Preparing a Financial Incentive Program to Improve Adherence to ART for Scale: Using an Implementation Science Framework to Evaluate an mHealth System in Tanzania

Background: Viral suppression is key to ending the HIV epidemic, yet only 58% of people living with HIV (PLHIV) in sub-Saharan Africa are suppressed. Cash transfers are an effective strategy to improve adherence, but little is known about optimization of implementation; for example, designing effective programs that integrate into existing clinic workows. We studied implementation of an mHealth system to deliver cash transfers to support antiretroviral medication (ART) adherence. Methods: We conducted an “implementation science-effectiveness” randomized controlled trial evaluating cash transfers conditional on visit attendance for viral suppression among Tanzanian PLHIV initiating ART. An mHealth system using ngerprint identication and mobile payments was used to automatically disburse mobile money to eligible PLHIV. We used Proctor’s framework, assessing implementation of the mHealth system from the perspectives of PLHIV and clinicians. We analyzed mHealth system data and conducted surveys (n=530) and in-depth interviews (n=25) with PLHIV, clinic and pharmacy staff (n=10), and structured clinic observations (n=2,293 visits). Results: 1,651 cash transfers were delivered to 346 PLHIV in the cash arms, 78% through mobile money. Among those in the cash arms, 81% registered their mobile money account with the mHealth system by study end, signaling high adoption. While acceptability for ngerprinting and mobile payments was high among PLHIV, interviews revealed mixed views: some had privacy concerns while others felt the system was secure and accurate, and provided some legitimacy to the clinical visits. Pharmacists praised system eciency, but concerns about duplicative recordkeeping and added work arose. Clinic staff voiced excitement for the system’s potential to bring the cash program to all patients and simplify workows; yet concerns about multiple systems, stang, and intermittent connectivity tempered enthusiasm, highlighting structural issues beyond program scope. Structured observations revealed a steep learning curve; repeat ngerprint scans and manual entry declined as the system improved. Conclusions: Biometric identication and mobile payments were acceptable to most patients and staff. This study uses a robust, implementation-effectiveness hybrid design to optimize the cash transfer implementation strategy for improving retention in care and viral suppression among ART initiates Our mixed-methods approach uncovered unanticipated benets and challenges to implementing cash transfers in a clinical setting – important insights for program directors and policy makers in low-resource settings considering cash transfer programs Our results illustrate how an mHealth system can streamline implementation of a cash transfer program, increasing the potential for sustainability and scale.

Checklists: StaRI (included with submission). Note CONSORT for cluster randomized trials was used for the main trial but is not directly applicable to this manuscript.

Contributions To The Literature
This study uses a robust, implementation-effectiveness hybrid design to optimize the cash transfer implementation strategy for improving retention in care and viral suppression among ART initiates Our mixed-methods approach uncovered unanticipated bene ts and challenges to implementing cash transfers in a clinical setting -important insights for program directors and policy makers in low-resource settings considering cash transfer programs Our results illustrate how an mHealth system can streamline implementation of a cash transfer program, increasing the potential for sustainability and scale.

