A novel automated participant-recorded dietary data collection method using low-cost mobile phones and Interactive Voice Response (IVR) with low-literacy women: a validation study in rural Uganda

Background: Dietary data gaps limit the design of effective food policies and programs in low-and middle-income countries (LMICs). Automated mobile phone-based tools offer an opportunity to fill data gaps using less resources compared with face-to-face methods, especially for high-frequency dietary quality monitoring in resource-constrained environments.

Objective: This study assessed the validity of automated participant-recorded dietary data via Interactive Voice Response (IVR) on basic mobile phones to assess dietary quality amongst rural women in a sub-Saharan African context, against same-day gold standard observed weighed food records (WFR).

Methods: Automated IVR collected list-based recalls of food groups consumed in the last 24 hour using yes/no push-button response from 156 randomly selected women in rural Northern Uganda (wet season). Inter-method agreement was assessed by comparing mean women’s dietary diversity scores (WDDS) using Weighted Cohen’s kappa, Bland-Altman plots and Wilcoxon signed-rank test. Agreement in percentage achieving minimum DDS for women (MDD-W) and consumption of unhealthy foods and beverages was tested using Cohen’s kappa and McNemar’s test.

Results: Most women (74.4%) completed the IVR, with completion associated with better network coverage and prior positive mobile phone experience. Compared with the WFR, agreement for the IVR was moderate for MDD-W (21.6% vs. 17.2%; kappa=0.48), fair for mean WDDS (3.3 vs. 3.5; weighted kappa=0.39), and moderate for unhealthy food (34.5% vs 23.3%; kappa=0.44) and beverage consumption (32.8% vs 31.9%; kappa=0.43).

Conclusion: This study highlights the potential of collecting data from low-literate women in rural resource-scarce settings using IVR via basic mobile phones to estimate population-level MDD-W, WDDS and percentage consuming unhealthy foods/beverages. Given the need for participant training, IVR may be best suited to high frequency monitoring of sentinel groups across time. Further refinement of the IVR may improve food group reporting, reducing trade-offs between operational simplicity and accuracy. With appropriate adaptation we expect this method is generalizable to other settings.