American Journal of Epidemiology Copyright 2006 by the Johns Hopkins Bloomberg School of Public Health All rights reserved;

; printed in U.S.A.

Vol. 163, No. 6 DOI: 10.1093/aje/kwj071 Advance Access publication January 27, 2006

Original Contribution Distinguishing the Temporal Association between Womens Intravaginal Practices and Risk of Human Immunodeciency Virus Infection: A Prospective Study of South African Women

Landon Myer1,2, Lynette Denny3, Michelle de Souza4, Thomas C. Wright, Jr.5, and Louise Kuhn2,6
1

Infectious Diseases Epidemiology Unit, School of Public Health and Family Medicine, Faculty of Health Sciences, University of Cape Town, Cape Town, South Africa. 2 Department of Epidemiology, Mailman School of Public Health, Columbia University, New York, NY. 3 Department of Obstetrics and Gynaecology, Faculty of Health Sciences, University of Cape Town, Cape Town, South Africa. 4 School of Public Health and Family Medicine, Faculty of Health Sciences, University of Cape Town, Cape Town, South Africa. 5 Department of Pathology, College of Physicians and Surgeons, Columbia University, New York, NY. 6 Gertrude H. Sergievsky Center, College of Physicians and Surgeons, Columbia University, New York, NY. Received for publication September 6, 2005; accepted for publication October 31, 2005.

Downloaded from aje.oxfordjournals.org by guest on April 15, 2011

Cross-sectional studies have suggested that intravaginal practices, such as douching or dry sex, may increase womens susceptibility to infection with human immunodeciency virus (HIV). The authors examined the temporal nature of this association in a cohort of South African women. At enrollment (20012002), 4,089 women were tested for HIV infection. Participants reported their intravaginal practices at a 6-month follow-up visit and were followed with repeat HIV testing for up to 24 months. Among the 3,570 women who were HIV-negative at enrollment, 26% reported some type of intravaginal practice, mostly washing inside the vagina with water and/or cloth as part of daily hygiene. During follow-up, 85 incident HIV infections were observed. Intravaginal practices were associated with prevalent HIV at enrollment (adjusted odds ratio 1.50, 95% condence interval: 1.22, 1.85), but during follow-up there was no association between intravaginal practices and incident HIV (adjusted hazard ratio 1.04, 95% condence interval: 0.65, 1.68). These ndings may be explained by a reversal of the causal sequence assumed for this association, since intravaginal practices may be undertaken in response to vaginal infections that occur more commonly among HIV-infected women. Intravaginal practices appear unlikely to be a cofactor in the male-to-female transmission of HIV in this setting. HIV; risk factors; sexually transmitted diseases; South Africa; vaginal discharge; vaginal douching

Abbreviations: CI, condence interval; HIV, human immunodeciency virus.

Reports published over the past decade have described womens intravaginal practices in different parts of subSaharan Africa and other regions (14). These practices involve the insertion of various materials or substances into the vagina, including liquids, paper, cloth, and medicinal preparations. In the context of human immunodeciency virus

(HIV) infection, intravaginal practices rst gained attention among sex workers seeking to remove moisture prior to intercourse, a practice referred to as dry sex (5, 6). Subsequent investigations have shown that these behaviors take place for a range of different purposes, including the removal of vaginal discharge and/or regular hygiene (2, 7). Research

Correspondence to Dr. Landon Myer, Infectious Diseases Epidemiology Unit, School of Public Health and Family Medicine, Faculty of Health Sciences, University of Cape Town, Anzio Road, Observatory 7925, Cape Town, South Africa (e-mail: lmyer@cormack.uct.ac.za).

552

Am J Epidemiol 2006;163:552560

Intravaginal Practices and Risk of HIV Infection

553

from different parts of Africa has suggested that intravaginal practices may be relatively common in general population samples, with reported prevalences ranging from 20 percent to 50 percent (713); slightly higher prevalences have been documented among female sex workers (5, 1416). Recent research has suggested an association between intravaginal practices and womens susceptibility to HIV infection. Several cross-sectional studies have found signicant associations between intravaginal practices and prevalent HIV infection (1, 5, 8, 13, 17). However, cross-sectional data are unable to distinguish the temporal sequence of this associationthat is, whether intravaginal practices increase susceptibility to HIV infection or whether HIV infection leads to increased intravaginal practices. The latter possibility requires careful consideration, because a range of different vaginal infections, such as Trichomonas vaginalis and Candida species, appear more commonly among HIV-infected women (18, 19). It is plausible that intravaginal practices may be undertaken to help relieve the symptoms of vaginal infection, including discharge and/or pruritus. There is also substantial evidence showing that intravaginal practices are positively correlated with different behavioral risk factors for HIV (e.g., number and type of sexual partners) (7, 2022). As a result, the prospect of confounding by sexual behaviors requires particular attention. Despite the importance of discerning the temporality of any association involving intravaginal practices and womens risk of HIV infection, there have been few prospective studies investigating this question. One Zambian study of 634 women visiting postnatal clinics found a substantially higher risk (relative risk 38.0) of incident HIV among women reporting intravaginal practices, but this association was not adjusted for any potentially confounding variables (17). Another study of 500 sex workers in Thailand showed no association (relative risk 1.1) between risk of HIV seroconversion and the use of policresulen vaginal suppositories (23). However, these suppositories are not widely used (24), and thus it is unclear how these results may be generalized to other types of intravaginal practices. Other prospective studies of intravaginal practices have not evaluated risk of HIV infection as an outcome but have focused instead on other sexually transmitted infections and vaginal epithelial disruptions (25, 26). If intravaginal practices truly increase womens susceptibility to HIV infection, these behaviors may be an important target for HIV prevention interventions. However, given the limited amount of prospective data, additional data are clearly required to establish the nature of this association before any intervention strategy can be recommended. We investigated intravaginal practices and risk of incident HIV infection in a cohort of women in Cape Town, South Africa. Previously, we observed a signicant adjusted association between intravaginal practices and prevalent HIV in this population (27). This analysis was undertaken to evaluate the temporal nature of the cross-sectional nding.
MATERIALS AND METHODS