Background
For people living with HIV (PLHIV), antiretroviral therapy (ART) is an effective strategy to clinically suppress the virus, providing the dual bene t of improved health and vastly decreased risk of transmission. Despite the robust evidence, this highly effective intervention has not yet reached all that could bene t. For example, in Tanzania, of the 1.6 million people living with HIV, only 62% are virally suppressed, and of those who are currently on HIV treatment, 87% are virally suppressed(1,2) -falling short of the '95-95-95 goals" that UNAIDS has set for 2030 for which 95% of those on HIV treatment should be virally suppressed. (2) To reach these goals, new and effective implementation strategies that can put evidence into widespread practice and bring sustained HIV treatment for viral suppression to all PLHIV are needed. mHealth systems, in combination with other evidence-based implementation strategies, have the potential to facilitate reaching the 95-95-95 goals by streamlining clinic-based recordkeeping, improving care coordination, and simplifying intervention implementation.
Cash incentives are increasingly recognized as one such evidence-based implementation strategy: these programs typically provide cash (or other incentives) to participants who meet a particular condition, for example testing for HIV, thus motivating certain behaviors that result in improved health. There is now a substantial evidence base demonstrating the effectiveness of cash transfer programs in improving outcomes along the HIV care continuum in low resource settings in a research context (2)(3)(4)(5)(6)(7)(8)(9)(10)(11)(12)(13)(14)(15)(16)(17)(18)(19). Despite this strong evidence and the increasing prevalence of large-scale government run cash transfer programs for social protection, few cash transfer programs to improve HIV outcomes have been scaled. One possible reason for this gap is that simpli ed, automated models for implementing these clinic-based programs at scale have yet to be developed and evaluated.
In order to begin to address this gap, we conducted a randomized controlled trial evaluating two cash transfer sizes (~$5 and ~$10) conditional on visit attendance, with the outcome of viral suppression at 6 months among PLHIV recently initiating ART in Shinyanga, Tanzania. Results of the trial indicated that the cash was effective in improving viral suppression at 6 months especially with the larger cash amount. (20) For a clinic-based cash transfer program contingent on visit attendance to be scalable and sustainable in resource constrained settings the delivery model must be simple. We developed an mHealth system designed to automate and simplify cash distribution through integration with mobile money providers. The mHealth system had the dual purpose of monitoring compliance (visit attendance) through biometric identi cation and automatically delivering cash transfers to those who were eligible, obviating the need for manual monitoring of compliance with clinic visits and manual delivery of cash. (21) In this manuscript we present the results of the implementation science portion of the trial, guided by Proctor's implementation science framework. (22) The overarching goal of the evaluation was to understand from both clinician and patient perspectives the level of t of the mHealth system with current health systems in Tanzania, and thus to provide recommendations for bringing the system to scale, consistent with Tanzania's country-level strategic goals relating to mHealth. (23)

Setting
This study took place within four HIV care and treatment clinics located in the Shinyanga region of Tanzania. Shinyanga is located in the Lake Zone region of Tanzania, a rural part of the country where over 4.5 million people live in poverty (32-35% of the population) (24). Adult HIV prevalence in Shinyanga is 5.9% (4.7% nationally); among those living with HIV in Shinyanga, approximately 40% are virally suppressed (52% nationally). (25) Intervention This paper reports on a mixed-methods, implementation science study nested within a randomized controlled trial, which has been described in detail elsewhere (20). Brie y, from April to December 2018, 530 adult PLHIV who were initiating ART were enrolled and randomized to three arms: standard of care, smaller cash transfer intervention (~$4.50 US) and larger cash transfer intervention (~$10 US). Participants in the intervention groups were provided with the cash transfer contingent on monthly appointment attendance during the rst 6 months of HIV treatment; the primary outcome was viral suppression as measured at the 6-month appointment. Attendance monitoring and cash transfers were implemented using a tablet-based mHealth system (described below). This sub-study focused speci cally on the successes and challenges of the intervention implementation strategy.

Description of the mHealth System
To confront the time-consuming process of patient attendance monitoring and verifying the conditions for cash transfers, and to simplify the logistics of the cash transfer payments, in collaboration with a local (Tanzania-based) technology rm, we created an mHealth system with the following key components: (1) pharmacy-based monitoring of patient visit attendance; (2) biometric identi cation using ngerprinting; and (3) cash disbursement for those in the intervention group integrated with all mobile money providers in Tanzania. Individual mobile money account details were registered in the mHealth system during enrollment for participants who had access and consented to mobile payments. The mHealth system was designed to be implemented in the pharmacies co-located within HIV primary care clinics; upon picking up medication, all participants were to use the ngerprint scanner to register in the system. The system then displayed a form in which the pharmacist or research assistant entered speci cs about the medication dispensed (type of ART and number of pills) as well as the next appointment date. For those in the intervention groups, upon completion of the pharmacy visit form, the cash transfer would then be automatically sent to the registered mobile phone number via mobile money (cash was available to those who did not have access to mobile money). Once received (typically within minutes), the participant receives an SMS con rming receipt of the transfer. The key design components of the system, the intended functionality of those components, their intended bene ts, and learnings from the implementation study are detailed in Table 1.
Following the launch of the mHealth system in the study clinics, we made iterative improvements to overcome ngerprint recognition challenges related to network outages, data storage and processing limitations, and low image quality due to hardware issues, patient unfamiliarity with ngerprinting, and worn or damaged ngerprint patterns among patients who perform manual work.

