Correlates of Adherence to the Dapivirine Vaginal Ring for HIV‑1 Prevention

Abstract

Understanding characteristics associated with adherence to pre-exposure prophylaxis (PrEP) methods for HIV-1 prevention may assist with optimizing implementation efforts. The dapivirine vaginal ring is a novel topical PrEP delivery method. Using data from a randomized, double-blind, placebo-controlled, phase III trial of the dapivirine vaginal ring conducted in four African countries, generalized estimating equation models were used to evaluate correlates of ring adherence. Two levels of quarterly dapivirine blood plasma, and dapivirine released from returned rings defined measures of adherence for recent and cumulative use, respectively. Time on study, calendar time, primary partner knowledge that the participant was taking part in the study, and use of long-acting contraceptive methods were associated with ring adherence whereas younger age, ring worries, condom use, episodes of menstrual bleeding and vaginal washing were associated with non-adherence. These findings may be useful for recruitment into future clinical studies and dapivirine ring implementation efforts.

Introduction

In 2018, 37.9 million people worldwide were living with HIV, approximately half (18.8 million) of whom were women [1]. Since 2010, among those aged 15 or older, the annual number of new infections has remained essentially constant at an estimated 1.9 million [2]. In sub-Saharan Africa, young women accounted for 25% of new HIV infections among adults and women of all ages accounted for 56% of new HIV infections [2]. The use of pre-exposure prophylaxis (PrEP) for the prevention of HIV-1 infection has been shown to be highly effective at reducing rates of HIV-1 infection in women and men when used with high adherence [3–5]. However, adherence has been variable across different populations and PrEP products preventing some trial results from showing protection against HIV-1 [6–8]. Long-acting forms of PrEP could improve user adherence and HIV-1 protection.

In two phase III trials among women, the Microbicides Trials Network (MTN)-020/ASPIRE study and The Ring Study, a vaginal ring containing dapivirine, worn for one month, reduced HIV-1 risk by ~ 30% overall and in ASPIRE, by > 50% among subgroups with greater adherence [9, 10]. This study examines participant characteristics correlated with adherence to vaginal ring use in the ASPIRE study using objective adherence markers of dapivirine levels in plasma, and dapivirine released in rings.

Methods

MTN-020/ASPIRE was a randomized, double-blind, placebo-controlled, phase III trial designed to test the safety and effectiveness of monthly use of a matrix vaginal ring containing dapivirine, a non-nucleoside reverse transcriptase inhibitor. The trial was carried out at 15 clinical sites in Malawi, South Africa, Uganda, and Zimbabwe from August 2012 through June 2015. Details of the design, methods and results of the trial (ClinicalTrails.gov number NCT01617096) were previously reported [10, 11]. Institutional review boards at each participating clinical site approved the study protocol, and participants provided written informed consent for participation and specimen storage.

Population and Study Procedures

Sexually active, non-pregnant, HIV-1 seronegative women aged 18–45 years were randomized 1:1 to use of a vaginal ring containing 25 mg of dapivirine or a matching placebo. At baseline, clinical staff collected demographic information and samples to test for sexually transmitted infections, including chlamydia, gonorrhea, trichomoniasis, and syphilis. In addition, participants were asked on case report forms about the number of sex acts she had in the past three months, whether she was worried about having the ring inside her every day for at least a year, and if she received money, material goods, gifts, drugs, or shelter in exchange for vaginal or anal sex in the past year (via an audio computer assisted self-interview). At monthly visits, women were asked about sexual practices with their male partners including condom use, partner knowledge of participation in the study and ring use, partner circumcision, partner HIV status and antiretroviral use, menstrual bleeding patterns during ring use, vaginal practices such as douching, whether she was bothered wearing the ring daily, and adherence to the use of the vaginal ring. An eligibility criteria was the use at enrollment (and intent to do so during study participation) of an effective form of contraception such as implants, intrauterine devices (IUDs), hormonal methods [i.e., intramuscular depot medroxyprogesterone acetate (DMPA-IM) and intramuscular injection norethisterone enanthate (NET-EN)], and sterilization, defined as tubal ligation/hysterectomy/laparoscopy/other surgical procedure that causes sterilization. For the purpose of this analysis, family planning methods were categorized based on the primary form of contraception in use at the visit, whereas if there were multiple methods reported, the one used the longest was taken. The most prevalent methods such as implants, IUDs and hormonal methods were evaluated as potential correlates to adherence compared to all other reported methods consisting of oral contraceptive pills, sterilization, male or female condoms, or no method.

Participants were instructed to insert and wear the ring continuously until their next scheduled monthly visit. Each participant was expected to complete at least 12 monthly visits up to a maximum of 33 months of follow-up. Vaginal rings were replaced at each monthly visit and blood specimens were collected for HIV-1 serology. At quarterly visits, blood plasma specimens were archived and subsequently tested for levels of dapivirine in active arm participants while preserving blinding of clinical staff and study participants [12]. Used rings were collected at monthly visits and sent to a central laboratory for residual drug testing. Residual drug measurements began approximately one year after the trial began due to a change from the initial plan of biofilm testing and time needed for IRB approvals at the clinical sites. The processing involved for biofilms rendered rings from the first year of the study unusable for residual drug testing.

Bi-annual groupings of 3 & 6 and 9 & 12 month visits were created for time on study to evaluate potential important phases of ring use; the first 6 month period was considered a learning phase, and the second 6 month period was considered a phase where participants pivoted from the learning phase to a phase of more sustained use. Last, the months of 15 through 33 were included to represent a maintenance period of ring use whereby participants did not change their adherence behavior dramatically [12]. Various categorizations were also explored to finalize combinations of visits into the most appropriate groupings.