This study was nested within a larger randomized controlled trial evaluating various cervical cancer screening
Am J Epidemiol 2006;163:552560

modalities in the African community of Khayelitsha, outside of Cape Town, South Africa (28). Women were recruited into the trial through community meetings and by health workers advertising a cancer prevention service. Volunteers were eligible if they: were between the ages of 35 and 65 years; had not previously had a Papanicolaou smear or treatment for cervical neoplasia; did not have any clinical signs of cervical cancer (including a four-quadrant highgrade squamous intraepithelial lesion); and were suitable for cryotherapy (e.g., had a cervix and did not have cervical polyps or prolapse). These criteria ruled out 6 percent of women who volunteered to participate, primarily because they were unsuitable for cryotherapy. All participants provided written informed consent before being enrolled in the study. Ethical approval for the study was received from the Institutional Review Board of Columbia University (New York, New York) and the Research Ethics Committee of the University of Cape Town (Cape Town, South Africa). Upon enrollment in the trial, women underwent clinical examination, completed a questionnaire on demographic characteristics and sexual behaviors, and provided blood samples for HIV testing. Samples for Neisseria gonorrhoeae and Chlamydia trachomatis were collected using endocervical cone-brushes and were tested using the Hybrid Capture GC/CT DNA Assay (Digene Corporation, Gaithersburg, Maryland). Endocervical cone-brush samples were also tested for human papillomavirus DNA using the Hybrid Capture II HPV DNA Assay (Digene Corporation). Wet-mount examinations were performed on-site during the clinical examination by trained study nurses to identify T. vaginalis. Bacterial vaginosis was assessed during the clinical examination and was considered present if at least three of the following four signs were noted: moderate or severe vaginal discharge; a positive whiff test after application of potassium hydroxide; the presence of clue cells upon wet-mount examination; and vaginal pH greater than 4.5. Blood specimens taken at each study visit were tested for HIVantibodies using the Abbott HIV 1/2 g 0 Kit on the Abbott AXSYM system (Abbott Laboratories, Abbott Park, Illinois). Positive results were conrmed using the Vironosticka HIV Uniform 2 Plus 0 Kit (Organon Teknika, Boxtel, the Netherlands). All HIV results were recorded using an anonymous, linked system, and separate voluntary counseling and testing services were provided for participants wishing to know their HIV serostatus. At a follow-up visit 6 months later, participants provided blood samples for HIV testing and completed a questionnaire which included detailed questions about intravaginal practices and sexual behaviors. This instrument was administered by trained study nurses and was designed for maximum sensitivity in detecting any type of intravaginal practice, including dry sex practices, douching or washing inside the vagina with liquids, and the insertion of any substances into the vagina. Because of concerns about the measurement of intravaginal practices, we conducted qualitative research (in-depth interviews) with women in this setting before the study began in order to investigate the different types of intravaginal practices that were known in this setting and the most appropriate terminology for referring to these practices. From this research, we developed

