Development of a real-time reporting system of the reference interval for gestational serum creatinine and estimated glomerular filtration rate using machine learning

Abstract

Abstract The evaluation of maternal serum creatinine (SCr) concentrations according to gestational week (GW)-specific reference intervals (RIs) could be helpful in predicting adverse pregnancy outcomes. From January 2010 to December 2020, 1,370 SCr measurements from 940 normal pregnant women were collected from electronic medical records. Data should be processed using the bootstrap resampling method as most of the sample sizes according to GW were too small for obtaining the RIs. To enable resampling, the GWs were divided into 12 gestational periods (GPs). Implementation of resampling, determination of the appropriateness of RIs from the resampled new datasets in every GP, and establishment of GW-specific SCr RI using polynomial regression model analysis of GP-specific SCr RIs were performed using machine learning techniques. As 100 means from two resampled SCr measurements without replacement were made at every GP, 1,200 resampled results were used for developing RIs. The regression equations used for calculating the upper and lower limit of GW-specific SCr RIs were y = 88.8 − 3.75x + 0.141x2 − 0.00157x3 and y = 42.3 − 1.48x + 0.0321x2, respectively. Gestational estimated glomerular filtration rate (eGFR) was defined as the rate of SCr hyperfiltration. The median regression equation for GW-specific eGFR RI was y = 99 + 5.71x − 0.184x2 + 0.00166x3, while the calculation process of SCr hyperfiltration at any GW was added to develop the gestational eGFR formula (GEF). As GW-specific SCr RI and eGFR by GEF with GW-specific eGFR RIs were reported in the laboratory information system in real time, this clinical application can be used as a screening tool for predicting the adverse pregnancy outcomes.