Measures of Adherence

Among women randomized to the dapivirine vaginal ring arm only, two biological measures were used to construct measures of adherence at quarterly visits, as follows: dapivirine levels in blood plasma (> 95 pg/mL and > 200 pg/mL), and the amount of dapivirine released from returned rings (based on residual drug levels remaining in the ring) over the expected 28 day monthly visit period (> 0.9 mg and ≥ 4.0 mg). For plasma, samples were sent to a central laboratory where testing employed ultra-high-performance liquid chromatographic-tandem mass spectrometric methods to determine dapivirine levels [13]. The > 95 pg/mL cut-point was observed in phase I/II trials to represent a level achieved with eight hours or more of continuous use. This level, however, could misclassify a woman as adherent who did not wear the ring consistently throughout the month but inserted the ring eight hours or more prior to the clinic visit (akin to “white-coat dosing”) [14]. Moreover, this level could also represent consistent use during the 28-day period and removal from 1 to 3 days prior to the clinic visit [15, 16]. A more stringent plasma cut point of > 200 pg/mL representing a higher bar for consistency of ring use compared to the > 95 pg/mL level was also chosen. This level is based on evidence in a phase I/II trial that showed mean plasma levels of 293.4 pg/mL and 238.9 pg/mL over 4 and 12 weeks, respectively, among 140 participants in the active dapivirine ring arm. Overall mean number of days of ring use was 83.3 (± 4) days [17]. Likewise, in two other studies of the same ring examining the pharmacokinetics of dapivirine, investigators found that with consistent use over a 24-h period plasma levels of dapivirine were > 200 pg/mL [15, 16].

Acetone extraction and high-pressure liquid chromatography was used to determine residual drug levels in the rings [18]. Manufacturing load levels minus residual drug levels were used to calculate the amount of drug released from each ring. Since study visits may not have been spaced exactly to 28 days, the ratio of the amount of drug released to the number of days since ring dispensed was calculated and normalized to estimate the amount of dapivirine released during a standard 28-day period. A cut point of < 0.9 mg was chosen to represent no or very low use of the ring, equating to one standard deviation of lab measurement error above 0 mg dapivirine released based on testing of unused rings. We consequently defined an outcome for moderate to high use of the ring at a level of ≥ 0.9 mg drug released during a 28-day period. In two phase I trials and one phase I/II trial [15–17], investigators concluded that an average of 4.0–5.0 mg of dapivirine was released over a 28-day period of continuous ring use, and thus in this study > 4.0 mg was also chosen as a more stringent measure, approximating consistent use [19]. Results from quarterly visits, instead of monthly visits, beginning one year after study start were used to be consistent with the timing of the blood plasma outcomes.

Statistical Analysis

Four separate generalized estimating equation (GEE) models with a logit link and exchangeable working correlation structure were built for participants randomized to the dapivirine ring arm only using a three-step process. Step one entailed constructing univariable GEE models for all potential correlates (data not shown), including baseline, demographic and time-dependent behavioral factors, to determine covariates to be included in building a multivariable model. Based on statistical significance of p ≤ 0.05 in univariable models and findings from previous analyses [10, 12] the following variables were selected to be included in all multivariable models: clinical site, age, time on study, and calendar time.

In step two, a high-performance statistical modeling selection routine [20] was employed using variables determined in step one. After forcing into each model the four variables described in step one, the routine assessed each additional covariate based on p ≤ 0.20 for entry and p ≤ 0.10 to remain in the model. Note that the standard error for the estimated beta coefficients using this model selection routine did not account for the intra-class correlation in measures. However, we expect the routine to provide more inclusive covariate selection since the standard errors of the beta estimates would be overestimated compared to when adjusting for intra-class correlation in the measures. We considered this approach acceptable given the large number of covariates tested and our objective of discovering any potential correlate to adherence for future hypothesis generation. Once candidate covariates were determined, step three involved constructing a final multivariable GEE model with a logit link function and exchangeable working correlation structure for each outcome. Statistical analyses were completed on participant-visits whereby at least one vaginal ring was previously dispensed. Vaginal rings not returned to the clinical sites were considered as not used. Associations are shown as odds ratios (ORs) with their corresponding 95% confidence intervals (CIs). Nominal p-values were not adjusted for multiple comparisons since this analysis was exploratory in nature. Statistical analyses were conducted in SAS, version 9.4 (SAS Institute, Cary, North Carolina).

Results

In total, 1313 women were randomized to the dapivirine arm of ASPIRE. The present analysis includes data from 1273 (97%) participants and a total of 8858 quarterly visits for two models with plasma outcomes and 1218 (93%) participants and a total of 7265 quarterly visits for two models with outcomes based on the amount of drug released from the rings. Participant follow-up was 12–33 months. As displayed in Table 1, owing to missing data across the covariates included in each model, the number of participant-visits varied from model to model. Among women randomized to the dapivirine ring arm, at baseline, approximately half were 18–26 years old, most had some secondary education or higher, and more than half were not married. Nearly all reported a primary partner in the three months prior to enrollment; three-quarters of participants reported disclosing their participation in the study to their partner(s) at baseline. Approximately half reported at baseline that they used a condom (46%) and 30% reported no condom use during their last act of vaginal sex in the past seven days, with the remainder reporting no recent sex (24%). Approximately 30% of participants reported at baseline that they were somewhat or very worried about wearing the vaginal ring every day for a year. In the prior three months, 77% used something (such as tissue, toilet paper, cloth, cotton wool, tampon, sanitary pad, water with soap inside the vagina or water without soap inside the vagina) to control or manage menstrual-cycle related spotting or bleeding, and 60% put something inside their vagina (such as for vaginal washing or drying). Approximately 6% received money, material goods, gifts, drugs, or shelter in exchange for sex in the previous year. Twenty-two percent were found to have a sexually transmitted infection at baseline: chlamydia (13%), gonorrhea (4%), trichomoniasis (7%), or syphilis (1%). At baseline, DMPA-IM was the most commonly used family planning method (39%), followed by implants (19%), the intramuscular injection NET-EN (15%), intrauterine devices (12%) and other forms of contraception such as oral contraceptive pills (10.4%) or sterilization (3.4%).