Downloaded from aje.oxfordjournals.org by guest on April 15, 2011

554 Myer et al.

a preliminary study instrument that was pilot-tested over a 2-month period. The results of this pilot study were used to rene the questionnaire structure and item phrasing for the nal study instrument. This instrument investigated 15 separate types of intravaginal practices (identied during qualitative research and the pilot study) and then employed open-ended questions to identify other practices that had not been specically asked about. Before asking questions on intravaginal practices, study nurses were trained to clearly distinguish intravaginal practices from wiping or cleaning of the external genitalia only. In addition to the type of practice and the substances involved, the questionnaire collected data on the reasons for womens use of intravaginal products, the frequency of various practices, and their proximity to sexual intercourse. Between March 2001 and December 2002, 4,668 women were enrolled in the study. Of these, 4,099 women (88 percent) completed the 6-month follow-up visit including the questionnaire on intravaginal practices; 4,089 had HIV test results available at baseline and at 6 months and were eligible for this analysis. All women were scheduled to be followed for at least 6 months within the cervical cancer trial (n 4,089). In addition, a subset of participants who either 1) were found to be positive upon direct visual inspection of the cervix at enrollment, 2) tested positive for human papillomavirus DNA at enrollment, and/or 3) were enrolled after January 2002 were scheduled by design to receive extended follow-up for 24 months after enrollment as part of a cervical cancer substudy (n 2,674). Thus, calculations of persontime and loss to follow-up in this analysis were based on women followed for 6 months and the subset of women scheduled for additional follow-up for 24 months. All 24month follow-up visits were completed (or participants were declared lost to follow-up) by March 2005. Data were analyzed using the statistical program SAS, version 9.0 (SAS Institute, Inc., Cary, North Carolina). We conducted separate analyses of ndings based on HIV incidence from the 6-month follow-up visit (including the entire study cohort), as well as the substudy cohort followed for 24 months; since no substantive differences were noted, we report only the results obtained at 24 months. Womens intravaginal practices were categorized according to their degree of hypothesized abrasiveness to vaginal epithelium, with a basic distinction between less abrasive practices (involving water, paper, and/or cloth only) and the insertion of other, potentially more abrasive products inside the vagina. Incident HIV infection was dened as a negative test result followed by a positive result at the next study visit, and dates of seroconversion were estimated using the midpoint of this interval (29). HIV seroconversion probabilities were compared among subgroups of the study sample using KaplanMeier methods. Proportional hazards regression models were used to estimate the association between selected covariates and HIV seroconversion. In addition, we repeated our previous analyses of the association between prevalent HIV infection and intravaginal practices using multiple logistic regression (27). For both proportional hazards and logistic regression models, variables were included in the analysis if they demonstrated an appreciable crude association with HIV infection; variables were removed from the

analysis if they showed no adjusted association with HIV and their removal did not alter the associations involving other covariates. The proportional hazards regression results are reported as hazard ratios with 95 percent condence intervals. All statistical tests were two-sided.

RESULTS

At enrollment, 519 (13 percent) of the 4,089 women with available data on intravaginal practices tested positive for HIV infection. Women with prevalent HIV infection at enrollment were signicantly more likely to report intravaginal practices than were women testing negative: 17 percent of women reporting intravaginal practices were HIV-infected as compared with 11 percent of women not reporting intravaginal practices (crude odds ratio 1.61, 95 percent condence interval (CI): 1.33, 1.95). This association persisted after adjustment for age, education, lifetime number of sexual partners, recent sexual activity, and prevalent infection with human papillomavirus, trichomonas, and chlamydia and/or gonorrhea (adjusted odds ratio 1.50, 95 percent CI: 1.22, 1.85). Prevalent HIV infection was strongly associated with an increasing amount of vaginal discharge, with no vaginal discharge being evident during clinical examination of 17 percent of HIV-negative women and 12 percent of HIV-positive women, while moderate or severe discharge was noted among 20 percent and 28 percent of HIV-negative and -positive women, respectively (ptrend < 0.001). A total of 4,641 person-years of observation were accrued among the 3,570 women who tested HIV-negative at enrollment. Within the cohort designated for follow-up beyond the 6-month visit, follow-up rates for subjects who were HIV-negative at enrollment were 83 percent (n 1,887) and 75 percent (n 1,718) at the 12- and 24-month visits, respectively. Over the course of follow-up, 83 participants missed a scheduled study visit but attended the following visit. Persons lost to follow-up were slightly younger than those who were successfully followed (mean age 43.4 years vs. 44.7 years; p 0.01), but those lost did not differ from those followed in terms of education, marital status, housing type, intravaginal practices, sexual activity, or number of sexual partners at enrollment, or by baseline prevalence of bacterial vaginosis, trichomonas, or chlamydia and/ or gonorrhea (data not shown). Baseline demographic characteristics, sexual behaviors, and prevalent sexually transmitted diseases among women who were HIV-negative at enrollment are described in table 1. The median age was 42 years (interquartile range, 3848); slightly more than half of the participants were married; and most of the women (70 percent; n 2,492) lived in shacks or informal housing (squatter camps), an indicator of low socioeconomic status. Although most women had been sexually active during the month prior to the study (72 percent; n 2,588), 98 percent reported having had only one sexual partner during this period (n 2,541), and condom use was low. Among women who tested HIV-negative at enrollment, 26 percent reported some type of intravaginal practice at
Am J Epidemiol 2006;163:552560

Downloaded from aje.oxfordjournals.org by guest on April 15, 2011

Intravaginal Practices and Risk of HIV Infection

555

TABLE 1. Demographic characteristics, sexual behaviors, and prevalent sexually transmitted infections, by self-reported intravaginal practices, among 3,570 human immunodeciency virus-negative women in Cape Town, South Africa, March 2001March 2005*
All participants No. or mean/median % or SDz/range Women reporting intravaginal practices No. or mean/median % or SD/range Women reporting no intravaginal practices p valuey No. or mean/median % or SD/range

