Procedures to select digital sensing technologies for passive data collection with children and their caregivers: Qualitative cultural assessment in South Africa and Nepal

Background: Populations in low-resource settings with high childhood morbidity and mortality increasingly are being selected as beneficiaries for interventions using passive sensing data collection through digital technologies. However, these populations often have limited familiarity with the processes and implications of passive data collection. Therefore, methods are needed to identify cultural norms and family preferences influencing the uptake of new technologies. Objective: Before introducing a new device or a passive data collection approach, it is important to determine what will be culturally acceptable and feasible. The objective of this study was to develop a systematic approach to determine acceptability and perceived utility of potential passive data collection technologies to inform selection and piloting of a device. To achieve this, we developed the Qualitative Cultural Assessment of Passive Data collection Technology (QualCAPDT). This approach is built upon structured elicitation tasks used in cultural anthropology. Methods: We piloted QualCAPDT using focus group discussions (FGDs), video demonstrations of simulated technology use, attribute rating with anchoring vignettes, and card ranking procedures. The procedure was used to select passive sensing technologies to evaluate child development and caregiver mental health in KwaZulu-Natal, South Africa, and Kathmandu, Nepal. Videos were produced in South Africa and Nepal to demonstrate the technologies and their potential local application. Structured elicitation tasks were administered in FGDs after showing the videos. Using QualCAPDT, we evaluated the following 5 technologies: home-based video recording, mobile device capture of audio, a wearable time-lapse camera attached to the child, proximity detection through a wearable passive Bluetooth beacon attached to the child, and an indoor environmental sensor measuring air quality. Results: In South Africa, 38 community health workers, health organization leaders, and caregivers participated in interviews and FGDs with structured elicitation tasks. We refined the procedure after South Africa to make the process more accessible for low-literacy populations in Nepal. In addition, the refined procedure reduced misconceptions about the tools being evaluated. In Nepal, 69 community health workers and caregivers participated in a refined QualCAPDT. In both countries, the child’s wearable time-lapse camera achieved many of the target attributes. Participants in Nepal also highly ranked a home-based environmental sensor and a proximity beacon worn by the child. Conclusions: The QualCAPDT procedure can be used to identify community norms and preferences to facilitate the selection of potential passive data collection strategies and devices. QualCAPDT is an important first step before selecting devices and piloting passive data collection in a community. It is especially important for work with caregivers and young children for whom cultural beliefs and shared family environments strongly determine behavior and potential uptake of new technology. Research was conducted in South Africa and Nepal. These countries are ideal sites to develop a procedure that can be used to select devices and data collection approaches that will be acceptable and feasible in settings with low-literacy populations unfamiliar with passive sensing technology. The countries have high rates of poverty and childhood morbidity and mortality with limited access to specialized child health and mental health services. Moreover, the countries are exemplified by large health disparities and poor outcomes with traditional health delivery approaches, thus necessitating greater use of technology. Moreover, there are preliminary successes for mHealth in these settings. In South Africa, mHealth is increasingly used for HIV/AIDS prevention and care, including home-based testing and counseling to promote treatment engagement and adherence [30,31]. Text messaging through mobile phones has been used to collect feedback on maternal-child health services throughout South Africa and resolve areas of poor-quality care [32]. mHealth in Nepal has worked in different capacities such as improving communication and coordination between health workers in rural areas and district hospitals, strengthening community-based surveillance systems, and improving maternal and neonatal health outcomes [33,34]. Furthermore, mHealth has been successful in achieving targets for reduced maternal mortality through increased health facility attendance and institutional delivery in Nepal [35]. However, in both countries, passive sensing data collection has received limited attention and would benefit from qualitative exploration before selecting and piloting new approaches. Furthermore, methods used to determine cultural acceptability and feasibility before selecting devices for pilot could be of great benefit throughout LMICs, other low-resource settings, and a context with diverse cultural