Table 1.

Correlates of adherence to the dapivirine vaginal ring, summary of multivariable models^a

		Plasma DPV >95 pg/mL	Plasma DPV >200 pg/mL	DPV Released from Ring ≥0.9 mg	DPV Released from Ring >4.0 mg
Characteristicb	Participant-visits (%) (Total: 8858)	Participant-visits (%) OR (95% CI) (p-value) (Participants: 1273, Participant-visits: 8597)	Participant-visits (%) OR (95% CI) (p-value) (Participants: 1273, Participant-visits: 8610)	Participant-visits (%) OR (95% CI) (p-value) (Participants: 1218, Participant-visits: 7222)	Participant-visits (%) OR (95% CI) (p-value) (Participants: 1218, Participant-visits: 7097)
Time on Study
3- & 6-Month Visit	2436 (27.5)	2376 (27.6) Ref	2381 (27.7) Ref	1285 (17.8) Ref	1267 (17.9) Ref
9- & 12-Month Visit	2296 (25.9)	2213 (25.7) 1.34 (1.15, 1.56) (p=0.0001)	2221 (25.8) 1.12 (1.00, 1.26) (p=0.04)	1943 (26.9) 1.03 (0.86, 1.23) (p=0.73)	1910 (26.9) 0.96 (0.81, 1.12) (p=0.58)
15 through 33 Month Visit	4126 (46.6)	4008 (46.6) 1.34 (1.08, 1.65) (p=0.007)	4008 (46.6) 1.35 (1.17, 1.56) (p<0.0001)	3994 (55.3) 0.90 (0.72, 1.13) (p=0.37)	3920 (55.2) 1.01 (0.84, 1.21) (p=0.93)
Calendar Time
Aug2012-Dec2012	60 (0.7)	57 (0.7) Ref	57 (0.7) Ref	N/Ac	N/A
Jan2013-Dec2013	2539 (28.7)	2455 (28.6) 1.82 (1.08, 3.08) (p=0.025)	2462 (28.6) 2.07 (1.19, 3.60) (p=0.01)	1137 (15.7) Ref	1113 (15.7) Ref
Jan2014-Dec2014	4424 (49.9)	4293 (49.9) 2.51 (1.45, 4.35) (p=0.001)	4296 (49.9) 2.30 (1.31, 4.04) (p=0.004)	4321 (59.8) 1.08 (0.90, 1.30) (p=0.41)	4244 (59.8) 1.18 (1.00, 1.39) (p=0.05)
Jan2015-June2015	1835 (20.7)	1792 (20.8) 2.01 (1.14, 3.55) (p=0.016)	1795 (20.9) 3.29 (1.86, 5.84) (p<0.0001)	1764 (24.4) 0.92 (0.72, 1.18) (p=0.51)	1740 (24.5) 0.89 (0.71, 1.11) (p=0.30)
Baseline Characteristics	N (%) at Baseline (Total N=1313)
Age
18–21	269 (20.5)	1730 (20.1) 0.56 (0.40, 0.78) (p=0.0005)	1723 (20.0) 0.72 (0.58, 0.88) (p=0.0015)	1388 (19.2) 0.66 (0.47, 0.92) (p=0.01)	1366 (19.3) 0.92 (0.73, 1.16) (p=0.49)
22–26	396 (30.2)	2551 (29.7) 0.63 (0.48, 0.82) (p=0.0006)	2554 (29.7) 0.86 (0.72, 1.03) (p=0.11)	2193 (30.4) 0.65 (0.49, 0.86) (p=0.003)	2132 (30.0) 0.93 (0.77, 1.12) (p=0.42)
27–45	648 (49.4)	4316 (50.2) Ref	4333 (50.3) Ref	3641 (50.4) Ref	3599 (50.7) Ref
Education
No secondary education	212 (16.2)	N/A	1383 (16.1) Ref	N/A	N/A
Some secondary education or higher	1101 (83.9)		7227 (83.9) 0.89 (0.70, 1.13) (p=0.35)
Marital Status
Not Married	786 (59.9)	N/A	N/A	N/A	N/A
Married	527 (40.1)
Alcoholic Drinks Per Weekd
No drinks	1141 (86.9)	7412 (86.2) Ref	7427 (86.3) Ref	6256 (86.6) Ref	6153 (86.7) Ref
1–6 drinks	139 (10.6)	992 (11.5) 1.16 (0.82, 1.65) (p=0.41)	990 (11.5) 1.19 (0.93, 1.51) (p=0.16)	806 (11.2) 0.90 (0.61, 1.34) (p=0.62)	790 (11.1) 0.87 (0.65, 1.15) (p=0.32)
7 or more drinks	32 (2.4)	193 (2.2) 0.68 (0.36, 1.29) (p=0.24)	193 (2.2) 0.70 (0.42, 1.17) (p=0.18)	160 (2.2) 0.52 (0.26, 1.04) (p=0.07)	154 (2.2) 0.63 (0.36, 1.11) (p=0.11)
Owns Mobile Phone
Does not own a phone	129 (9.8)	N/A	791 (9.2) Ref	N/A	N/A
Owns a phone	1184 (90.2)		7819 (90.8) 1.16 (0.88, 1.52) (p=0.29)
Travel Time to Clinic
< 30 mins	306 (23.3)	2054 (23.9) 0.89 (0.39, 2.01) (p=0.78)	N/A	N/A	N/A
30–60 mins	655 (49.9)	4305 (50.1) 1.02 (0.47, 2.22) (p=0.96)
1–2 hours	310 (23.6)	1976 (23.0) 0.98 (0.45, 2.14) (p=0.97)
> 2 hours	42 (3.2)	262 (3.1) Ref
Earns Own Income
Does not earn own income	707 (53.9)	4490 (52.2) Ref	N/A	N/A	N/A
Earns own income	606 (46.2)	4107 (47.8) 0.93 (0.71, 1.20) (p=0.57)
