Validation of an Automated Wearable Camera-Based Image-Assisted Recall Method and the 24-h Recall Method for Assessing Women’s Time Allocation in a Nutritionally Vulnerable Population: The Case of Rural Uganda

Accurate data are essential for investigating relationships between maternal time-use patterns and nutritional outcomes. The 24 h recall (24HR) has traditionally been used to collect time-use data, however, automated wearable cameras (AWCs) with an image-assisted recall (IAR) may reduce recall bias. This study aimed to evaluate their concurrent criterion validity for assessing women’s time use in rural Eastern Ugandan. Women’s (n = 211) time allocations estimated via the AWC-IAR and 24HR methods were compared with direct observation (criterion method) using the Bland–Altman limits of agreement (LOA) method of analysis and Cronbach’s coefficient alpha (time allocation) or Cohen’s κ (concurrent activities). Systematic bias varied from 1 min (domestic chores) to 226 min (caregiving) for 24HR and 1 min (own production) to 109 min (socializing) for AWC-IAR. The LOAs were within 2 h for employment, own production, and self-care for 24HR and AWC-IAR but exceeded 11 h (24HR) and 9 h (AWC-IAR) for caregiving and socializing. The LOAs were within four concurrent activities for 24HR (−1.1 to 3.7) and AWC-IAR (−3.2 to 3.2). Cronbach’s alpha for time allocation ranged from 0.1728 (socializing) to 0.8056 (own production) for 24HR and 0.2270 (socializing) to 0.7938 (own production) for AWC-IAR. For assessing women’s time allocations at the population level, the 24HR and AWC-IAR methods are accurate and reliable for employment, own production, and domestic chores but poor for caregiving and socializing. The results of this study suggest the need to revisit previously published research investigating the associations between women’s time allocations and nutrition outcomes. This study was nested within a cross-sectional study of women with a child aged between 12 and 23 months inclusive (n = 211), to examine the impact of a labor-saving technology on women’s time for childcare, food preparation, and dietary practices. The study was conducted between January and February 2018 in Bugiri and Kamuli Districts, Eastern Region, Uganda. In our study, women’s time allocation was assessed, for the same day, using three concurrent methods: (1) direct continuous observation (15 h), (2) 24HR, and (3) IAR using photos captured via an AWC. An IAR is a method using photographs, either automatically generated from a wearable camera or taken by the participant, as an autobiographical memory cue (recall trigger) to help respondents reconstruct key details from their previous day [100,101,102,103]. Data were collected over five consecutive days, following one of two possible patterns, as presented in Supplementary Table S1. Specifically, for both patterns, on day 1, eligibility was confirmed, a structured questionnaire was administered, and anthropometric data were collected for all participants. For half of the study participants, on day 2, time allocation data were collected using direct observation and recorded on the AWC attached to the respondent. On day 3, a 24HR was administered, followed by an IAR using photos captured on day 2 by the AWC. On day 4, time allocation data were again recorded via AWC only (i.e., no observation). On day 5, an IAR was administered using photos captured on day 4 by the AWC. The other half of the study participants began with the AWC only (i.e., days 2 and 4 were switched) and ended with all three methods (i.e., days 3 and 5 were switched). For all participants, on the 5th day, a final structured questionnaire was also administered. Time allocation data collection was distributed across all days of the week at the population level to account for a day-of-the-week effect, and for each respondent, the enumerator assigned to conduct the direct observation was different from the enumerator assigned to administer the 24HR and IAR. Ethical approval was obtained from the [location masked for blind review] (A24ES), [location masked for blind review] Research Ethics Committee (Project ID: 1420), and [location masked for blind review] Ethics Committee (Project ID: B0501). Following community sensitization, verbal explanation of the study, and demonstration of the AWC, written consent (signature or thumb print) was obtained from all respondents who participated in our study. Twenty-two villages were purposefully selected, for this study, of which eleven had access to labor-saving technology and eleven did not. These villages participated in the Sasakawa Global 2000 Uganda (SG2000 Uganda) country program (the local implementing partner for the parent study). The sample size calculation (n = 264; 22 communities, 12 households per community) was based on requirements of