groups. In South Africa, the study was conducted in the Sweetwaters region of the Greater Edendale Area of Pietermaritzburg, KwaZulu-Natal. This location has been the site of ongoing research on public health initiatives led by the Human Sciences Research Council (HSRC). The Sweetwaters area is emblematic of rural regions in South Africa that have suffered high rates of maternal HIV and mother-to-child transmission. However, through public health programs, these rates have dramatically reduced over the past decade. In Nepal, the study was conducted in Sankhu, Manamaiju, and Phutung, which are all located approximately 30 to 60 min away from Kathmandu. These areas were heavily affected by the 2015 Nepal earthquake. Sankhu was devastated with more than 100 persons killed on the day of the first earthquake. Sankhu, although having access to Kathmandu, lacks child health specialty services, and there are no local mental health services. Manamaiju was comparatively closer to the city but did not have mental health services in their local health posts. Research in these sites was conducted in 2016 to 2017. We developed the QualCAPDT procedure by adapting methods commonly used in cultural anthropology [36,37]. We had 2 objectives in the development of QualCAPDT. The first was to have a systematic process that could be replicated in other settings to evaluate the acceptability and perceived utility of different passive data collection strategies before development or adaptation of the technology. The goal was to avoid selection of data collection platforms before receiving input from end-user communities, which would lead to potential waste of resources for technologies that would not be adopted. The second was to gain information about how participants viewed, among other issues, the ethical risks they would be exposed to by participating in a study that collected passive digital data about their behavior in their homes. A crucial outcome was to protect people’s ethical rights in research because passive data collection of daily family life is an invasive process. Our hope was that QualCAPDT would provide insight into privacy, confidentiality, and the major ethical responsibilities researchers have in this new technological age of health interventions. To frame our 2 objectives, we used the following domains, adapted from Buenaflor and Kim’s 6 human factors [22], to guide all participant discussions about the devices and data under review: Narrative focus group discussions (FGDs) are a technique commonly used in cultural anthropology and public health [38-40]. In narrative FGDs, participants are typically read a story or scenario at the beginning of the session, then they comment according to probes provided by the facilitator. Given the lack of familiarity with passive data collection devices among community health workers and caregivers in rural South Africa and Nepal, we felt that narratives about the devices would have been insufficient for participants to understand the technology and types of data capture. Therefore, videos ranging in length from 3 to 6 min were produced, demonstrating the technologies in the local settings. In South Africa, the production of the videos was conducted by a group of research assistants over a period of a month. In Nepal, the video production task was assigned to a team of independent contractors, and they took around 1 month to produce the final cuts. Five videos, each demonstrating a technology, were produced in both countries. Separate videos were produced for South Africa and Nepal so that participants could relate to the experiences of persons in the video. Videos were in the local languages of the participants, isiZulu with English subtitles in South Africa and Nepali in Nepal. The video content was similar between the 2 countries. The videos provided images of the technology, a scenario in which a researcher explains the technology to a rural family with a small child, and, when possible, an example of the output of the technology that could be used for health interventions. In Nepal, the video was prefaced with information on the role of female community health volunteers (FCHVs) as these would be the individuals managing the devices. Screenshots from the South Africa and Nepal videos are provided in Figure 1. Individual videos are indexed below as Multimedia Appendices. Screenshots from videos demonstrating passive data collection devices in South Africa and Nepal. After the videos were produced, they were integrated into the FGDs with community health workers and caregivers (see Figure 2 for timeline). FGDs were conducted in isiZulu in South Africa and Nepali in Nepal. In both countries, the FGDs for community health workers began with general questions about health needs of children and their daily routines, including any health and development concerns that were common in that community. The FGDs then transitioned to showing videos. After the videos, the groups rated attributes using anchoring vignettes (step 2b) followed by card ranking tasks (step 2c), and then, supplemental interviews were conducted with other stakeholders (step 3). Qualitative Cultural Assessment