Body Mass Indexd
Underweight: ≤ 18.5	41 (3.1)	253 (2.9) Ref	N/A	N/A	212 (3.0) Ref
Normal weight: = 18.5–24.9	561 (42.7)	3639 (42.3) 1.03 (0.56, 1.91) (p=0.92)			3043 (42.9) 0.78 (0.47, 1.28) (p=0.32)
Overweight: = 25–29.9	338 (25.7)	2320 (27.0) 1.08 (0.57, 2.03) (p=0.82)			1923 (27.1) 0.64 (0.39, 1.07) (p=0.09)
Obesity: ≥ 30	356 (27.1)	2385 (27.7) 0.77 (0.41, 1.46) (p=0.43)			1919 (27.0) 0.73 (0.44, 1.21) (p=0.22)
Number of Vaginal Sex Acts in Past 3 Months
Mean (Std Dev)	26.5 (24.7)	8597 (100.0) 0.99 (0.98, 1.00) (p=0.003)	N/A	N/A	N/A
Median (25th, 75th %tile)	20 (7,36)
Received Money, Material Goods, Gifts, Drugs, or Shelter in Exchange for Vaginal or Anal Sex in the Past Yeard
No	1230 (93.7)	8170 (95.0) Ref	8187 (95.1) Ref	N/A	N/A
Yes	73 (5.6)	427 (5.0) 0.70 (0.47, 1.06) (p=0.095)	423 (4.9) 0.79 (0.56, 1.13) (p=0.20)
Any Sexually Transmitted Infectione
No (All Negative)	1028 (78.3)	N/A	N/A	N/A	N/A
Yes (Any positive)	285 (21.7)
Chlamydia
Negative	1138 (86.7)	N/A	7533 (87.5) Ref	N/A	N/A
Positive	175 (13.3)		1077 (12.5) 0.85 (0.69, 1.05) (p=0.14)
Gonorrhea
Negative	1255 (95.6)	N/A	N/A	N/A	N/A
Positive	58 (4.4)
Trichomoniasisd
Negative	1221 (93.0)	N/A	N/A	N/A	N/A
Positive	91 (6.9)
Syphilis
Negative	1297 (98.8)	N/A	8518 (98.9) Ref	N/A	N/A
Positive	16 (1.2)		92 (1.1) 1.30 (0.77, 2.19) (p=0.33)
Worried about Having Ring Inside Her Every Day for At Least A Year
Not at all worried	926 (70.5)	5981 (69.6) Ref	5988 (69.6) Ref	N/A	N/A
Somewhat worried	366 (27.9)	2495 (29.0) 0.74 (0.57, 0.95) (p=0.02)	2501 (29.1) 0.87 (0.72, 1.04) (p=0.12)
Very worried	21 (1.6)	121 (1.4) 0.79 (0.36, 1.76) (p=0.57)	121 (1.4) 0.61 (0.33, 1.13) (p=0.12)
Time-dependent Characteristics
Percentage of Vaginal Sex Acts Protected by a Condom in Past 7 Days
0%	349 (26.6)	2842 (33.1) Ref	N/A	N/A	N/A
1–49%	66 (5.0)	336 (3.9) 1.11 (0.74, 1.66) (p=0.61)
50–99%	105 (8.0)	531 (6.2) 0.90 (0.64, 1.27) (p=0.56)
100% (includes no sex)	793 (60.4)	4888 (56.9) 1.15 (0.79, 1.68) (p=0.46)
Male or Female Condom Use During Last Act of Vaginal Sex in Past 7 Daysd
No Sex in Past 7 Days	315 (24.0)	2138 (24.9) Ref	2142 (24.9) Ref	N/A	1776 (25.0) Ref
Neither	390 (29.7)	3040 (35.4) 0.98 (0.68, 1.42) (p=0.92)	3037 (35.3) 0.88 (0.78, 1.00) (p=0.05)		2595 (36.6) 0.88 (0.75, 1.03) (p=0.10)
Male, Female or Both	602 (45.9)	3419 (39.8) 0.89 (0.75, 1.06) (p=0.18)	3431 (39.9) 0.85 (0.76, 0.95) (p=0.005)		2726 (38.4) 0.94 (0.81, 1.09) (p=0.40)
Anal Sex in Past 3 Monthsd
No	1287 (98.0)	N/A	N/A	N/A	N/A
Yes	25 (1.9)
Number of Male Sex Partners in Past 3 Months
0–1	1102 (83.9)	N/A	N/A	N/A	N/A
2	156 (11.9)
3 or more	55 (4.2)
Sex with Primary Partner in Past 3 Months
No	2 (0.2)	N/A	N/A	N/A	N/A
Yes	1305 (99.4)
No Primary Partner	6 (0.5)
Primary Partner Knows Participant is Taking Part in the Study
No/not sure	331 (25.2)	1318 (15.3) Ref	1326 (15.4) Ref	1055 (14.6) Ref	N/A
Yes	976 (74.3)	7072 (82.3) 1.45 (1.04, 2.04) (p=0.03)	7073 (82.2) 1.25 (0.99, 1.58) (p=0.06)	5966 (82.6) 1.16 (0.87, 1.54) (p=0.32)
No primary partner	6 (0.5)	207 (2.4) 0.77 (0.43, 1.38) (p=0.38)	211 (2.5) 1.22 (0.85, 1.75) (p=0.28)	201 (2.8) 1.96 (1.07, 3.60) (p=0.03)
Primary Partner Knows Participant is using Vaginal Ring
No/Unsure	472 (36.0)	1912 (22.2) Ref	1921 (22.3) Ref	N/A	N/A
Yes	835 (63.6)	6478 (75.4) 0.90 (0.66, 1.23) (p=0.52)	6478 (75.2) 0.87 (0.71, 1.07) (p=0.20)
No Primary Partner	6 (0.5)	207 (2.4) 1.00 (N/A) (p=N/A)f	211 (2.5) 1.00 (N/A) (p=N/A)f
