This study was approved by the institutional review boards of the University of Georgia (ID PROJECT00000571), Makerere University (protocol 756), and the Uganda National Council for Science and Technology (HS656ES). All participants provided written informed consent in either English or Luganda, based on their language preference. The consent process included assurances regarding the privacy and confidentiality of the data collected during the study. Participants received approximately US $10 as compensation for transportation and the time spent participating in each visit.

The “DOT Selfie” study was a parallel-group, open-label randomized controlled trial involving 144 adult patients with TB at selected treatment clinics in Kampala, Uganda. Participants were randomized into a control group (n=72), which received UCDOT, and an intervention group (n=72), which received VDOT for adherence monitoring. Enrollment into the RCT occurred between July 13, 2020, and follow-up continued through October 25, 2021. The details of the trial’s design and methods were described in our published protocol [33].

Eligible participants were adults aged 18-65 years with a confirmed diagnosis of drug-susceptible TB in the new or retreatment category, who had initiated treatment within the past month and planned to reside in Kampala for the entire 6-month treatment period to facilitate close study follow-up. Participants were excluded if they could not speak and read either Luganda (the local dialect) or English, were too ill to withstand the study procedures at enrollment, or had a self-reported motor, visual, hearing, or cognitive disability that would hinder proper use of a smartphone. Additionally, patients residing in areas without cellular network coverage or access to electricity for charging their smartphones were excluded.

The primary study site was the Lubaga TB clinic, with secondary recruitment sites being public clinics that provide free TB services in Kampala. All TB clinic sites are located about 10-15 km from the Kampala city center and collectively treat approximately 10,000 patients with TB annually. The study team worked closely with clinic nurses, who helped identify potentially eligible patients at the selected public clinic sites. Trained research assistants used a screening checklist to assess study eligibility, and then provided a brief description of the study. The details of the design and methods of the randomized controlled trial are described in our published protocol [33]. The baseline questionnaire was administered after consent, and a copy is provided in Multimedia Appendix 1. Participants were then randomized into the intervention or control study group. The study flow is shown in Figure 1.

We randomized 144 participants in a 1:1 ratio to receive either VDOT or UCDOT, as shown in Figure 1. We chose equal allocation to optimize the power of the study [34]. Randomization was stratified by sex to ensure equal representation of men and women in each study group [34]. Sex stratification was justified because sex is a known factor associated with adherence to TB treatment [34,35]. Permuted block randomization with block sizes of 4 and 6 was used to ensure a balanced assignment of participants to the intervention or control groups and to minimize the predictability of study arm assignment. Each block contained a specified number of randomly ordered treatment assignments, as recommended in clinical trial methods by Friedman et al [34]. The detailed process followed during randomization is published in the protocol elsewhere [33]. Given the nature of the intervention, participants and study staff were not masked to study group assignment. Study investigators were not blinded to allocation information due to ongoing scientific reviews and the possibility of making clinical decisions about participants’ treatment if it became necessary to withdraw them from the intervention. The trial statistician was blinded to study group assignment during the analyses.

VDOT is a licensed Health Insurance Portability and Accountability Act (HIPAA)–compliant system originally developed by SureAdhere Mobile Technology, an initiative that started at the University of California, San Diego [36]. It consists of 3 main components, as described previously by Garfein and colleagues [19] and in our published protocol paper [33]. The components include (1) the patient-facing part, which is the smartphone app used by the patient to record medication videos; (2) the secure cloud-based server, which stores the uploaded medication videos; and (3) the provider-facing part, which is the web-based browser accessed through a secure, password-protected log-in via a computer, laptop, or tablet to review the submitted videos and daily adherence reports [33]. At the end of each video recording, an automatic, encrypted, time-stamped video was uploaded to the secure cloud server for storage and playback (Figure 2). Upon submission, the video was removed from the smartphone, ensuring that patients could not retrieve it.

The VDOT system was adapted to the local Ugandan context in 3 ways. First, we expanded the English-based SMS text message reminders to include motivational messages encouraging patients to continue taking their medications and submitting videos. Second, we translated the SMS text messages into Luganda (a local dialect) to facilitate understanding for patients who did not speak English. The following are examples of the SMS text message reminders and motivational messages we used: “It’s time to take your pills and send a video” and “Taking your pills will help you to get better.” Third, an additional SMS text message reminder was sent automatically 8 hours after the standard SMS text message reminder, which was sent to participants at 6:00 AM, if a medication video had not been received in the VDOT system. After adapting the VDOT system to the local context, it was dubbed “DOT Selfie.”

