Does bacterial vaginosis modify the effect of hormonal contraception on HIV seroconversion: a prospective cohort study

Abstract

Objectives:

A recent study of HIV serodiscordant couples found that depot medroxyprogesterone acetate (DMPA) and oral contraceptive pills (OCPs) were associated with increased HIV risk in the presence, but not in the absence, of bacterial vaginosis (BV). We assessed whether BV is an effect modifier of the association between hormonal contraception and HIV seroconversion in female sex workers (FSWs) in Mombasa, Kenya.

Design:

Prospective cohort study.

Methods:

Data collected from HIV-negative FSWs from 1993–2017 were analyzed. Cox proportional hazards models were used to assess the relationship between HIV seroconversion and use of DMPA, OCPs, or hormonal contraceptive implants (Norplant, Jadelle).

Results:

A total of 1,985 women contributed 7,127 person-years of follow-up; 307 women seroconverted to HIV (4.32/100 person-years). DMPA was significantly associated with elevated risk of HIV seroconversion in women with (aHR 1.56, 95% confidence interval [CI] 1.08–2.25; p=0.02) and without (aHR 2.08, 95%CI 1.46–2.97; p<0.001) BV (interaction p=0.4). Similarly, OCP use was associated with increased HIV risk both in the presence (aHR 1.50, 95%CI 0.94–2.39; p=0.09) and absence (aHR 1.61, 95% CI 0.99–2.64; p=0.06) of BV (interaction p=0.9), though neither stratum reached statistical significance. Implants were not associated with HIV seroconversion overall, (aHR 0.99, 95%CI 0.40–2.45; p=0.9), or in women with (aHR 0.65, 95% CI 0.16–2.72; p=0.6) and without (aHR 1.39, 95% CI 0.43–4.46; p=0.6) BV (interaction p=0.5).

Conclusions:

Bacterial vaginosis had no effect on the associations between hormonal contraceptives and HIV seroconversion in this cohort. Contraceptive implants were not associated with increased HIV risk compared to no contraception.

INTRODUCTION

Hormonal contraception is an effective method for preventing unintended pregnancy, and reduces maternal morbidity and mortality [1]. Several studies have shown an association between hormonal contraception and increased risk of HIV seroconversion, although results have varied between cohorts, and by type of hormonal contraceptive used [2, 3]. Depot-medroxyprogesterone acetate (DMPA) has been associated with increased risk of HIV seroconversion in several studies, including two meta-analyses [2, 3]. Despite this, nearly an equal number of reports have failed to find an association between DMPA and HIV [2, 3]. One high-quality study demonstrated an association between use of oral contraceptive pills (OCPs) and HIV risk [4]. However, this finding has not been replicated in other cohorts [2, 3]. Use of implantable hormonal contraceptives (implants) has not been associated with HIV seroconversion in two studies [5, 6]. However, both analyses were limited in their ability to draw firm conclusions due to small sample sizes.

Recently, a study of serodiscordant couples in Zambia was the first to demonstrate that use of DMPA or OCPs increased the risk of HIV seroconversion when bacterial vaginosis (BV) was present, but not when BV was absent [7]. Analyses of data from the Mombasa Cohort, a long-term open cohort study of Kenyan female sex workers (FSWs) initiated in 1993, were among the first to show an association between use of DMPA or OCPs and HIV seroconversion [8]. These associations have remained consistent in updated analyses over time [4, 5, 8]. The present study evaluates the hypothesis that BV is an effect modifier of the association between hormonal contraception and HIV seroconversion in the Mombasa Cohort.

MATERIALS AND METHODS

HIV-seronegative FSWs participating in the Mombasa Cohort between February 1993 and April 2017 were included in this analysis. Detailed study procedures have been published [4, 5, 8, 9]. In brief, women aged 18–50 years old presenting to the Ganjoni Clinic in Mombasa for sexually transmitted infection (STI) screening, and reporting exchange of sex for cash or in-kind payment, were eligible for enrollment. All participants provided informed consent. The study was approved by the Institutional Review Boards at the University of Washington, the Fred Hutchison Cancer Research Center, and Kenyatta National Hospital.