Participants
The implementation science study participants included a subset of PLHIV enrolled in the RCT, as well as clinical staff from the four study clinics. Participants in the cash groups were to share preferences on mobile money and a subset of all participants was asked for more detailed feedback about the entire mHealth system (described below). In-depth interviews were also conducted with 25 PLHIV who were selected using purposeful sampling (equal distributions of men and women in both of the two cash groups). We also recruited 10 clinic and pharmacy staff at the four study clinics; clinic staff were purposively sampled to represent pharmacists, nurses, doctors, and heads of clinics.

Data Collection -Surveys
At the 6-month follow-up visit, a subset of participants were asked about two aspects of the mHealth system using an adapted version of the Health Information Technology Usability Evaluation Scale (HITUES)(26): the biometric identi cation feature (n = 104 participants from all study groups) and the automatic mobile money disbursement (n = 53 participants from the intervention groups). The HITUES is divided into four domains to assess the impact, usefulness, ease of use, and user control of the system from the perspective of the patient (see supplemental Table S1 for all questions, by domain). Possible responses were on a 5-point scale of agreement, with higher scores relating to strong agreement with the presented statement. In addition, all participants in the cash groups (n = 346) were asked about their preferences for mobile money versus cash in hand. All surveys were conducted using Qualtrics o ine surveys.

Data Collection -In-depth Interviews
We conducted semi-structured in-depth interviews (IDIs) with PLHIV (n = 25) and clinic and clinical staff (n = 10). PLHIV were asked about their experience with the biometric ngerprinting system of the mHealth system and about their experience with the cash transfer (via mobile money or cash). These interviews were conducted at the 6-month follow-up visit. Clinical staff were asked about the mHealth system as a whole and their experience using it as part of clinic operations in the context of the study. Interviews with clinical staff were conducted at the end of the study. All qualitative interviews were conducted in Kiswahili and recorded, transcribed and translated into English.

Data Collection -Structured Observations
Four months into the study, we initiated structured observations in all of the study clinics (n = 2,293 visits) to document functionality of the biometric identi cation system and the mHealth system. Data were collected on a per visit basis (not containing identi able information), and documented using Google Forms. Data were then imported into STATA for analysis.

Implementation Outcomes
We explored outcomes from Proctor's implementation science framework(ref) from both the PLHIV and clinic staff perspectives. Speci cally, we triangulated data from multiple sources to evaluate the following outcomes: appropriateness, acceptability, delity, adoption, coverage and sustainability. Table 2 lists each outcome, the adapted de nition of that measure for this study context, the indicator used to assess the outcome, the data source used, and which population the outcome applies to, e.g., PLHIV or clinical staff. (The subsequent results section is organized by implementation science outcome and participant perspective.) Analysis Descriptive analyses of quantitative data collected through surveys and structured observations were analyzed using STATA statistical software (27). Qualitative data were analyzed using an inductive data analysis approach(28), and themes were documented by two coders as they emerged. Coding was completed by one member of the US team and one member of the Tanzania team; codes were compared and coding discrepancies were discussed and resolved. All coding was conducted using Dedoose (29). Table 3 shows the demographic information for the 530 PLHIV enrolled in the full trial, the subset of the 104 PLHIV who responded to the HITUES survey and the subset of 25 PLHIV who participated in the indepth interviews. The mean age ranged between 34 and 36 years, and across the three groups the majority of participants (~60%) were women. Most were in a monogamous marriage, and initiated ART at WHO clinical HIV stage 1. The majority reported having access to a mobile phone at baseline (88%-100%), slightly fewer reported having access to mobile money at baseline. By the end of the study, 81% of participants had registered their mobile money account with the mHealth system.

Page 7/24
Fingerprint Scanning: Patient Perspective We found that most PHIV were comfortable with the ngerprint scanning and the mHealth system overall, especially given the growing ubiquity of biometric identi cation in Tanzania. Several mentioned that biometrics are the way things are moving in the country, and that many businesses are already using it.
The good thing [about the ngerprinting] is about con rmation, because if the ngerprint is not yours then the system won't con rm…. I think ngerprints are clearer therefore service provision becomes easier… I knew it was something useful in most sectors that's why I didn't see any problem or feel bad.
Some PLHIV felt that ngerprint scanning added not only accuracy in con rming their identity, but also legitimacy, security, and enhanced reliability in tracking visits and medication pick up.
The current system which we used to scan should continue…this system is assured…the data remain safe and in a good system. By the previous system, you can forget, you can also lose the papers; different from the current system where information is kept electronically.
Participants noted that the ngerprint system simpli ed the appointment process and remarked that they spent less time at the clinic once they started using the ngerprinting and mHealth system. While the mHealth system did not completely obviate the need for paper les, the perception among some was that the system did streamline the clinic process and procedures. However, in contrast to those who talked about the e ciency or simplicity of the system, some participants brought up that they felt the system caused delays at the clinic because of its dependency on a reliable network connection, and delays related to staff who were not pro cient with the system.
What I can say is sometimes when you come [to the clinic] and meet one person. When you come another time you nd a new person. So, some pharmacists were good and some I can say didn't know how to use it [mHealth system], so if you meet with the one who is well experienced you don't spend a lot of time but another might tell you the system is not working because they don't know how to use it.