The trial was powered at 80% to detect a 22% difference in medication adherence, the primary outcome, between the 2 groups: VDOT (85%) and UCDOT (63%), based on a 2-sided significance level of 5%. The adherence levels for the comparison groups were derived from previous DOT and VDOT studies conducted in Kampala, Uganda [25,39]. To achieve a chi-square test with a difference of 0.22 and an odds ratio of 0.30, the estimated total sample size was 124. We assumed an attrition rate of 14%, based on a randomized controlled trial comparing usual care and digital adherence interventions in Kenya, where the overall loss to follow-up rate was 11.7% (9.9% in UCDOT and 1.76% in VDOT) [40]. The calculated final sample size was 144 participants, with 72 per group. When calculating power and sample size, we treated the response as binary (0 or 1), with a cutoff value for nonadherence as the FEDO <80%, and adherence as the fraction ≥80%. This binary response variable for adherence was used in the final regression analysis to facilitate comparison with previous clinical trials [16]. Sample size tables were used to estimate precision, as described in detail elsewhere in our protocol paper [33]. All calculations were performed using the SAS statistical software package, version 9.4 (SAS Institute).

We conducted an intention-to-treat analysis according to the study group to which patients were originally randomized. Superiority was determined by a 22% difference in the proportion of patients achieving the primary outcome (63% vs 85%). VDOT treatment observation was classified as completed if patients’ medication doses were observed via video. UCDOT treatment observation was classified as completed if the patient reported that a designated treatment supporter observed the ingestion of prescribed doses. Univariate analysis was performed to describe the baseline characteristics of the study population. Multivariable logistic regression analysis was conducted to test for significant associations between the study groups (as the main exposure) and the primary binary outcome of adherence (no <80%, yes ≥80%). Age and HIV status were selected as potential confounders and were adjusted for in all models. Other covariates considered for adjustment during the primary analysis included education level, income, and smartphone ownership at baseline. We did not adjust for sex because randomization was stratified by this variable. Crude and adjusted risk ratios with 95% CIs are presented, and statistical significance was considered at P values less than .05.

Of the 228 patients with confirmed TB who were screened, 144 met the eligibility criteria and were randomized into 1 of the 2 study groups, each with 72 participants. Two participants were excluded from the intention-to-treat analysis (n=142): 1 participant in the VDOT group refused to initiate TB treatment shortly after study enrollment, and another participant in the UCDOT group declined participation after initiating treatment. The 2 participants were included in the baseline analysis but excluded from any further analyses because they did not have the study outcomes of interest (Figure 1 and Multimedia Appendix 4 [41]). The median age of the participants was 34 (IQR 26-45) years. There was a significant difference in median age (P=.005) between the study groups, with participants in the UCDOT group being older; 120 of 144 (83.3%) participants owned a cell phone, while 53 of 144 (36.8%) had smartphones. Within the VDOT group, 29 of 72 (40%) owned a smartphone. Overall, 46 of 144 (31.9%) participants were HIV positive. All the measured baseline characteristics were balanced between the study groups, except for age (Table 1).

A visual illustration of the patterns of treatment observation among study participants is shown in Figure 3. For the VDOT group, the light gray bars represent the videos received via the technology system, while the dark gray bars represent the videos that were not submitted (ie, missed). For the UCDOT group, the light gray bars show the doses that were observed (as reported by the patients) and the dark gray bars show the unobserved doses. We observed clear differences in the patterns of observation between the 2 study groups. However, when the same data are further stratified by sex, there are no obvious differences in the patterns of doses observed between men and women within each group.

There was a significant difference in the median number of doses observed between VDOT and UCDOT (Table 2). However, there were no significant differences in the number of doses when comparing male and female participants within the same study arm (Table 3).

We evaluated adherence using the FEDO, expressed as a percentage, as shown in Table 4. The overall median FEDO was 80% (IQR 30%-100%), but the median adherence in the VDOT arm was 100% (IQR 80%-100%), which was significantly higher (P<.001) than that of the UCDOT group. When using a cutoff of ≥80% for FEDO to define optimal adherence and <80% for nonadherence, only 63 of 142 (44.4%) participants, on average, achieved optimal adherence with their expected doses observed. However, in the VDOT group, 56 of 71 (79%) participants achieved an optimum FEDO, which was significantly higher (P<.001) than in the UCDOT group. The median FEDO did not significantly differ by sex in either study arm (P=.16 for VDOT and P=.96 for UCDOT).

