Recruitment and Sample
We screened 782 mothers, of whom 320 were eligible for inclusion based on age criteria of the mothers and infants. We consented the eligible mothers and screened them for depression. Approximately 92% of the eligible mothers scored below the depression cut-off (n=294) and 8% scored above the cut-off (n=26). We serially enrolled non-depressed mothers. Approximately two-thirds of the non-depressed mothers and their families provided consent and were enrolled, with a final sample of 27 non-depressed mothers consenting to participate in the passive data collection. All mothers subsequently screened who scored below the depression cut-off were not eligible for the study. Of the 26 mothers who screened positive for depression, 11 of the mothers and their families provided consent and enrolled in the study. Table 1 contains demographic information of the participants. Case memos revealed reasons for non-participation of both non-depressed and depressed mothers, and memos were also used to document reasons for withdrawing from the study. Reasons for not participating included mothers moving outside of the study area, inability of the research team to contact mothers following initial screening in the health facilities, families not-consenting to participate in the study, and mothers too busy to participate due to family obligations. Two participants withdrew from the study before completing the full 2 weeks of passive sensing data collection (additional details on why they withdrew are provided below).
Passive Sensing Data collection
We collected activity, audio, GPS and infant proximity data through passive sensor readings (every 15 minutes from 4am to 9pm, daily). Table 2 provides the total passive data collected between 4 AM - 9 PM among depressed, non-depressed, and the total sample. Audio and activity data were captured more often than GPS and proximity data on average. For the 38 young mothers, the cumulative maximum number of passive sensor readings (every 15 minutes from 4am to 9pm, daily) was 9,605, see Table 2. This total of 9,605 was obtained because data collection was for two weeks (14 days) for 18 hours a day for 38 moms (38*18*14=9,576). Another 29 readings resulted from some participants with atypical start times of their passive sensing collection or delayed cessation of their collection based on when research assistants could meet them.
Per participants, there were, on average, 57.4% of audio, 50.6% of activity, 41.1% of proximity, and 35.4% of GPS data readings. Figure 2 shows the average passive data readings by the time of the day. Data collection was lower than the total possible readings in the early morning across all sensors and tapered off at night, but was generally consistent from 10 AM to 6 PM. We explored possible explanations for differences in successful data collection by time of day and sensor type along with description of qualitative results to illuminate these differences.
Four issues emerged during the implementation phase of this pilot that were fixed and rolled out in subsequent releases of the app. The issues were:
- The proximity chart was not displaying correctly on the J2Ace phone Samsung A7 tablet and the graphing software had to be updated.
- Users requested that the proximity chart display order be reversed.
- Added a daily routine award card
- Fixed an issue with the Nepali Date Calendar which was not displaying the correct Gregorian date/time when converted.
User engagement was tracked and 30% of app usage time was spent in the participant details page viewing data, awards, and posts. Another 20% of time was spent reading the posts (after clicking on the summary in the participant details page). Scrolling, filtering, and searching made up the other 50% of time spent in the app. The 4-digit pin password used was entered correctly on the first try in 94% of login attempts.
We categorize these findings below based on the 6 qualitative domains for feasibility and acceptability (see Table 3).
Mobile phone battery charge, data usage, and positive and negative family involvement were the main technical feasibility issues identified that limited data collection throughout the day. Passive sensing requires the phone to be turned on. Morning data (and some evening data) were most likely to be missing because mothers typically were instructed to turn their phones off at night before bed and then turn it back on when awakening in the morning. Subsequently, there were gaps in data collection in the evening and early morning hours.
“I used to switch off this watch at night and connect to the charger and switch on in the morning. … In the daytime I used to switch off this watch again and connect it to the charger.”
- (19-year old depressed mother)
Common feasibility challenges likely reduced data capture across all sensor collection. These feasibility challenges included lack of electricity, mothers forgetting to charge the phone, devices not retaining charge (especially smart watches), technical difficulties such as phones not connecting to the charger properly, or environmental factors such as protecting the device from rain. In general, there were more technical issues reported by mothers who used smartwatches with frequent battery draining. We tested the use of smartwatches instead of mobile phones with three participants but 0% of the activity data was collected due to the smartwatch not supporting activity data collection (see Text Box 1). The use of smartwatches with no activity data contributed to reduce the average across all participants to 60% at best, with the greatest data capture during midday.
