From Pruritus to Cholestasis: Building a Statistical Model and Online Application to Predict a Diagnosis Prior to Bile Acid Determination

Objective This study aimed to create a statistical model using clinical and laboratory parameters to predict which patients presenting with pruritus in pregnancy will have elevated total bile acids (TBA) and thus, have a high risk of intrahepatic cholestasis of pregnancy (ICP). Study Design Retrospective cohort study of patients presenting with pruritus in pregnancy and had TBA sent from a single public hospital from January 1, 2017, to December 31, 2017. Primary outcome is TBA ≥ 10 μmol/L. Multivariate logistic regression with stepwise and backward variable selection were used to create predictive models. Four models were compared using Akaike information criterion (AIC), C-statistic, and the DeLong nonparametric approach to test for differences between area under the curve (AUC) of receiver operating characteristic (ROC) curves. Internal validation was performed via fivefold cross-validation technique on the best-fitting, most parsimonious model. Results Of the 320 patients with pruritus, 153 (47.8%) had elevated bile acid levels ≥10 μmol/L. Sixty-nine variables were assessed for association with the primary outcome. Five variables were significantly associated with elevated TBA: pruritus of palms and soles (adjusted odds ratio [aOR]: 2.35 [95% confidence interval, CI: 1.22, 4.54]), gestational hypertension (aOR: 0.10 [95% CI: 0.02, 0.60]), log of total bilirubin (aOR: 4.71 [95% CI: 2.28, 9.75]), systolic blood pressure (aOR: 0.97 [95% CI: 0.94, 0.99]), and alanine aminotransferase (aOR: 1.05 [95% CI: 1.02, 1.07]). The final model was chosen for being parsimonious while having the lowest AIC with highest AUC (0.85; 95% CI: 0.81, 0.89). Internal validation using a probability threshold of 50% demonstrated a sensitivity of 65.5%, specificity of 83.5%, and accuracy of 75.1%. Conclusion We provide a predictive model using five simple variables to determine the probability that a patient presenting with pruritus in pregnancy carries the diagnosis of ICP. This tool, available via a web app, is designed to aid providers and enhance clinical judgment in difficult triage situations. Key Points Currently, no standard method to triage pruritus in pregnancy exists. We present a predictive statistical model using five readily available clinical variables. Final calculator yields probability of having intrahepatic cholestasis of pregnancy.