Mobile Money: Patient Perspective
Using the survey and enrollment data, we found high levels of acceptability of the mobile money system and the ngerprint registration process among PLHIV. Nearly all (99%) of those who had access to a mobile bank account consented to automatic cash disbursement through mobile money. Further, we found that 98% of eligible participants consented for the study (consenting included ngerprint scanning).
In both IDIs and the structured survey we asked participants (in the cash groups) about their experiences receiving the cash, and regardless of how they received the cash during the study, whether they would in general prefer to receive the transfers in cash or delivered through mobile money. The interviews point to considerable variation of preferences for delivery in cash as compared to delivery as mobile money, and considerable variation in the rationale cited for these preferences. Speci cally, one theme that emerged related to the acceptability of mobile money included accuracy, safety, and reliability. For example, those preferring the cash transfer delivery via mobile money noted that this method was more secure and more accurate: I can say there is accuracy because when money is sent from the machine to my phone it means there is a report that will be sent. Therefore, there is no security in giving someone cash in hand because the sender may not get the money to the intended person or they might. So, there are two possibilities but the system is more accurate.
A second theme that was discussed related to control over spending and ability to save money. Interestingly, several PLHIV talked about how they used delivery via mobile money as a way to save or control spending; they were less likely to simply spend the cash on the way home if the money was in their phone versus in their pocket, for example. …another thing is when you have the money in hand you may end up spending all of it in things that are of no importance., But when it's in the phone you can leave the clinic …and the money remains in your phone. When you get a problem you withdraw it but if you have it in hand you get tempted to buy random things that aren't important. So keeping it there helps, it's like a small bank, your personal bank.
A third theme that was brought up by PLHIV related to privacy concerns. Some noted that mobile money offered increased privacy compared with receiving cash; in particular, mobile money was discreet enough that their participation in the study could remain private and was not revealed to others attending the clinic, for example. However, others noted that they felt that the cash was more private that the mobile money option -some expressed concern that someone would see the message on the phone that appears when the money has been delivered and know they received money. Related to this, some women noted that cash allowed for more decision-making power -for example they may share a phone with their partner, and if the money is delivered through the phone, their partner will see the SMS message, and they will no longer be able to decide how to spend the money on their own. In addition, others noted that they were concerned that the SMS message alerting them that the mobile money had been delivered might reveal something about their HIV status or that they were participating in a study.
I: Why you like to receive [the cash] at hand? R: Because I live with my husband there at home and he is not supportive to children, sometime my children may need a small amount of money to use at school but he will not help. Sometime they need books, so it helps me but if sent through the phone, he must know it when he read the messages in my phone and he will start to question and this will be a problem.
Other themes that emerged related to preference for cash over mobile money included concerns about the technology infrastructure (e.g., the unreliable nature of the network) and fees related to the use of mobile money services (see supplemental tables for sample quotes).

Clinical Staff Perspective
To understand the level of acceptability of the mHealth system among clinical staff, we evaluated themes emerging from the in-depth interviews with clinicians, pharmacists and pharmacy staff. We de ned acceptability from the clinical perspective as the degree to which the mHealth components are considered reasonable and satisfactory given current clinical working environment. One of the themes that emerged in discussing the mHealth system with the clinical staff related to the perception that of the system into the clinics would result in additional work for the staff.
…on the side of staff, it [mHealth system] will be something new which has been added to them, it will be a new task which used not to be there, this is what I see will be like a challenge to workers because they will see it as an extra work, it can bring like a sort of resistance to change but with time they will cope it.
A second theme that emerged was related to training on the system -some noted that the system was easy to use once they had adequate training, that it simpli ed their work, and helped with managing patient ow within the clinic. …at rst, I didn't know how to use this system but now I can use it very well. At rst, I was afraid to use this system and I told them that I can't work with it and they told me that you will know it, so they directed me how to use it and I can now use it.
Finally, a theme related to system fatigue emerged in our discussions with the clinical staff. Some clinicians noted that many systems come and go as part of research or government programs, and the clinics often do not see any lasting bene t. R: Because these systems are always present…and other new systems will keep coming, because there are even some new systems that we did not nd here before, but the kept coming and we captured them and we are still working with them till now and they are good, so even if this one comes o cially then I hope that it will be captured by the servants and they will just understand.

