Traditional recall approaches of data collection for assessing dietary intake and time use are prone to recall bias [1,2,3]. Prospective methods, which avoid recall bias, such as self-report diaries are not practicable in rural low-income country contexts due to low literacy, whereas direct observation is labor intensive. An alternative prospective approach is the use of automated wearable cameras. These devices are inexpensive technologies that prospectively and unobtrusively record activities as they are performed. Automated wearable cameras have been used to collect human behavior data in middle- and high-income countries, but their feasibility in rural, low-income country settings has not yet been determined.
Automated wearable cameras have been evaluated in middle- and high-income countries as a method for improving individuals’ recalls of dietary intakes (i.e., food and beverage consumption, eating episodes, and energy intakes) [4,5,6,7,8,9,10,11,12], the food environment (i.e., food and beverage marketing exposure, purchase, and consumption context) [6, 8], and time allocated to daily activities [13,14,15]. In studies using automated wearable cameras, the captured images have been coded by topical experts [16,17,18,19,20,21,22,23,24,25,26,27,28,29,30,31,32,33,34,35,36,37,38,39,40], artificial intelligence [41], or an enumerator with the assistance of the participant via an image-assisted recall [4, 6,7,8,9, 12,13,14,15, 42,43,44,45,46,47] (Fig. 1). In an image-assisted recall, photographs which have been taken either automatically via an automated wearable camera or by the participant themselves are used as an memory cue (i.e., recall trigger) to help respondents reconstruct key details from their previous day [14, 48,49,50]. Most image-assisted recall studies provide participants the opportunity to review and delete the images captured by the device privately, before being viewed by the researchers [51].
The "feasibility" of automated wearable cameras for collecting health data comprises an array of perceived and objective measures. Perceived measures include the emotional burden on study participants and the people they interact with; ease-of-use of the device; acceptability of image content captured by the wearable camera; and utility of the images captured by the wearable camera in aiding recall. Objective measures include participation refusal, non-compliance, and study withdrawal; device malfunction; observed or reported interactions regarding the camera with members of participants’ households and communities; image quality and fit for purpose; time and other resources required for image processing, coding, and analysis; and device cost. Feasibility issues can further be categorized by audience, i.e., from a participant, community, and/or researcher perspective.
Feasibility from a participant perspective
In several studies, some participants found the automated wearable camera cumbersome to wear, especially during physical activity [4, 8, 14, 20, 29, 31, 45, 47, 52, 53]. Studies in which participants were responsible for operating the device (e.g., turning the automated wearable camera on and off at the start and end of the data collection day), commonly reported that participants forgot to wear or charge the device [23, 26, 29, 31, 35, 44], or had difficulty pressing the devices’ small buttons [5, 17, 52, 53]. Participants have also reported emotional discomfort due to wearing the device, especially in public [6, 12, 29, 34, 53]. Heightened awareness of an automated wearable camera may result in a reactive change in the study’s behavior of interest [4, 12, 16, 20, 29, 34]. In six studies participants reported having modified their behavior in reaction to being recorded.
Concerns about either wearing the camera or what it might capture may also negatively influence the rates of recruitment and completion. Response rates varied substantially across automated wearable camera-based studies (16% to 89% where reported) [4, 6, 20, 29, 35, 47]. Several of the studies, which explicitly investigated the impact of an automated wearable camera on response rates, attributed recruitment challenges to the device [6, 20, 29, 35, 53, 54]. Study withdrawal [8, 47] and non-compliance [14, 20, 31] have also been attributed to the use of automated wearable cameras.
For studies using the image-assisted recall method, participants across all age groups reported that viewing the images captured by the automated wearable camera helped them to recall pertinent details of the data collection period [4,5,6,7,8, 12, 14]. Participants reported that neither the length of time nor the process of reviewing their automated wearable camera-captured images (i.e., the image-assisted recall) was onerous [4, 14].
Among the 28 studies reporting on automated wearable camera feasibility from a participant perspective, only three were conducted outside of high-income country contexts [7, 40, 55]. The evidence from these three studies, which were conducted in middle-income countries, is sparse but consistent with the results already reported. There were no issues related to recruitment or retention, and neither the automated wearable camera nor the image-assisted recall was overly burdensome, however the battery life of the device was insufficient.
