TABLE 2.

Cox proportional hazards models for incidence of underweight for children aged 3–12 months by prenatal DDS in the Uganda birth cohort, 2014–2016

	Maternal diet diversity score (DDS)
	Q1 (0–2)	Q2 (3)	Q3 (4)	Q4 (5–9)	p for trend
Underweight a
n/N	66/720	61/1128	50/842	34/589
Univariate HR (95% CI)
HR (95% CI)	1	0.63 (0.56, 0.70)***	0.64 (0.57, 0.73)***	0.65 (0.56, 0.74)***	<0.001***
Multivariable model HR (95% CI) b
HR (95% CI)	1	0.65 (0.57, 0.72)***	0.67 (0.59, 0.75)***	0.70 (0.62, 0.80)***	<0.001***
Multivariable model controlling for birthweight c
HR (95% CI)	1	0.66 (0.59, 0.74)***	0.67 (0.60, 0.76)***	0.70 (0.61, 0.80)***	<0.001***

Note: Women's food consumption was assessed using 24‐h recall. Dietary diversity scores (DDS) were computed as the number of food groups consumed. Test for trend conducted using median DDS for diet quartiles.

Abbreviations: 95% CI, 95% confidence interval; DDS: dietary diversity scores; HR: hazard ratio.

^a We estimated the hazard ratio (HR) of underweight using Cox regression models. HR below 1 implies that the incidence of the outcome was lower among infants of women with more diverse diets. For underweight models, the interaction of maternal diet diversity with child age was not significant, and the proportional hazards assumption is not violated. Models for underweight do not include an interaction term, and we do not present HR by child age.

^b The multivariable model for underweight adjusts for region (north/south‐west), child sex (male/female), maternal height, breastfeeding status (yes/no), paternal education status (0–6 years/7 or more years), marital status (married/not married), season of dietary intake (rainy/dry) and household wealth index (tertiles). The model also adjusts for participation in the community connector programme (treatment/control) and for clustering by subcounty.

^c Multivariable model controls for confounders above and also controls for birthweight in sensitivity analysis.

^* p < 0.05.

^** p < 0.01.

^*** p < 0.001.