Passive sensing on mobile devices to improve mental health services with adolescent and young mothers in low-resource settings: the role of families in feasibility and acceptability

Background: Passive sensor data from mobile devices can shed light on daily activities, social behavior, and maternal-child interactions to improve maternal and child health services including mental healthcare. We assessed feasibility and acceptability of the Sensing Technologies for Maternal Depression Treatment in Low Resource Settings (StandStrong) platform. The StandStrong passive data collection platform was piloted with adolescent and young mothers, including mothers experiencing postpartum depression, in Nepal. Methods: Mothers (15–25 years old) with infants (< 12 months old) were recruited in person from vaccination clinics in rural Nepal. They were provided with an Android smartphone and a Bluetooth beacon to collect data in four domains: the mother’s location using the Global Positioning System (GPS), physical activity using the phone’s accelerometer, auditory environment using episodic audio recording on the phone, and mother-infant proximity measured with the Bluetooth beacon attached to the infant’s clothing. Feasibility and acceptability were evaluated based on the amount of passive sensing data collected compared to the total amount that could be collected in a 2-week period. Endline qualitative interviews were conducted to understand mothers’ experiences and perceptions of passive data collection. Results: Of the 782 women approached, 320 met eligibility criteria and 38 mothers (11 depressed, 27 non-depressed) were enrolled. 38 mothers (11 depressed, 27 non-depressed) were enrolled. Across all participants, 5,579 of the hour-long data collection windows had at least one audio recording [mean (M) = 57.4% of the total possible hour-long recording windows per participant; median (Mdn) = 62.6%], 5,001 activity readings (M = 50.6%; Mdn = 63.2%), 4,168 proximity readings (M = 41.1%; Mdn = 47.6%), and 3,482 GPS readings (M = 35.4%; Mdn = 39.2%). Feasibility challenges were phone battery charging, data usage exceeding prepaid limits, and burden of carrying mobile phones. Acceptability challenges were privacy concerns and lack of family involvement. Overall, families’ understanding of passive sensing and families’ awareness of potential benefits to mothers and infants were the major modifiable factors increasing acceptability and reducing gaps in data collection. Conclusion: Per sensor type, approximately half of the hour-long collection windows had at least one reading. Feasibility challenges for passive sensing on mobile devices can be addressed by providing alternative phone charging options, reverse billing for the app, and replacing mobile phones with smartwatches. Enhancing acceptability will require greater family involvement and improved communication regarding benefits of passive sensing for psychological interventions and other health services. Registration International Registered Report Identifier (IRRID): DERR1-10.2196/14734 The study protocol is outlined in detail elsewhere [29]: International Registered Report Identifier (IRRID): DERR1-https://doi.org/10.2196/14734. The procedures and results described here refer to Component 2 of the original study protocol. For details on the study according to the EHEALTH extension to CONSORT guidelines [30] see attached Additional file 1: File 1 and RE-AIM framework [31] in Additional file 2: File 2. Recruitment and data collection occurred between November 2018 through April 2019. In brief, young mothers (15–25 years of age) with infants (< 12 months of age) were recruited from vaccination clinics in rural Nepal. Both depressed and non-depressed young mothers were recruited. The mothers then participated in 2 weeks of passive sensing data collection capturing her physical activity, geographic movement, the auditory environment, and mother’s proximity to her infant. Technologies piloted for passive sensing included Android smartphones, smartwatches, and Bluetooth Low Energy beacons. The study was conducted in a setting that exemplifies limited health resources. This site was Chitwan district, a southern region of Nepal. The total population of Chitwan is 579,984. The under 5 mortality rates for Chitwan is 38.6 per 1,000. The literacy rate is 78.9%, with considerable gender disparities because fewer girls are sent to and complete schooling [32]. Chitwan district was selected because of a longstanding established partnership with the local health system and a district-wide scaling-up of community-based mental health services that was being conducted [33]. Study participants were young mothers (15–25 years old) with infants (< 12 months), including both mothers with and without postpartum depression. Recruitment of mothers was conducted at infant immunization camps held at seven health facilities in rural areas of Chitwan. Camps were typically attended by 136 mothers on average every month. Inclusion criteria were mothers between 15 and 25 years of age with an infant aged between 1 and 12 months living in the study area, and willing to be screened for postnatal depression. There were no inclusion requirements for computer/internet literacy on the part of mothers because of the passive nature of the mobile sensing data collection. The intention was to assess