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

Abstract

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

Machine learning prediction of gestational age from metabolic screening markers resistant to ambient temperature transportation: Facilitating use of this technology in low resource settings of South Asia and East Africa

“…When the published ML model [ 19 ] was used to predict GA in the simulated data set, a similar distribution pattern of predicted GA was observed comparing the original and the simulated database from the sub-sample (Figure S1 in the Online Supplementary Document ). The simulated data set from 330 samples, with samples shipped on dry ice resulted in RMSE of 1.07 (95% confidence interval (CI) = 0.96-1.21) as compared to RMSE of 1.02 (95% CI = 0.91-1.14) for the parent data set (1318 children shipped and stored at -80°C and shipped on dry ice.)…”

“…Baseline characteristics of the 330 children contributing the samples provided in Table 1 , were evaluated against 1318 children [ 19 ], from which this sample was sub-selected. Percentage difference for each of the metabolites, between samples shipped in dry ice and ambient temperature were calculated across all the samples ( Figure 2 ).…”

“…A normal probability distribution ( dnorm ) was used so that the means and standard deviation for the vector points remained consistent. To evaluate if the 330 selected samples and their simulated data were in fact an unbiased representation of the parent data set used in the validation exercise [ 18 , 19 ], we used a published machine learning model to predict the GA in the simulated data set, and compared the results with those, obtained from 1283 participants in published validation study [ 19 ]. Stacked percentage and distribution plots were made to compare the predictability.…”

“…One of these models, based on conventional multivariable linear and logistic regression methods, have been internally validated in US population [ 17 ], and externally validated in Alliance for Maternal and Newborn Health Improvement (AMANHI) LMICs cohort, demonstrating satisfactory performance [ 18 ]. We have further demonstrated improvement in validity with an error of around 1 week by using machine learning models [ 19 ]. During external validation however, we did recognize the need for temperature control if the samples are to be shipped to a central laboratory within or outside the country (which would be most often the case if this method was to be used in LMIC settings).…”

Current Status and Future Directions of Neuromonitoring With Emerging Technologies in Neonatal Care

“…85.21 % (95% CI 72.31–94.65) of GAs were properly predicted to within one week. The model also showed high sensitivity (100%) and specificity (92.6%) for differentiation of preterm and term birth ( 127 ). Meenakshi et al ( 128 ) showed good results of a model to automatically estimate the GA at third trimester by using biparietal diameter and kidney length.…”