Same Primary Partner in Past 3 Months
No	45 (3.4)	172 (2.0) Ref	N/A	N/A	143 (2.0) Ref
Yes	1262 (96.1)	8452 (95.6) 0.64 (0.43, 0.94) (p=0.02)			6762 (95.3) 1.43 (0.94, 2.15) (p=0.09)
No primary partner	6 (0.5)	207 (2.4) 1.00 (N/A) (p=N/A)f			192 (2.7) 1.76 (1.00, 3.09) (p=0.05)
HIV Status of Primary Partner in Past 3 Months
Negative	707 (53.9)	4847 (56.4) Ref	N/A	N/A	N/A
Positive	22 (1.7)	205 (2.4) 0.71 (0.32, 1.59) (p=0.40)
Don’t Know	578 (44.0)	3338 (38.8) 1.11 (0.95, 1.30) (p=0.19)
No Primary Partner	6 (0.5)	207 (2.4) 1.00 (N/A) (p=N/A)f
Primary Partner is Taking Antiretroviral Drugsd
No	1101 (83.9)	7210 (83.9) Ref	7214 (83.8) Ref	N/A	N/A
Yes	15 (1.1)	140 (1.6) 1.67 (0.60, 4.63) (p=0.33)	140 (1.6) 1.29 (0.79, 2.12) (p=0.31)
Don’t know	190 (14.5)	1040 (12.1) 1.14 (0.91, 1.42) (p=0.27)	1045 (12.1) 0.99 (0.87, 1.14) (p=0.93)
No Primary partner	6 (0.5)	207 (2.4) 1.00 (N/A) (p=N/A)f	211 (2.5) 1.00 (N/A) (p=N/A)f
Primary Partner is Circumcisedd
No	705 (53.7)	N/A	4343 (50.4) Ref	3641 (50.4) Ref	N/A
Yes	551 (42.0)		3864 (44.9) 1.00 (0.87, 1.15) (p=1.00)	3238 (44.8) 0.88 (0.72, 1.09) (p=0.24)
Don’t Know	51 (3.9)		192 (2.2) 1.06 (0.75, 1.49) (p=0.76)	142 (2.0) 1.27 (0.83, 1.94) (p=0.27)
No Primary Partner	6 (0.5)		211 (2.5) 1.00 (N/A) (p=N/A)f	201 (2.8) 1.00 (N/A) (p=N/A)f
Bothered Wearing Ring Every Dayg,h
No		N/A	N/A	N/A	N/A
Yes
Number of Times Started or Had Her Menstrual Period in the Last 3 Monthsh
0		4161 (48.4) Ref	4166 (48.4) Ref	N/A	3438 (48.4) Ref
1		992 (11.5) 0.92 (0.75, 1.13) (p=0.45)	995 (11.6) 0.91 (0.78, 1.07) (p=0.26)		751 (10.6) 0.92 (0.76, 1.12) (p=0.42)
2		1272 (14.8) 1.00 (0.80, 1.26) (p=0.99)	1275 (14.8) 0.94 (0.79, 1.11) (p=0.44)		1036 (14.6) 0.78 (0.63, 0.97) (p=0.03)
3		2172 (25.3) 1.21 (0.94, 1.57) (p=0.15)	2174 (25.3) 0.85 (0.71, 1.01) (p=0.07)		1872 (26.4) 0.75 (0.60, 0.95) (p=0.01)
Used Somethingi to Control Spotting or Bleeding in Last 3 Monthsd,j
No	7 (0.5)	N/A	55 (0.6) Ref	N/A	41 (0.6) Ref
No Spotting/Bleeding	295 (22.5)		2741 (31.8) 0.94 (0.60, 1.48) (p=0.79)		2247 (31.7) 0.89 (0.47, 1.67) (p=0.71)
Yes	1008 (76.8)		5814 (67.5) 0.76 (0.48, 1.19) (p=0.22)		4809 (67.8) 0.70 (0.37, 1.32) (p=0.27)
Put Anything Inside Vagina in Last 3 Monthsd,j
No	518 (39.5)	5773 (67.2) Ref	5788 (67.2) Ref	5032 (69.7) Ref	4949 (69.7) Ref
Yes	793 (60.4)	2824 (32.9) 0.86 (0.74, 1.00) (p=0.06)	2822 (32.8) 0.90 (0.81, 1.01) (p=0.08)	2190 (30.3) 0.83 (0.71, 0.97) (p=0.02)	2148 (30.3) 0.95 (0.82, 1.10) (p=0.51)
Primary Family Planning Methodk
Implants	255 (19.4)	1902 (22.1) 1.72 (1.20, 2.45) (p=0.003)	1909 (22.2) 1.40 (1.10, 1.77) (p=0.005)	1699 (23.5) 1.61 (1.17, 2.21) (p=0.004)	1685 (23.7) 1.61 (1.22, 2.12) (p=0.0007)
Intrauterine device	161 (12.3)	1697 (19.7) 1.24 (0.87, 1.75) (p=0.23)	1697 (19.7) 0.88 (0.70, 1.10) (p=0.25)	1517 (21.0) 1.40 (1.00, 1.98) (p=0.05)	1484 (20.9) 0.77 (0.57, 1.06) (p=0.11)
Norethisterone enanthate	200 (15.2)	986 (11.5) 1.51 (1.09, 2.09) (p=0.01)	987 (11.5) 1.53 (1.18, 1.98) (p=0.001)	758 (10.5) 1.12 (0.80, 1.56) (p=0.52)	736 (10.4) 1.48 (1.07, 2.05) (p=0.02)
Depot medroxyprogesterone acetate	515 (39.2)	2838 (33.0) 1.53 (1.14, 2.04) (p=0.004)	2840 (33.0) 1.13 (0.91, 1.40) (p=0.27)	2253 (31.2) 1.67 (1.25, 2.22) (p=0.0004)	2213 (31.2) 1.96 (1.51, 2.56) (p<0.0001)
Other (includes oral contraceptive pills, sterilization, male or female condoms only, no family planning method)l	182 (13.9)	1174 (13.7) Ref	1177 (13.7) Ref	995 (13.8) Ref	979 (13.8) Ref