Text Box 1: Limitations of passive sensing data collection with a smartwatch
For a 21-year old non-depressed mother with a 4-month old infant, it was her first experience of using a smartwatch. She did not have a problem attaching the beacon on the child’s clothing. However, she felt uncomfortable wearing the watch on her wrist. This was largely due to her workload at home -- she washed clothes and did dishes multiple times a day, which necessitated her to get her hands and arms wet. She was concerned about possible water damage to the watch. She also had trouble keeping the watch charged because it lost battery quickly. Her husband helped her complete the study, providing reminders and assistance charging the watch. She eventually completed the study with her husband’s support who helped her charge the smartwatch. We collected 62.5% and 51.9% audio and proximity data but were unable to collect GPS or activity data due to limited functionality of the smartwatch.
Of the total possible readings, about 47% of the GPS data was collected during the high period of data capture from 7AM - 7 PM, see online Supplemental File 4). A likely reason for less successful data capture compared to audio was that GPS data collection was not triggered the same way as audio, proximity, or activity data were. The GPS data collection required WiFi or 3G signal for collection. Therefore, data collection was low because mothers were mostly indoors which limited this GPS connection. Initially, we used a combined method that included both mobile data and direct GPS connection to collect GPS location of the mothers. However, mothers and their families quickly used available data, typically by watching YouTube videos, which then led to not enough prepaid data for GPS collection. We switched to direct connection to the satellite to address the loss of data due to low mobile data by turning on high accuracy in the GPS settings.
Domain 2: Interference
We explored daily interference related to mobile phone and Bluetooth beacon use. Consistent daily mobile phone use was acceptable to most mothers. However, carrying two mobile phones (one for the study and one personal phone) was considered interfering with daily activities. Additionally, mothers shared their concern about the probability of losing mobile phones if they had to carry two phones at all times. Mothers working in a shop or office outside the home and mothers who were farming found it even more challenging to carry two mobile phones. Additionally, it was difficult for the study team to contact mothers who worked during the day if she had technical issues with the devices, which caused further disruption in data collection [Text Box 2].
Text Box 2: Working mothers and passive data collection
A 23-year old non-depressed mother with a 10-month old infant was a small business owner of a shop in which she made toy dolls and trained others in this trade. She received assistance from her family to use the smartwatch. Her husband helped her charge the watch when the battery was low, reminded her to wear the watch, and put the beacon on the child (in the secured pouch). As she was busy with her business’s work, it was difficult for her to have enough free time for regular study team visits. So, she suggested recruiting housewives instead of employed mothers so that they could dedicate their time to the study. She completed the study, and we were able to collect 62.7% of audio and 54.1% of proximity data. (No GPS or activity data were recorded due to technical issues in the smartwatch.)
The participants who did not find mobile phones interfering typically wore clothes with pockets or used the mobile phones for non-study related activities such as watching videos or listening to music. Bluetooth beacons presented more challenges with daily interference. This may have been due to beacon novelty to the mothers so they felt more concerned initially about the device than about mobile phones. One of the mother’s major concerns was physical discomfort to the baby. Mothers generally put the beacons on the baby and took it off at night, or during oil massages and baths. Some mothers found the routine of putting beacons on the baby tedious after the first few days with some mothers delaying or forgetting to put the beacon on the baby after bathing or oil massages.
The proximity data was consistently lower than both audio and activity data with about 60% data collection at midday, even though it was triggered at the same time as audio and activity. One possible reason could be the participants turned off the Bluetooth signal on their mobile phones.
Domain 3: Confidentiality
We collected about 75 % of audio data during the daytime hours of 7 AM to 7 PM (median value for all participants, see Supplemental File 4). Audio recordings were the data type most concerning to the participants from a confidentiality perspective. At least 11 mothers in our study expressed concerns of being audio recorded. One of the reasons was fear that study staff would listen to the audio clips and share private information with other people in the community, or that the community would know about the family disputes. For instance, two participants reported:
"If I talk about difficulty, mostly I am worried that other people will listen to all our discussions. They will know all our family problems. And they will talk about our problems everywhere."
- (24-year old depressed mother)
Interviewer [I]: Do you like this beacon and this mobile? Do you like to use this mobile and fix this beacon on your baby?
Participant [P]: I like using the devices but I feel worried because you might know all our family matters and our conversations.
I: We don’t listen to those recordings.
P: You will not share those recordings with other people?
- (21-year old depressed mother)
Some participants reported changing their behavior such as spending more time with the baby, talking lovingly or softly around the baby, shutting the mobile off during family arguments, or asking family members to not use bad words. Each of these behaviors may have reduced or biased the audio capture. Mothers with family members that abused alcohol and mothers in conflict with their mothers-in-law usually expressed the greatest privacy concerns. Family members of such households also asked mothers to switch their mobile phones off when they were under the influence of alcohol, or delete audio recordings that had their voices (See Textbox 3).
