This study was a prospective cohort analysis of participants who were recruited into the PREPTA study, which was partly described in our previous publications [17,18]. Briefly, the PREPTA is a research project with the overall aim of assessing the effectiveness of an mHealth intervention in promoting retention in PrEP care and adherence to PrEP among female sex workers and men who have sex with men in the following two cites of Tanzania: Dar es Salaam (intervention arm) and Tanga (control arm). The project is being implemented by the Muhimbili University of Health and Allied Sciences, Tanzania, and the University of Oslo, Norway. This study analyzed app usage after 6 months among female sex workers who were recruited in the intervention arm of the trial.

A participatory design approach was adopted when developing the Jichunge app [17]. The app provides users with information about PrEP, reminds them to take their daily pill, provides an opportunity to consult a doctor and a peer educator via the web, and operates a web-based forum for anonymous discussions among PrEP users (Figure 1). The overall aim of the app is to improve adherence to PrEP and retention in the PrEP program. Further details about the Jichunge app can be found in our previous publications [17,18].

We included women aged at least 18 years who had sold sex in the past 3 months. Participants were residents of Dar es Salaam (they had an address in the city and had lived there for the past 6 months), owned a smartphone, and were prepared to start PrEP treatment. The sample size for this analysis was estimated by using a formula for cohort studies [19,20]. Since the retention in PrEP services and adherence to PrEP were not known at the design stage of this study, we used a proportion (p) of 50%, which has been shown to provide an optimum sample size when the actual proportion is between 10% and 90% [21]. A precision of 5% and a design effect of 2 were used in estimating the sample size to achieve a statistical power of 80%. After adjusting for a 20% potential loss to follow-up, the minimum sample size was estimated to be 423 female sex workers.

Participants were recruited between March and June 2021, using respondent-driven sampling (RDS). RDS is a method for sampling from populations for which there is no available sampling frame [19,22]. On the first day, 3 initial participants (referred to as “seeds”) were recruited purposively by researchers and peer educators who had prior experience with biobehavioral surveys among female sex workers in Tanzania. The first seed was aged 20 years and was a street-based sex worker. The second seed was 32 years old and was a phone- and internet-based sex worker. The third seed was 28 years old and was a bar- and pub-based sex worker. Before the end of the recruitment phase, an additional 6 seeds were added to speed up recruitment, as individuals in the initial seeds' recruitment chains were facing difficulties in recruiting more participants. After having completed the study procedures, each seed was given 3 recruitment coupons and asked to pass them on to her peers and invite them to participate in the PREPTA study. Of the 9 seeds, only 1 did not recruit any other participant to the study. Every new study participant was asked to do the same until we had reached the desired sample size.

All participants were screened for PrEP eligibility by a health clinic, as per national guidelines [23]. PrEP eligibility criteria included a negative HIV test result, no signs of acute HIV infection, a serum creatinine clearance of >60 mL/min, and the willingness to start PrEP. Those who qualified were provided with PrEP pills for 30 days and thereafter invited to take part in the PREPTA study. After consenting to participate in the study, participants received the Jichunge app (free of charge), attended an app onboarding session, and thereafter participated in a face-to-face baseline interview with a trained interviewer. The Jichunge app was only available to study participants, and it was not available in mainstream app stores. If study participants needed to reinstall the app, they had to contact the project site managers and be given a link, through which they could download the app again. Upon completion of the interview, each participant was paid the modest amount of TZS 8000 (US $3.18) as compensation for transport costs and the time spent at the study site. In addition, participants received TZS 4000 (US $1.59) for each peer they referred to the study, as per the RDS protocol.

All participants were invited to participate in a 6-month follow-up interview regardless of whether they had used PrEP and the Jichunge app. For the 6-month follow-up data collection, we attempted to contact each participant by phone at least 3 times on different days before they were considered lost to follow-up.