All women Demographic characteristics Mean age (years) (with SD) Married Greater than primary school education Living in informal housing Sexual behaviors Median lifetime no. of sexual partners (with range) Sexual activity in the previous month More than one sexual partner in the previous month New sexual partner in past 5 months{ Partner had sex with other women in past 5 months{ Ever use of condoms Prevalent sexually transmitted infections Trichomonas Chlamydia and/or gonorrhea Bacterial vaginosis Human papillomavirus

3,570 43.4 1,878 1,947 2,492

100 7.0 53 55 70

945 42.6 463 538 640

26 6.4 49 57 68

2,625 43.7 1,415 1,409 1,852

74 7.1 54 54 71 <0.001 0.010 0.085 0.105

4.0 2,588 47 100 741 249

1100 72 2 3 21 7

4.0 731 19 39 226 93

1100 77 3 4 24 10

3.0 1,857 28 61 515 156

150 71 2 2 20 6

<0.001 0.001 0.061 0.004 0.005 <0.001

Downloaded from aje.oxfordjournals.org by guest on April 15, 2011

326 168 727 591

9 5 20 17

102 47 176 158

11 5 19 17

224 121 551 433

9 5 21 17

0.039 0.654 0.121 0.874

* All data are numbers and percentages unless otherwise specied. y p values were calculated from Students t tests (for means), Wilcoxon rank-sum tests (for medians), and chi-squared tests (for proportions). z SD, standard deviation. Among those reporting sexual activity in the prior month. { This information was obtained at the 6-month follow-up visit.

the 6-month follow-up interview (n 945). The majority (63 percent; n 592) of these women described potentially less abrasive practices, primarily involving the insertion of water, paper, and/or cloth only inside the vagina (table 2). Within this group, 453 women reported practices involving cloth specically. Among the 348 women reporting potentially more abrasive practices, the most common products inserted into the vagina included household or laundry soaps (n 158; 17 percent of all women reporting intravaginal practices), household antiseptics (n 121; 13 percent), vinegar (n 18; 2 percent), salt or saltwater (n 19; 2 percent), and industrial detergents (n 14; 1 percent) (participants could report using more than one product). The majority of women reporting intravaginal practices stated that these activities took place as part of regular washing (88 percent; n 835), while smaller proportions reported engaging in these practices before or after sexual intercourse (6 percent; n 60) or in association with menses (5 percent; n 50). Only nine women (1 percent) reported engaging in intravaginal practices prior to intercourse.
Am J Epidemiol 2006;163:552560

Slightly more than half reported engaging in these practices daily (51 percent; n 486), and 75 percent reported undertaking intravaginal practices during the week prior to being interviewed (n 713). The majority of women gave hygiene as the motivation for their intravaginal practices (69 percent; n 650), with a lower proportion giving removal of excess vaginal discharge as their primary reason (21 percent; n 202). There were no associations between the reasons given for intravaginal practices and the specic practices involved (data not shown). There were 85 incident HIV infections observed during the 24-month follow-up period (incidence rate 1.83 infections per 100 person-years). The seroconversion probabilities at 6, 12, and 24 months were 0.009, 0.016, and 0.037, respectively. In the crude analysis, rates of incident HIV infection did not differ signicantly among women reporting any type of intravaginal practice (incidence rate 1.97 per 100 person-years, 95 percent CI: 1.13, 2.94) and women reporting no intravaginal practices (incidence rate 1.78 per 100 person-years, 95 percent CI: 1.36, 2.29;

556 Myer et al.

TABLE 2. Distribution of intravaginal practices and their relation to human immunodeciency virus (HIV) seroconversion among 3,570 HIV-negative women in Cape Town, South Africa, March 2001March 2005
No. of incident infections No. of women reporting Probability of seroconversion* Log-rank p value* Crude hazard ratioy 95% condence interval

No intravaginal practices reported Any intravaginal practice reported Less abrasive practicesz Cloth and/or water only Cloth with/without water Fingers with/without water Other less abrasive practices More abrasive practicesz Use of household or laundry soaps Use of household disinfectants Use of vinegar Use of salt or saltwater Use of industrial detergents Use of other substances{ Timing of intravaginal practices As part of washing/hygiene During menses Before or after sexual intercourse Frequency of intravaginal practices Occasionally/sometimes Every day or frequently

61 24 19 10 10 9 0 5 3 1 1 0 0 0 20 2 2 8 16

2,625 945 592 474 453 101 21 348 158 121 18 19 14 61 835 50 60 383 562

0.036 0.040 0.052 0.033 0.048 0.129 0.021 0.032 0.013 0.700 0.163 0.901 0.961 <0.001 0.492 0.253 0.739 0.275 0.257 0.558 0.648 0.981 0.036 0.036 0.071 0.034 0.044 0.929 0.345 0.379 0.813 0.438