Feasibility from a community perspective
Study participants reported removing or covering the automated wearable camera at school [8, 13, 14, 34], work [14], home [20, 31], and in public [45]. Three studies reported participants being approached about the automated wearable camera by members of the public, but they were not requested to remove it [44, 52, 53].
No studies outside of high-income country contexts have assessed the feasibility of using an automated wearable camera from the community perspective.
Feasibility from a researcher perspective
Lost data due to device inoperability (e.g., insufficient battery life or another malfunction) is among the most commonly reported challenges to the use of automated wearable cameras as a research method [4,5,6, 9, 12, 16, 20, 30, 31, 34,35,36, 38, 41, 47, 56]. Reported data losses due to device inoperability, as a proportion of intended image capture, ranged from 11–50% [5, 6, 12, 35,36,37, 47, 57]. Most studies report that the images generated by the automated wearable camera are of sufficient quality to enable analysis for the intended purpose. However, several image quality issues are commonly reported across a variety of contexts, including sub-optimal camera angle and positioning, inadequate image capture frequency, and key events that occur off-camera; [4, 6, 8, 9, 12, 16, 18, 19, 21, 25, 31, 36, 38] dark images caused by low or artificial lighting or obscured lens; [5, 8, 12, 21, 23, 24, 28, 31,32,33,34,35,36, 44, 47, 52] and blurry or scrambled images [16, 21, 22, 27, 31,32,33, 36, 37]. Furthermore, automated wearable camera images have been reported to be unsuitable for detailed analyses, for specific research areas, such as determining specific items of clothing worn by children far away from the camera [21], or detecting low intensity activities (e.g., fidgeting or activities performed while sitting down) [19]. The proportion of automated wearable camera images reported to be "uncodable" ranged from 1–35% [17, 20,21,22,23,24, 27, 28, 30,31,32,33,34, 36, 39, 52].
The results were again sparse but consistent for the two studies conducted in middle-income countries with results related to automated wearable camera feasibility from a researcher perspective [40, 55]. In these studies, the cause of data losses was indeterminant and, although the image quality was acceptable, in one study the images captured were unfit for research purpose (determining the quantity of food consumed) [40].
The use of automated wearable cameras for research data collection aims to maximize reporting accuracy while minimizing participant burden. Part of this burden is shifted to the research team, and several studies highlight the heavy time burden required to manually code the automated wearable camera images for analysis [6, 14, 18, 24, 28, 29, 33, 34, 36, 38, 47], and its susceptibility to human error [24, 29]. Not all studies quantified the amount of time entailed, but where reported, the estimated time required to code automated wearable camera images range from approximately 1 to 2 h per participant day [7, 14, 17, 21, 23, 24, 27, 30, 31, 33,34,35,36, 38, 39]. Little information on other costs of automated wearable camera-based research is available. Only Kelly, et al. (2015) reported on the cost of the device (Autographer, £300 each), adding that it was "resource intensive" [14].
Although many studies acknowledged some feasibility limitations, especially for use in large-scale studies, nearly all concluded that automated wearable cameras are a promising method for collecting objective health behavior data in a free-living setting. Furthermore, despite the challenges described above, studies in high-income countries provide evidence that automated wearable cameras may help to improve study participants’ recall of foods consumed [9], and daily activities performed [14].
The available evidence for automated wearable camera feasibility, however, almost exclusively derives from studies conducted in high-income and upper-middle income countries. Key characteristics of rural women residents of low-income countries, such as literacy, exposure to technology and social norms, are quite different compared to any of the populations targeted in the automated wearable camera research published thus far. The research environment in rural low-income countries also poses different challenges including, for example, limited access to electricity for lighting the activity space or charging devices, and higher chance of device exposure to dirt or liquids, and lack of enumerators having pertinent skills. Feasibility needs to be explored in low-income countries, especially in rural contexts, given that the environmental conditions (which may affect device operability and image quality), social norms (which may affect acceptability by participants and the public), and familiarity with technological devices (which may affect ease-of-use) are all quite different than in high-income and/or upper-middle-income countries.
This study was therefore undertaken to assess the feasibility of using an automated wearable camera for data collection, in rural Eastern Uganda, on the dietary practices of women and young children and time-use patterns of women. The results can inform future automated wearable camera studies conducted in similar contexts.