of Passive Data collection Technology (QualCAPDT) process timeline of piloting in South Africa and Nepal. In South Africa, the FGDs were organized through a community outreach team that works and lives within the community where the participants were sampled. An equivalent of US $5 was issued out to each participant as reimbursement for the time that they spent in the FGD. The FGDs were conducted within the community in a central community hall. All FDGs were conducted by qualitative interviewers using IsiZulu as a medium of communication. The videos were projected onto a wall using a projection system connected to a laptop. The 4 FGDs took a period of 9 days for completion due to the availability of both the community hall and the participants in these groups. In Nepal, the FGDs were scheduled with the help of FCHV. One of the FCHV was contacted by phone 2 days before the FGD and were asked to gather participants for FGDs. The participants were provided nonmonetary compensation, for example, household items such as soap, toothpaste, brush, and lunch, for their time and effort. For FCHV FGDs, the data were collected in a quiet office room in a health care facility. Caregiver FGDs took place in 1 of the caregiver’s homes. One of the authors (KT) moderated the discussions along with a research assistant who did note-taking. The discussions were held in the local language. We also audio recorded the FGDs after receiving consent from all the participants. The FGD guide was semistructured. The videos were shown in a laptop where participants were seated just in front so that everyone could see and hear what was happening in the video. In addition to the use of anchoring vignettes (described below), several probes were used during the FGDs to elicit responses from different participants. Some examples of follow-up questions post video including probes used in both sites are shown below: Anchoring vignettes are a structured elicitation tool commonly used in anthropological research, health behavior studies, and public opinion polls [41-43]. As individuals may have a social desirability bias toward responding in the affirmative, anchoring vignettes can be used to normalize a range of responses [44]. When initially piloting preferences for the devices in this study (unpublished), we found that few women had negative responses. Therefore, we created 2 vignettes to anchor the response options. For each criterion, there was 1 vignette that ranked the device high on the criterion and 1 vignette that ranked the device low on the criterion. For example, in Nepal, we referred to Maya and Asha, with Maya vignettes having a concern about the device and Asha being supportive about the device on that attribute. Then, participants were asked to say whether they felt that women in their community were more likely to be like Maya or Asha (see Figure 3). The anchoring vignettes were presented in isiZulu in South Africa and in Nepali in Nepal. Anchoring vignette elicitation technique to rate devices by attribute domains. (The anchoring vignettes, including all data collection materials and videos, were presented in the local language of participants: isiZulu in Kwa-Zulu Natal, South Africa, and Nepali in Kathmandu Valley, Nepal. Names and illustrations should be adapted to local cultural context). When doing the rating according to the anchoring vignettes, participants are encouraged to describe their thought process and discuss as a group why they are making certain decisions. This process is based on techniques from anthropological research on cultural domain analysis [45]. In this process, participants prompted to describe attributes that lead to categorizing in a certain way. For example, what are the attributes of a device that lead to it being categorized more closely to Maya’s perspective in one domain but then closer to Asha’s perspective in another domain. This type of prompting reduces the likelihood that a device is ranked all toward Maya on every domain or all toward on Asha on every domain. Through this prompting, the participants consider each domain independent of the others and identify the attributes that contribute to the device’s categorization in each domain. This is rich qualitative information, which often came in the form of participants debating the ways in which community members think more like Maya or Asha. After rating with anchoring vignettes, we used a card sort ranking task at the end of the FGD [36]. In this activity, each device had a unique card with a photograph of the device, and participants were asked to sort those cards according to each of the attributes. For example, the 6 devices were ranked in order from most to least confidentiality, and similarly, all devices were ranked from the most to least useful for child health promotion. This is a forced-choice approach in which participants have to make cognitive decisions to up- or down-rank certain devices. During this process, participants are encouraged to describe the thought process and decision making that influences their ranking. The cards were images of the devices and did not include written language. Discussions during the