We performed a multivariable logistic regression analysis to evaluate the factors associated with optimal adherence levels (≥80% doses observed), as measured by the FEDO, with the results presented in Table 5. The results from the unadjusted regression models indicated that being in the VDOT group and identifying as Muslim were independent factors significantly (P=.03) associated with achieving adherence of ≥80%. We found that participants in the VDOT group were 8.4 times more likely to have ≥80% of their expected doses observed compared with those in the UCDOT group, after adjusting for confounding factors such as age, religion, current cell phone ownership, and HIV status. Current cell phone ownership was significantly (P=.04) associated with reaching the 80% adherence level, based on expected doses observed, compared with those who did not own a cell phone, after adjusting for the study group and other covariates.

In the intention-to-treat analysis of 142 participants, the secondary TB outcomes (Table 6) evaluated showed treatment completion at 90.1% (128/142) for all participants, with significantly higher completion rates in the VDOT group compared with the UCDOT group. The rate of loss to follow-up was 6.3% (9/142) overall, but 7 of 71 (10%) participants in the UCDOT group were lost to follow-up, compared with 2 of 71 (3%) in the VDOT group. Overall, there were 5 deaths; 4 of these were HIV coinfected and 3 were male (see details in Table 7).

Based on the standard 6-month regimen of 168 doses and the enrollment date, we expected a total of 11,928 videos. Of these, 11,041 (92.56%) were successfully submitted by participants included in the intention-to-treat analysis over the study period. Of the 887 missed videos (7.44%), the most common reason reported by participants was a lack of a charged phone battery, accounting for 183 of the 887 (20.6%) missed videos. Technical or technology issues accounted for 450 (50.7%) of the missed videos. Patient-related issues accounted for 348 (39.2%) missed videos, with the most common reason (136/887, 15.3%) being forgetting to record the videos while swallowing the medication. Additional reasons for missed videos are provided in Table 8.

In this study, we found that VDOT was significantly more effective in observing medication ingestion and monitoring adherence among patients with TB. Our findings align with a recent systematic review and meta-analysis that included studies conducted in Australia, China, Moldova, the United Kingdom, and the United States. The review concluded that monitoring treatment using VDOT was significantly associated with increased adherence and better microbiological outcomes compared with UCDOT [32]. Additionally, we found that cell phone ownership was significantly associated with a higher level of observed doses, regardless of the study group. Although VDOT was effective, some participants occasionally failed to submit their expected videos of medication intake. The most common reasons for missed videos were lack of a charged phone battery, forgetting to record the medication event, and loss of the smartphone. This finding is consistent with results from our previous pilot study conducted in Kampala, Uganda [25]. To our knowledge, this study is one of the few randomized controlled trials conducted to evaluate the effectiveness of VDOT in Uganda and Africa. More studies are needed to assess the effectiveness of VDOT in rural African contexts and other low-resource settings. Additionally, it is important to identify which subgroups of patients are most likely to benefit from this technology.

There are a few published randomized controlled trials that have compared various forms of VDOT intervention packages with in-person DOT thus far [16,28,29]. Story and colleagues [16] published the first multicenter trial evaluating the efficacy of video-observed treatment compared with usual care in 2019. The study was conducted in the United Kingdom and involved a larger sample of 226 patients from 22 centers. The sample was also highly diverse, with 58% of participants having a history of drug and alcohol abuse, homelessness, and incarceration. In the UK study, 70% of patients on VDOT achieved optimal (≥80%) scheduled observations during the first 2 months, compared with 31% in the UCDOT group. The difference observed in the VDOT arm could be attributed to the difference in the timing of the measurement of the primary outcome, which was 2 months versus 6 months after randomization. The Moldova trial by Ravenscroft and colleagues [28] compared patients assigned to VDOT and clinic-based DOT. The study measured the primary outcome as the observed medication adherence every 2 weeks for the entire 6 months of treatment. The study concluded that VDOT significantly decreased nonadherence compared with standard DOT and reduced the time and money patients spent during their treatment [28]. Although there were some differences in the delivery of the VDOT intervention package and the measurement of the primary outcome, the conclusions on the effectiveness of VDOT are quite similar across studies.

