Development of prediction models for antenatal care attendance in Amhara region, Ethiopia

Abstract

ABSTRACTBackgroundIn low-resource settings, coverage of at least four antenatal care (ANC) visits remains low. As a first step towards enhancing ANC attendance, this study aims to develop a series of predictive models to identify women who are at high risk of failing to attend ANC in a rural setting in Ethiopia.MethodsThis is a cohort study conducted in the Birhan field site, Amhara region. Using data of a surveillance system and a pregnancy cohort, we developed and internally validated a series of logistic regressions with regularization (LASSO), and ensembles of decision trees.Discrimination was estimated using the area under the receiving operator characteristic curve (AUC). Three prediction time points were considered: conception, and gestational weeks 13 and All models were internally validated using 5-fold cross validation to avoid overfitting.ResultsThe study sample size was 2195. Mean age of participants was 26.8 years (Standard Deviation (SD) 6.1) and mean gestational age at enrolment was 25.5 weeks (SD 8.8). A total of 582 women (26.5%) failed to attend ANC during cohort follow-up. We observed AUC in the range of 0.61-0.70, with higher values for models predicting at weeks 13 and 24. All AUC values were similar with slightly higher performance for the ensembles of decision trees.ConclusionThis study presents a series of prediction models for ANC attendance with modest performance. The developed models may be useful to identify women at high risk of missing their ANC visits to target interventions to improve attendance rates. This study opens the possibility to develop and validate easy-to-use tools to predict health-related behaviors in settings with scarce resources.SUMMARY BOXNo published studies to date have developed risk prediction models for ANC attendance.The presented models show modest performance, but may be useful to identify pregnancies at a high risk of not initiating ANC.This type of models could be used by countries with strong community health programs to identify high-risk women to target specific interventions aiming to improve ANC attendance rates, increasing feasibility and cost-effectiveness of those interventions.Our models were internally validated using cross-validation to avoid overfitting, and despite not being tested in other populations, they are useful to inform local and regional health authorities.This study demonstrates that it is possible to develop predictive models for behavioral outcomes using data from surveillance systems and pregnancy cohorts in settings with scarcity of resources.