Minimum Dietary Diversity for Women: precision of national surveys and accuracy of brief data collection instruments.

Background: Minimum Dietary Diversity for Women (MDD-W) has been identified as a promising indicator for monitoring diets globally. MDD-W questionnaires have been integrated into, amongst others, the Demographic and Health Surveys (DHS) and the Gallup World Poll (GWP). Disagreement between estimates of healthy diet metrics for the same country collected in the same year is not desirable.

Objectives: To determine the measurement agreement of MDD-W estimates collected through DHS and GWP (i.e., precision), to assess the impact of survey characteristics on potential discordance, and to examine the comparative validity of the brief data collection instruments used to estimate MDD-W (i.e., accuracy).

Methods: Using meta-data from DHS and GWP, we quantified the percentage points (pp) difference in food group consumption and MDD-W prevalence. Furthermore, we qualitatively examined the differences of four survey characteristics: food groups and sentinel foods used in the MDD-W questionnaire, sampling framework, survey timing, and data collection modality. In addition, using data from two non-inferiority studies in Ethiopia - which used either a weighed food record (WFR) or quantitative 24-hour recall (24-HR) as the reference method - we simulated the total magnitude of errors associated with non-quantitative open or list-based 24-HRs, and subsequently compared the pp differences in simulated food group consumption and MDD-W prevalence.

Results: MDD-W estimates from the GWP were significantly higher than those from the DHS in five of nine country-year sets, one difference was non-significant, and three estimates could not be statistically compared due to lack of reporting on margins of error. The absolute difference between MDD-W estimates from the DHS and GWP were >|5| pp for all country-year sets (range: -17 to + 21 pp). There was poor agreement between the DHS and GWP questionnaires regarding the choice and number of sentinel foods used for each food group in the same country (range: 21 to 65%). In general, GWP data collection covered fewer months and questionnaires were enumerated in fewer languages than the DHS, but the number of sentinel foods used per food group was more standardized across countries. Simulations indicated that the magnitude of errors associated with pilot tested non-quantitative open and extensive list-based 24-HRs were marginally lower than sentinel list-based 24-HRs in Ethiopia.

Conclusion: For global monitoring, standards must be defined for the preferred data collection instrument and survey platform for each healthy diet metric. This would facilitate cross-country comparability and help mitigate misinterpretations of change over time within countries and the selective reporting of national statistics. A harmonized methodology for developing, pilot testing, and continuously updating sentinel food lists is needed to further improve the accuracy of MDD-W questionnaires.