the feasibility among representative mothers in the community, which includes women with limited technology literacy. To determine eligibility, trained research assistants approached mothers at immunization to ask their age and infant’s age, after which they conducted the consent procedures. For mothers 15–17 years old, assent was obtained, and a guardian provided consent. Because of passive sensing data collection that captured information about the household, a meeting was held with the mother’s household representative to describe the study and data collection procedures. If mothers and their family members agreed to the study, passive sensing data were collected for 2 weeks (14 consecutive days; details on passive sensing described below). The age range of 15–25 years was selected because this is based on the United Nations definition of youth which includes 15–24 years of age [34]. Similarly, the lower age limit of 15 years old was used because data are routinely collected on pregnancy for mothers 15 years of age and above, such as is in the Nepal Demographic Health Survey (DHS). The women’s module of UNICEF’s Multiple Indicator Cluster Surveys (MICS) also collects data beginning at 15 years of age [35]. We included 25 years of age as well because of prior research on suicide deaths in Nepal which showed the greatest burden of suicide mortality among women was ≤ 25 years of age [36]. A sample size calculation was not conducted because this is a pilot study and the recruitment was done based on feasibility of using these devices by the participants [37, 38]. Our goal was to recruit 25 depressed and 25 non-depressed mothers. To allow for potential dropouts, we allowed for recruitment of up to 27 consenting mothers in each category. We enrolled more mothers than our target because we anticipated dropouts due to novelty of the study and potential reluctance from the participants in using the technology. This sample was based on feasibility of the number of mothers in the youth age range attending local clinics during the study period. Mothers depression status was determined with the Patient Health Questionnaire (PHQ-9), which is 9-item self-report tool for depression screening widely used in both high-income countries and LMIC [39]. The PHQ-9 has been validated for use in Nepal [21]. We categorized the mothers as “depressed” and “non-depressed” based on their PHQ-9 total score. For the purposes of this study, mothers scoring below 9 were classified as ‘non-depressed,’ and those with a score of 9 or above as ‘depressed’. Among Nepali adults presenting to outpatient services, a cut-off of 9 has a sensitivity of 94% and specificity of 69%, positive predictive value (PPV) of 0.33 and negative predictive value (NPV) of 0.99. For non-depressed, we used a cut-off of less than or equal to 7, which has a sensitivity of 100%, specificity of 55%, PPV of 0.26, and NPV of 1.00; this means that an adult presenting to primary care centers with a score of 7 or below has an extremely low probability of having depression (i.e., low probability of being a ‘false negative’). Of note, specific psychometric values for Nepali postpartum mothers aged 15–25 are not available. Additional details on study measures are available in the published protocol [29]. Passive sensing was collected through two devices: mothers were given a low-cost Android Samsung J2 Ace smartphone and a Bluetooth Low Energy beacon to be attached to her infant’s clothing. The devices used in this study were selected following extensive ethnographic inquiry regarding acceptability and feasibility in the study site [26]. The two devices selected for this study (smartphone and Bluetooth beacon) were considered culturally acceptable and feasible based on that formative work. The Samsung J2 Ace smartphone is a mobile phone (US $160) that is popular in the study setting. We selected the Samsung J2 Ace phone because it is widely available for purchase within Nepal and it was the cheapest option that could effectively run all the features and apps required for the study. Common, low-end mobile phones in Nepal cost US $70-$120, and therefore the device selected in the study was slightly more expensive than commonly-used devices. In the study area, most individuals owned mobile phones or have family members who own mobile phones. Hence, there is limited risk of stigmatization because of phone use in the study. For a subset of four mothers, we also piloted the use of smartwatches in place of the smartphone. Models included the Zeblaze Thor 4 Android smartwatch and Lemfo Lem8 Android smartwatch. The cost of the smartwatches was approximately $200. The smartwatches are not yet commonly used compared to the mobile phones but they have the potential of providing the same data in a more convenient device. We gave it only to the subset of the mothers as an exploratory component of the study that was added after the original design and implementation commenced. The Bluetooth Low Energy beacon was the RadBeacon dot ($10–15) developed by Radius Networks [40]. We used a closed cloth pouch (Nepali: thaili) to hold the beacon around the baby’s waist and prevent the infant from being able to remove or play with the device. Because the beacon was sewn into the pouch around the waist, the baby could not access the beacons, or get hold of the device. The research