^aThere are four separate multivariable models. Site (not shown), age, study time, and calendar time were included in all models and additional covariates were based on model selection criteria. The number of participant-visits varied across models due to missing data. Models using DPV released from rings as the outcome include data from vaginal rings collected starting approximately one year after the study was initiated, and thus fewer participant-visits were available for analysis compared to models using plasma outcomes. Adjusted odds ratios controlling for all other variables are displayed for each model

^bCovariates with a 95% confidence interval (CI) lower bound > 1.00 are highlighted in dark green (positive association with adherence), and those with a 95% CI upper bound < 1.00 are highlighted in dark orange (negative association with adherence). Covariates flagged as having a “marginal” positive association with adherence are highlighted in light green (defined by a 95% CI lower bound ≥ 0.90 and ≤ 1.00), and those flagged as having a “marginal” negative association with adherence are highlighted in light orange (defined by a 95% CI upper bound ≥ 1.00 and ≤ 1.10). Given our desire to broadly identify participant-characteristics that may be important for future research in this study cohort, we identified “marginal” associations recognizing that we may have been under-powered to detect a statistically significant association

^cN/A indicates that the characteristic was evaluated in univariable models but was not selected into the multivariable model

^dCategory with missing data, N (%): alcoholic drinks per week: 1 (0.1%), Body Mass Index: 17 (1.3%), anal sex in past 3 months: 1 (0.1%), male or female condom use during last act of vaginal sex in past 7 days: 6 (0.5%), primary partner is taking antiretroviral drugs: 1 (0.1%), used something to control spotting or bleeding in last 3 months: 3 (0.2%), put anything inside vagina in last 3 months: 2 (0.2%), received money, material goods, gifts, drugs, or shelter in exchange for vaginal or anal sex in the past year: 10 (0.8%), trichomoniasis: 1 (0.1%)

^eIncludes baseline results only for chlamydia, gonorrhea, trichomoniasis and syphilis

^fDue to low numbers and lack of variability in this category compared to the reference group, the OR was equal to 1.00, and thus the confidence interval and p-value were inestimable

^gThis variable is missing if a participant did not have a ring in her possession during the visit-month

^hNo data was collected at baseline

ⁱIncludes tissue, toilet paper, cloth or cotton wool put inside the vagina, or tissue, toilet paper, cloth or cotton wool placed in underwear/clothing, or tampon alone, or sanitary pad alone, or water without soap inside the vagina, or water with soap inside the vagina, or anything else

^jMissing data was imputed using the convention of last value carried forward, however, some missing values could have remained in the analysis dataset if all quarterly data was missing

^kFamily planning method considered to be the primary method in use at the visit

^lAt baseline, frequency of use is as follows: oral contraceptive pills [N = 137 (10.4%)], sterilization (tubal ligation/hysterectomy/laparoscopy/other surgical procedure that causes sterilization) [N = 45 (3.4%)], male or female condoms only (N = 0), no family planning method (N = 0)

In multivariable analyses, longer time in follow-up for individual women and later calendar time for the study was significantly positively associated with plasma adherence outcomes of > 95 pg/mL and > 200 pg/mL. However, this result was not replicated in the outcomes based on the amount of dapivirine released from the ring. Younger age was also negatively associated with the plasma adherence outcomes of > 95 pg/mL and > 200 pg/mL, and the outcome of ≥ 0.9 mg DPV released from ring. As shown in Table 1, compared to women 27–45 years old, 18–21 and 22–26 year old participants were less likely to be adherent with ORs (95% CIs) of 0.56 (0.40, 0.78) and 0.63 (0.48, 0.82), respectively, for the outcome of plasma > 95 pg/mL, and 0.72 (0.58, 0.88) and 0.86 (0.72, 1.03) for the plasma > 200 pg/mL outcome. Likewise, the ORs (95% CIs) were 0.66 (0.47, 0.92) and 0.65 (0.49, 0.86), respectively, for the outcome of ≥ 0.9 mg dapivirine released from the ring.

Additional baseline variables found in model selection for the plasma > 95 pg/mL outcome were the number of alcoholic drinks per week, travel time to the clinic, whether the participant earns her own income, and BMI. For the plasma > 200 pg/mL outcome, other baseline covariates included education, the number of alcoholic drinks per week, and whether the participant owns a mobile phone. For the outcome of ≥ 0.9 mg dapivirine released from ring, seven or more alcoholic drinks per week compared to no drinks per week was found to be marginally associated, whereas higher BMI compared to being underweight was marginally associated with the > 4.0 mg outcome.

An increase by one in the number of vaginal sex acts in the prior three months from baseline was marginally negatively associated with the plasma > 95 pg/mL adherence outcome. Time-dependent condom use, and no condom use during the last act of vaginal sex in the past seven days compared to no sex were marginally negatively associated with three of the four adherence outcomes. Primary partner knowledge of participants involvement in the study over time was positively associated with plasma adherence outcomes. Primary partner knowledge that the participant is using the vaginal ring was selected into the plasma adherence models only and was marginally negatively associated with the plasma > 200 pg/mL outcome. Compared to not having the same primary partner in the past three months, participants reporting no primary partner or those having the same primary partner was associated with the outcome of > 4.0 mg dapivirine released from the ring with ORs (95% CIs) of 1.43 (0.94, 2.15), and 1.76 (1.00, 3.09), respectively. For the plasma > 95 pg/mL outcome, the association was in the opposite direction with an OR (95% CI) of 0.64 (0.43, 0.94) for those reporting that they had the same primary partner.

