A Simple data-driven dietary pattern associated with lower risk for adverse pregnancy outcomes.

Abstract

Background: Periconceptional diet has been associated with the risk of adverse pregnancy outcomes (APOs). Various types of diets have been studied to evaluate their potential protective effect against APOs. However, many of these diets are based on regional dietary preferences, are complex, and are composed of many different food categories. A simpler index may increase compliance and facilitate better clinical translation.

Objective: Evaluate whether alignment to a parsimonious index derived from periconceptional diet data is associated with reduced risk for APOs.

Design: We analyzed prospective data from 7,798 pregnant individuals from the USA in the Nulliparous Pregnancy Outcomes Study: Monitoring Mothers-to-be (nuMoM2b) study. We applied a sparse regression model to data from food frequency questionnaires and used it to select a small number of food categories that were jointly associated with low risk. We validated this index internally in the nuMoM2b cohort and externally on similar data on 2,240 participants from the Environmental influences on Child Health Outcomes (ECHO) study.

Results: The data-driven dietary index (DDI) included legumes, citrus fruits, dark green vegetables, and whole grains as food categories associated with lower risk for APOs versus processed meats, potatoes, and refined grains as associated with higher risk. High alignment with the DDI was associated with 22% relative risk (RR) reduction for 1 or more APO (adjusted RR (aRR) [CI] of 0.78 [0.70,0.87]), preeclampsia (0.74 [0.56,0.99]), and gestational hypertension (0.81 [0.67,0.98]). Validation in the ECHO dataset showed that high alignment to the DDI was associated with a 31% reduction in APOs (crude RR of 0.69 [0.50,0.89]).

Conclusions: We demonstrated an approach for devising a simple data-driven dietary index, showed that it is associated with reduced APO risk when accounting for confounders, and validated it in an independent dataset. Future studies could validate the DDI for prospective interventions, as well as apply our approach to other populations or clinical settings.