van der Straten, Ariane PhD*†; Shiboski, Stephen PhD‡; Montgomery, Elizabeth T MHS*; Moore, Jie PhD§; De Bruyn, Guy MBChB‖ ; Ramjee, Gita PhD¶ ; Chidanyika, Agnes MBA#; Kacanek, Deborah PhD**; Padian, Nancy PhD*; and the MIRA Team
The Methods for Improving Reproductive Health in Africa (MIRA) study, a randomized controlled trial among Southern African women, evaluated the effectiveness of the diaphragm used with a noncontraceptive lubricant gel (Replens) in preventing male to female transmission of HIV and other sexually transmitted infections (STIs). Although the findings did not indicate a protective effect against HIV, chlamydia or gonorrhea, they did not rule out a possible modest protective effect of the diaphragm, particularly in light of significant lower condom use in the intervention arm.1-4 Another trial is examining the effect of the diaphragm used with a microbicide on cervical STI reinfection.5 Cervical barriers also continue to be studied as dual purpose methods (to prevent pregnancy and disease) and as reusable microbicide delivery mechanisms.6 This analysis explores diaphragm adherence in greater depth over the course of the MIRA study among women in the intervention arm and assesses predictors of various diaphragm use patterns.
Achieving and maintaining product adherence is essential to accurately assess the safety and effectiveness of HIV prevention methods. Previous studies, conducted in a variety of populations and settings, have reported a large range of diaphragm adherence in the context of contraceptive use and also for disease prevention, where diaphragms have been used alone, with inactive lubricants, or in combination with potential microbicides. In 2 contraceptive studies, approximately 60% of women reported consistent use (at every sex act).7,8 When diaphragms have been researched as potential disease prevention methods, adherence has ranged from less than 20% to more than 95%.9 Some of these studies examined predictors of consistent diaphragm use, including older age,7 various measures of condom use at baseline10 or follow-up,9,11 partner approval,7 and covert use.9,12 Being married and living with a partner was associated with discontinuation and inconsistent use of the diaphragm in a multisite contraceptive study7 and among sex workers and clinic attendees in Kenya.10 All of these studies had relatively short follow-up periods.
In the MIRA study, however, we followed a large sample of sexually active, HIV-negative, Southern African women for up to 2 years. In the intervention arm, women reported that 73% of their last sex acts were protected by a diaphragm, and 58% reported always using diaphragms since their last visit (note that these responses did not always overlap).1 The current analysis explores in greater depth the pattern of diaphragm (and gel) use among women randomized to this arm by examining a more rigorous measure of adherence: reporting always using a diaphragm over the past quarter “and” use at last sex act. We also conducted trajectory analyses to identify behavioral groups associated with specific diaphragm adherence patterns, and we characterized women affiliated with the various behavioral trajectory groups, using semiparametric group-based modeling techniques, which allow for the examination of individual-level patterns of use.
Participants, Study Design, Screening, and Enrollment Procedures
The MIRA study was a phase III randomized, controlled, open-label trial of the diaphragm and lubricant gel in preventing heterosexual HIV/STI acquisition in women. The trial was conducted at 3 study sites in Southern Africa: Harare, Zimbabwe; and Durban and Johannesburg, South Africa. The sites spanned semirural, periurban, and urban settings. Trained outreach workers recruited participants from family planning, well-baby, and general health clinics, through print and radio advertisements, community outreach, and word of mouth.
Enrollment took place between September 2003 and September 2005; follow-up concluded in December 2006. The methods, study procedures, and eligibility criteria for this trial have been described in detail elsewhere.1 Briefly, to be eligible for the study, women had to test negative [or be treated in the case of Chlamydia trachomatis (CT), and Neisseria gonorrhea (GC)] for HIV, CT, GC, and pregnancy; be sexually active (defined as an average of at least 4 sex acts per month); be aged 18-49 years; have no allergy to latex or history of toxic shock syndrome; have a healthy cervix; and be willing and able to correctly insert and remove the diaphragm at the clinic. Eligible women were randomized in a 1:1 ratio to receive an Ortho All-Flex diaphragm, Replens gel, and male condoms (intervention arm) or to receive male condoms only (control arm). All study participants received a comprehensive HIV prevention package that included screening and treatment for curable STIs, quarterly counseling and testing for HIV, intensive risk reduction and product use counseling, and provision of free male condoms. This analysis only considers women randomized to the intervention arm.