card ranking were conducted in isiZulu in South Africa and in Nepali in Nepal. Example prompts and probes during the card ranking task were as follows: In addition to the FGDs structured around the videos with anchoring vignettes and card sort ranking, we conducted additional key informant interviews to explore other themes that would reflect community norms, preferences, and perceived utility and feasibility of the devices. In some of these individual interviews, the videos were also used to elicit discussion. Example questions from the supplemental interviews were as follows: In South Africa, the supplementary interviews were conducted with local organizations that work in the early childhood development sphere. The participants were recommended by the South African principal investigator who had extensive experience working in the area or by research assistant based on collaborating organizations working in the area of child health and/or mobile technology. These participants were not reimbursed for their time because they were professionals often collaborating in research and health initiatives in the area. The participants were contacted through the community outreach team for available times when the interview could be conducted. Supplemental interviews in South Africa were conducted in English because they were often with organizational staff educated in English who were South African but may not have been native to Sweetwaters. In Nepal, for FCHV interviews, the contact information of the FCHV was obtained from local health facility. They were contacted 2 days before the day of data collection and scheduled the time. For caregivers’ interviews, we obtained the contact information from FCHV. The interviewees were provided nonmonetary compensation as with the FGDs, for example, household items such as soap, toothpaste, and brush, for their time and effort. The data were collected in a quiet room in a FCHV or caregiver’s home. One of the authors (KT) and 2 research assistants were involved in conducting interviews. We also audio recorded the entire interview after written consent from the interviewee. The interview guide was semistructured. When videos were shown, a laptop was used. The information generated from interviews differed from FGDs in that interviews helped us to identify attitudes and perceptions of individuals in depth as opposed to coming to a general consensus in FGDs. In addition, the interviews may have also differed from other standard semistructured interviews in that participants watched videos of technologies without group discussions and the use of multiple probe questions. Supplemental interviews were all conducted in Nepali. We used the QualCAPDT method to assess the suitability and acceptability of 8 approaches to collecting passive digital sensor data from 5 devices about caregivers and children in the home. They ranged from invasive approaches that produce rich data to less-invasive approaches that result in less rich data. There were 2 forms of video recording presented to participants. One was a continuous recording with a camera mounted in the living room of the home. Families were shown that data would be stored on secure digital cards that would be removed and reviewed by the researcher together with the family. The participants were also told that a time-lapse version was possible in which video would be captured every 15 min for 30 seconds. The participants were told that neither data collection platform would record sound. The South African video demonstrating video recording is available in Multimedia Appendix 1 and the video from Nepal is provided in Multimedia Appendix 2. Similarly, participants were presented with technology that could collect audio from the home environment. It was explained to participants that the technology could be set to record continuous or episodic audio, with an example of the latter being recordings made every 15 min for 30 seconds. The videos demonstrated a fixed recording device in South Africa (Multimedia Appendix 3) and a mobile recording device in Nepal (Multimedia Appendix 4). A wearable time-lapse camera was demonstrated as a technology that could be used to capture images from the child’s point of view. Videos demonstrated how children could wear the devices (see screenshots in Figure 4). The images captured on the devices in the children’s daily life were presented in the video. The South Africa wearable camera video is provided in Multimedia Appendix 5, and the video from Nepal is provided in Multimedia Appendix 6. Screenshots of videos demonstrating wearable time-lapse camera for children. Red arrows point toward the wearable device on the child. A Bluetooth beacon was displayed in a video as a way to evaluate when the target caregiver and target child are in close proximity. This was illustrated through a small coin size plastic toy that could be attached to the child and that sent out a signal that could be received by a mobile phone carried by the caregiver. A mock output was shown in the video to simulate data that could be reported on time the caregiver and child spend together. The video demonstration for South