Patient Perspective
We used the HITUES to assess the appropriateness of the automated mobile money system and the ngerprint components of the mHealth system among PLHIV (Table 4). We adapted the HITUES to explore domains of appropriateness, de ned here as the perceived t of the mHealth components (mobile money, ngerprinting) within the existing clinical care context for PLHIV. Those four domains included impact, usefulness, ease of use, and user control (see supplementary materials for full question text). The overall average score for the scale as it related to using ngerprinting for biometric identi cation was 4.1 (out of a total possible 5.0). Scores for the ngerprinting showed even less variation than those for the automatic mobile money disbursement, ranging from 4.0 (usefulness domain) to 4.2 (user control domain).
The overall average score for the scale as it related to the automatic mobile money disbursement (compared to receiving cash in hand) was 4.2 out of a possible 5, indicating that participants found the system both acceptable and useful, with minimal variation by domain. The impact domain had the highest average score (4.3 out of a possible 5 points).

Clinical Staff Perspective
We explored appropriateness of the mHealth system with clinical staff through the in-depth interviews. We de ned appropriateness as the perceived t of the mHealth system components within the existing clinic context. Discussions converged around the following themes: technical di culties, staff shortages and staff turnover, facilitation of patient follow-up, bene ts for the patients, and spillover effects for patients not enrolled in the study.
In terms of technical di culties, some staff expressed frustration, as the system did not always work, and was dependent on having a reliable network connection.
Maybe what gives me a challenge is your system of research, for example a customer may come, you will pass him/her through the system, if the customer put his/her nger prints, the system fails to show recognition, you may try the left hand but it does not respond, the same with the right hand also, so you have to use the ID number, so this is what challenges me.
Other clinicians discussed the di culties of implementing the systems when there are substantial staff shortages and staff turnover. Such shortages meant that job duties and roles changed regularly, as managers shifted staff around to cover gaps.
My opinions are, rst let me start with the positive ones, the admission of the system went hand in hand with offering of bene ts to the clients and it even helped us in reducing the number of lost follow-up clients meaning on those days that a client is on appointment, he does not show up, though it had its challenges.
Additionally, others noted that the system helped with reminders for when patients were due for viral load testing, and some pointed out the potential for the system to help with patient follow-up across clinics if the system were to be implemented nationally or even regionally.
My feeling is that [with the system], we will rst reduce LTF [loss to follow-up] and there will be frequent tests because if a patient scan his/her details will automatically come as the way they are and I think that a person can test himself or herself in another place, If he/she has already being recruited in this system, I think if he/she just scan, the details will appear automatically, so we will reduce LTF because most of the LTF appeared because of repetition of testing, you may nd that he/she was taking medicine from another place and then he/she come to our place and nd it di cult to follow his/her transfer or he/she has lost the card so he/she decide to do new registration. I think re-registration of patients will end.
Several clinicians talked about how the mHealth system bene tted patients during the study, and discussed how the mHealth system had the potential to bene t all patients in the clinic were the system to be implemented broadly. Speci cally, they noted that the system helped patients transition from every month prescription pick-ups to 3-month prescription pick-ups as it helped with visit attendance and thus with adherence.
Others recognized that the system helped patients come to appointments on schedule and remarked that the system led to more complete viral load testing as it helped to reduce loss to follow up. Finally, some clinicians recognized that the system had positive spillover effects even for those who were not enrolled in the study.
…because the system is there and there is a close follow up, it helps to remind us that the patient is required to conduct a test, it has motivated us and increase our attention in making follow up, not only for those who are in the system but also for all patients in general, it has helped us improve the service in the HVL [HIV viral load testing] in general.

Adoption Patient Perspective
To measure adoption, we explored the proportion of PLHIV study participants who had their mobile bank accounts linked to the mHealth system at the end of the study. At the end of the study, 88% of study participants reported that they had access to a mobile phone, 78% reported that they had access to a mobile money account, and among those in the cash groups, 81% had registered their mobile money account with the mHealth system.