The study protocol was reviewed and approved by 6 Institutional Review Boards and regulatory authorities at the University of California, San Francisco, in the United States, and at all local institutions and collaborating organizations. The MIRA study is registered with ClinicalTrials.gov (number NCT00121459).
As described previously,1 at screening, participants completed a demographic form, received pretest and posttest HIV/STI counseling, and were tested for HIV, CT, GC, Trichomonas vaginalis (TV), and pregnancy. The enrollment visit was scheduled within 2 weeks for participants who met the initial eligibility criteria. Participants were tested again for pregnancy, for syphilis and herpes simplex virus 2, as described by Padian et al1 and were diagnosed and treated (if applicable) for any reproductive tract infections. All women were fitted for a diaphragm (ranging from size 55 mm to 95 mm) by a trained clinician using diaphragm fitting kits and a standardized technique. They were then required to insert and remove the diaphragm at the clinic, and the clinician verified correct placement and removal.
Women who could not successfully complete this diaphragm procedure within 5 attempts were not eligible for randomization (n = 38, 0.75%). Twenty-six women could not insert and remove the diaphragm after 5 attempts; 1 woman experienced injury at her perineum after attempting to remove her diaphragm because of her long fingernails; 10 women could not be fitted because no correct size was found (n = 3) or they had a clinical condition [ie, pelvic relax (n = 3), cystocele (n = 2), obesity (n = 2)]; and 1 woman had mental illness.
Women enrolled in the trial completed an audio computer-assisted self-interviewing (ACASI) baseline questionnaire in their native language, including demographics, sexual behavior, and current and previous use of the diaphragm, gel, and condoms.
After randomization and ACASI, women in the intervention arm were counseled to vaginally insert the diaphragm at any time that was convenient to them before coitus and to leave it in place for at least 6 hours after sex, but no more than 24 hours. They were given detailed instructions on maintenance, cleaning, and storage of their diaphragm. Women received a 3-month supply of gel and were asked to (a) empty an applicator of gel (approximately 2.5 g) in the dome of the diaphragm at the time of insertion, (b) spread gel onto the rim to facilitate insertion, and (c) insert another applicator of gel in the vagina before each act of vaginal sex. At each quarterly visit, women received a resupply of gel and were asked to use the diaphragm with gel every time they had sex. Women who had problems using the gel were provided with a hierarchical message that included (a) using a different amount of gel for either the diaphragm or the vaginal application and (b) using the diaphragm only. Use of gel alone was never recommended. Staff counseled participants to use condoms during every sex act (whether they used the diaphragm and gel or not) because the effectiveness of the investigational products for the prevention of HIV/STIs was not known. Participants were also told that they should not use the diaphragm and lubricant gel for contraception. The diaphragm is approved for contraception only when used with a spermicide.13 Use of effective contraception was encouraged, and hormonal contraceptives were provided at no cost through the clinic.
Participants were scheduled for quarterly follow-up visits (with a window of 14 days before and 77 days after the scheduled visit) for a period ranging from 12 to 24 months, depending on how late in the accrual period they enrolled.1 At each quarterly visit, product use and sexual behavior were assessed by ACASI. We also assessed recent medical history, which included a question about the most recent date of use of each of the products (diaphragm, gel, and condoms). Women were tested for HIV/STIs, and treatment for laboratory diagnosed, curable STIs was provided when appropriate. Finally, participants had individual HIV risk-reduction counseling and product adherence counseling, and they received a resupply of gel and condoms.