1.0 1.108 1.440 0.959 1.021 3.659 0.591 0.822 0.337 0.691, 1.777 0.861, 2.411 0.492, 1.872 0.522, 1.996 1.817, 7.369 0.237, 1.471 0.258, 2.619 0.047, 2.430

Downloaded from aje.oxfordjournals.org by guest on April 15, 2011

1.023 1.947 1.865 0.915 1.243

0.618, 1.695 0.476, 7.963 0.456, 7.636 0.438, 1.912 0.717, 2.155

* Seroconversion (failure) probabilities and log-rank tests were from Kaplan-Meier analyses. y Hazard ratios presented throughout the table used women reporting no intravaginal practices as the reference category. z The categories of less abrasive practices and more abrasive practices are not mutually exclusive. There were four other unique combinations of less abrasive practices reported. { There were seven other types of more abrasive practices reported, including insertion of traditional medicinal products, tobacco, and other products into the vagina.

unadjusted hazard ratio 1.11, 95 percent CI: 0.69, 1.78), nor were more abrasive practices related to seroconversion (table 2). In one small subgroup of less abrasive practices wiping inside the vagina with ngers, with or without waterthe incidence rate of HIV infection was signicantly higher than that among women reporting no intravaginal practices (crude hazard ratio 3.66, 95 percent CI: 1.82, 7.37). However, within all of the other subgroups, there were no signicant differences between women in subgroups of intravaginal practices and women reporting no intravaginal practices. Other factors associated with incident HIV infection in unadjusted analyses included younger age, being unmarried, trichomonas infection at enrollment, and chlamydia and/or gonorrhea infection at enrollment (table 3). In stratied analyses, the null association between intravaginal practices and incident HIV infection persisted across all levels of these factors. In a proportional hazards regression model adjusting for demographic characteristics, sexual behaviors, and prevalent sexually transmitted infections at baseline, the overall null association between intravaginal practices and HIV in-

cidence persisted (hazard ratio 1.04, 95 percent CI: 0.65, 1.68) (table 4). Increased hazards of incident HIV infection were associated with being unmarried and being younger. In another model (data not shown) containing separate terms for women reporting less abrasive practices and women reporting more abrasive practices (compared with women reporting no practices), there was no adjusted association between less abrasive practices and incident HIV infection (hazard ratio 0.71, 95 percent CI: 0.29, 1.77).

DISCUSSION

This was a large prospective study investigating intravaginal practices as a risk factor for incident and prevalent HIV infection in a general population sample of women. Our results demonstrate an adjusted association between intravaginal practices and prevalent HIV infection, in keeping with a previous analysis (27), but no association between intravaginal practices and incident HIV infection. This evidence may appear at rst to be contradictory, with the crosssectional result suggesting that intravaginal practices may
Am J Epidemiol 2006;163:552560

Intravaginal Practices and Risk of HIV Infection

557

TABLE 3. Crude probabilities of human immunodeciency virus (HIV) seroconversion according to selected risk factors and associated hazard ratios among 3,570 HIV-negative women in Cape Town, South Africa, March 2001March 2005
No. of incident infections No. of women reporting Probability of seroconversion* Log-rank Crude hazard p value* ratio 95% condence interval

Age (years) 3539 4049 5065 Completed primary education No Yes Marital status Not married Married Housing type Formal Informal/shack New sexual partner in past 5 monthsz No Yes Partner had sex with other women in past 5 monthsz No Yes Dont know No partner Baseline trichomonas infection Negative Positive Baseline clinical evidence of bacterial vaginosis No Yes Baseline chlamydia and/or gonorrhea infection Negative Positive 76 9 3,402 168 0.035 0.085 0.011 1.0y 2.396 1.201, 4.782 60 25 2,843 727 0.033 0.053 0.055 1.0y 1.573 0.986, 2.508 72 13 3,244 326 0.035 0.064 0.040 1.0y 1.837 1.018, 3.317 17 24 29 15 907 741 1,141 781 0.032 0.047 0.043 0.027 0.077 0.259 0.909 1.0y 1.739 1.410 1.020 0.935, 3.236 0.775, 2.565 0.721, 1.444 80 5 3,470 100 0.037 0.063 0.076 1.0y 2.217 0.898, 5.473 55 30 24 61 1,692 1,878 1,078 2,492 0.049 0.026 0.031 0.040 0.506 0.002 1.0y 0.493 1.0y 1.174 0.732, 1.882 0.316, 0.769 38 47 1,623 1,947 0.036 0.038 0.974 1.007 0.657, 1.545 43 35 7 1,345 1,537 688 0.049 0.035 0.019 0.010 0.004 1.0y 0.688 0.569 0.440, 1.075 0.381, 0.848

Downloaded from aje.oxfordjournals.org by guest on April 15, 2011

* Seroconversion (failure) probabilities and log-rank tests were from Kaplan-Meier analyses. y Reference category. z This information was obtained at the 6-month follow-up visit.

increase womens susceptibility to HIV infection and the prospective data suggesting no association. However, taken together, these results may be explained by a reversal of the causal sequence commonly assumed for this association. Specically, intravaginal practices in this setting may increase as a result of womens HIV infection.
Am J Epidemiol 2006;163:552560

