Loss to Follow-up and Visit Nonadherence in an HIV Incidence Cohort in Kisumu, Kenya

Abstract

Background:

Adherence to study visits enhances safety and validity of human research studies. We identified characteristics associated with missed visits in a Kenyan HIV incidence study to inform retention strategies.

Methods:

We enrolled volunteers 18–35 years-of-age, without HIV, who reported 2+ sex partners into a study with visits every 3 months for 24 months. We used zero-inflated Poisson regression to assess sociobehavioral and demographic characteristics associated with missed visits and loss-to-follow-up. Latent class analysis (LCA) identified clusters of characteristics associated with each outcome.

Results:

Among 619 participants without HIV, 278 (44.9%) were women with median age of 24 (interquartile range: 21–28). During follow-up, 134 (21.6%) missed 1+ and 84 (13.6%) 2+ visits. Missed visits were more common among participants with 3+ sexual partners compared to 2 (adjusted risk ratio [RR]: 1.50, 95% confidence interval [CI]: 1.13–1.97) and less common in participants who completed secondary school (aRR: 0.74, 95%CI: 0.56–0.99, compared to some primary school or less). Females were more likely to be lost to follow-up (aRR: 1.85, 95%CI: 1.05–3.26) LCA showed that missed visits were more likely among females aged 24+ years with less than secondary education who reported transactional sex compared to largely younger unmarried males with less than secondary education and 3+ sexual partners (RR: 1.24, 95%CI: 1.01–1.54) and poorly educated, largely older married males (RR: 1.41, 95%CI: 1.05–1.90).

Conclusion:

Older unmarried females with less education, more sexual partners, and transactional sex may benefit from targeted retention strategies like optimizing accessibility to testing and prevention services and frequent visit reminders.

Introduction

Globally, there were 39.9 million people with HIV (PWH) at the end of 2023 with 1.3 million people acquiring human immunodeficiency virus (HIV) that same year (1). Sub-Saharan Africa accounts for only about 13% of the world’s population, yet represents greater than 70% of PWH worldwide (2) and 50% of new HIV cases. Kenya ranks among the top fifteen countries in the world with the highest prevalence of HIV. It has a mixed epidemic with geographic and population variations; approximately 30% of new HIV diagnoses occur in key and priority populations including men who have sex with men (MSM), female sex workers, adolescent girls, and young women. Key populations constitute a small proportion of the general population but are at an elevated risk of acquiring HIV due, in part, to social exclusion, mental health issues, unfair treatment, and decreased engagement with healthcare services (3–6). Although some highly effective HIV prevention tools such as pre-exposure prophylaxis (PrEP) (7) are available, the HIV epidemic remains a public health concern in many African countries (1).

Hence, several clinical trials of HIV vaccines and other preventive methods are being conducted in Kenya. Participant retention is essential to adequately achieve the scientific goals of HIV trials and to protect participant safety. Retention in trials or cohort studies is typically measured by tracking the percentage of individuals who remain actively engaged in the study over the protocol-specified period. Active engagement is assessed through adherence to required study visits, completion of protocol-specified activities, and follow-up contact or visits. Although there have been numerous studies on retention and antiretroviral treatment (ART) adherence in clinical programs among PWH (8–10), retention among participants in observational cohort studies has been less frequently evaluated (11, 12).

Factors associated with non-adherence and loss to follow-up (LTFU) in cohort studies conducted in African countries vary by population and geography. In an observational cohort study conducted in fishing communities in Uganda, dropout factors included younger age, male sex, marijuana use, ethnicity, number of sexual partners, and time of travel away from home (13). The most common reason for missing study visits was migration (13). Consistent with these findings, LTFU was highest among women, younger persons, recreational drug users, and persons who engage in transactional sex in an HIV vaccine preparedness study in Masaka, Uganda (14). In ART adherence studies, the most frequently identified barriers for PWH were forgetting to take their medication, lack of access to adequate food, sexual behavior stigma and unfair treatment, side effects, and being outside the house or travelling (15). Men are more difficult to retain than women due to a historical focus on HIV testing and treatment as a maternal and child health issue (11). It is expected that some of these barriers will contribute to non-adherence in prevention studies, but area-specific studies are necessary.