Among sexually active participants, follow-up ACASI measures of product use since last quarterly visit included the following: ever use of diaphragm during vaginal sex and ever use of gel (yes-no). Women were also asked to indicate which product(s), if any, they used at last sex act. Additionally, all women who were sexually active since their last visit were asked a summary diaphragm question: “since your last visit, how often have you used a diaphragm when having sex?” (never, ≤50%, >50%, always; collapsed for this analysis into: never, sometimes, always). Women were asked the same summary questions about male condom use and about gel use (overall) and specifically about vaginal application of gel. Women who reported always using the diaphragm and always using gel (overall) since their last visit were classified as “always using diaphragm and gel.” Among diaphragm users, we also assessed how often these women had applied gel in the dome of the diaphragm before insertion (never, sometimes, always).
For this analysis, the primary time-dependent outcome variable of diaphragm adherence was created by combining responses to the 2 recall periods among participants who were sexually active since their last quarterly interview: the summary diaphragm question (frequency of use since last visit) and reported diaphragm use at last sex act. For each quarterly visit, we classified women as “diaphragm adherent” if they responded at that visit that they “always” used the diaphragm since their last visit and that they also reported that they used the diaphragm at last sex act. Women who reported discrepant responses (ie, reporting always use, but not at last sex act, or use at last sex act but inconsistent or no use during the quarter) were considered to be nonadherent. Correct removal time of the diaphragm was assessed at study exit through face-to-face interview by asking the following question: “How many hours after sex do you usually remove the diaphragm?” (response categories collapsed into: 5 hours or less, 6-23 hours, and 24 hours or more).
We examined the following baseline variables (gathered through face-to-face interview or ACASI) from 3 separate domains as potential predictors in the multivariable model (described below): (a) sociodemographic characteristics-age, education, parity, employment, income, and marital status; (b) risk behavior/history-having a positive test for any curable STI (CT, GC, TV, syphilis), positive serology for herpes simplex virus 2, and indicator variables both for participant's high-risk behavior and for having a high-risk partner (Table 1 for definitions); and (c) contraceptive/barrier method use-current contraceptive method. The same procedure used to create the outcome (described above) was used to create a baseline measure of condom adherence (“always” used the condoms in the past quarter and used a condom at last sex act), which was examined as one of the baseline predictors of diaphragm adherence.
Preliminary analyses focused on tabulation of observed frequencies of adherence measures at each visit. Although these results provide information about the average patterns of diaphragm and gel use observed among women in the intervention arm, trends in individual-level results may be obscured. Further, group average adherence results cannot be linked easily to baseline characteristics of individual women. Because of the infinite number of possible individual-specific adherence patterns, we used group-based trajectory modeling14 to identify clusters of women with similar observed temporal patterns of diaphragm use. Similar to latent class analysis,15 this method-developed first in developmental psychology-assumes that the population is composed of a mixture of distinct groups defined by their behavioral trajectories. Given a specified number of groups, the model estimates membership probabilities in each group for every participant, based on the participant's observed patterns of adherence over time. It should be noted that it is not assumed that any woman belongs strictly to any of the identified trajectory groups.
This procedure allows joint modeling of the probability of group membership and group-specific trajectories. The latter are modeled semiparametrically via flexible polynomial functions fitted to individual adherence responses over time. Group membership probabilities are allowed to depend on covariates via a multinomial logistic regression model. Estimated variability in parameter estimates accounts for uncertainty in group membership and between-individual and within-individual variation in observed longitudinal responses. Selection of the optimal number of groups is based on comparison of model fits using the Bayesian information criterion.