The UCDOT group had a low level (21/71, 30%) of reported observed doses of medication over the 6 months of treatment. This result is very similar to the findings from the UK trial, despite differences in the study populations, delivery of DOT, and treatment durations [16]. This finding further reinforces the fact that in-person DOT is generally difficult to implement, particularly in low-resource settings [39,42]. A Cochrane review of 11 clinical trials concluded that TB cure and treatment completion rates were not improved by DOT [13]. In South Africa, a recent study found that only about 25% of patients with TB received formal DOT support, either home-based or clinic-based [43].

Cell phone ownership was significantly associated with achieving optimal observation of the expected doses of medications, regardless of the study group. Moreover, we found that 120 of 142 (84.5%) patients owned a cell phone, while 53 of 142 (37.3%) had personal smartphones at baseline. This finding suggests that having access to a cell phone could serve as an indirect facilitator of adherence support for patients on treatment. In African settings, smartphone ownership and cellular networks are rapidly proliferating, presenting opportunities to leverage modern technology to improve health care and service delivery. According to the National Institutes of Health, “there is a need to stimulate research utilizing mHealth tools aimed at the improvement of adherence to treatment, effective patient-provider communication, and self-management of chronic diseases in underserved populations” [44]. There is increased utility and effectiveness of accessible mobile technologies in enhancing the monitoring of patients with chronic diseases, which can be attributed to the affordability and reliability of smartphones in both high- and low-income settings [45,46].

The main secondary outcomes, including treatment completion and loss-to-follow-up, were statistically better in the VDOT group compared with the UCDOT group. These findings are consistent with the conclusions of a recent systematic review and meta-analysis [32]. The notably higher percentage of TB/HIV coinfected individuals among the deaths aligns with epidemiologic trends [2]. Treatment failure is an important outcome, but it is often difficult to ascertain, especially with follow-up periods of less than 1 year. To determine treatment failure, we used the records at the end of treatment at 6 months, although it should be routinely assessed at 5 months. We relied on the TB clinic records of sputum results at month 5 because our study protocol included only follow-up visits at months 2, 4, and 6. The study team did not have direct control over the standard clinical evaluations that occur in routine TB management. The clinic records were grossly deficient; therefore, we cannot conclusively state that there were no treatment failures. Larger studies are needed to compare these outcomes across the groups to ascertain the true differences.

Overall, technical and technology-related issues were the most commonly cited reasons for failure to submit videos. The lack of a charged phone battery was the single most common reason for failure to submit videos, as sometimes phones turned off during video recording. This finding is consistent with results from our pilot study in Kampala [25,47]. To minimize the phone battery issue, patients were trained and encouraged to ensure that their phones were always charged at night or before recording videos. The loss of a smartphone occurred among a few participants, resulting in a substantial number of missed videos. The study team mitigated the disruption in data collection by replacing lost phones as soon as they were made aware. Missing adherence records were filled using self-reports from patients or clinic records on prescription refills. However, the self-report method falls short because it lacks an objective way to validate the doses taken [48].

The main patient-related reasons that mostly prevented the submission of videos included forgetting to record the medication intake, traveling to locations without access to electricity to charge the phone, or being too ill to record videos. These findings were similar to those reported in our first pilot study [25] and in other studies elsewhere [19,26]. Typically, when patients travel to rural areas in Uganda, there is a high likelihood that they will not have access to electricity, further complicating adherence monitoring. When videos were missed due to travel, the research team followed up to gather the missing information using self-reported adherence. A small number of patients reported a lack of a convenient place to record videos, which was a concern similar to our previous study [25]. This could suggest an underlying stigma associated with TB treatment. However, a study done in Turkey, which compared the levels of stigma among patients with TB using in-person DOT versus VDOT, concluded that patients who received VDOT reported experiencing less stigma [49]. The role of VDOT in mitigating stigma among patients with TB warrants further research, especially in diverse contexts.

First, we evaluated the enhanced VDOT intervention, which is the only digital adherence technology offering remote visual observation of doses taken, making it comparable to the UCDOT. Other commonly used digital adherence technologies for TB monitoring, such as 99 DOTS and digital or smart pillboxes, only provide indirect signals as proxies for doses of medication intake [18]. Second, we successfully evaluated a culturally adapted VDOT in a randomized controlled trial (RCT) within the context of a low-income setting. In our study, we demonstrated that more than one-third of the participants owned a personal smartphone, and these numbers are growing globally. Third, we conducted the study in Kampala, Uganda, where the TB rates are estimated at 250 cases per 100,000, which is much higher than in the majority of settings where VDOT has been previously evaluated [16,19,29]. Fourth, the study facilitated the delivery of a patient-centered approach that was more convenient and less intrusive during the critical period of the COVID-19 pandemic. This also catered to the public health mitigation measures aimed at curtailing the transmission of COVID-19. Lastly, the RCT results generated in our study provide a basis for the systematic evaluation of the implementation process of VDOT. Although there is no gold standard for using digital adherence technologies to monitor TB treatment, our study adds to the mounting evidence that supports the digital observation of pill-taking as an alternative to in-person DOT in usual care.