Information on sociodemographics and other structural factors at baseline was collected by using a questionnaire, which was administered by trained research assistants. At 6 months following enrollment, participants were invited to participate in a follow-up survey consisting of an interviewer-administered questionnaire that focused on their experiences with PrEP and the Jichunge app. From the back end of the app, we also continuously collected data on participants’ use of the app’s different functionalities (ie, opening the app, registering medicine taking, reading editorial contents, accessing a web-based consultation, and entering the web-based discussion forum). Data from the back end of the app were collected from all participants regardless of whether they participated in the 6-month follow-up survey. In this study, we included user statistics pertaining to participants' use of the Jichunge app after 6 months (ie, the 60-day period following the first 150 days of being in the intervention arm).

This study was conducted in accordance with the Declaration of Helsinki. Details on how we worked to protect participants’ privacy were provided in a previous article [18]. Briefly, data from the questionnaires were stored on a secure server that was developed particularly for the handling of sensitive research data (a service for sensitive data called “TSD” [24]) at the University of Oslo, Norway. The Jichunge app does not store, transfer, or expose sensitive data. Further, no element in the user interface of the app refers to any information that can identify a participant. Written informed consent was obtained from all participants involved in this study for the collection and use of both questionnaire data and app data. The protocol for this study was reviewed and approved by the National Health Research Ethics Committee in Tanzania (protocol code: NIMR/HQ/R.8a/Vol. IX/3454) and by the Regional Committee for Medical and Health Research Ethics in Norway (protocol reference number: 33675).

We started by generating the distribution of sociodemographic characteristics and descriptive statistics related to the use of the Jichunge app. Using the chi-square test, we assessed the association of 6-month retention in the app with sociodemographics, sex work characteristics, and other sociostructural factors. Since the outcome variable was binary and relatively common (proportion of retained participants: 63/470, 13.4%), we used a generalized linear model with a log link and binomial distribution (log-binomial regression) to examine independent predictors of 6-month retention in the app [29]. Log-binomial regression was chosen because it estimates the risk ratios directly, as opposed to logistic regression, which gives odds ratios that can overestimate the association if the outcome variable is common (proportion of >10%) [30]. We started with a bivariate analysis to obtain crude estimates of risk ratios. All variables with a P value of <.25 in the bivariate analysis were included in the multivariable model to obtain adjusted estimates of risk ratios. All analyses were performed using Stata version 17 (StataCorp LLC), and a P value of <.05 was considered statistically significant.

A total of 470 participants (age: median 26, IQR 22-30 years) were recruited at baseline. Of these, 277 (58.9%) reported to have had at least some secondary education, and 361 (76.8%) had never been married. Out of all 470 participants, we managed to reach 340 (72.3%) participants for the 6-month follow-up survey, and 284 (60.4%) participants were interviewed.

Participants who came to the 6-month follow-up interview (n=284) were asked if they still had the Jichunge app on their phones; nearly three-quarters (206/284, 72.5%) no longer had the app. When asked about the reasons for this, a large majority of these 206 study participants (n=193, 93.7%) said that they no longer had access to the app because of issues related to their phones, such as the phone having been stolen or lost (n=89, 43.2%), the phone having been sold or changed (n=53, 25.7%), or the phone having technical problems (n=51, 24.8%); only a few (n=13, 6.3%) reported to have uninstalled the app from their phones.

More than three-quarters of the participants (376/470, 80%) had accessed the app at least once after enrollment into the PREPTA study. Figure 2 shows the proportion of study participants who had opened the app and used different Jichunge services at least once per month during the 6-month observation period. App use was high in the first month and declined continually in the subsequent months. For all months, pill registration was the most used service, while web-based consultation was the least used service.

Overall, of the 470 participants, 63 (13.4%) were retained in the sense that they had opened the app at least once after 6 months of enrollment. Specifically, of the 470 participants, 52 (11.1%) had registered their daily pill taking, 40 (8.5%) had accessed PrEP editorial contents, 35 (7.4%) had used the discussion forum, and 14 (3%) had used the Jichunge web-based consultation services.

