ACASI and FtF-interview) on the same day. For ease of clinic organization, participants were either allocated to take ACASI first followed by FtF-interview, or vice versa, in periods of approximately 6 months during the study period. All ACASI and FtF-interviews were conducted before voluntary testing and counselling for HIV-1.

Risk assessment
We developed an ACASI questionnaire with identical questions to the structured questionnaire used for the FtF-interview. Ordered questions appeared in text on the computer screen, accompanied by a spoken recording in either English or Swahili. Automatic text and voice prompts included standard definitions of terms (e.g. '' a new sex partner is someone you have only known for one week or less''). The questionnaire included questions on number of regular, casual, and new partners in the last week and month, anal sex (practice, role and condom use), gender of their sexual partners (from Sept 2006), transactional sex, substance misuse, and sexual violence.
ACASI study participants were accompanied to a small computer room and instructed in the working of ACASI using sociodemographic questions. Upon completion of these, ACASI was operated alone and in privacy to answer risk assessment questions in either English or Kiswahili. The ACASI programme included builtin skip patterns, and logic checks. Upon completion of each ACASI interview, six acceptability questions were asked in ACASI only.
Risk assessment in FtF-interview was conducted after sociodemographic questions were obtained. Counsellors had been trained on asking risk questions and providing explanations to certain questions during the start of the cohort (July 2005), and were familiar with the questionnaire as it also was used for follow up visits.

Data management and analysis
Questionnaire data was entered and stored in a secure database. Quality of data recorded was scrutinised by external data monitors, twice a month. Prior to data analysis, all site database entries were individually compared to source documents. Data cleaning, coding and analysis were conducted using Stata 9.2 [26].
The Chi-squared test and Student t-test were used to determine associations between categorical and continuous values, respectively. Differences in paired ACASI and FtF-responses were compared using the McNemar's chi squared test for binary data, and the Wilcoxon matched-pairs signed-rank test or paired Student t-test for continuous variables as appropriate [27]. Intermethod agreement was assessed by calculation of the Cohen's kappa statistic for binary and categorical variables, and Spearman's correlation coefficient for continuous variables. For categorical variables that had significant differences between interview methods the effect of interview order was assessed using simple test of proportions. For continuous responses Bland Altman plots were used to inspect for differences in variability between methods and the effect of the order of interview methods [28].