At monthly follow-up visits, women provided demographic and behavioral data, and underwent speculum-assisted pelvic examination with collection of genital specimens for assessment of genital tract conditions, including BV. Blood was collected for laboratory testing. Cervical mucopus was defined as purulent discharge from the endocervix [5]. Vulvovaginal candidiasis and Trichomonas vaginalis infection were diagnosed by wet mount. Gram stain was used to diagnose BV using the method of Nugent and Hillier [10]. Cervicitis was defined as cervical Gram stain showing an average of >30 polymorphonuclear neutrophils (PMNs) in 3 high-power oil immersion fields. Infection with Neisseria gonorrhoeae was diagnosed by culture on modified Thayer-Martin media. Serologic testing for herpes simplex virus type-2 (HSV-2) was performed using ELISA (HerpeSelect 2, Focus Diagnostics, Cypress, CA, USA) on archived samples, with an optical density (OD) ≥2.1 considered a positive test [9].

Screening for HIV was performed by enzyme linked immunosorbent assay (ELISA) using the Detect HIV 1–2 test (Biochem Immunosystems, Montreal, Canada) from 1993 through January 2010 and the Pishtaz HIV 1.2 ELISA (Pishtaz Teb Diagnostics, Tehran, Iran) from February 2010 through April 16, 2014. Positive screening tests were confirmed by ELISA as follows: i) Recombigen HIV 1–2 (Cambridge Biotec, Worchester, MA, USA) from 1993 through August 2004; ii) Biorad HIV 1–2 (Biorad Laboratories, Hercules, CA, USA) from September 2004 through May 2006 and; iii) Vironostika HIV-1 Uni-Form II Ag/Ab (bioMérieux, Marcy I’Etoile, France) from June 2006 through April 16, 2014. On April 17, 2014, all tests were converted to 4^th generation HIV 1/2 rapid testing platforms, with the Determine Assay used for screening (Alere International Ltd, Galway, Ireland) and the Unigold assay (Trinity Biotech, Bray, Ireland) used for confirmation.

HIV seroconversion was the primary outcome for this analysis. The primary exposure was self-reported use of hormonal contraceptives (DMPA, OCPs, or contraceptive implants [Norplant/Jadelle], with each category of contraception evaluated separately). Contraceptive users were compared to women not using hormonal contraception (reference category). The primary hypothesis was that the risk of HIV seroconversion in women using hormonal contraception will be increased when BV is present (defined as Nugent score ≥7), but not when BV is absent [10]. As with prior analyses in this cohort, it was assumed that HIV infection occurred at the midpoint between study visits, which were scheduled at 30 day intervals [4, 5, 8]. Seroconversion was estimated to appear 25 days after infection [11]. Thus, HIV diagnosis was estimated to occur 45 days after infection. The effect of hormonal contraception on HIV seroconversion after discontinuation was estimated to be 70 days [4, 5, 8, 12, 13]. For BV, the duration of the effect on HIV seroconversion was assumed to be 15 days from the clinic visit at which BV was diagnosed [4, 5, 8]. Because vaginal microbiota can vary over time due to menstruation, vaginal washing, and condomless sex [14–16], we re-classified vaginal microbiota status for each interval based on the visit at the start of that interval, as in our prior analyses [4, 5, 8].