“We have different conditions in our home. Sometimes people quarrel and we have arguments in our house. This mobile might record all those things so I have to switch off this mobile.”
- (17-year old non-depressed mother)
P: [Smiling] This mobile records sound. And all the recordings were stored in its memory so my husband told me to delete his recordings.
I: When did you switch off this mobile at night?
P: Sometimes at 6 or sometimes at 7 pm.
I: And what about your sisters and your own mother?
P: Sometimes when I start talking about my mother-in-law, my mother tells me not to talk about her, this mobile might record the voice and save it.
- (16-year old non-depressed mother)
Text Box 3: Deletion of data and other reasons for low data capture
The family of a 21-year old depressed mother lived in a temporary squatter settlement without electricity g near the jungle. They agreed to participate, but due to lack of electricity they charged their mobile phones at a neighbor’s house. The participant was concerned that the device may get stolen and worried about needing to cover its expenses. The study team provided her with a power bank to charge the mobile and assured her and her family not to worry if something happened to the device, there would be no financial consequence. The provision of a power bank helped to keep the mobile running for a longer period of time.
We collected 73.9% of activity, 41.6% of audio, 10.5% of GPS, and 29.0% of proximity data from the mother. The low GPS data collection was a result of excessive data usage. During data collection, we relied only on mobile data to collect GPS. This and other similar situations where mothers ran out of prepaid data prompted us to change the connectivity so that the phones had direct connection to the satellite. There was lower audio and proximity data collection in comparison to her activity data. In our qualitative interviews, the mother shared that her husband and mother-in-law listened to the audio files and deleted the ones that had their voices. Another reason for her relatively low data capture was that the proximity data collection was interrupted when the Bluetooth was turned off on the phone, which was another reason for the low data collection. The participant’s husband used to turn off the Bluetooth and keep the mobile for his own entertainment purposes due to which proximity data was interrupted.
Domain 4: Safety concerns
Three types of safety concerns were highlighted by participants: (1) safety of the infant when the Bluetooth beacon was attached to his/her clothing, (2) mother’s safety when using mobile phones, and (3) physical safety of the devices. Among the three, child safety was the most concerning to the mothers. One major child safety issue was physical discomfort that the beacon could cause to the infant, such as device poking the baby during sleep.
“I think that this beacon might poke my baby and make it difficult for him to sleep.”
- (21-year old non-depressed mother)
To avoid physical discomfort, mothers were instructed to remove the beacons when the baby was sleeping or mothers moved the device over multiple layers of clothes to avoid poking. The feeling of potential discomfort to the child hindered the consistent use of beacons and may have been one factor for reduced proximity data capture.
“While using these technologies, I thought this beacon might poke my baby and sometimes I removed the beacon.”
- (17-year old non-depressed mother)
Despite child safety concerns, there were two major facilitators that propelled mothers to continue using the device - trust in the study staff and no adverse effect to the baby after the first few days of use. Because the study staff reached out to the mothers in health posts where mothers went for regular checkups and immunization, they trusted the health workers in the health facility. The study staff coordinated with the health workers, and were therefore looked at by the mothers as trustworthy. Second, despite initial concern, when the baby did not get sick or have adverse effects in the first few days of use, the mothers were convinced that the beacons were safe:
I: Did you think that it might affect your baby or your baby might feel difficulty due to this beacon?
P: In the beginning I had those types of [negative] thoughts but after using [the devices] regularly for many days I didn’t have that type of thought anymore.
I: What types of thoughts did you have in the beginning?
P: That my baby might get sick, or it might have some health effects.
- (23-year old non-depressed mother)
Mother’s safety was less of a concern in comparison to child’s safety. In general, all the mothers thought the mobile phones and beacons did not have an adverse effect on the mothers. A factor facilitating the use of smartphones was mother’s prior experience with mobile phones. Because mothers were familiar with smartphones, they did not think it would affect their health.
The final safety concern we explored was potential theft or breakage of the devices. Mothers, especially the ones from poor economic backgrounds, were scared that the study devices could get stolen or broken. Despite assurance from the study team that they were not liable in case of theft or accident, mothers were still anxious, especially for the first few days. The study team provided support and reassurance to the mothers during subsequent home visits to assuage remaining anxiety related to device safety.