In Table 1, we present the distribution of participants’ baseline characteristics by retention in the Jichunge app at 6 months. Retention increased significantly and linearly with age; 10.6% (21/199) of women aged 18 to 24 years were retained, whereas 12.8% (29/226) of women aged 25 to 34 years and 28.9% (13/45) of women aged ≥35 years were retained (χ²₂=10.8; P=.01). The proportion of retained users was also higher among women with high PrEP awareness (41/232, 17.7%) than that among women with low PrEP awareness (22/238, 9.2%; χ²₁=7.2; P=.01). Retention was higher among female sex workers who had experienced financial difficulties due to health care spending (43/244, 17.6%) than that among female sex workers who had not experienced such difficulties (19/224, 8.5%; χ²₁=8.2; P=.004).

Table 2 shows the results of the log-binomial regression of the independent predictors of retention in the Jichunge app. The findings revealed that older age, high PrEP awareness, and financial difficulties due to health care spending independently predicted retention in the app. Participants aged at least 35 years were more than 2 times more likely to be retained in the app than those aged 18 to 24 years (adjusted risk ratio [aRR] 2.3, 95% CI 1.2-4.1; P=.01). Similarly, participants who demonstrated high PrEP awareness at baseline were more likely to be retained in the app after 6 months than those who demonstrated low PrEP awareness (aRR 1.8, 95% CI 1.1-3; P=.01). Furthermore, female sex workers who had experienced financial difficulties due to health care spending were approximately 2 times more likely to be retained in the app after 6 months than those who had not experienced such difficulties (aRR 1.9, 95% CI 1.2-3.2; P=.01).

Although several mHealth interventions have been designed to support the delivery of health care services, few studies have evaluated retention in those interventions. Understanding the retention in mHealth apps is crucial when planning and designing interventions. In this study, we examined the extent and predictors of 6-month retention in an interactive mHealth app (Jichunge app) designed to promote adherence to PrEP use and retention in PrEP care within Tanzania, a lower-middle–income country in sub-Saharan Africa.

The proportion of retained female sex workers (those who still used the Jichunge app after 6 months) was only 13.4% (63/470). This is much lower than the the optimal 46.4% (218/470) of users we found 1 month after the app had been installed [18]. The main reason for the lower retention in the app after 6 months did not appear to have much to do with the app itself but rather with the fact that a very large proportion of users had lost access to the app due to their phones having been lost, stolen, sold, changed, or damaged. Issues related to phones clearly represented a main challenge for implementing this mHealth intervention among female sex workers in Tanzania.

We have not found any previous study that evaluated the extent of retention in mHealth apps for HIV prevention. However, our results are in line with those of a study conducted in Poland, which found that less than one-quarter of participants used mHealth to monitor diet, weight, and physical activity in 2022 [31]. On the other hand, our estimate was lower than those reported in studies of the engagement with mHealth apps for self-management among patients with cancer in Sweden [8] and among adolescent and young adult cancer survivors in Pennsylvania [32]. It was also lower than the engagement with text message–delivered support among patients with type 2 diabetes in Tennessee [9]. The difference in the extent of use could be due to variations in contextual, population, and intended health outcomes for each of the interventions. Importantly, the studies mentioned for comparison entailed people diagnosed with chronic medical conditions, while PrEP is used for prevention among healthy people; therefore, the motivation to use the app might be different. Thus, when designing mHealth apps, strategies for attracting users and promoting long-term retention should take into consideration the contextual setting and the specific needs of the targeted population.