We selected baseline predictors of group membership from 3 distinct domains: sociodemographic, risk behavior, and contraceptive/barrier method use (as detailed above in “measures”). To construct the final multivariate model, within each domain, we selected factors that were significantly associated at the univariate level with the probability of group membership. The effects of selected baseline characteristics on the probability of group membership were estimated via multinomial logistic regression and summarized as odds ratios (ORs) with 95% confidence intervals (CIs). ORs estimate the odds of membership in each group, relative to odds of membership in the selected reference group. In addition to these OR estimates, we estimated the percentages of women in each trajectory group. Model results were also summarized graphically by the predicted and observed adherence probabilities in the groups. The former are estimated directly from the model coefficients for the group-specific trajectory models. Observed results are based on the average adherence probabilities among women assigned to a particular group. Group assignments are based on posterior model-based probability calculations, with actual assignment based on the maximal group membership probability for each participant. All analyses were conducted using SAS version 9.1 (SAS Institute, Inc, Cary, NC) using a custom procedure (PROC TRAJ) to estimate the models described above.16
Study Population and Baseline Characteristics
Of the 2521 eligible women randomized into the MIRA intervention arm, 2429 (96%) had at least 1 follow-up visit where they were sexually active on or before HIV seroconversion. These participants constitute the analytical sample (total person-visits = 12,871). Only 382 (3%) visits were censored because they occurred after a woman seroconverted, after which we expected women's product use behavior to change. Cumulatively, at 95% of participant visits (range: 83%-99%), women reported having had sex since their last quarterly interview. Among women who were sexually active, participants reported a median of 4 vaginal sex acts per week during follow-up. Women were followed for a median of 7 quarterly visits or 21 months (range: 1-8 quarterly visits); and 2277 women (94%) completed their expected time in the study.
As shown in Table 1, at baseline, almost 40% of the women were aged 18-24; the median number of lifetime partners was 1; and they reported a median of 3 weekly sex acts over the past 3 months. Over two-thirds were living with a husband or primary partner, 58% had 2 or more children, nearly 57% earned an income in the past year, and about 55% had less than a high school education. Although about 68% were classified as having a history of high-risk partner(s), only 28% were classified as having high-risk behavior themselves (Table 1 footnote for definition); about 67% of participants reported having used a condom at their last sex act, and about 30% said they always used a male condom in the previous quarter. Oral contraceptive pills were the most common contraceptive reported by about 36% of women; about 26% used male condoms, and nearly 25% used injectables; 6% used a long-term method (implants, tubal ligation, intrauterine device, or vasectomy); and nearly 7% used a traditional method of contraception or no method.
Reported Product Use
Only 1 woman had ever used a diaphragm before entering the study (Table 1). Among the total study sample (n = 2429), 31 (1.3%) never used the diaphragm during the entire follow-up period (data not shown). At study exit, the majority of women (67.6%) correctly reported usually removing the diaphragm 6-23 hours after sex (Table 2). When examining the total proportion of women reporting diaphragm use since last visit, this proportion was stable over time, across multiple measures (Fig. 1). Over a cumulative total of 12,871 person-visits, ever use of the diaphragm since last quarterly visit was reported 94.4% of the time, use of the diaphragm at last sex act was reported 73% of the time, “always” using the diaphragm was reported 58% of the time, and “diaphragm adherence” (consistent responses across time frame modality-that is, reporting “always” use and use at last sex act) was reported 49% of the time. Thus, in 84% of the person-visits where the diaphragm was reportedly “always” used, the diaphragm was also used at last sex act. On a per-participant basis, we observed the following patterns of adherence: 25% of women were diaphragm adherent at none of their visits, whereas 15% (n = 365) were adherent at all their visits, the median were adherent at 50% of their visits; 32% were adherent at 0%-20% of their visits, whereas 31% were adherent at 75%-100% of their visits (data not shown).
Across all visits, ever use of gel was reported 96.1% of the time (data not shown). Gel and diaphragm use were highly correlated: women reported always using both diaphragm and gel since last quarterly visit 52% of the time (Fig. 1). Additionally, women reported having used gel on the last date the diaphragm was used 91.4% of the time (data not shown). Adherent diaphragm users reported always using gel when having sex since last quarterly visit 91.0% of the time; they reported always having applied gel in the dome of the diaphragm 91.3% of the time and they reported always applying the gel vaginally 84.2% of the time (Table 2). Thus, for simplicity and because the diaphragm was the key product hypothesized to protect against HIV (as it covers the cervix), we limited our examination of consistent product use to diaphragm adherence for the remaining analyses.