Domain 5: Perceived utility
In general, mothers and families did agree to and continue using the phones throughout the period because of perceived benefits. Some of the perceived benefits aligned with the study goals, while some benefits were non-study related. Among the study-aligned perceived utility, mothers mentioned using the beacon and mobile phone to know the distance between them and their babies throughout the days. They could see the actual distance in smart devices. They also knew the mobile recorded their sounds, movement, and activities. Some mothers went back and listened to their audio clips. For perceived utility not related to the study purpose, mothers reported that they used the phone for listening to music, taking pictures and videos, using Facebook and watching YouTube videos. Perceived utility, however, varied across participants. For example, when asked about the utility of the beacons, some mothers said that the study showed how much they loved their babies. Other participants speculated that the data could help understand growth and brain development of the baby, or help in conflict resolution at home.
I: - I gave you this watch and fixed this beacon on your baby for two weeks. How was your experience these two weeks? What was your experience while using this watch and beacon? Please share something about that.
P: - I think I got a chance to provide more care to my baby. I got a chance to learn many things from the technologies that you provided me. This watch helps to find out whether we are speaking the truth or not. This watch records our voices continuously for a long time. One thought is continuously stuck in my heart-mind: through these recordings we can find out if someone is hiding something.
- (21-year old non-depressed mother)
Some mothers thought the technology could record the time spent with the baby or identify their mood changes during the day.
I: - Do you know anything about why you are using this beacon and mobile? Though I had already told you about its use, what do you think about this technology?
P: - These technologies are for observing the changes in a mother like being irritated, distance between mother and the baby and problems in family relations. In the future, our daughter-in-law’s granddaughters would benefit from this technology. That’s why I agreed to use this technology.
- (24-year old depressed mother)
Domain 6: Communication
Communication facilitated passive data collection through three subdomains - study team engagement, tech literacy, and autonomy of using devices. The importance of the study team engagement was critical when explaining the technology and addressing any queries that mothers had during the study duration. Mothers enjoyed their interactions with the study staff, especially when the study staff asked them about their children and family. They enjoyed study staff visiting them every few days to discuss any new queries and talk to the family members about the technology. We also provided the mothers with a study brief handout in Nepali, as a support tool, so that other family members and neighbors could read and understand about the study themselves during and after the consent process. The study tool supported mothers in answering family’s or neighbors’ questions about the study when the study team was not physically present to answer those questions.
One of the major social influences for acceptability was collaboration with local health posts for screening and recruitment at the community level. The recommendation from the health workers helped the study team to establish rapport with the participant and then follow up through the home visits. With the study team’s consistent technical support and clear communication of the study findings, the mothers felt more involved in the data collection process. They also felt more empowered to censor data collection by turning on/off the mobile device or beacon if needed. For example, 11 mothers described that they switched the phones off or left the phone in another room during family discussions, particularly to avoid recording any disputes or bad language.
Although 32% of study mothers were new to smartphone technology, all mothers confidently described their ability to navigate and operate the varying features by the second week. Mothers made decisions when or whether to attach the beacons on the child’s clothing as well. To support autonomy of using the devices, we found family engagement and consent to be important facilitators in both the research process and successful implementation of passive data sensing (See Text Box 4). Mothers felt more confident and comfortable when the study team explained the technology and study objectives to their families, especially to the family members in decision making roles such as husbands and mothers-in-law. As per our protocol, the study team visited the mother’s family after the initial screening at the vaccination clinics. Family consent helped the family understand the technology better and ask questions to the study team. Mothers generally said they were able to answer questions on the study objectives independently, but the family consent helped them get support from family members when the mothers had to explain the technology to non-family members.
Text Box 4: Other reasons for low data collection--religious concerns
No participants refused to participate because of religious beliefs with the exception of one family that was concerned that the technology was used for Christian religious conversion.
A 22-year old non-depressed mother withdrew from the study after a few days. Through follow up qualitative work, we later learned that the main reason the mother’s family asked her to withdraw was that they suspected that the technological devices were being used to convert them to Christianity (due to a legacy of coercive organizations in the study region). We also learned that the participant wanted to continue the study but was forced to drop out by her husband and father-in-law.
We collected 14.5% of activity, 65.5% of audio, 6.3% of GPS, and 7.5% of proximity data from the mother. This mother was also one of the earliest participants we gave the devices to, as we were still making changes to the technology for appropriate data collection. The lower GPS data could be due to the mobile phone running out of data. Other data collection could have been affected by social factors. For example, the mother’s family later told the study team that they were reluctant to use the devices, including the beacon on the child. The lower proximity data collection could mean that the Bluetooth on the mobile phone was switched off most of the time. The higher audio data indicates that the mobile phone was still switched on most of the time, although functionality such as Bluetooth was likely disabled or turned off.