Aim: Intrahepatic cholestasis of pregnancy (ICP) is a pregnancy-specific liver disease associated with a significant risk of fetal complications including pre-term delivery and fetal death. Typically, it was diagnosed in the third trimester of pregnancy. This study utilized characteristics from routine maternal examinations in the first 20 weeks' gestation to predict ICP in pregnant women.
Methods: This is a retrospective case-control study. 13,329 medical records were collected on pregnant women presenting to the West China Second University Hospital between December 2017 and December 2018. After screening according to strict criteria, a total of 487 patients, 250 intrahepatic cholestasis of pregnancy cases, and 237 controls were selected for this study. We collected seven maternal characteristics indices for analysis and forty-three routine blood examination indices were obtained from routine hepatic, renal, and coagulation function examinations. The least absolute shrinkage and selection operator regression was applied for variable selection. Classification and regression trees, logistic regression, random forests, and light gradient boosting machines were fit for predictive modeling. We randomly divided 25% of the original data as testing set to conduct internal validation of the performance of the prediction model. The area under the receiver operating characteristic curves (AUC) was used to compare methods.
Results: Eight variables were selected out as potentially significant predictors that could reliably predict ICP. The sensitivity, specificity, accuracy, and AUC of the final prediction model obtained by light gradient boosting machines were 72.41, 79.69, 76.23, and 79.77%, respectively. Significantly higher platelet large cell ratio, alanine aminotransferase, glutamyl transpeptidase, and fibrinogen levels were found in cases as compared to healthy controls, while activated partial thromboplastin time and mean corpuscular hemoglobin concentration levels were significantly lower (p < .001).
Conclusions: The combination of alanine aminotransferase, glutamyl transpeptidase, fibrinogen, platelet large cell ratio, activated partial thromboplastin time, lactate dehydrogenase, creatinine, and mean corpuscular hemoglobin concentration levels can effectively predict ICP in the first 20 weeks of gestation. These could help provide direction for earlier detection and prevention of ICP.
Keywords: Intrahepatic cholestasis of pregnancy; machine learning; maternal characteristics; prediction; screening.