Cox proportional hazards regression models were used to assess the relationship between HIV seroconversion and use of DMPA, OCPs, or implants, compared to no hormonal contraception. To investigate whether hormonal contraception had a stronger association with HIV seroconversion when BV was present versus absent, interaction terms for BV and each type of hormonal contraceptive were added to the multivariate model. Effect modification was considered present if the p-value for the interaction terms was <0.05. Adjusted analyses were performed using potential confounding factors identified a priori based on literature suggesting associations with HIV seroconversion. Time varying adjustments were performed for the following variables: age (continuous); <1 versus ≥1 sex acts in the past week (binary); any unprotected sex in the past week (yes or no, [binary]); years of sex work (≤1, >1 to <5, ≥5 to <10, ≥10 [categorical]); HSV-2 serostatus (binary) and; incident genital tract infections, each analyzed as a separate, binary variable (N. gonorrhoeae, cervicitis, cervical mucopus, Trichomonas vaginalis infection, and vulvovaginal candidiasis). Data on education ≤8 vs. >8 years (binary) was collected at enrollment and also included in adjusted analyses [8, 17]. All models (including univariate) only include visits with complete data for all variables. For the 36,557 visits analyzed, only the following variables had missing data: cervical mucopus (13 visits, 0.03%), cervicitis (53 visits, 0.14%); N. gonorrhoeae by culture (6 visits, 0.02%), vulvovaginal candidiasis (793 visits, 2.20%), T. vaginalis (7 visits, 0.02%) and Gram stain for BV (28 visits, 0.08%).

Sensitivity analyses were performed to assess if the presence of any abnormal microbiota modified the effect of hormonal contraception on HIV seroconversion. Abnormal microbiota were defined as a Nugent score ≥4 [10]. Sensitivity analyses were otherwise performed using the same approach as the primary analyses.

RESULTS

Data collected between February 1993 and April 2017 were analyzed for a total of 1,985 enrolled women. Baseline characteristics are reported in Table 1. Participants’ median age was 27 years (interquartile range [IQR] 23–32), the median number of years of sex work was 1 (IQR 0.1–4.0), and unprotected sex in the past week was reported by 648 (32.7%) women. In terms of contraceptive use, 1,178 (59.3%) women reported use of condoms only or no contraception, 410 (20.7%) reported use of DMPA, 227 (11.4%) reported use of OCPs, 69 (3.5%) reported use of hormonal contraceptive implants, and 100 (4.7%) reported using another form of contraception. Bacterial vaginosis was identified at baseline in 645 (32.5%) women.

Table 1: Baseline Characteristics.

Data are presented as N (%) or median (interquartile range). Abbreviations: KSh, Kenyan shilling; DMPA, depo-medroxyprogesterone acetate; IUD, intrauterine device

Demographics	All participants (N = 1985)
Age	27 (23, 32)
Years of schooling completed	8 (7, 10)
Years of prostitution	1.0 (0.1, 4.0)
Charge for sex, KSh	300 (9, 925)
Sexual history
Frequency of vaginal sex in the past week	2 (1, 3)
Unprotected sex in the past week	648 (32.7%)
Number of sex partners in the past week	1 (1, 3)
Parity	2 (1,3)
Contraceptive Method
Oral Contraceptive Pills	227 (11.4%)
DMPA	410 (20.7%)
Implant	69 (3.5%)
IUD	51 (2.6%)
Tubal ligation	39 (2.0%)
Other1	10 (<0.1%)
None/condoms	1178 (59.3%)
Vaginal Washing
Reports vaginal washing (yes/no)	1851 (93.2%)
Method of Vaginal Washing
Water only	501 (25.2%)
Soap and water	729 (36.7%)
Commercial Products	129 (6.5%)
Other	491 (24.7%)
Genital Tract Conditions
Genital Ulcer Disease	33 (1.7%)
Cervical Mucopus2	76 (3.8%)
Cervicitis3	211 (10.6%)
Cervical Ectopy	270 (13.6%)
Laboratory data
Gonorrhea4	78 (3.9%)
HSV-25	1464 (73.8%)
Vulvovaginal candidiasis6	85 (4.3%)
Trichomonas vaginalis6	93 (4.7%)
Abnormal microbiota7	1125, (56.7%)
Bacterial Vaginosis8	645 (32.5%)

¹Other included spermicide (n=1), hysterectomy (n=5), or unspecified (4)

²Defined as detection of purulent endocervical drainage on pelvic examination

³Defined as cervical gram stain showing >30 PMNs in 3 oil immersion fields

⁴Diagnosed by culture on Thayer-Martin media

⁵Determined by serologic testing using ELISA (HerpeSelect 2) on archived samples; a positive test was OD ≥2.1 [9]