While retention in HIV treatment programs has been widely studied, there are limited data about adherence and LTFU in HIV prevention cohorts. We identified factors associated with visit non-adherence and LTFU in a Kenyan HIV incidence cohort. We further used latent class analysis to identify participant subgroups at higher risk for missed visits. Our findings offer insights for tailoring retention strategies in future HIV prevention trials targeting vulnerable populations.

METHODS

Study Design

We conducted secondary analyses of data collected in the RV393 study specifically focused on identifying factors associated with visit non-adherence and LTFU. RV393 was a prospective HIV incidence cohort study conducted in Kisumu County, Kenya (16). From January 2017 to May 2018, the study enrolled males and females aged 18–35 years without HIV who planned to reside within Kisumu for at least 24 months; were literate in English, Luo, or Kiswahili; and reported at least two sexual partners in the three months prior to screening for study eligibility. Study visits were conducted every three months for up to 24 months. At every visit, HIV status was determined using a rapid test serial algorithm. HIV status was assessed using the First Response^® Rapid HIV 1–2–0 card test (Premier Medical Corporation, Maharashtra, India). In cases where rapid test results were inconclusive, additional testing was conducted using an enzyme-linked immunosorbent assay (ELISA) with the GenScreen ULTRA HIV Ag-Ab Combo Kit (Bio-Rad Laboratories, Hercules, CA). Final confirmation of HIV was performed using the Geenius HIV-1/2 Supplemental Assay (Bio-Rad Laboratories, Hercules, CA). Participants who tested positive for HIV at screening for study eligibility were excluded from these analyses, though some were offered enrollment into RV393 to serve as a masking group for study implementation (masking the HIV status of study participants) and comparator for some analyses. Every six months, behavioral questionnaires were administered by trained staff that collected data on socio-demographic characteristics, sexual activity, travel, drug and alcohol use, and healthcare engagement. Study staff attempted to contact all participants who missed study visits and, where possible, recorded the reason for withdrawal or termination from the study.

Measures

These analyses explored two outcomes: LTFU and visit non-adherence. Visit non-adherence was calculated as the number of missed visits out of eight possible follow-up visits. A study visit was determined as missed if the participant did not complete any study procedures within the protocol-allowed window for that scheduled visit. HIV acquisition represented the end of observation time in the HIV incidence cohort, so participants who acquired HIV during follow-up were classified as having completed all study visits after the visit at which HIV was diagnosed. Participants who did not attend their final study visit at month 24 were considered LTFU.

Demographic and behavioral variables were collected via self-administered questionnaires at screening. The following variables were selected based on a literature review, summarized above. Behavioral characteristics included: total number of sexual partners during the three months prior to screening; whether the participant had sex with someone in exchange for money, goods, gifts, or favors during the three months prior to screening; whether the participant had a history of using needles to inject illegal drugs; and how often the participant traveled outside of the Kisumu County area and stayed overnight within the preceding six months.

Statistical Analyses

To explore factors potentially associated with visit non-adherence, each variable was tested in univariable and multivariable generalized linear models with a zero-inflated Poisson distribution; this approach was selected because a high proportion of participants did not miss any visits. To explore factors potentially associated with LTFU, each variable was tested in univariable and multivariable generalized linear models with robust Poisson standard errors; this approach was selected to enable direct estimation of risk ratios (RRs) and 95% confidence intervals (CIs) for a non-rare outcome.

We implemented latent class analysis (LCA) (17–19) to identify latent subgroups based on sex, age, education, relationship status, number of sexual partners, transactional sex, injection drug use, and travel outside Kisumu in the 3 months prior to screening. Dichotomous cut-points were implemented across all LCA variables to aid interpretability and anchor clustering on meaningful thresholds, considering response rates. Variables were dichotomized as age (18 to 23 years vs. 24+ years), education (Primary complete/Secondary incomplete/or less vs. Secondary complete/more than secondary) and total number of partners in the last three months (1–2 vs. 3+ sexual partners). We used Bayesian Information Criterion (BIC) and entropy to determine the best fitting class solution. Participants were assigned to the class with the highest posterior probability.

Median number of missed study visits between each LCA subgroup was compared using Mann-Whitney U tests. We then assessed risk of non-adherence using a generalized linear model with a Zero-Inflated Poisson distribution and a log link function to estimate RRs and 95% CIs across LCA clustering variables. Next, comparative risk of visit non-adherence was implemented using pairings between each of the LCA subgroups. We also assessed LTFU using a generalized linear model with a Poisson distribution and robust standard errors to estimate RRs and 95% CIs across LCA clustering variables. Comparative risk of LTFU status was implemented using the same LCA subgroup pairings. LCA was performed in R (version 4.2.2) using the poLCA package. All other analyses were conducted in SAS 9.4 (SAS Institute, Inc.; Cary, NC).