Understanding predictors of retention in mHealth apps is essential in tailoring the design of mHealth for its intended users. In this study, we found that retention in the Jichunge app was associated with older age, high PrEP awareness, and financial difficulties due to health care spending. Contrary to the evaluation of the use of the Jichunge app after 1 month [18], education level and social support were not associated with the retention in the app. This implies that the predictors of early mHealth use are not necessarily the same as those for retention in mHealth apps. Thus, continual evaluation of mHealth interventions is essential in understanding the users’ early use and long-term retention behaviors.

In this study, older female sex workers were more likely to be retained in the Jichunge app than younger women after 6 months. Our findings are similar to those in a study by Chang et al [33], who found that older persons were more likely to use mobile phones for opioid use disorder telemedication during the COVID-19 pandemic in New York. Contrary to this, Bauer et al [34] found that younger age was associated with the use of mHealth tools among primary care patients in the Northwest United States. The use of the Jichunge app does not require sophisticated technical skills, and this might have contributed to easier use among older women. On the other hand, the app does not contain entertaining features, such as those found in many commercial social media apps, and this might have discouraged younger persons from engaging with the app over time. Therefore, future mHealth app development may consider the inclusion of features that may speak to the different needs of younger and older populations.

Participants who demonstrated high PrEP awareness at baseline were more likely to be retained in the Jichunge app after 6 months. A lack of prior awareness of medication is among factors that can affect the willingness to use PrEP and hence can affect the acceptability and use of a PrEP-related app [35]. Therefore, people with prior awareness of PrEP medication might have been motivated to use the Jichunge app to learn more about the medication. The primary goal of Jichunge was to promote adherence to PrEP and retention in PrEP care; hence, it is likely that participants who stopped using PrEP might have lost interest in the Jichunge app as well. The findings underline the need to create awareness about PrEP before launching innovative interventions to support PrEP adherence and retention.

One of the benefits of many mHealth interventions is that they can reduce the financial costs related to health care services [36,37]. In this study, we found that participants who reported to have experienced financial difficulties due to health care spending were more likely to be retained in the Jichunge app than those who did not have such difficulties. In Tanzania, PrEP is provided free of cost, and at the time of this study, only a few clinics were designated to prescribe PrEP. Participants with financial challenges could have been using Jichunge services, such as the web-based consultation that is provided free of cost, as an alternative to physical consultations for PrEP-related and other concerns or conditions. Our findings are contrary to those reported in a retrospective comparative study conducted in Boston by Xiong et al [38], who found that patients insured through Medicaid were less likely to use telemedicine to connect with orthopedic surgery services than privately insured patients. On the basis of our study, mHealth apps seem to have the potential to help improve universal health coverage for underserved populations by removing some of the financial barriers associated with access to health care services.

First, among the strengths of this study is that it provides evidence on retention in mHealth apps within a real-world setting. Second, the large sample size ensured enough statistical power and stable estimates. Third, we evaluated the outcome variable (6-month retention) by using objective measures (electronic records of actual app use), thereby avoiding the bias that could have resulted from using subjective measures (such as self-reported use of the app). One limitation of this study is that we did not carry out any in-depth investigation of users’ experiences with the Jichunge app, which limits our understanding of why our participants did or did not engage with the mHealth intervention. Furthermore, we only measured 1 aspect of user engagement with the app (retention), which limits our understanding of other aspects, such as time spent in the app, total sessions, and subjective measures.

Overall, retention in the Jichunge app after 6 months was low and was predicted by older age, high PrEP awareness, and financial difficulties due to health care spending. A large majority of the sex workers (193/206, 93.7%) who took part in the 6-month follow-up interview reported that they no longer had access to the Jichunge app because their phones had been lost, sold, stolen, or changed or were in disrepair. Clearly, female sex workers in Tanzania cannot be expected to use the same phone over long periods of time, and this may underscore some of the complexities associated with mHealth interventions in this population. Thus, the user patterns and “social lives” of smartphones should be taken into consideration when designing mHealth interventions. We recommend qualitative studies that could provide in-depth information on the users’ experiences with the use of mHealth interventions for PrEP services.