Clinical Staff Perspective
We measured adoption from the clinical staff perspective by examining the proportion of time that the pharmacist was operating the mHealth system -the intended design. As the mHealth system was rolled out in the clinics, research assistants provided signi cant levels of support to the pharmacist in registering patients with the mHealth system, however, by the end of the study, the pharmacist was operating the system on average 72% of the time (as opposed to the research assistant being the primary operator).

Fidelity Patient Perspective
Among PLHIV in the cash award groups (n=346), 331 (96%) received at least one cash transfer during the 6-month study, and the average number of cash transfers per study participant was 4.7 (out of a possible 6 transfers). Out of a total of 1,651 cash transfers delivered to study participants, 1,283 (78%) were delivered through mobile money (the remainder were paid to participants in cash) and 2.5% of those sent through mobile money required manual re-sending due to network failures.

Clinical Staff Perspective
To assess delity of implementation from the clinical staff perspective, we looked at the proportion of visits for which the mHealth system was used. Over 3,067 total clinical visits across all four health facilities during the 6 months following study enrollment, 172 (5.6%) were not captured in the mHealth system; 94.4% of all visits were registered into the mHealth system at the time of visit. The proportion of visits for which the mHealth system was used ranged from 94.6% to 89.1% by clinical site.

Feasibility -Clinical Staff Perspective
To assess feasibility, we focused on the technological aspects of the mHealth system implementation, exploring how frequently the ngerprint scanning system and mobile money distribution systems failed.
The clinic pharmacy structured observations captured data on 2,293 patient visits over the course of the study. Observations were focused on the use of the mHealth system during clinic visits and included the number of ngerprint scans required until the mHealth system successfully identi ed the patient, and whether or not the ngerprint scan was eventually successful in identifying the patient (regardless of the number of scans required). It should be noted that as more PLHIV enrolled in the study and more ngerprints were added to the database, nding correct ngerprint matches became more complex, and required several iterations of the matching algorithm. Overall, ngerprint recognition succeeded for 74.1% of visits while 25.9% required manual entry of the patient's unique identi cation number due to poor image quality. The success rate for ngerprint recognition increased over time; by the nal month of structured observations, the success rate was 87.3% ( Figure 1). Overall, the average number of ngerprint scans required for the mHealth system to successfully identify the patient was 2.04; this also varied considerably by study month, and by the end of the study, the average number of scans was 1.8.

Sustainability -Clinical Staff Perspective
In an effort to explore scale-up and sustainability, we asked the clinicians about what challenges they might anticipate encountering should the mHealth system offset some paper-based aspects of the current recordkeeping system. Many of the themes that emerged had already been mentioned, including the importance of training, considerations related to staff turnover, and the need for a salary top-up as many staff saw use of the system as an added task rather than as a way to simplify or facilitate their existing work.
The potential bene t for patients was another theme that arose when discussing sustainability of the mHealth system. Several clinicians noted recognized the potential bene ts and mentioned that with adequate training, they would be prepared for the new system.
We are positively ready for it [mHealth system] as I have said that we have observed its high impact within this short time of using it. So, for us we have received it with all our hands and we wish if it would come earlier, it would be more better because data shows, data shows us that it has good results so we are well prepared about this, and if there will be changes which will make us to change our ow pattern or service provisions in general we are positive about it because we have observed the positive impacts so as the clinic we are ready for the changes which will come.