RESULTS

Population Characteristics

Of the 671 participants enrolled, 619 participants tested negative for HIV and were included in these analyses. Among the 619 participants analyzed, there were 278 females (44.9%), and the median (Q1–Q3) age of study participants was 24 years (21–28). Overall, 335 (54.1%) had less than secondary education, 478 (77.2%) were unmarried or divorced, and 444 (71.8%) engaged in transactional sex (Table 1). Additionally, 11 participants acquired HIV during the study (1.8%).

Table 1.

Select Participant Demographic and Behavioral Characteristics at Screening Stratified by Non-Adherence and Loss to Follow-up

	Total N=619 (100%)	0 Missed Visits (N=485, 78.4%)	1+ Missed Visits (N=134, 21.6%)	p-value	No Loss to Follow-Up (N=549, 88.7%)	Lost to Follow-Up (N=70, 11.3%)	p-value
Sex				0.084			0.094
Male	341 (55.1)	276 (56.9)	65 (48.5)		309 (56.3)	32 (45.7)
Female	278 (44.9)	209 (43.1)	69 (51.5)		240 (43.7)	38 (54.3)
Age				0.239			0.716
18 to 23 years	275 (44.4)	207 (42.7)	68 (50.8)		243 (44.3)	32 (45.7)
24 to 30 years	234 (37.8)	188 (38.8)	46 (34.3)		206 (37.5)	28 (40.0)
30 to 35 years	110 (17.8)	90 (18.6)	20 (14.9)		100 (18.2)	10 (14.3)
Education				0.434			0.640
Some primary or less	93 (15.0)	74 (15.3)	19 (14.2)		80 (14.6)	13 (18.6)
Primary complete/secondary not complete	242 (39.1)	195 (40.2)	47 (35.1)		217 (39.5)	25 (35.7)
Secondary complete/more than secondary	284 (45.9)	216 (44.5)	68 (50.8)		252 (45.9)	32 (45.7)
Marital status				0.080			0.987
Unmarried/ Separated/Divorced/Widowed	478 (77.2)	367 (75.7)	111 (82.8)		424 (77.2)	54 (77.1)
Married/Cohabiting (monogamous/polygamous)	141 (22.8)	118 (24.3)	23 (17.2)		125 (22.8)	16 (22.9)
Total number of partners in the last 3 months				0.202			0.608
2 sexual partners	147 (23.7)	110 (22.7)	37 (27.6)		134 (24.4)	13 (18.6)
3–5 sexual partners	261 (42.2)	215 (44.3)	46 (34.3)		228 (41.5)	33 (47.1)
6–10 sexual partners	93 (15.0)	69 (14.2)	24 (17.9)		84 (15.3)	9 (12.9)
More than 10 sexual partners	118 (19.1)	91 (18.8)	27 (20.2)		103 (18.8)	15 (21.4)
Engaged in transactional sex in the last 3 months				0.622			0.716
No	174 (28.2)	135 (27.8)	40 (29.9)		153 (27.9)	21 (30.0)
Yes	444 (71.8)	350 (72.2)	94 (70.1)		395 (72.1)	49 (70.0)
History of non-injectable drug use				0.906			0.928
No	524 (84.7)	411 (84.7)	113 (84.3)		465 (84.7)	59 (84.3)
Yes	95 (15.3)	74 (15.3)	21 (15.7)		84 (15.3)	11 (15.7)
Traveled 1+night out of the county				0.431			0.705
No	207 (33.4)	166 (34.2)	41 (30.6)		185 (33.7)	22 (31.4)
Yes	412 (66.6)	319 (65.8)	93 (69.4)		364 (66.3)	48 (68.6)

Note. Visit non-adherence was defined as the number of missed study visits. Loss to follow-up was defined as a missed final study visit at 24 months.

Visit Adherence and Retention

During study follow-up, 485 (78.4%) participants did not miss any study visits, 67 (10.8%) participants missed 1–2 study visits, 20 (3.2%) missed 3–4 study visits, 18 (2.9%) missed 5–6 study visits, and 29 (4.7%) missed 7–8 study visits. In total, 549 (88.7%) participants completed the study, and out of 5,571 possible study visits per protocol across all participants, 5,096 (91.5%) were completed. Overall, 70 participants (11.3%) were lost to follow up.

