Using AMANHI-ACT cohorts for external validation of Iowa new-born metabolic profiles based models for postnatal gestational age estimation

Post-natal gestational age assessment using targeted metabolites of neonatal heel prick and umbilical cord blood: A GARBH-Ini cohort study from North India

Background

Accurate assessment of gestational age (GA) and identification of preterm birth (PTB) at delivery is essential to guide appropriate post-natal clinical care. Undoubtedly, dating ultrasound sonography (USG) is the gold standard to ascertain GA, but is not accessible to the majority of pregnant women in low- and middle-income countries (LMICs), particularly in rural areas and small secondary care hospitals. Conventional methods of post-natal GA assessment are not reliable at delivery and are further compounded by a lack of trained personnel to conduct them. We aimed to develop a population-specific GA model using integrated clinical and biochemical variables measured at delivery.

Methods

We acquired metabolic profiles on paired neonatal heel prick (nHP) and umbilical cord blood (uCB) dried blood spot (DBS) samples (n = 1278). The master data set consists of 31 predictors from nHP and 24 from uCB after feature selection. These selected predictors including biochemical analytes, birth weight, and placental weight were considered for the development of population-specific GA estimation and birth outcome classification models using eXtreme Gradient Boosting (XGBoost) algorithm.

Results

The nHP and uCB full model revealed root mean square error (RMSE) of 1.14 (95% confidence interval (CI) = 0.82-1.18) and of 1.26 (95% CI = 0.88-1.32) to estimate the GA as compared to actual GA, respectively. In addition, these models correctly estimated 87.9 to 92.5% of the infants within ±2 weeks of the actual GA. The classification models also performed as the best fit to discriminate the PTB from term birth (TB) infants with an area under curve (AUC) of 0.89 (95% CI = 0.84-0.94) for nHP and an AUC of 0.89 (95% CI = 0.85-0.95) for uCB.

Conclusion

The biochemical analytes along with clinical variables in the nHP and uCB data sets provide higher accuracy in predicting GA. These models also performed as the best fit to identify PTB infants at delivery.

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

Background

Knowledge of gestational age is critical for guiding preterm neonatal care. In the last decade, metabolic gestational dating approaches emerged in response to a global health need; because in most of the developing world, accurate antenatal gestational age estimates are not feasible. These methods initially developed in North America have now been externally validated in two studies in developing countries, however, require shipment of samples at sub-zero temperature.

Methods

A subset of 330 pairs of heel prick dried blood spot samples were shipped on dry ice and in ambient temperature from field sites in Tanzania, Bangladesh and Pakistan to laboratory in Iowa (USA). We evaluated impact on recovery of analytes of shipment temperature, developed and evaluated models for predicting gestational age using a limited set of metabolic screening analytes after excluding 17 analytes that were impacted by shipment conditions of a total of 44 analytes.

Results

With the machine learning model using all the analytes, samples shipped in dry ice yielded a Root Mean Square Error (RMSE) of 1.19 weeks compared to 1.58 weeks for samples shipped in ambient temperature. Out of the 44 screening analytes, recovery of 17 analytes was significantly different between the two shipment methods and these were excluded from further machine learning model development. The final model, restricted to stable analytes provided a RMSE of 1.24 (95% confidence interval (CI) = 1.10-1.37) weeks for samples shipped on dry ice and RMSE of 1.28 (95% CI = 1.15-1.39) for samples shipped at ambient temperature. Analysis for discriminating preterm births (gestational age <37 weeks), yielded an area under curve (AUC) of 0.76 (95% CI = 0.71-0.81) for samples shipped on dry ice and AUC of 0.73 (95% CI = 0.67-0.78) for samples shipped in ambient temperature.

Conclusions

In this study, we demonstrate that machine learning algorithms developed using a sub-set of newborn screening analytes which are not sensitive to shipment at ambient temperature, can accurately provide estimates of gestational age comparable to those from published regression models from North America using all analytes. If validated in larger samples especially with more newborns <34 weeks, this technology could substantially facilitate implementation in LMICs.

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

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.