Two aspects of these data support this nding. First, the vast majority of women who reported intravaginal practices stated that these behaviors were part of their regular hygiene or were done to relieve symptoms of itching or excess discharge. While we found no difference in intravaginal practices according to degree of vaginal discharge at enrollment,

558 Myer et al.

TABLE 4. Multivariate relative hazards of incident human immunodeciency virus (HIV) infection among 3,570 HIV-negative women in Cape Town, South Africa, March 2001 March 2005
Hazard ratio 95% condence interval

Any intravaginal practice (yes 1, no 0) Age (years; continuous variable) Marital status (not married 1, married 0) New sexual partner in past 5 months* (yes 1, no 0) Partner had sex with other women in past 5 months* (yes 1, no/dont know/no partner 0) Baseline trichomonas infection (yes 1, no 0) Baseline chlamydia and/or gonorrhea infection (yes 1, no 0) * This information was obtained at the 6-month follow-up visit.

1.043 0.930 2.250 1.921 1.228 1.575 1.965

0.647, 1.681 0.893, 0.968 1.430, 3.540 0.751, 4.913 0.750, 2.011 0.864, 2.874 0.977, 3.953

women entering the study with prevalent HIV infection were signicantly more likely to have moderate or severe vaginal discharge than women without HIV (27). In this light, it is plausible that HIV infection over time causes or exacerbates lower genital tract infections as immunosuppression increases, leading to increased vaginal discharge and related symptoms, which in turn motivate intravaginal practices. This is supported by data demonstrating an increased prevalence of discharge-producing infections among HIV-positive women, including vulvovaginal candidiasis, trichomonas, and bacterial vaginosis (19, 30, 31), as well as data showing that increased disease progression is linked to increased incidence and persistence of vaginal candidiasis (32) and bacterial vaginosis (33). Second, intravaginal practices prior to sexual intercourse appeared to be uncommon in this population, suggesting that these behaviors were not temporally associated with sexual intercourse and possible exposure to HIV infection. The potential for intravaginal practices to disrupt mucosal membranes prior to sex is the most commonly hypothesized mechanism for intravaginal practices increasing susceptibility to HIV (16); other mechanisms of action that may not require a close temporal association with sexual intercourse remain unclear. While there have been few prospective investigations of intravaginal practices and HIV infection, data from research carried out among Thai sex workers (34) appear to be broadly consistent with our ndings. A preliminary cross-sectional study of these brothel-based sex workers found a link between intravaginal practices and prevalent HIV infection (34), suggesting that these behaviors may increase womens susceptibility to HIV. However, later investigations revealed no association with incident HIV infection (23, 35). Taken together, these results point to similar interpretations: that intravaginal practices may be motivated by factors related to HIV infection, rather than these behaviors increasing womens risk of becoming infected with HIV. In addition to the hypothesis that HIV infection leads to increased intravaginal practices, there are several alternative explanations for the observed data. First, if intravaginal practices do truly increase womens risk of HIV infection, the disagreement between the prevalent results and the incident results could be explained by the highest-risk wom-

ens being lost to follow-up. However, the rates of follow-up in this cohort were relatively high, and there were few systematic differences between the women retained in the cohort and those who were lost to follow-up. To evaluate the possibility of differential loss to follow-up as a possible explanation for the observed data, we conducted a basic sensitivity analysis (data not shown). This analysis suggested that in order to explain these study results, persons lost to follow-up would have required at least twice the HIV incidence of those retained in the cohort and would have had to be more than ve times as likely to report intravaginal practices as those retained in the cohort. Neither of these parameters is consistent with the available data, which show that women lost to follow-up were slightly younger (mean difference in age 1.3 years) than those retained but did not differ from those retained according to any other measured variable, including reported intravaginal practices. In this light, differential loss to follow-up is unlikely to have played a signicant role in explaining the differences between incident and prevalent results. Related to this, there were no associations between any of the criteria for inclusion in the substudy cohort followed to 24 months and either womens intravaginal practices or incident HIV infection, suggesting that the possibility of selection bias is unlikely to explain these ndings. Second, with 85 seroconversions, this analysis had reasonable statistical power to detect associations of the magnitude observed for intravaginal practices and prevalent HIV. Along these lines, the 95 percent condence interval for the adjusted odds ratio for the prevalent association (95 percent CI: 1.20, 1.82) was not consistent with the adjusted hazard ratio for the incident association (hazard ratio 1.04), suggesting that random sampling error is unlikely to account for these differences. Third, it may be possible that unmeasured confounding created a spurious difference between prevalent and incident results. In particular, we did not have detailed information on high-risk sexual behaviors occurring throughout the follow-up period, and we were able to adjust for new sexual partners only at the 6-month follow-up visit (by which time the majority of seroconversions had taken place). We also did not have data on the duration and nature of menses,
Am J Epidemiol 2006;163:552560