In unadjusted models, lower visit non-adherence was observed among educated participants (Table 2). Greater visit non-adherence was observed among females, participants with 3+ sexual partners, and participants who reported transactional sex. In the adjusted model, lower visit non-adherence is seen in participants with only primary education (RR: 0.65, 95% CI: 0.49–0.87, p=0.003) and secondary education (0.74, 0.59–0.99, p=0.044) compared to those who have not completed primary education. Greater visit non-adherence was observed among those who had 3–5 sexual partners in the past three months (1.5, 1.13–1.97, p=0.004) compared to those with 2 or fewer partners. In unadjusted robust Poisson models assessing LTFU, no significant associations were observed between participant characteristics and study LTFU. However, in the adjusted model, they were more likely to be lost to follow up (1.85, 1.05–3.26, p=0.034).

Table 2.

Unadjusted and Adjusted Risk Ratios for Factors Potentially Associated with Visit Non-Adherence and Loss to Follow-up

	Visit Non-Adherence				Loss to Follow-Up
	Unadjusted		Adjusted		Unadjusted		Adjusted
Characteristic	RR (95% CI)	p	RR (95% CI)	p	RR (95% CI)	p	RR (95% CI)	p
Sex
Male	Reference	-	Reference	-	Reference	-	Reference	-
Female	1.24 (1.02, 1.51)	0.028	0.96 (0.72, 1.28)	0.786	1.46 (0.94, 2.27)	0.096	1.85 (1.05, 3.26)	0.034
Age
18 to 23 years	Reference	-	Reference	-	Reference	-	Reference	-
24 to 30 years	1.18 (0.96, 1.45)	0.122	1.18 (0.91, 1.52)	0.211	1.03 (0.64, 1.66)	0.909	0.92 (0.56, 1.50)	0.727
30 to 35 years	0.98 (0.73, 1.32)	0.897	1.00 (0.70, 1.43)	0.998	0.78 (0.40, 1.53)	0.473	0.68 (0.34, 1.38)	0.287
Education
Some Primary or Less	Reference	-	Reference	-	Reference	-	Reference	-
Primary complete/secondary not complete	0.62 (0.48, 0.82)	0.001	0.65 (0.49, 0.87)	0.003	0.74 (0.40, 1.38)	0.344	0.76 (0.40, 1.44)	0.407
Secondary complete/more than secondary	0.66 (0.52, 0.85)	0.001	0.74 (0.56, 0.99)	0.044	0.81 (0.44, 1.47)	0.482	0.83 (0.43, 1.60)	0.581
Marital status
Unmarried/ Separated/Divorced/Widowed	Reference	-	Reference	-	Reference	-	Reference	-
Married/Cohabiting (monogamous/polygamous)	0.82 (0.62, 1.09)	0.165	0.73 (0.52, 1.03)	0.076	1.00 (0.59, 1.70)	0.987	1.11 (0.61, 2.03)	0.737
Total number of partners in the last 3 months
2 sexual partners	Reference	-	Reference	-	Reference	-	Reference	-
3–5 sexual partners	1.51 (1.16, 1.96)	0.002	1.50 (1.13, 1.97)	0.004	1.43 (0.78, 2.63)	0.250	1.49 (0.80, 2.78)	0.214
6–10 sexual partners	1.41 (1.04, 1.92)	0.026	1.22 (0.87, 1.72)	0.251	1.09 (0.49, 2.46)	0.827	1.10 (0.47, 2.56)	0.823
More than 10 sexual partners	1.52 (1.13, 2.03)	0.005	1.37 (0.97, 1.94)	0.073	1.44 (0.71, 2.90)	0.311	1.31 (0.58, 2.98)	0.521
Transactional sex in the last 3 months
No	Reference	-	Reference	-	Reference	-	Reference	-
Yes	1.31 (1.05, 1.64)	0.018	1.25 (0.95, 1.66)	0.115	0.91 (0.57, 1.48)	0.715	0.70 (0.40, 1.24)	0.217
History of non-injectable drug use
No	Reference	-	Reference	-	Reference	-	Reference	-
Yes	1.23 (0.97, 1.57)	0.092	1.24 (0.94, 1.63)	0.129	1.03 (0.56, 1.88)	0.928	1.21 (0.64, 2.28)	0.556
Traveled 1+night out of the county
No	Reference	-	Reference	-	Reference	-	Reference	-
Yes	0.87 (0.71, 1.06)	0.166	0.86 (0.70, 1.07)	0.176	1.10 (0.68, 1.76)	0.705	1.09 (0.67, 1.79)	0.718

RR, relative risk; CI, confidence interval; p, p-value.

