Machine learning guided postnatal gestational age assessment using new-born screening metabolomic data in South Asia and sub-Saharan Africa

• DOI: 10.1186/s12884-021-04067-y
• Publication Date: 2021-09-07T13:02:48Z
• Authors: Sunil Sazawal; Kelli K. Ryckman; Sayan Das; Rasheda Khanam; Imran Nisar; Elizabeth Jasper; Arup Dutta; Sayedur Rahman; Usma Mehmood; Bruce Bedell; Saikat Deb; Nabidul Haque Cho...; Amina Barkat; Harshita Mittal; Salahuddin Ahmed; Farah Khalid; Rubhana Raqib; Alexander Manu; Sachiyo Yoshida; Muhammad Ilyas; Ambreen Nizar; Said Mohammed Ali; Abdullah H. Baqui; Fyezah Jehan; Usha Dhingra; Rajiv Bahl
• Keywords: Male; Fetal Programming; Cohort Studies; Machine Learning; 0302 clinical medicine; Pregnancy; Prospective Studies; Statistics; Obstetrics and Gynecology; Neonatal Brain Injury and Developmental Consequences; 3. Good health; Obstetrics; Environmental health; Premature Birth; Medicine; Female; Algorithms; Asia; Population; Gestational Age; Ultrasonography, Prenatal; 03 medical and health sciences; Neonatal Screening; Developmental Origins of Adult Health and Disease; Birth weight; Health Sciences; Machine learning; FOS: Mathematics; Genetics; Humans; Metabolomics; Developing Countries; Biology; Africa South of the Sahara; New born screening; Research; Infant, Newborn; Low Birth Weight; Gestational age; Gynecology and obstetrics; Computer science; Pathophysiology and Management of Preeclampsia; Pre-term births; ROC Curve; FOS: Biological sciences; Pediatrics, Perinatology and Child Health; RG1-991; Mean squared error; Reproductive medicine; Mathematics; Random forest

Babies born early and/or small for gestational age in Low and Middle-income countries (LMICs) contribute substantially to global neonatal infant mortality. Tracking this metric is critical at a population level informed policy, advocacy, resources allocation program evaluation an individual targeted care. Early prenatal ultrasound examination not available these settings, (GA) estimated using new-born assessment, last menstrual period (LMP) recalls birth weight, which are unreliable. Algorithms developed metabolic screen data, provided GA estimates within 1-2 weeks of ultrasonography-based GA. We sought leverage machine learning algorithms improve accuracy applicability approach LMICs settings.This study uses data from AMANHI-ACT, prospective pregnancy cohorts Asia Africa where ultrasonography weight metabolite screening subset 1318 new-borns were also available. utilized opportunity develop (ML) algorithms. Random Forest Regressor was used randomly split into model-building model-testing dataset. Mean absolute error (MAE) root mean square (RMSE) evaluate performance. Bootstrap procedures estimate confidence intervals (CI) RMSE MAE. For pre-term identification ROC analysis with bootstrap exact estimation CI area under curve (AUC) performed.Overall model had MAE 5.2 days (95% 4.6-6.8), similar performance SGA, 5.3 4.6-6.2). correctly 1 week 85.21% 72.31-94.65). preterm classification, AUC 98.1% 96.0-99.0; p < 0.001). This performed better than Iowa regression, Difference 14.4% 5-23.7; = 0.002).Machine models applied metabolomic dating offer ladder providing accurate population-level settings. These findings point investigation region-specific models, more focused feasible analyte broad untargeted metabolome investigation.