⁶Diagnosed using wet mount preparation

⁷Defined as a Nugent score ≥4

⁸Defined as Nugent score ≥7

Participants remained in the cohort for a median of 1.71 years (IQR 0.48–4.88), generating 7,127 person-years of follow-up time over 34,573 visits. Women reported use of condoms only or no contraception at 22,459 (61.4%) visits, DMPA at 7,109 (19.4%) visits, OCPs at 3,543 (9.7%) visits, and implants at 1,084 (3.0%) visits. Women were diagnosed with BV at 15,573 (42.6%) visits.

During the study period, 307 women seroconverted to HIV infection (4.32/100 person-years). Use of DMPA was associated with increased risk of HIV seroconversion compared to use of condoms alone or no contraception (HR 1.99, 95% confidence interval [CI] 1.55–2.53, p<0.001) (Table 2). This association remained statistically significant after adjustment for potential confounding factors (aHR 1.72, 95%CI 1.34–2.20, p <0.001). Similarly, use of OCPs was associated with HIV seroconversion compared to the reference group in both unadjusted (HR 1.73, 95%CI 1.24–2.41, p=0.001) and adjusted (aHR 1.48, 95%CI 1.05–2.09, p=0.02) analyses. In contrast, when compared to women using condoms alone or no contraception, the risk of HIV seroconversion was not increased in women reporting use of implants in unadjusted (HR=0.63, 95%CI 0.26–1.55; p=0.3) or adjusted (aHR=0.99, 95%CI 0.40–2.45; p=0.9) analyses.

Table 2: Hazard ratios for HIV seroconversion based on contraceptive use for all participants and stratified by presence or absence of bacterial vaginosis.

Contraceptive use was based on self-report at monthly visits. Time-varying exposure periods for hormonal contraception and BV are defined in the methods.

Abbreviations: OCPs, oral contraceptive pills; DMPA, depot-medroxyprogesterone acetate, BV, bacterial vaginosis.

Contraceptive Method	Seroconversions (n)1	Person-years	Incidence of HIV (100 person-years)	HR (95% CI)	p-value	Adjusted HR (95% CI)2	p-value
All participants
None3	151	4597.69	3.28	1.0		1.0
DMPA	100	1328.08	7.53	1.99 (1.55, 2.53)	<0.001	1.72 (1.34, 2.20)	<0.001
OCPs	40	543.15	7.36	1.73 (1.24, 2.41)	0.001	1.48 (1.05, 2.09)	0.02
Implant	5	203.19	2.46	0.63 (0.26, 1.55)	0.3	0.99 (0.40, 2.45)	0.9
Other4	13	476.87	2.73	0.81 (0.47, 1.40)	0.4	0.82 (0.47, 1.43)	0.5
BV present5
None3	88	1779.58	4.94	1.0		1.0
DMPA	43	381.29	11.28	1.88 (1.32, 2.67)	<0.001	1.56 (1.08, 2.25)	0.02
OCPs	20	171.58	11.66	1.74 (1.10, 2.75)	0.02	1.50 (0.94, 2.39)	0.09
Implant	2	74.41	2.69	0.45 (0.11, 1.82)	0.3	0.65 (0.16, 2.72)	0.6
Other4	6	204.88	2.93	0.54 (0.24, 1.23)	0.1	0.54 (0.24, 1.23)	0.1
BV absent5
None3	63	2818.10	2.24	1.0		1.0
DMPA	57	946.80	6.02	2.38 (1.68, 3.36)	<0.001	2.08 (1.46, 2.97)	<0.001
OCPs	20	371.57	5.38	1.87 (1.15, 3.02)	0.01	1.61 (0.99, 2.64)	0.06
Implant	3	128.78	2.33	0.89 (0.28, 2.83)	0.8	1.39 (0.43, 4.46)	0.6
Other4	7	271.99	2.57	1.16 (0.54, 2.48)	0.7	1.20 (0.53, 2.68)	0.7
Abnormal microbiota present6
None3	128	2706.38	4.73	1.0		1.0
DMPA	69	659.13	10.47	1.62 (1.02, 2.57)	0.039	1.54 (0.96, 2.48)	0.08
OCPs	33	292.39	11.29	1.47 (0.85, 2.54)	0.2	1.34 (0.76, 2.38)	0.3
Implant	3	98.34	3.05	0.40 (0.06, 2.83)	0.4	0.61 (0.08, 4.36)	0.6
Other4	9	300.02	3.00	0.42 (0.13, 1.32)	0.1	0.39 (0.12, 1.27)	0.1
Abnormal microbiota absent6
None3	23	1891.30	1.22	1.0		1.0
DMPA	31	668.95	4.63	2.22 (1.66, 2.96)	<0.001	1.83 (1.36, 2.46)	<0.001
OCPs	7	250.76	2.79	1.80 (1.18, 2.72)	0.006	1.53 (1.00, 2.35)	0.052
Implant	2	104.85	1.91	0.76 (0.28, 2.07)	0.6	1.21 (0.44, 3.31)	0.7
Other4	4	176.85	2.26	1.01 (0.54, 1.91)	1.0	1.07 (0.56, 2.04)	0.8