Note. Visit non-adherence was defined as the number of missed study visits. Loss to follow-up was defined as a missed final study visit at 24 months. Bold values indicate significant p-values at p≤0.05. All characteristics were assessed cross-sectionally at screening for study eligibility.

The 3-class LCA solution demonstrated the lowest BIC with acceptable entropy (Supplementary Table 1). Class 1 (n=246, 40%, Table 3) primarily represented males (94%), aged 18–23 years old (74%), who were unmarried or cohabitating (98%), completed at least secondary school (71%) and had 3 or more sexual partners in the last 3 months (71%), including 2 who acquired HIV. Class 2 (n=242, 39%) was entirely comprised of females who engaged in transactional sex, including 6 people who acquired HIV; Class 3 (n=131, 21%) included mostly males (83%) 24 years old or older (91%) who were married or cohabitating (87%) and completed primary school or less (83%) including 3 people who acquired HIV (Table 3). Class 2 had the greatest visit non-adherence with higher proportions of missed study visits; 11 (4.6%) participants had 3–4 missed visits, 10 (4.1%) participants had 5–6, and 15 (6.2%) participants had 7–8 missed study visits (Figure 1). The median (Q1–Q3) number of missed visits was 2 (1–5) for Class 1, 3.5 (1–6) for Class 2, and 2 (1–5) for Class 3.

Table 3.

Frequencies of Select Participant Demographic and Behavioral Characteristics at Screening in 3-Class Solution from Latent Class Analysis

Characteristic	Total N=619 (100%)	Class 1 N=246 (39.7%)	Class 2 N=242 (39.1%)	Class 3 N=131 (21.2%)
Sex
Male	341 (55.1)	232 (94.3)	0 (0)	109 (83.2)
Female	278 (44.9)	14 (5.7)	242 (100)	22 (16.8)
Age
18 to 23 years	275 (44.4)	183 (74.4)	80 (33.1)	12 (9.2)
24 to 35 years	344 (55.6)	63 (25.6)	162 (66.9)	119 (90.8)
Education
Primary complete/secondary incomplete/or less	335 (54.1)	71 (28.9)	155 (64.0)	109 (83.2)
Secondary complete/more than secondary	284 (45.9)	175 (71.1)	87 (36.0)	22 (16.8)
Marital status
Unmarried/Separated/Divorced/ Widowed	478 (77.2)	240 (97.6)	221 (91.3)	17 (13.0)
Married/Cohabiting (monogamous/ polygamous)	141 (22.8)	6 (2.4)	21 (8.7)	114 (87.0)
Total number of partners in the last 3 months
1–2 sexual partners	147 (23.7)	71 (28.9)	15 (6.2)	61 (46.6)
3+ sexual partners	472 (76.3)	175 (71.1)	227 (93.8)	70 (53.4)
Engaged in transactional sex in the last 3 months
No/No Answer	175 (28.3)	114 (46.3)	0 (0)	61 (46.6)
Yes	444 (71.7)	132 (53.7)	242 (100)	70 (53.4)
History of non-injectable drug use
No	524 (84.7)	188 (76.4)	230 (95.0)	106 (80.9)
Yes	95 (15.3)	58 (23.6)	12 (5.0)	24 (19.1)
Traveled 1+ night out of the county
No	207 (33.4)	80 (32.5)	94 (38.8)	33 (25.2)
Yes	412 (66.6)	166 (67.5)	148 (61.2)	98 (74.8)
Non-adherence
0 missed visits	485 (78.4)	197 (80.1)	180 (74.4)	108 (82.4)
1–2 missed visits	67 (10.8)	27 (11.0)	26 (10.7)	14 (10.7)
3–4 missed visits	20 (3.2)	6 (2.4)	11 (4.6)	3 (2.3)
5–6 missed visits	18 (2.9)	6 (2.4)	10 (4.1)	2 (1.5)
7–8 missed visits	29 (4.7)	10 (4.1)	15 (6.2)	4 (3.1)
Lost to follow-up?
No	549 (88.7)	223 (90.7)	210 (86.8)	116 (88.6)
Yes	70 (11.3)	23 (9.3)	32 (13.2)	15 (11.4)