Downloaded from aje.oxfordjournals.org by guest on April 15, 2011

Intravaginal Practices and Risk of HIV Infection

559

which could be correlated with intravaginal practices and could act as a confounder of the observed associations. While some unmeasured confounding may be plausible, we did examine a number of potential demographic, behavioral, and biologic confounding variables. It is not clear what biologic or behavioral variables could act as confounders of either the cross-sectional analysis or the prospective analysis but not both, and thus help to explain the observed associations. We measured intravaginal practices only at the 6-month follow-up visit, and we used this measure to reect behaviors occurring throughout the follow-up period. While our preliminary qualitative research (described above) suggested that these behaviors were stable through time, it is possible that intravaginal practices uctuated for some women over the course of the study, introducing some degree of measurement error into the analysis. Similarly, the reporting of intravaginal practices may be awkward for women and thus subject to some degree of systematic reporting error. However, in order for either form of misclassication in the assessment of intravaginal practices to create a bias that would explain the observed discrepancy between incident and prevalent ndings, seroconverters in the prospective study of HIV would need to substantially underreport their intravaginal practices, relative to women who would not seroconvert in the prospective study and all women in the cross-sectional study (36). In this light, it seems unlikely that these forms of mismeasurement could systematically affect the incident analysis but not the prevalent analysis. Given the wide variation in intravaginal practices reported around the world, the generalizability of these ndings should be viewed critically. Although they were still at considerable risk of HIV infection, this sample of women was substantially older (mean age 42 years) than the populations examined in other studies. More importantly, there is variability across populations in the motivations for and types of intravaginal practices. Although the forms of wiping and/or douching with cloth and/or water for hygienic purposes that appeared to be most common in this study population have been documented in several other settings (7, 12, 26), other behaviors appear to be more common in samples of women from different countries (1, 10). Intravaginal practices designed to remove vaginal moisture before intercourse (dry sex), which have been the focus of several previous studies, appeared to be uncommon in this population. The relation between intravaginal practices and HIV infection may vary with the specic practices involved, the reasons for those practices, and their proximity to sexual intercourse; as a result, our overall interpretation may not be applicable to all types of intravaginal practices in all places. In summary, these results indicate that previous crosssectional studies may have overestimated the association between womens intravaginal practices and HIV infection. The types of intravaginal practices reported in this study population appear unlikely to be an important cofactor in the spread of HIV. Instead, existing HIV infection may motivate increased intravaginal practices, spuriously inating the association between intravaginal practices and prevalent HIV. Targeting intravaginal practices for HIV prevention is
Am J Epidemiol 2006;163:552560

unlikely to be a useful strategy for reducing new infections in this setting.

ACKNOWLEDGMENTS

This research was funded by the Bill and Melinda Gates Foundation and EngenderHealth. Conict of interest: none declared.

REFERENCES 1. Dallabetta GA, Miotti PG, Chiphangwi JD, et al. Traditional vaginal agents: use and association with HIV infection in Malawian women. AIDS 1995;9:2937. 2. van de Wijgert JH, Mbizvo M, Dube S, et al. Intravaginal practices in Zimbabwe: which women engage in them and why? Cult Health Sex 2001;3:13348. 3. Karim QA, Karim SS, Soldan K, et al. Reducing the risk of HIV infection among South African sex workers: socioeconomic and gender barriers. Am J Public Health 1995;85: 15215. 4. Halperin DT. Dry sex practices and HIV infection in the Dominican Republic and Haiti. Sex Transm Infect 1999;75:4456. 5. Mann JM, Nzilambi N, Piot P, et al. HIV infection and associated risk factors in female prostitutes in Kinshasa, Zaire. AIDS 1988;2:24954. 6. Civic D, Wilson D. Dry sex in Zimbabwe and implications for condom use. Soc Sci Med 1996;42:918. 7. Beksinska ME, Rees HV, Kleinschmidt I, et al. The practice and prevalence of dry sex among men and women in South Africa: a risk factor for sexually transmitted infections? Sex Transm Infect 1999;75:17880. 8. Mbizvo MT, Mashu A, Chipato T, et al. Trends in HIV-1 and HIV-2 prevalence and risk factors in pregnant women in Harare, Zimbabwe. Cent Afr J Med 1996;42:1421. 9. La Ruche G, Messou N, Ali-Napo L, et al. Vaginal douching: association with lower genital tract infections in African pregnant women. Sex Transm Dis 1999;26:1916. 10. Sandala L, Lurie P, Sunkutu MR, et al. Dry sex and HIV infection among women attending a sexually transmitted diseases clinic in Lusaka, Zambia. AIDS 1995;9(suppl 1): S618. 11. Runganga AO, Kasule J. The vaginal use of herbs/substances: an HIV transmission facilitatory factor? AIDS Care 1995;7: 63945. 12. Orubuloye IO, Caldwell P, Caldwell JC. A note on suspect practices during the AIDS epidemic: vaginal drying and scarication in southwest Nigeria. Health Transit Rev 1995; 5:1615. 13. Gresenguet G, Kreiss JK, Chapko MK, et al. HIV infection and vaginal douching in central Africa. AIDS 1997;11:1016. 14. Morar NS, Ramjee G, Abdool Karim SS. Vaginal insertion and douching practices among sex workers at truck stops in KwaZulu-Natal. (Letter). S Afr Med J 1998;88:470. 15. Fonck K, Kaul R, Keli F, et al. Sexually transmitted infections and vaginal douching in a population of female sex workers in Nairobi, Kenya. Sex Transm Infect 2001;77:2715. 16. Brown RC, Brown JE, Ayowa OB. The use and physical effects of intravaginal substances in Zairian women. Sex Transm Dis 1993;20:969.