Whether the participant was somewhat worried about having the ring inside her every day for at least a year compared to not being worried at all was negatively associated with the plasma adherence outcomes [OR (95% CI): 0.74 (0.57, 0.95) for the > 95 pg/mL, and 0.87 (0.72, 1.04) for the > 200 pg/mL adherence outcomes]. Reporting more menstrual bleeding episodes in the last three months compared to no bleeding was marginally associated with lower adherence for plasma > 200 pg/mL [OR (95% CI): 0.91 (0.78, 1.07) for one episode, and 0.85 (0.71, 1.01) for three episodes], and statistically significantly associated with lower adherence for the outcome of > 4.0 mg dapivirine released from the ring [OR (95% CI): 0.78 (0.63, 0.97) for two episodes and 0.75 (0.60, 0.95) for three episodes]. Participants reporting having put something inside the vagina in the last three months compared to not were less likely to be adherent for all outcomes. Whether a participant reported receiving money, material goods, gifts, drugs, or shelter in exchange for vaginal or anal sex was marginally negatively associated with the > 95 pg/mL plasma outcome [OR (95% CI): 0.70 (0.47, 1.06)]. A positive test result for chlamydia at baseline showed a marginal negative association for the plasma > 200 pg/mL outcome only [OR (95% CI): 0.85 (0.69, 1.05)].

The use of implants, NET-EN, or DMPA-IM as the primary family planning method compared to a combined reference group of oral contraceptive pills, sterilization, male or female condoms, and no family planning method were each statistically significantly associated with positive adherence for all outcomes. The use of an IUD compared to the reference group, however, showed marginal positive association with the lower plasma outcome, and dapivirine released from the ring adherence outcome. However, a marginal negative association was seen with IUD use and the higher plasma outcome, and dapivirine released from the ring adherence outcome.

Discussion

In this study of the dapivirine vaginal ring, a novel topical HIV-1 PrEP agent, demographic, behavioral, and time-dependent factors were related to objective measures of adherence. Our analyses suggest that vaginal ring use improved over time for individual women as measured by plasma outcomes, possibly indicating improving acceptance, better understanding of proper use, and improving acceptability by their partners.

Two sets of objective measures, each with two cut-offs, were utilized in this study to define adherence to the dapivirine vaginal ring. No perfect measure of adherence to the dapivirine vaginal ring has yet been defined and thus the four measures used in this study each have strengths and weaknesses. Plasma dapivirine levels measure recent use, with the higher value of > 200 pg/mL potentially more strongly defining use than the > 95 pg/mL value. If the ring was consistently used, the plasma measures would reflect adherence; however, neither plasma cut-off can definitively indicate whether the ring had been in place consistently over the prior month. Residual dapivirine levels in returned rings provide a measure of adherence across a month of use, with the ≥ 0.9 mg marker ruling out no or very little use of the ring and the > 4.0 mg marker likely indicating consistent use. However, some adherent women would be expected to release < 4.0 mg even with the ring in place at all times. Thus, all four markers carry some level of uncertainty in classifying adherence, and thus may be subject to misclassification in use specifically when HIV-1 exposure may have been present (i.e., a consistent user could remove a ring for sex, or an inconsistent user may consistently use the ring around the time of sex). For these reasons and lack of power for some of the categories of correlates, in addition to the fact that the first year of data was unavailable for the models using the residual drug levels in the rings, we observed variation in results for some correlates across the adherence measures. Thus, our results suggest factors associated with ring use, which could be confirmed or refuted in future studies of adherence to the dapivirine vaginal ring across populations and geographies.

In line with the TRIO study [21] where women utilizing highly-effective, long-acting reversible contraception (LARC) such as implants or IUDs were more likely to state choice of the ring over injections or tablets, our findings show that implants and injectable hormonal contraception (DMPA-IM or NET-EN) were positively associated with ring use. This could be due to participants being early adopters of new technology and thus more willing to integrate the use of a novel strategy like the ring for HIV-1 prevention [21]. Or, perhaps women who have had experience using a device inside the body, such as an implant, may be more comfortable with the use of a vaginal ring.

Our findings of the association of time on study, and calendar time, with adherent use of the vaginal ring as measured by plasma levels of dapivirine are consistent with prior exploratory analyses conducted in the ASPIRE cohort [10]. Qualitative work in ASPIRE found that participants became more comfortable with wearing the ring once they adjusted to using it and fears of negative side effects dissipated with counseling and support from study staff and peers over time [22]. Other studies support the hypothesis that participants initially may have concerns about the novelty of the ring and over time develop a sense of comfort [17, 23, 24]. The association with calendar time likely reflects in part implementation of a real time adherence monitoring intervention, as previously reported [12, 32]. Used rings were not available for this analysis from the first year of the study, which may explain why a similar positive trend was not seen for the ring residual drug outcomes. Our findings are in stark contrast to other HIV prevention and treatment strategies in trials that have found waning adherence over time and may suggest favorable persistence to the dapivirine vaginal ring, once initial barriers to use are overcome [25, 26].

Some participant characteristics appeared to have played a role in adherence to use of the ring. Younger age compared to older age is associated with less consistent adherence to ring use, as was seen in earlier analyses from ASPIRE and in other PrEP studies [9, 10, 35]. This finding is consistent with studies showing that younger women may have challenges maintaining full adherence to HIV-1 treatments, prevention agents, and other similar user-dependent strategies like contraceptives [27–30]. Importantly, our finding suggests that some younger persons will need additional adherence support in order to realize the ring’s HIV-1 risk reduction benefits. Moreover, safety of ring use, perhaps due to intentions to get pregnant and fear of effects on a future pregnancy, should be an area of further study since unproven safety of an investigational product could be a possible explanation for the observed reduction of adherence among young participants in the ASPIRE cohort.