First, the study was conducted during the COVID-19 pandemic; therefore, some unintended disruptions in the health system could have altered the practical delivery of usual DOT care in Kampala, Uganda. This may have led to an overestimation of the effect of VDOT. However, a study comparing VDOT and UCDOT before and during COVID-19 found significantly higher adherence to TB treatment when using VDOT during both periods [50]. It is possible that adherence in the UCDOT group could have been greater if not for social distancing mandates and lockdowns. The potential effect of the pandemic period may limit the generalizability of our findings. Second, our study did not specifically measure clinical endpoints, such as sputum conversion at follow-up visits. Although we attempted to extract this information from routine clinic records, the data were incomplete. Third, our study was conducted in an urban setting, which may limit its generalizability to rural settings due to differences in key factors such as cellular network availability, internet coverage, and access to electricity for charging phone batteries. Lastly, the enhanced VDOT intervention included multiple components; however, this study was not designed to isolate the effects of specific components that may have contributed to the differences in adherence observed. Future studies could use a factorial design to assess the impact of different intervention components. There are cross-cutting limitations common to all digital health technologies, such as technology infrastructure, cost, and access to smartphones, which will impact adoption, scale-up, and sustainability, particularly in low-income settings [18].

We proposed several considerations for the broader adoption, scale-up, and sustainability of VDOT in low-resource settings. Our findings support the need to carefully address the known facilitators and barriers to VDOT acceptability, such as those highlighted in the qualitative exit study of patients who used VDOT in Uganda [30]. First, VDOT requires upfront investment in the health system and local technology infrastructure to facilitate scale-up. A robust system would include the hardware—computers or smartphones—and the software—access to the internet, cellular networks, and maintenance of the VDOT system. There is also an overarching need for adequate access to electricity. Second, national TB programs could leverage public-private partnerships with entities such as telecommunication companies that manage vital resources such as internet coverage for sustainability. Third, standardized training for health providers is essential to facilitate the effective use and adoption of digital technology. Fourth, there is a need to integrate the VDOT system into the general health management information system to ensure interoperability and the timely use of adherence outcome data to improve patient care. The VDOT system could eventually be optimized for monitoring treatment adherence across multiple diseases, such as HIV/AIDS, diabetes, and hypertension. Lastly, patients would need basic training in the use of the VDOT app to ensure ease of use. In our qualitative exit interviews with users, training in the use of the smartphone app was cited as a major facilitator of ease of use and acceptability for VDOT [30]. In general, we expect that health providers and patients will grow more experienced and comfortable with technology over time.

Confidentiality, privacy, and stigma concerns are overarching issues that must be carefully addressed to alleviate fears of intentional or unintended disclosure of a patient’s disease status. The flexibility of making video recordings at the patient’s desired time and place greatly enhances privacy [51]. In a VDOT study conducted in the United States and Mexico, patients agreed that VDOT promotes autonomy and a sense of control over their health [19]. The ethical concerns related to digital technologies can vary widely, particularly in relation to collecting, securely storing, and accessing videos that contain personally identifiable information, such as patients’ faces. These concerns may also have legal and sociocultural contexts that need to be considered [52]. A recent qualitative study focusing on digital health technologies in sub-Saharan Africa concluded that threats to scale-up include the lack of buy-in from both patients and providers, insufficient human resources and local capacity, inadequate governmental support, overly restrictive regulations, and a lack of focus on cybersecurity and data protection [53]. Further research is needed to evaluate the impact of VDOT on patient-centered outcomes, such as satisfaction, quality of life, user engagement with the technology, and retention to prevent dropout. Other important areas of research include the effect of VDOT on health system workflows and workload, as well as its costs and cost-effectiveness compared with usual care.

The enhanced VDOT was more effective in increasing adherence observation to treatment than UCDOT among patients with TB in Uganda. This evidence supports the potential of digital technologies to improve monitoring and support of treatment adherence in high TB burden settings with limited human resources. Further research is needed to evaluate which subgroups are most likely to benefit from this technology.