Trajectory Analysis and Baseline Predictors of Trajectory Group Affiliation
We used trajectory analysis to identify temporal group patterns of diaphragm adherence. Women were clustered into trajectory groups based on their highest estimated membership probability. Although we considered models including different number of groups, we selected a model with 4 trajectory groups as optimal (Fig. 2). This choice was based both on considerations of fit (determined by comparison of Bayesian information criterion values between models), parsimoniousness, and interpretability of the resulting groups. We also estimated the percentages of women belonging to each trajectory group in the study population. Group 1 (labeled “low adherers”) had an estimated group membership rate of 31% and was characterized by a consistently low probability of diaphragm adherence at all study visits. Group 2 (labeled “increasing adherers”) had an estimated group membership rate of 9.3% and was characterized by an increasing probability of diaphragm adherence over time. Group 3 (labeled “decreasing adherers”) had an estimated group membership rate of 28.9% and was characterize by a decreasing probability of diaphragm adherence. Group 4 (labeled “high adherers”) had an estimated group membership rate of 30.8% and was characterized by a consistently high probability of diaphragm adherence. There was generally good agreement between the observed and predicted visit-specific adherence probabilities (as shown in Fig. 2), indicating that the selected models for the trajectories provided a reasonable fit to the adherence data. The fits seemed to be particularly good for the groups with the highest and lowest adherence levels.
As described above, the trajectory analysis method also allows investigation of the effects of selected covariates on the probability of group membership via multinomial logistic regression. This procedure identified a set of baseline characteristics associated with a higher probability of being in a particular trajectory group. As shown in Table 3, adjusting for all other variables in the models, younger women were more likely to be low adherers (group 1), as compared with any of the other groups (increasing, decreasing, and high adherers, groups 2, 3, and 4, respectively). Study site was significantly associated with group membership too, with participants from the Harare site (as compared with the Durban site) being more likely to be decreasing adherers (group 3; AOR = 36.97, 95% CI: 3.80 to 360) or high adherers (group 4: AOR = 2.82, 95% CI: 1.89 to 4.20); whereas participants from the Johannesburg site (as compared with the Durban site) were less likely to be increasing adherers (group 2: AOR = 0.42, 95% CI: 0.26 to 0.70) or high adherers (group 4: AOR = 0.51, 95% CI: 0.34 to 0.77). Women with high-risk behaviors were less likely to be high adherers (group 4: AOR = 0.51, 95% CI: 0.34 to 0.77), as were women with a history of high-risk partners (AOR = 0.58, 95% CI: 0.43 to 0.78). Finally, women who reported baseline condom adherence were more likely to be high adherers (group 4: AOR = 2.00; 95% CI: 1.47 to 2.71). Baseline (nonbarrier) contraceptive use was not associated with group membership in the multivariate model (data not shown).
In this exploratory analysis of diaphragm adherence in the MIRA trial, we used a relatively stringent measure of diaphragm adherence, combining reports from 2 recall periods collected by ACASI. Overall, using this combined adherence measure, we observed moderate levels of diaphragm adherence: on average, about half the sample was adherent, and only 15% of women were diaphragm adherent at every single visit. Although by some of the self-reported measures, product adherence in MIRA was in the same range as that reported in previous diaphragm studies1,6,17 and other recent HIV prevention trials of female-initiated methods,18,19 it nevertheless fell short of the 80% expected adherence level required to retain power to detect a protective effect (if any) of the intervention.20 Motivating consistent use of an intervention of unknown efficacy among healthy HIV-negative participants is a key challenge in these trials, which is particularly acute for coitally dependent user-controlled methods. Although in the MIRA trial, women were counseled to apply gel in the dome and insert the diaphragm at any time convenient to them, the second (vaginal) application of gel was requested within 1 hour of sex. Making diaphragms coitally independent by promoting continuous use may help increase adherence, as has been reported in previous contraceptive studies21,22 and in a pilot study of sex workers.23
Another key limitation in trials like MIRA is that of no gold standard biological measures exist, so measuring adherence and sexual behavior relies solely on self-reports. In MIRA, we attempted to minimize social desirability bias in 2 ways. First, most sexual behavior and product use questions were asked by ACASI, which is reported to encourage higher reports of sensitive behaviors in these populations.24,25 Second, participants were asked to complete the ACASI before any product adherence or risk-reduction counseling. Here, we chose to combine the 2 time frames to increase accurate measurement of individual adherence behavior over time.