¹The sum of HIV-1 seroconversions is greater than the total numbers for the cohort as a whole because two women were exposed to more than one contraceptive method prior to seroconversion due to switching contraceptive methods.

²Adjusted for HSV-2, STIs (N. gonorrhoeae, cervicitis, cervical mucopus, T. vaginalis, vulvovaginal candidiasis), age, education >8 years, >1 sex act in past week, >1 sex partner in the past week, unprotected sex in past week, and years of sex work (≤1, >1 to <5, ≥5 to <10, ≥10).

³None includes no contraceptive method or condoms only.

⁴Other included: intrauterine device, spermicide, diaphragm, tubal ligation, hysterectomy, and unspecified.

⁵BV was defined as Nugent score ≥7.

⁶Abnormal microbiota was defined as Nugent score ≥4.

Diagnosis of BV was independently associated with increased risk of HIV seroconversion (HR 1.73, 95%CI 1.38–2.16, p<0.001). This relationship remained significant after adjustment for potential confounding factors (aHR 1.65, 95%CI 1.32–2.07, p<0.001). However, no significant interaction was seen between BV and use of DMPA (p=0.4), OCPs (p=0.9), or implants (p=0.5). Specifically, the association between use of DMPA and HIV seroconversion was not significantly altered when BV was present (aHR 1.56, 95%CI 1.08–2.25; p=0.02) compared to when BV was absent (aHR 2.08, 95%CI 1.46–2.97; p<0.001) (Table 2). Similarly, the association between use of OCPs and HIV seroconversion was not modified in the presence (aHR 1.50, 95%CI 0.94–2.39; p=0.09) or absence (aHR 1.61, 95%CI 0.99–2.64, p=0.06) of BV. No association between implants and HIV seroconversion was seen when BV was present (HR 0.65, 95%CI 0.16–2.72; p=0.6) versus absent (aHR 1.39, 95%CI 0.43–4.46; p=0.6).

In a sensitivity analysis, no significant interaction was seen between abnormal microbiota (Nugent score ≥4) and use of DMPA (p=0.5), OCPs (p=0.9), or implants (p=0.6). In stratified analysis, the association between use of DMPA and HIV seroconversion was not increased when abnormal microbiota were present (aHR 1.54, 95%CI 0.96–2.48; p=0.08) versus absent (aHR 1.83, 95%CI 1.36–2.46; p<0.001) (Table 2). Sensitivity analysis also did not demonstrate a relationship between OCPs and HIV seroconversion in the presence (aHR 1.34, 95%CI 0.76–2.38; p=0.3) versus absence (aHR 1.53, 95%CI 1.00–2.35; p=0.052) of abnormal microbiota. Use of implants was not associated with HIV seroconversion when abnormal microbiota were present (aHR 0.61, 95%CI 0.08–4.36; p=0.6) or absent (aHR 1.21, 95%CI 0.44–3.31; p=0.7). The relationship between HIV seroconversion and all covariates used in the study is shown in supplementary tables 1–5.

