Development of a prediction model for infant hospitalization and death using clinical features assessed by community health workers during routine postnatal home visits in Dhaka, Bangladesh

Abstract Objectives Community health worker (CHW) identification of life-threatening illnesses among young infants (<2 months) during home visits and referral to hospital are critical to reducing infant morbidity and mortality in low-resource settings. We aimed to develop a prediction model for hospitalization and/or death among young infants in Dhaka, Bangladesh using clinical features assessed by CHWs during routine home visits. Methods This was a secondary analysis of data from generally healthy infants prospectively enrolled at birth and assessed by CHWs at 11 scheduled home visits from 3-60 days of age. Time-varying Cox regression with backward selection was used to identify clinical features associated with time to first hospitalization and/or death. Prediction models were developed and internally validated using 5-fold cross-validation. We evaluated model discrimination (C-statistic and time-varying area under the curve) and calibration (calibration plots). We also evaluated discrimination and calibration of a Cox model based on World Health Organization (WHO)-recommended eight danger signs to identify sick infants requiring referral during home visits. Results Among 1906 infants, 176 (9.2%) had an event (173 hospitalizations and 3 deaths). The best-performing model consisted of three baseline covariates (any perinatal/delivery complication, umbilical cord care, gestational age) and four clinical features (nasal congestion, cough, jaundice, skin rash). The best-performing model discrimination (C-statistic=0.71; 95% CI 0.68-0.75), and discrimination of the best-performing model’s four clinical features added to WHO danger signs (C-statistic=0.70; 95% CI 0.67-0.74), were slightly higher than that of WHO danger signs alone (C-statistic=0.56; 95% CI 0.53-0.60), but calibration was similar. Conclusions A prediction model for hospitalization and/or death using baseline covariates and clinical features assessed during home visits may support identification of infants in need of facility-level care. Adding four clinical features to the WHO danger signs algorithm may improve its predictive performance by capturing a broader spectrum of severe infant illnesses requiring hospitalization.