the main study within which this current study was nested. This sample size was deemed sufficient for the current validation study, using the Bland–Altman (BA) method of analyses [106,107,108]. The sampling frame, for each village, was a household listing of all women with a child born between 1 January 2017 and 1 May 2017 (children aged 12 to 23 months at the time of data collection). These lists were generated by the SG2000 community-based facilitators. Twelve mother–child dyads in each village were randomly selected to participate in the study. Substitutions were made, as needed, until 12 mother–child dyads who met the inclusion/exclusion criteria were recruited. Mother–child dyads were excluded if the child was less than 12 months or greater than 23 months of age, was not yet eating solid foods on a regular basis, or was a multiple-birth child; the mother was unable to communicate in Lusoga, Luganda, or English; either the mother or child had a severe disability; the mother was not the biological mother of the child; the mother was a co-wife with a selected mother; or either the mother or child was not available for the duration of the study. The enumerators administered two structured questionnaires to the respondent. The first questionnaire collected information on household socio-demographics and assets, and factors related to women’s empowerment. The second questionnaire collected information on household mobile phone access and ownership, and perceptions of their experiences with each of the three time allocation data collection methods. For the criterion time allocation assessment method (i.e., direct observation), enumerators recorded all activities undertaken by the respondent in 15 min intervals (“timeslots”) from approximately 06:00 to 21:00, using a structured instrument comprising 44 activities. On the day after the observation day, a multiple-pass 24HR was administered to the respondent to collect information on all activities undertaken by the respondent on the previous day. In the first pass, the respondent was asked to list everything she did the previous day; in the second pass, additional details about each activity and any concurrent activities were recorded. The time and duration of each activity were recorded in 15 min increments. In the third pass, the enumerator confirmed with the respondent that each activity was recorded accurately. The 24HR protocol was based on a module developed for the WEAI, which was itself based on the Lesotho Time Budget Study [65,109]. On the observation day, a small, lightweight AWC (iON SnapCam Lite, dimensions 42 × 42 × 13 mm3) was attached to a t-shirt worn by the respondent at approximately 06:00 and removed at approximately 21:00. Participants were instructed to wear the AWC while continuing their usual activities, covering or removing the camera as needed for privacy. The AWC automatically recorded a picture every 30 s, storing all photos (approximately 1800) on a memory card. The following day, an enumerator first reviewed the photos captured by the AWC on a tablet and annotated the activities she thought—based on the photos—were undertaken by the respondent, i.e., the enumerator image interpretation (EII). Based on her interpretation of the photos, the enumerator demarcated the series of activities for review later that day with the respondent. Upon meeting with the respondent, the enumerator first administered the 24HR. The enumerator then administered the IAR by first reviewing the AWC photos with the respondent on the tablet (16GB Samsung Galaxy Tab 3 with a 10” screen, using Simple Gallery software for image display). During this interview, the enumerator used “verbal probing” to elicit from the participant additional relevant information about the activities performed, for example, to elaborate on what she was doing, who she was with, where she was going, and why [110,111]. The enumerator revised her original annotations (i.e., the EII) of activities undertaken by the respondent, as needed, based on the respondent’s feedback. The IAR protocol was adapted from one described by Kelly et al. (2015). The protocol followed ethical guidelines for AWC research to ensure privacy of the participants was maintained [112]. All protocols were pilot tested and refined prior to the start of the study. The number of minutes allocated to each of 44 activities recorded over the fifteen-hour period was calculated for each respondent and for each of the 3 data collection methods, in 15 min intervals (“timeslots”). The discrete activities were categorized into the nine mutually exclusive ICATUS-2016 major divisions (“activity groups”): (1) employment and related activities (“employment”), (2) production of goods for own final use (“own production”), (3) unpaid domestic services for household and family members (“domestic chores”), (4) unpaid caregiving services for household and family members (“caregiving”), (5) unpaid volunteer, trainee, and other unpaid work, (6) learning, (7) socializing and communication, community participation, and religious practice (“socializing”), (8) culture, leisure, mass media, and sports practices (“leisure”), and (9) self-care and maintenance (“self-care”), as presented in Supplementary Table S2 [70]. When individuals were observed performing more than one activity concurrently, the activities were given equal weight such that no activity was deemed “primary” or “secondary”. Of the 44 activities tracked, four were considered to be “simultaneous”, i.e., they could be performed while also performing other activities: care of the index child, care of other children or adults, chatting with friends or family, and watching TV or listening to the radio. When just one activity was performed in a timeslot, the activity performed counted for the entire 15 min. The simultaneous activities were always credited the full 15 min. However, for all other activities, when more than one activity was performed per timeslot, the 15 min were evenly distributed across the activities performed. For example, if in a 15 min timeslot, the participant was snacking (self-care) and then started preparing food (domestic chores) while feeding the index child (caregiving), caregiving—a simultaneous activity—was credited 15 min and self-care and domestic chores were each credited 7.5 min. The proportion of the study population living below USD 1.25/day was calculated using the Uganda 2012 Poverty Probability Index (PPI) with data collected via the respondents’ questionnaires [113]. The primary outcome variables analyzed were the total minutes allocated to each of the nine ICATUS-2016 activity groups and the median number of concurrent activities performed across all 15 min timeslots. Data were analyzed using Stata/SE version 15.1. p-values less than 0.05 were considered significant for all tests. Cases with incomplete data for any of the three methods (observation, 24HR, or IAR) were eliminated from analysis. Key socio-demographic characteristics for participating and missing households were compared using the Mann–Whitney U two-sample statistic for continuous data, and the Fisher exact test for categorical data. Due to inter-participant differences in actual observation start and end times and technical challenges with insufficient light in the early morning and evening, the analyses were limited to the 12 h period from 8 a.m. to 8 p.m. to retain as many cases as possible with complete data. The Wilcoxon signed rank sum test was used to compare the distributions of time allocation obtained via the criterion method (observation) versus the 24HR, IAR, or EII. The median time allocated for only those women partaking in each activity was also calculated and compared using the Wilcoxon signed rank sum test. Women’s time allocations estimated via EII and IAR were also calculated for the non-observation day and compared to the corresponding estimates for the observation day using the Wilcoxon signed rank sum test. The Wilcoxon signed rank sum test was also used to compare the distributions of the median number of concurrent activities obtained via the criterion method (observation) versus the 24HR or IAR. The inter-tool agreement between the criterion method (observation) and 24HR or IAR was assessed using the Bland–Altman limits of agreement (LOA) method for each ICATUS-2016 major division (minutes/d) [106]. Specifically, for each individual, the differences between the methods (the criterion measure of time allocation minus the time allocation estimated using either 24HR or IAR) versus the mean of the two methods were plotted; the bias and the 95% LOA (mean difference ± 2 SD of the differences) were estimated. The numbers of participants for whom the differences between the two methods were greater or less than zero were also calculated. The Bland–Altman LOA approach was used to assess inter-method agreement for estimating the median number of concurrent activities. Time allocations estimated via the 24HR and IAR methods against the reference method were compared using Cronbach’s (reliability) coefficient alpha. Cronbach’s coefficient alpha was interpreted as follows: <0.70 unacceptable; >0.70 acceptable; >0.80 moderate; 0.90–0.95 high; >0.95 suspect [114]. The inter-method reliability (24HR and IAR methods against the criterion method) for estimating the median number of concurrent activities was compared using the weighted Cohen’s κ coefficient. Cohen’s κ coefficient was interpreted as follows: <0·00 poor agreement; 0·00–0·20 slight agreement; 0·21–0·40 fair agreement; 0·41–0·60 moderate agreement; 0·61–0·80 substantial agreement; 0·81–1·00 almost perfect agreement [115,116].