assistants informed the mothers about the safe use of beacons, including caution while using the device, so the device would not cause physical discomfort to the baby. The use of the RadBeacon Dot was approved by the Nepal Health Research Council for the purpose of this study. The RadBeacon Dot also has United States Federal Communications Commission (FCC) certification. FCC is a body that oversees the permissible exposure level for all devices with radio frequency. In the United States, the Food and Drug Administration (FDA) relies on FCC for inputs on medical devices [41]. According to FDA, devices such as activity trackers are general wellness devices because these devices have “(1) an intended use that relates to maintaining or encouraging a general state of health or a healthy activity or (2) an intended use that relates the role of healthy lifestyle with helping to reduce the risk or impact of certain chronic diseases or conditions and where it is well understood and accepted that healthy lifestyle choices may play an important role in health outcomes for the disease or condition,” [42]. Additionally, in our study, RadBeacon is not a medical device used for treatment or for the transmission of health information (e.g., temperature, pulse, respiration) from the infant. It is only used to track proximity between mother and infant during daytime hours. Regarding the safety of exposure for infants, the FCC limit for radiation from devices is 1600 mW/kg [43], which equates to approximately 800 mW for a 5 kg infant. The RadBeacon Dot specifications are + 4 to − 20 dBm, which equates to 2.5 to 0.1 mW. These ranges are comparable to an infant in a house with a standard wireless network and Bluetooth devices. The mothers were provided with the phone and beacon for the duration of the study. They returned the devices after completing data collection. The smart devices collected 4 types of data—proximity, episodic audio, physical activity, and geographic location. To collect these data, we installed our custom-built Electronic Behavior Monitoring app (EBM version 2.0). The EBM app passively collected data for 30 s every 15 min between 4:00AM and 9:59PM (i.e., 18-h intervals of data collection per day). The EBM app starts automatically whenever the device is turned on. A folder, NAMASTE, was created automatically once the EBM app was downloaded on the smartphone (see EBM screenshots in Fig. 1). All data were stored in the folder. Because all sensing and data collection was done passively, mothers did not need to interact with the app in any way to enable data collection. Details of each passive sensing domain are provided below: Screenshots of Electronic Behavior Monitoring app (EBM version 2.0). a EBM package installer; b EBM permission controller; c EBM privacy timer; d EBM privacy timer running; e EBM settings; f EBM settings sources During the consent process with mothers and families, the research assistants explained the various components of the passive data collection, how the information would be used, and demonstrated using the technology. The research assistants explained that this was an initial study to learn about collecting this information and that study would inform how this information could be used in the future to improve health interventions for mothers and infants for mental and physical health. The research assistant demonstrated the proximity beacon information to record when mothers and infants were together. The research assistants explained that the study was not advocating that mothers and infants should or should not be together but rather the goal was to learn about these patterns in their community. The GPS was demonstrated to show how the study would learn about where the mother went in her daily activities. The research assistants explained that the study was not interested in specific locations but rather how much a mother traveled around outside of the house. Again, the research assistants explained that travel outside the house was not considered good or bad for mother’s and infant’s health but rather it was something that the team wanted to learn about for different mothers. Similarly, the activity data was explained as capturing when mothers were resting or physically active. This also was framed as not recommending that mothers rest or be active but rather to learn about what mother’s patterns were. Finally, the audio recording was demonstrated. The research assistants clearly stated that the team was not interested in what mothers and family members said. Instead, the focus was to learn how much of the time there was speaking around the mother, as well as other sounds such as cleaning and cooking activities, vehicle sounds, animal sounds, and other things the recordings would capture. The research assistants explained that the recordings would be used to teach computer programs to better distinguish among these sounds so that future health interventions could use this type of information to improve care for women, infants, and families. All mothers and families were given an opportunity to ask questions about the mobile data collection and data usage. In addition, research assistants followed up with mothers and families during subsequent home visits as additional questions arouse. Anticipating the privacy concerns, we had a “privacy timer” in the EBM app that allowed