DISCUSSION

As in prior analyses of the Mombasa Cohort, associations were seen between use of DMPA or OCPs and HIV seroconversion. In contrast, use of hormonal contraceptive implants was not associated with HIV seroconversion. Bacterial vaginosis did not alter the effect of any form of hormonal contraception on HIV risk.

These results contrast with findings from a recent publication that demonstrated strong associations between DMPA or OCPs and HIV seroconversion when BV was present, but not when BV was absent, in a cohort of serodiscordant couples from Zambia [7]. There are several possible explanations for the difference in results between the present paper and those from the women in Zambia. It is possible that effect modification by BV in the Zambian cohort was seen by chance (type I error). It is also possible that the Kenyan study failed to detect effect modification (type II error), although this seems unlikely given the large sample size and high statistical power based on the large number of HIV seroconversions in this study. Another explanation is that the populations differed. Women in the Zambian cohort were married or cohabitating with an untreated HIV-seropositive partner, while women in the Mombasa Cohort were FSWs. The lack of knowledge about the HIV status of partners in the Mombasa Cohort could lead to imprecision in the measurement of exposure to HIV. Measurement of unprotected sex also differed between the two cohorts. Specifically, condomless sex was reported as a monthly time-varying measure based on self-reported total sex acts with and without a condom in the past week in the Mombasa Cohort. In contrast, the Zambian cohort used a measure of any reported instance of condomless sex over the last three months at quarterly visits [6, 7]. This distinction is important in interpreting the results of both studies, as BV is a marker for unprotected sex [16]. Finally, BV was diagnosed by Nugent score in the Mombasa Cohort, and by wet-preparation for detection of clue cells, KOH whiff test and/or Gram stain in the Zambian cohort. Differences in diagnostic techniques may have resulted in misclassification of BV, and it is difficult to predict how this might have influenced the results.

An additional important contribution of the results presented here is their confirmation of the lack of association between use of hormonal implants and HIV risk, previously seen in smaller studies [2, 5, 6]. This study included 203 person-years of follow-up in women exposed to hormonal contraceptive implants, making it one of the largest analyses to date exploring the possible association between implants and HIV seroconversion [2, 5, 6].

This study has a number of strengths. The Mombasa Cohort is a large, long-standing cohort with comprehensive data on hormonal contraception, BV, and HIV seroconversion. This is particularly important in an evaluation for effect modification, which essentially explores the effect of hormonal contraceptives in subsets of the cohort stratified by BV status. The large sample size also facilitated adjustment for potential confounding factors [8, 17]. In addition, monthly visits allowed for relatively precise evaluation of time-varying factors including the primary exposures, outcomes, and potential confounding factors.

These results should also be interpreted in the context of several limitations. Although the analyses controlled for multiple variables, unmeasured confounding factors are invariably present in this type of study, and could impact the results. Additionally, adjustment for confounding factors was restricted to variables that were available throughout the life of the cohort. Although this limitation could bias analyses of the association between hormonal contraception and HIV, it would be unlikely to impact the ability of the analysis to examine BV as an effect modifier. In addition, the dates of HIV seroconversion were estimated, which can lead to errors in categorizing women with and without exposure to hormonal contraception and BV. However, errors in measurement of hormonal contraceptive use and BV at the time of HIV seroconversion are likely to be similar between women who remained HIV negative versus those who seroconverted for HIV, which would tend to reduce the magnitude of the associations observed in the study.

In conclusion, a strong effect of BV on the association between hormonal contraceptives and HIV risk, if present, would be an important and potentially actionable finding. However, in this cohort, DMPA and OCPs were associated with similarly increased HIV risk regardless of the presence or absence of BV. If supported by the results of other observational studies and the ECHO trial (ClinicalTrials.gov ID NCT02550067), these results suggest that contraceptive implants could be recommended as a safer contraceptive option in women at increased risk for HIV acquisition.