When examining individual adherence trajectories within this sample, 4 groups were identified: low adherers, increasing adherers, decreasing adherers, and high adherers. Younger women were more likely to be in the low adherence group compared with all other groups. Conversely, women who had lower risk behaviors, who had a history of lower risk partner(s), and who reported baseline condom adherence were more likely to be in the high adherence group. Taken together, this suggests that women with baseline characteristics that typically put them at lower risk of HIV infection were more likely to report high diaphragm adherence throughout the trial. If indeed individuals at lower risk are more likely to adhere to a prevention intervention, although those at higher risk are less likely to adhere, this may bias effectiveness results toward the null, as fewer “informative” coital events will occur (those in which there was a high likelihood of HIV exposure and an investigational product was used). Alternatively, self-reported lower risk behavior and high adherence may simply be the reflection of strong self-presentation bias in a subset of participants.
In addition to the baseline factors predicting membership in adherence trajectory groups examined here, it is likely there are many time-varying determinants of adherence as well. For example, postrandomization use of condoms was not independent of observed levels of diaphragm adherence in the MIRA study.1,2 Indeed, because condom use was differential between study arms in the MIRA trial, condom use during the trial will most likely be associated with diaphragm trajectories. However, incorporating the effects of time-varying predictors in trajectory analyses introduces complications for both model estimation and interpretation, and it is not clear that existing methods effectively address time-dependent confounding factors. For these reasons, we plan to assess the effect of condoms and other time-varying covariates on diaphragm use in future analyses and limit our conclusions here to baseline predictors.
The findings indicate differences in diaphragm trajectories by study site, with women from the Johannesburg site being the most likely to be in the decreasing adherence or low adherence groups. In contrast, women from the Harare site were most likely to be in the high adherence and the decreasing adherence groups. In this analysis, we adjusted for potential heterogeneity at the 3 study sites but did not explore the nature of the difference among them. These will be best explored through a forthcoming analysis of focus group discussions conducted with women and their partners who exited the trial at the 3 sites.
This study revealed interesting findings about characteristics of diaphragm adherence and temporal trends that may not have been apparent when using across-time average analytical approaches. This method could help select “high compliers” for enrollment in future trials (if this population also incurs enough exposure to HIV). Furthermore, it may help to identify the types of participants who may need more product support and/or intensive adherence counseling during the course of a trial.
We would like to thank the women who participated in this study. For A.S., N.P., and E.T.M., most work for this study was conducted at the University of California San Francisco, Department of Obstetrics, Gynecology, and Reproductive Sciences. Special thanks to Helen Cheng for statistical support during final revisions of the article, Mark Weaver for providing statistical advice with trajectory analysis and reviewing an earlier version of the article, Kelly Blanchard for article review, Jeff Novey and Caitlin Gerdts for editorial assistance.
1. Padian NS, van der Straten A, Ramjee G, et al. Diaphragm and lubricant gel for prevention of HIV acquisition in southern African women: a randomised controlled trial. Lancet
2. Rosenblum M, Jewell NP, van der Laan MJ, et al. Analyzing Direct Effects in Randomized Trials With Secondary Interventions
. Berkeley, CA: University of California, Berkeley; 2007. Paper 223.