Note. Visit non-adherence was defined as the number of missed study visits. Loss to follow-up was defined as a missed final study visit at 24 months. Class 1: mostly young (18-23, 74%) unmarried (98%) males with higher education (at least secondary school complete, 71%) and 3+ sexual partners (71%); Class 2: mostly older (24+, 67%) single (91%) females with less education (less than secondary school, 64%) who had 3+ sexual partners (94%) and engaged in transactional sex (100%); Class 3: mostly older (24+, 91%) married (87%) males (83%) with less education (less than secondary school, 83%).

Figure 1. Visit Non-Adherence Described as Percentage of Missed Visit Categories Stratified by Latent Class Membership.

Note. Class 2 had the greatest percentage of missed study visits compared to any other latent class across 3–4, 5–6, and 7–8 missed visits.

Classes. Class 1: mostly young (18–23, 74%) unmarried (98%) males with higher education (at least secondary school education, 71%) and 3+ sexual partners (71%); Class 2: mostly older (24+, 67%) unmarried (91%) females with less education (less than secondary school, 64%) who had 3+ sexual partners (94%) and engaged in transactional sex (100%); Class 3: mostly older (24+, 91%) married (87%) males (83%) with less education (less than secondary school, 83%).

In robust Poisson models comparing visit non-adherence by class, Class 2 had 24% more missed study visits than Class 1 (1.24, 1.01–1.54, p=0.045) and 40% more than Class 3 (1.40, 1.04–1.89, p=0.025). Number of missed study visits did not significantly differ between Class 1 and Class 3 (p=0.445) (Table 4). In robust Poisson models comparing study LTFU by class, no significant differences were observed between each class group in their LTFU status.

Table 4.

Comparisons of Visit Non-Adherence and Lost to Follow-Up Risk by Latent Class Membership

Class	Non-Adherence		Lost to follow-up
	Risk Ratio (Confidence Interval)	p-value	Risk Ratio (Confidence Interval)	p-value
Class 1	0.81 (0.65, 1.00)	0.045	0.71 (0.43, 1.17)	0.179
Class 2	Reference	-	Reference	-
Class 1	1.13 (0.83, 1.55)	0.445	0.82 (0.44, 1.51)	0.518
Class 3	Reference	-	Reference	-
Class 2	1.40 (1.04, 1.89)	0.025	1.15 (0.65, 2.05)	0.624
Class 3	Reference	-	Reference	-

Note. Visit non-adherence was calculated as the number of missed visits out of eight possible follow-up visits. A study visit was considered missed if no study procedures were performed within the protocol-allowed window for that scheduled visit. Participants were considered lost to follow-up if the final study visit at 24 months was missed. Bold values indicate significant p-values at p≤0.05. Classes. Class 1: mostly young (18–23, 74%) unmarried (98%) males (94%) with higher education (at least secondary school complete, 71%) and 3+ sexual partners (71%); Class 2: mostly older (24+, 67%) unmarried (91%) females (100%) with less education (less than secondary school, 64%) who had 3+ sexual partners (94%) and engaged in transactional sex (100%); Class 3: mostly older (24+, 91%) married (87%) males (83%) with less education (less than secondary school, 83%).

DISCUSSION

The aim of these analyses was to identify factors associated with visit non-adherence and LTFU in the RV393 HIV incidence cohort in Kisumu, Kenya. With over 90% of visits completed and nearly 90% of participants completing the final study visit, visit non-adherence and LTFU in this study were much lower than the 20–40% reported from cohorts of PWH enrolled for ART treatment in sub-Saharan Africa (10). In Kenya, LTFU as high as 57% has been reported among young PWH (9, 10, 20). Higher LTFU has also been observed in studies of PrEP implementation in Kenya, ranging from 19% in a PrEP program for MSM and sex workers (21) to 42.5% in a study of MSM only (22). A meta-analysis of 70 studies, mostly from sub-Saharan Africa, estimated 20–25% LTFU at 24 months after ART initiation in PWH (23).