Downloaded from aje.oxfordjournals.org by guest on April 15, 2011

560 Myer et al.

17. Hira SK, Mangrola UG, Mwale C, et al. Apparent vertical transmission of human immunodeciency virus type 1 by breast-feeding in Zambia. J Pediatr 1990;117:4214. 18. Cu-Uvin S, Ko H, Jamieson DJ, et al. Prevalence, incidence, and persistence or recurrence of trichomoniasis among human immunodeciency virus (HIV)-positive women and among HIV-negative women at high risk for HIV infection. Clin Infect Dis 2002;34:140611. 19. Moodley P, Connolly C, Sturm AW. Interrelationships among human immunodeciency virus type 1 infection, bacterial vaginosis, trichomoniasis, and the presence of yeasts. J Infect Dis 2002;185:6973. 20. Blythe MJ, Fortenberry JD, Orr DP. Douching behaviors reported by adolescent and young adult women at high risk for sexually transmitted infections. J Pediatr Adolesc Gynecol 2003;16:95100. 21. Foxman B, Aral SO, Holmes KK. Interrelationships among douching practices, risky sexual practices, and history of self-reported sexually transmitted diseases in an urban population. Sex Transm Dis 1998;25:909. 22. Koblin BA, Mayer K, Mwatha A, et al. Douching practices among women at high risk of HIV infection in the United States: implications for microbicide testing and use. Sex Transm Dis 2002;29:40610. 23. Palanuvej T, Kilmarx PH, Limpakarnjanarat K, et al. Lack of demonstrable association between use of cervicovaginal mucosal irritant and HIV infection in female commercial sex workers in Thailand. (Letter). AIDS 1998;12:22301. 24. Kilmarx PH, Limpakarnjanarat K, Supawitkul S, et al. Mucosal disruption due to use of a widely-distributed commercial vaginal product: potential to facilitate HIV transmission. AIDS 1998;12:76773. 25. van de Wijgert JH, Chirenje ZM, Iliff V, et al. Effect of intravaginal practices on the vaginal and cervical mucosa of Zimbabwean women. J Acquir Immune Dec Syndr 2000;24: 627.

26. van de Wijgert JH, Mason PR, Gwanzura L, et al. Intravaginal practices, vaginal ora disturbances, and acquisition of sexually transmitted diseases in Zimbabwean women. J Infect Dis 2000;181:58794. 27. Myer L, Denny L, De Souza M, et al. Intravaginal practices, HIV and other sexually transmitted diseases among South African women. Sex Transm Dis 2004;31:1749. 28. Denny L, Kuhn L, De Souza M, et al. Screen-and-treat approaches for cervical cancer prevention in low-resource settings: a randomized controlled trial. JAMA 2005;294: 217381. 29. Clayton D, Hills M. Statistical models in epidemiology. New York, NY: Oxford University Press, 1993. 30. Sobel JD. Gynecologic infections in human immunodeciency virus-infected women. Clin Infect Dis 2000;31:122533. 31. Warren D, Klein RS, Sobel J, et al. A multicenter study of bacterial vaginosis in women with or at risk for human immunodeciency virus infection. Infect Dis Obstet Gynecol 2001;9:13341. 32. Duerr A, Heilig CM, Meikle SF, et al. Incident and persistent vulvovaginal candidiasis among human immunodeciency virus-infected women: risk factors and severity. Obstet Gynecol 2003;101:54856. 33. Jamieson DJ, Duerr A, Klein RS, et al. Longitudinal analysis of bacterial vaginosis: ndings from the HIV Epidemiology Research Study. Obstet Gynecol 2001;98:65663. 34. Siraprapasiri T, Thanprasertsuk S, Rodklay A, et al. Risk factors for HIV among prostitutes in Chiangmai, Thailand. AIDS 1991;5:57982. 35. Kilmarx PH, Limpakarnjanarat K, Mastro TD, et al. HIV-1 seroconversion in a prospective study of female sex workers in northern Thailand: continued high incidence among brothelbased women. AIDS 1998;12:188998. 36. Myer L, Morroni C, Link BG. Impact of measurement error in the study of sexually transmitted infections. Sex Transm Infect 2004;80:31823.

Downloaded from aje.oxfordjournals.org by guest on April 15, 2011

Am J Epidemiol 2006;163:552560