An Image-Voice Dietary Assessment System for Estimating Individual Nutrient Intakes in Cambodian Women and Children: Relative Validity, Reliability, and Acceptability Study.

Background: Individual-level dietary intake data are fundamental for developing nutrition policy and programs. In low- and lower-middle-income countries, proxy measures of individual intake (household consumption and expenditure surveys and food balance sheets) are often used, with limited implementation of new technology-assisted applications.

Objective: We aimed to determine the relative validity, test-retest reliability, and acceptability of the Voice-Image Solution for Individual Dietary Assessment (VISIDA) system in a sample of Cambodian women and their children aged ≤5 years.

Methods: Mothers and one of their children were recruited from 3 locations (rural, semirural, and urban) in Siem Reap province, Cambodia. Dietary intake data were collected for each participant using 2 methods across 3 recording periods over approximately 4 weeks. In week 1, intake was recorded using VISIDA for 3 nonconsecutive days, followed by 3 interviewer-administered, multiple-pass 24-hour recalls collected in weeks 2 to 3. In week 4, VISIDA was used again to collect a 3-day food record. After the third intake recording period, the mothers completed a feedback survey. Differences in estimated nutrient intakes for the 3 recording periods for mothers and children were examined using a linear mixed model approach.

Results: The analysis included 210 participants (n=119, 56.7% mothers and n=91, 43.3% children). Estimated mean nutrient intakes reported in both VISIDA recording periods were mostly lower compared to intakes reported using the 24-hour recalls. Compared to the 24-hour recalls, statistically significant differences were found for the VISIDA recording periods for 80% (16/20) of nutrients for mothers and 32% (6/19) of nutrients for children. Nutrient intakes estimated from both VISIDA recording periods showed no statistically significant differences for mothers and children. For mothers, the differences of model weighted marginal means in energy intakes (kcal) were -296 (95% CI -410 to -181; VISIDA period 1 minus 24-h recall), -274 (95% CI -390 to -158; VISIDA period 2 minus 24-h recall), and -22 (95% CI -131 to 87; VISIDA period 1 minus VISIDA period 2). For children, the differences in model weighted marginal means in energy intakes (kcal) were -158 (95% CI -227 to -89; VISIDA period 1 minus 24-h recall), -127 (95% CI -198 to -57; VISIDA period 2 minus 24-h recall), and -31 (95% CI -98 to 37; VISIDA period 1 minus VISIDA period 2). Most mothers reported that the VISIDA smartphone app was "easy to use" (68/108, 63%) or "very easy to use" (23/108, 21.3%) for collecting dietary intake data.

Conclusions: The VISIDA system produced lower estimates of nutrient intakes when compared to the 24-hour recalls in a sample of mothers and children in Siem Reap province, Cambodia. However, the estimated nutrient intakes for the 2 VISIDA recording periods were similar. The participating mothers reported high acceptability for using the VISIDA smartphone app to collect intake data.