Africa is provided in Multimedia Appendix 7, and the video demonstration for Nepal is provided in Multimedia Appendix 8. Finally, a room-based environmental sensor was presented as a technology that could report on temperature, humidity, and air quality within the home. The devices were fixed in the home, and families were told they could be placed in any room in the household that they preferred to assess air quality. The video demonstrating the environmental sensor in South Africa is provided in Multimedia Appendix 9, and the video for Nepal is provided in Multimedia Appendix 10. In South Africa, 38 participants were recruited for 4 FGDs. Participants included health organization workers, community health workers, and caregivers. Caregivers were deemed eligible if they had a child in their care who resided with them in the household. Of the 4 FGDs conducted, 2 FGDs were with health organization workers and community caregivers (n=17) and 2 were with caregivers (n=21). In Nepal, 69 participants were recruited. There were 5 FGDs for FCHV, inclusive of 32 participants. Caregivers (all women) participated in 3 FGDs (n=18 participants). Data for the ranking task were also collected from the health volunteer and caregiver focus groups (n=50). In addition, 10 individual key informant interviews were conducted with FCHV, and 9 individual key informant interviews were conducted with caregivers. FCHV and caregivers were selected in Sankhu, Manamaiju, and Phutung. FCHV were identified from the roster of local government health facilities, where they report activities monthly. Every FCHV from 2 health facilities was selected for the study. The caregivers were selected based on referrals from FCHVs. The audio recordings of FGDs and interviews were translated into English. The translated transcripts were loaded in NVivo version 12 (QSR International Pvt Ltd) for qualitative data analysis [46]. A framework coding analysis approach was used [47]. A priori codes included the devices (continuous audio recording, continuous video recording, environmental sensor, episodic audio recording, episodic video recording, proximity beacon, and wearable beacon) and device attributes initially used for the anchoring vignettes (acceptability, confidentiality, noninterference, safety, and utility or benefit). Additional a priori codes included themes from the qualitative interviews (community health activities: child development and behavior, child health program, community health worker activities, and information collection and technology readiness: mobile phone usage and other technology usage). Using the a priori codes, 2 coders (authors BAK, who was familiar with the Nepal context, and KV, who was familiar with the South Africa context) read and coded the same 4 transcripts (1 FGD from South Africa, 1 interview from South Africa, 1 FGD from Nepal, and 1 interview from Nepal). The coders then used coder comparison within NVivo to determine areas of common versus discrepant coding. This was used to redefine the codes where needed. Additional code themes also emerged. These included feasibility and health risks and injury under attributes. Feasibility was added to address comments regarding whether the tools could be used but did not address acceptability or other attributes. Health risks was added to distinguish safety as an area where a person may be endangered by using the device through theft or assault versus health risks such as radiation or other perceived health consequences. Under community health activities, codes for context of community and facilities and family and caregiver behaviors were added. Under technology readiness, barriers to technology use and facilitators of technology use were added. All referenced quotations are provided in a supplemental file (see Multimedia Appendix 11). The 2 coders then coded 4 new transcripts with the same breakdown by country and qualitative type to establish interrater reliability. The 2 authors achieved 0.80 interrater reliability. Subsequently, the coders reviewed half of the remaining qualitative dataset. Additional information on the qualitative process is available in Multimedia Appendix 12, using the consolidated criteria for reporting qualitative research framework [48]. Statistical analysis was performed on the card sort data for South Africa and Nepal datasets using median and interquartile range with inference testing using the Wilcoxon Rank Test. This approach was selected because the data elicited through the ranking tasks were nonparametric in distribution. Statistical analyses were performed with Statistical Package for Social Sciences version 24 [49]. Ethical approval for the study was provided by the HSRC Research Ethics Committee in South Africa (REC6/18/05/16) and the Nepal Health Research Council (#241/2016) in Nepal. In addition, Duke University (Pro00074454) provided ethical approval for data analysis by US-based team members working with deidentified data from the sites. Participants in both sites completed written consents forms, which were also read to participants by research assistants because of low literacy rates among some groups in the study regions.