mothers to pause the app for as long as they wanted. Once the privacy timer was on, the EBM app stopped collecting data until the timer was closed. In addition, strategies to maintain confidentiality, such as deleting audio files was piloted with mothers using similar devices in South Africa [53]. On the devices, mothers can delete audio files at any time. They can also ask the research assistants to not transfer the data if they do not wish to share it with the team. Mothers were also instructed to turn off their phone anytime they chose in order to stop data collection. Because of our prior work piloting passive sensing technologies and evaluating mothers’ ability to delete data [26, 53], we did not conduct further formal usability testing with mothers for this activity within the current study. A female research assistant briefed each mother and her family on the technical use of the phone and beacon. All research assistants self-identified as employees of the non-governmental organization Transcultural Psychosocial Organization (TPO) Nepal. The research assistant visited the mother’s home on average 3–5 times over the two-week period, which included study briefing and collecting consent, day one of data collection for technology delivery and training, day three of data collection for technology troubleshooting, and then weekly, with intermittent phone calls to additionally troubleshoot and provide any needed technology support. Mothers were instructed to keep their mobile phones with them as much as possible and attach the beacon to their infant’s clothing throughout the day. Mothers were asked to turn the mobile phones off and remove the beacon from the child during the night. Mothers’ identifiable information (name, phone numbers) were stored in a secure server. The app was password protected, with counselors needing a passcode to access the app. All the passive sensing data and qualitative data were stored under a unique participant code (without identifying information) and stored in a secure cloud-based server. Mothers and family members had the opportunity to become comfortable with the researcher assistants because of their repeated visits to participants’ homes. We had previously produced a video to explain these data collection processes to potential study participants [26]. No prompts or reminders were provided to mothers electronically once the EBM was installed. Research assistants visiting the home would check the functioning of EBM app and detection of the Bluetooth beacon, then they would conduct troubleshooting as needed. Once the mother was comfortable sharing the data, research assistants copied the data from the phones in a portable device. The data did not contain any identifiable information, and contained passive sensing data in .csv and .m4a format. The data folder was coded with a de-identified ID to anonymize the data. It was then uploaded to a secure cloud server, through a secured connection and removed from the local devices within 24 h. Two types of app errors were collected: app exceptions and user engagement issues. App exceptions and failures can have multiple causes. Commonly they include errors in code logic which can be introduced, for example, when code runs on different versions of Android, or when hardware interfaces are implemented in a non-standard way by device manufacturers. User engagement issues include trouble remembering a password, trying to perform an unsupported function, and/or struggling to find a function. In this study we tracked fatal app exceptions and user-login challenges. To assess feasibility and acceptability of passive data collection, we triangulated several sources of data including in-depth interviews (IDIs) performed at the end of 14 days of passive sensing data collection, field notes recorded by research assistants from each participant encounter, and memos documenting the significant events (e.g., drop outs, service outages). Female research assistants conducted IDIs using a semi-structured interview guide lasting between 20 and 45 min. Questions elicited maternal experiences and perceptions of the technology and EBM application, covering feasibility, social acceptability, confidentiality, utility, and recommendations for improvement. Our inquiry focused on confidentiality and social acceptability given important ethical considerations of passive data collection. Questions regarding these domains were elicited both from the mother as well as from her family throughout the study period. Importantly, the research assistant had established meaningful rapport with both the mother and her family (on average visiting the mother’s home 3–5 times), permitting more comfort and allowing detailed and frequent field notes to capture examples and texture not captured by the IDI, as well as notes related to confidentiality concerns from either the mother or her family members. Qualitative interviews were audio-taped, transcribed, and translated before coding and analysis. The interviews were first transcribed verbatim in Nepali and then translated to English by a bilingual translator. We followed a standardized Nepali mental health glossary for translation of emotional and psychological terms into English [54]. The Consolidate Criteria for Reporting Qualitative Studies (COREQ) checklist is included in Additional file 3: File 3 [55].