A pilot trial using EMA messaging to influence certain health-related behaviors was conducted in Rakai, Uganda in 2017 [9]. In addition to supporting healthful behaviors (eg, fruit and vegetable consumption), an EMA app sent real-time messages to support users in reducing smoking, drinking alcoholic beverages, and having condomless sex with a non-long-term partner. Participants aged 18-49 years with a telephone number and who completed at least secondary education were purposively recruited from the Rakai Community Cohort Study (RCCS), an ongoing population-based cohort, between February 2016 and March 2017 [5]. Participants were given a password-protected smartphone with the trial’s app (emocha Health Inc) installed, along with a charger, power bank, and funds to purchase 525 megabytes of data monthly [9]. As part of the informed consent process, participants were told that the smartphone would track their location over the study period and that the same phone would be used to log their health behaviors—all of which would be kept confidential in locked files or secure computers by authorized project personnel.

This substudy used 2 data types: GPS locations and user-initiated reports. The smartphone app logged GPS coordinates every 1-2 hours if the phone was turned on and connected to the internet, otherwise, data were logged once the phone was connected. Using the app, participants were instructed to submit self-initiated behavioral reports within 1 hour of engaging in any of the behaviors. Each participant’s user-initiated reports were linked to logged GPS coordinates by nearest timestamp. Sociodemographic information from the most recently available RCCS survey (2015-2018) was used to describe the study population [10]. Participants’ home GPS coordinates were logged during RCCS household enumeration.

For each participant, the great circle distance was measured in kilometers between each recorded GPS point and the previously occurring point [11]. GPS points were excluded if they were outside of sub-Saharan Africa (<1% of GPS coordinates per participant). Interpoint distances were totaled over each participant’s roughly 90-day observation period [12]. Mobility was summarized as the average distance traveled per week to account for certain days of the week with outliers of distance traveled (eg, staying home on Sundays) that can bias centrality estimates [13]. Two other measures were also described: the number of unique places visited (deduplicated points within 25 m of one another, based on smartphones’ 5-20-m GPS point accuracy [12]), and the percent of GPS points recorded away from home [7]. To calculate the latter measure, GPS points were dichotomized as being located within or outside a home buffer region. A 100-m Euclidian distance circular buffer around participants’ home GPS coordinates reflected the home location, accounting for physical landscape and GPS accuracy [14]. We also calculated 400-m and 800-m buffers reflecting home neighborhoods that might include neighbors’ residences, shops, and transit stops [15].

We further evaluated the number of unique locations where behaviors occurred and the number of behavior locations away from the home or home neighborhood among participants reporting a particular behavior, using the same methods as above [16].

The Ugandan Virus Research Institute Research and Ethics Committee and the Johns Hopkins School of Medicine institutional review board approved the pilot trial, registered on ClinicalTrials.gov (NCT04375423). Written informed consent was obtained from interested participants at the first study visit. Any identifying information (eg, contact information) was stored in locked files or on secured computers. Once study data were collected, each participant was assigned a unique identifier and all data for analyses were deidentified. Participants were compensated for their time with 10,000 UGX (approximately US $3 at the time) and refunded for their travel to each study visit [9].

EMA trial participants were highly mobile, with over 75% of participants traveling at least 150 km/week. Most participants reporting cigarette smoking, alcohol use, and condomless sex with a non–long-term partner recorded those behaviors in 2 or more unique locations. mHealth studies using mobile phone apps could leverage GPS-based mobility data from consenting participants to inform future interventions.

The high mobility observed in this sample reflects the regional context, known for transient fisherfolk communities, localized market trading, and transport corridors connecting Kampala to the Tanzanian border [5]. As expected, participants traveled between Rakai town and Masaka city (approximately 65 km), Kampala (approximately 197 km), and fish landing sites around Misozi (approximately 60 km). In comparison, residents of a rural agricultural district of Zimbabwe traveled a fraction of the distance per week [4]. Place-based interventions should consider the local socioeconomic context when deciding where and how to deliver services, as it seems to impact the scale and geographic patterning of individual mobility [18].

Since most participants reported health behaviors in 2 or more unique locations, EMA interventions could consider using strategies, like geofencing, to deliver behavior change messages when users enter geographies in which they have historically been more likely to report certain behaviors [19,20]. For example, individuals may prefer to drink alcohol in locations where consumption is acceptable, affordable, and available [8]. Geofencing is understudied in LMICs but has been used in South Africa, where CareConekta (a GPS-based mHealth app) links people living with HIV to health facilities while traveling outside their regular clinic catchment area [21].

Increased availability of and access to mHealth technologies with GPS capabilities, including increasing mobile connectivity in LMICs, can make spatial analyses possible [22]. Our study demonstrates successful data collection using study-supplied mobile phones with an EMA app [23]. This means spatial data collection and analysis can be more easily integrated into mHealth studies than others that provide a separate wearable GPS device for short periods [24].

This study has limitations. First, the sample may not accurately represent the underlying population as only individuals with secondary education or higher were eligible to ensure they could be trained on how to use the smartphone app. For example, these findings may overrepresent the mobilities and health behaviors of people in higher socioeconomic levels. Second, under half of the participants reported each behavior, and often only once, limiting the analysis. Third, great circle distance measures may underestimate travel distances [25], though short distances between GPS points (median 0.8 km) likely minimized this bias. Finally, Euclidian buffer regions may not accurately reflect neighborhoods. Road networks and geographic features, or participant interviews, could also be used to define areas that more accurately reflect what individuals perceive as their “neighborhood” [1].

As a growing number of mHealth interventions in LMICs attempt to apply place-based strategies to improve health outcomes [26-30], more work is needed to understand the mobility of potential users. Among residents in rural Uganda, as part of a pilot trial, an EMA app successfully captured high mobility and health-related behaviors in multiple locations. Future mHealth interventions in similar settings could leverage individual-level GPS data to inform place-based strategies for promoting healthy behavior change.