3. Ramjee G, van der Straten A, Chipato T, et al. The diaphragm and lubricant gel for prevention of cervical sexually transmitted infections: results of a randomized controlled trial. PLoS ONE
4. Jewell N, van der Straten A, Montgomery E, et al. Diaphragms and lubricant gel for prevention of HIV. Lancet
. 2007;370:1823-1824. Authors' reply.
5. Behets F, Van Damme K, Turner A, et al. Evidence-based planning of a randomized controlled trial on diaphragm use for prevention of sexually transmitted infections. Sex Transm Dis
6. Ballagh S, Brache V, Mauck C, et al. A phase 1 study of the functional performance, safety, and acceptability of the BufferGel Duet. Contraception
7. Bulut A, Ortayli N, Ringheim K, et al. Assessing the acceptability, service delivery requirements, and use-effectiveness of the diaphragm in Colombia, Philippines, and Turkey. Contraception
8. Barnhart KT, Rosenberg MJ, MacKay HT, et al. Contraceptive efficacy of a novel spermicidal microbicide used with a diaphragm: a randomized controlled trial. Obstet Gynecol
9. van der Straten A, Moore J, Napierala S, et al. Consistent use of a combination product versus a single product in a safety trial of the diaphragm and microbicide in Harare, Zimbabwe. Contraception
10. Luchters S, Chersich MF, Jao I, et al. Acceptability of the diaphragm in Mombasa, Kenya: a 6-month prospective study. Eur J Contracept Reprod Health Care
11. van der Straten A, Kang M, Posner SF, et al. Predictors of diaphragm use as a potential sexually transmitted disease/HIV prevention method in Zimbabwe. Sex Transm Dis
12. Ortayli N, Bulut A, Nalbant H, et al. Is the diaphragm a viable option for women in turkey? Int Fam Plan Perspect
13. Cook L, Nanda K, Grimes D. Diaphragm versus diaphragm with spermicides for contraception. Cochrane Database Syst Rev
14. Nagin D. Analyzing developmental trajectories: semi-parametric, group-based approach. Psychol Methods
15. Goodman L. Exploratory latent structure analysis using both identifiable and unidentifiable models. Biometrika
16. Jones B, Nagin D, Roeder K. A SAS Procedure based on mixture models for estimating developmental trajectories. Sociol Methods Res
17. Feldblum P, Adeiga A, Bakare R, et al. SAVVY vaginal gel (C31G) for prevention of HIV infection: a randomized controlled trial in Nigeria. PLoS ONE
19. Watson-Jones D, Weiss H, Rusizoka M, et al. Effect of herpes simplex suppression on incidence of HIV among women in Tanzania. N Engl J Med
20. Committee on the Methodological Challenges in HIV Prevention Trials. Methodological Challenges in Biomedical HIV Prevention Trials
. Lagakos S, Gable A, eds. Washington, DC: Institute of Medicine of the National Academies; 2008. Available at: http://www.nap.edu/catalog.php
?record_id = 12056. Accessed June 2008.
21. di Giacomo do Lago T, Babose R, Kalckmann S, et al. Acceptability of the diaphragm among low-income women in Sao Paulo, Brazil. Int Fam Plann Perspect
22. Stim E. The nonspermicide fit-free diaphragm: a new contraceptive method. Adv Plan Parent
23. Behets F, Turner A, Van Damme K, et al. Acceptability and feasibility of continuous diaphragm use among sex workers in Madagascar. Sex Transm Infect
24. Hewett P, Mensch B, Erulkar A. Consistency in the reporting of sexual behaviour by adolescent girls in Kenya: a comparison of interviewing methods. Sex Transm Infect
25. Minnis AM, Muchini A, Shiboski S, et al. Audio computer-assisted self-interviewing in reproductive health research: reliability assessment among women in Harare, Zimbabwe. Contraception
© 2009 Lippincott Williams & Wilkins, Inc.