Discussion
In general, we found that the mHealth system was overall successfully implemented with high levels of acceptance and usability from both patients and providers; the system we designed functioned as intended. While ngerprint recognition encountered some feasibility limitations in the rst months of testing, payments via mobile money were highly successful. Clinic management staff voiced excitement for the system's potential to bring the cash program to patients and simplify work ows; yet concerns about multiple systems, sta ng shortages, and intermittent connectivity tempered enthusiasm, highlighting structural issues beyond the scope of the program.
Still, there were unintended bene ts and consequences that we learned about along the way, many of which could impact scale-up efforts of clinic-based incentives for improving HIV care outcomes. Speci cally, ngerprint identi cation plus automatic disbursement of mobile payments have strong potential as a means to e ciently implement clinic-based cash incentives in low-income country settings. In fact, utilization of mobile money has increased signi cantly in low-resource settings in particular (over 50% of Tanzanians have a mobile money account), and continues to grow quickly (30). As a result, social protection programs in low resource settings that utilize cash transfers are increasingly using mobile money instead of cash due to its multiple advantages, including the potential for increasing nancial inclusion, decreased travel and time requirements for recipients, and increased accuracy and accountability (31,32). However, based on our results, the following should be considered: a) upfront effort needs to be expended to ensure that patient concerns about privacy are addressed, for example by allowing participants to opt-out of SMS noti cations, or by providing assurance that SMS messages do not reveal anything about HIV status; b) while most participants had access to mobile phones and mobile money, providing access for all will be needed to successfully implement this system and consideration given to how to weigh this need against the ability to bring the intervention to scale;; c) implementation support in the clinic including universal training, incentive payments for staff to use the system, and ensuring there is a solid understanding of the bene ts conferred to patients (e.g., decreasing loss to follow up) will be crucial; d) getting the ngerprint system right took more time and was more complicated than expected -issues relating to image quality, computing storage and creating the threshold precision level for ngerprint recognition resulted in a steep learning curve; e) unexpected bene ts of the system included lending a sense of legitimacy and belonging for patients, a strong understanding among clinical staff of how such a system could help patients and simplify clinic operations, and as a result of the mobile money, the potential for increased savings and increased control over spending, as well as increasing intrahousehold bargaining power for women. These ndings are supported by other research and data showing that mobile money payments (as compared to cash) provide an early entry point into the formal nancial sector, and provide the opportunity for women to have more control over their money and how they spend it. (33)(34)(35) It should be noted that some characteristics inherent to many clinical settings in resource constrained settings may limit the impact of any mHealth system. Many of these were mentioned by clinical staff in discussing the appropriateness of the mHealth system in this context; these include infrastructure (reliable WiFi networks, replacement tablets, secure storage), sta ng shortages and turnover, and the need for universal training and sustained training support for using the system. These inherent limitations underscore the need for developing a system that easy to use, is easily scaled and does not rely on paper-based tracking systems and manual identi cation of eligible bene ciaries for a cash transfer program. In addition, working with a local technology rm who understands well the context and unique challenges that arise in these settings is essential, a type of partnership that bene tted this research.
This study has several limitations. For example, despite the implementation science focus of this work, gaining a true understanding of how this mHealth system might function outside of the study context can be challenging, particularly with signi cant levels of research staff involvement during the implementation phase. We attempted to gain an understanding of this dynamic by assessing the proportion of visits that were handled by the pharmacist versus the research staff at study end; these data showed a positive trend toward implementation as designed, with the pharmacist operating the system, allowing research staff to step away from the process. Additionally, we will have the opportunity to explore this aspect of implementation in more detail during phase 2 of our study(36). Phase 2, scheduled to begin in early 2021, is a cluster RCT in 32 clinics in Tanzania, and will provide ample opportunity to evaluate implementation outcomes in a setting that approaches "real world," where we can assess heterogeneity in implementation and its impact on the primary study outcome(36). Additionally, in-depth interviews conducted with a small subsample of study participants may not be representative of the full study population, and some data collection methods (e.g., structured observation in the clinics and the HITUES questions) were introduced toward the end of the study, so these data may not represent the full study experience of participants. Other limitations include the potential for social desirability bias to impact reported perceptions of the mHealth system among both participants and clinical staff, and the lack of data from PLHIV participants in the cash groups who received payments in cash rather than through mobile money.

Conclusions
We found that biometric identi cation and mobile payments were acceptable to most patients and staff in this semi-rural region of Tanzania. Biometric identi cation and mobile payments may provide a scalable mechanism to improve patient tracking and e ciently implement cash incentives in low-income country settings, and mobile money payments may result in the unintended bene t of increasing savings and providing a means to control spending.

Consent for publication: Not Applicable
Availability of data and materials: The datasets used and analyzed during the current study are available from the corresponding author on reasonable request.
Competing interests: The authors declare that they have no competing interests Funding: This study was funded by the National Institute of Mental Health (NIMH) grant number: R01MH112432-01A1 Authors' contributions: LP analyzed the data and drafted the manuscript; CF assisted with data analysis and interpretation, data collection, and manuscript review and revisions; AM and AK conducted interviews, analyzed and interpreted qualitative data, and provided comments on the manuscript; SM designed the study, assisted with interpretation of data, and provided revisions and comments on the manuscript; PN oversaw all data collection efforts, contributed to the design of the study, and reviewed and provided comments on the manuscript.