Other participant characteristics such as the higher number of alcoholic drinks per week and higher BMI were suggestive of less adherence with respect to the drug release markers. The finding that partner knowledge of the participant taking part in the study is associated with higher adherence to the ring is consistent with qualitative findings indicating that when participants disclosed their participation in the study to their male partners they were often met with support to use the ring and appreciation for how the ring might help them or their families in the future [22]. In serial in-depth interviews, participants reported that their partners’ attitudes changed over time from one of concern or opposition to greater acceptance and use of the ring. Qualitative results also showed that some women were proactive in removing the ring prior to sex, and preemptively disclosed ring use to their partners to avoid inciting abusive behavior that might follow inadvertent disclosure [22]. Acceptability measures related to interference with sex were also found to be associated with non-adherence to ring use [31]. Moreover, women who were in stable relationships disclosed ring use to obtain acceptance by their primary partners, whereas women who were not married, or their relationship was new, transactional or casual were less likely to disclose to partners that they were using the ring [22]. Our finding that adherence was lower with any condom use compared to no sex is suggestive that women may have been less interested in ring use if other HIV-1 protection methods were being used, or simply that they removed the ring prior to sex whether a condom was used or not.

We found that transactional sex, being positive for chlamydia and ring worries at baseline, and number of menstrual bleeding episodes in last 3 months, partner circumcision and putting anything inside the vagina in the last three months played a role in whether women were adherent to the use of the vaginal ring. It is perhaps not surprising that women who were worried about using the novel ring were less likely to use the ring on a regular basis, as was seen in a sub-study examining ring worries among ASPIRE participants. Women who experienced more menstrual bleeding episodes compared to none in the last three months were found to be less adherent to vaginal ring use. All women in this study used a form of contraception whereby some could have had side effects resulting in either less (including amenorrhea) or more (like intermenstrual spotting) vaginal bleeding. In this context, therefore, fewer vaginal bleeding episodes reported would be expected among women using contraceptive methods that inherently prevented break-through bleeding, such as long-acting injectables. This finding is consistent with the observation that family planning methods where long-acting forms of contraception was used were correlated with higher levels of ring use when compared to oral pills and sterilization [32]. Women who put things in their vagina during the month may have perceived that the use of the ring was interfering. Previous qualitative findings corroborate these results wherein some women reported that they removed the ring during menses because they felt unclean or wanted to wash it. Likewise, women reported that they would remove the ring if they were concerned with side effects such as vaginal discharge, odor, or itching to see if symptoms resolved [22]. In another qualitative sub-study, reasons for ring removal included hygiene, beliefs that the ring blocked the flow of menstrual blood, and fears that the ring would come out with blood or during tampon removal [33].

The significant independent positive association between all four adherence measures and the use of longer-acting hormonal-based family planning methods was a result not directly seen in other studies of the dapivirine ring. Although, in one study comparing stated preference between tablets, injections and the ring, ring use was preferred among women reporting use of implants or IUDs [21]. Women reporting use of longer-acting methods were more likely to adhere to ring use, which was broadly seen for all adherence markers. It is possible, and further research is warranted, to examine whether women who were willing to use longer-acting contraceptive methods made more prevention-oriented decisions, and thus, may have been less concerned about an unintended pregnancy occurring while using the investigational ring. Likewise, women who used an IUD may have had concerns about integrating another device internally. One study evaluated the effect of dapivirine exposure on the effectiveness of hormonal contraception but did not evaluate its influence on adherence to the use of the ring [34]. Although, notably, a qualitative analysis of a subset of ASPIRE participants found that perception of “vaginal overload” with tampon, intrauterine contraceptive device, and ring use during menses was an impetus for ring removal [33].

Although this analysis proved useful for identifying possible correlates to adherence to ring use, there were several limitations. First, there is no perfect objective measure of dapivirine vaginal ring use and thus there is potential for misclassification of the adherence outcomes [18]. However, we believe that objective markers like plasma and ring residual drug levels as measures of adherence are likely to provide a more sensitive estimate for women’s true level of adherence than self-reported measures [35]. Therefore, our analysis likely provides reasonable hypothesis-generating results especially given they show congruence with qualitative findings on perceived factors associated with adherence (e.g., sexual behavior, menses, partner disclosure, ring worries, use of contraceptive methods, etc.). Second, given the quantity of covariates to be assessed, constructing all possible subsets of models was impossible and thus we relied on a high-performance modeling routine to generate the best multivariable models. Last, misclassification bias may be present for any one of the adherence outcomes impacting which covariates were ultimately selected in the multivariable models. Measurement error and misclassification of the outcomes, assuming present at both levels and non-differential (exists at the same level for the outcomes for all levels of the covariates) would result in attenuation of odd ratios (i.e., toward the null). However, given that we examined four different outcomes, two objective measures with low and high bars to define adherence, we believe this approach may have also provided a more robust insight into correlates when for example we found similar effects across the adherence outcomes.

Correct and consistent use of the dapivirine vaginal ring, like other PrEP, and HIV-1 treatment, interventions, is key to effectiveness. Future studies examining similar vaginal rings may benefit from knowing characteristics of women that are correlates to adherence. This new knowledge could be used in recruitment into future clinical studies, real-time adherence monitoring and dapivirine ring implementation efforts, especially if corroborated by other studies. Evaluating multiple biomarkers of adherence provides a more thorough understanding of correlates of ring adherence and could strengthen the design of future adherence marker studies. Likewise, these results may help inform implementation of the dapivirine vaginal ring were it to be approved for use in high HIV prevalence settings.