In our multivariable analysis of visit non-adherence, participants with higher education levels were less likely to miss study visits and participants with 3–5 sexual partners were more likely to miss study visits. Prior studies have shown that increasing education level is associated with improved adherence, engagement in care, and retention across a variety of settings (24–26). Education level is a key determinant of health literacy, which in turn is strongly associated with adherence and retention (27). People with multiple sexual partners – especially people who engage in sex work or same-sex sexual practices – often face stigma related to their sexual behavior that can lead to internalized shame, fear of disclosure, and avoidance of healthcare settings, all of which may reduce adherence and retention (28–32). People with multiple partners may also have mobile lifestyles that contribute to frequent changes in sexual partners and also make it harder to adhere to study visits.

In our multivariable analysis of LTFU, only female sex was independently associated with the outcome. This contrasts with prior literature that generally suggests better engagement and retention in care by females than males (25, 33–35). However, adolescent girls and young women face multiple barriers to engagement and retention, including stigma, lack of social support, and low perceived HIV risk (36). Our study population was mostly young and this may partially explain increased LTFU among female participants. No other factors were significantly associated with LTFU and absence of differential attrition supports the internal validity of our analyses of visit adherence by reducing the risk that differences in outcomes between groups were due to systematic differences in who dropped out, rather than participant characteristics we evaluated.

LCA identified a subgroup of older single females with less than secondary education, more sexual partners, and engagement in transactional sex that had the greatest visit non-adherence with higher proportions of missed study visits. Females who exchange sex for money and other necessities/goods in prevention studies experience health disparities and engaging these populations requires both policy reform and community engagement (37–40). Community engagement strategies include building strong relationships with community organizations to identify, reach, and retain high-risk populations and optimizing accessibility to testing and prevention services like mobile clinics in remote or underserved areas. Other strategies to improve retention and study adherence include utilizing study visits’ reminders and immediate (≤7 days) follow-up of missed visits via phone and/or home visits and offering transportation services or flexible reimbursement for transport costs (8, 11, 41, 42). Addressing individual needs and barriers through personalized counseling and support services could also be an option. Women who engage in transactional sex can have diverse motivations and experiences that necessitate diverse interventions (43). Women who exchange sex for money or other goods have a particularly high risk of HIV acquisition in eastern and southern Africa (44–46). This is important because it is essentially the same subgroup with poorer adherence and retention in this study. It highlights the importance of focusing on this vulnerable subpopulation of women who engage in transactional sex and implementing tailored interventions that would improve their study retention and adherence.

These analyses were strengthened by a participant population selected based on eligibility criteria specifically designed to simulate criteria for enrollment into a potential future clinical trial of an HIV vaccine candidate or other preventive intervention, so findings about retention of this population should be directly transferable to retention efforts in future HIV prevention clinical trials. We used multiple analytic approaches to evaluate the related outcomes of visit non-adherence and loss-to-follow-up, strengthening the likelihood of true associations that were observed across multiple analyses, such as findings that women and participants with multiple sexual partners were at risk for poorer retention outcomes. However, the findings presented here should be interpreted in the context of some limitations. First, the behavioral characteristics used for the LCA analysis were collected at the time of screening for study eligibility and may have changed over time; however, focusing on enrollment characteristics as predictors of visit adherence and retention yielded inferences with particular relevance to future clinical trials that must set predefined eligibility criteria and identify participants that might benefit most from retention interventions based on their enrollment characteristics. Our study consisted of visits at 3-month intervals, which may not emulate the schedule of evaluations for future clinical trials, which often include more frequent visits around the time of investigational product administration. Our participants were relatively young and we are unable to generalize our findings to older populations, but our study population did reflect the age and other demographic characteristics likely to be enrolled into future clinical trials of HIV prevention interventions. It is possible that other measures not included in our LCA model may be associated with visit non-adherence. However, the parsimonious set of variables included in this LCA is informative for identifying participant groups more likely to be non-adherent. Lastly, we were unable to assess further social and structural determinants that may have provided additional insight into identified classes, including economic insecurity and prevention behaviors with sexual partners.

Conclusions

We enrolled a study population with particularly strong visit adherence and retention, making them potentially well-suited for future clinical trials of HIV prevention interventions. However, within this mostly adherent population, we identified a subgroup with room for improvement: older single females with less education, more sexual partners, and engagement in transactional sex. This is a group that is also at particularly high risk for HIV acquisition, so it is doubly important to develop and implement interventions to improve retention of these individuals in both clinical research and clinical care.

Data Availability

The data included in these analyses are available to qualified researchers upon reasonable request.