Development and Internal Validation of a Prediction Model for Preeclampsia Using Simulated Data: Integration of Hematological Inflammatory Indices and Serum Uric Acid

Preeclampsia is a hypertensive disorder of pregnancy that results in numerous adverse outcomes for the mother and fetus. Changes in inflammatory markers and blood cell counts as well as uric acid may help further define risk in preeclampsia. Determining risk of preeclampsia using different parameters may be more powerful than using a single biomarker. The goal of this study was to create a preeclampsia risk assessment tool using age, body mass index (BMI), blood cell counts and inflammatory markers, platelet count, and serum uric acid. We created a dataset consisting of 1,000 pregnant women of which 158 developed preeclampsia. We used this data to simulate a predictive modeling study. The goal of our study was to test and validate a predictive model, not to evaluate the real-world risk of developing preeclampsia. The parameters of the model included age, BMI, blood cell counts (neutrophils, lymphocytes, and platelets), inflammatory markers (SII, PLR), and uric acid. We first analyzed the parameters of interest in a univariable analysis. Next, a multivariable logistic regression was performed to create a predictive model. We used the area under the receiver operating characteristic curve (AUROC) to evaluate model discrimination. The Brier score was calculated to assess the expected error for the prediction. To validate the model, we performed 10-fold cross validation and 300 bootstrap resamples. In the final multivariable model, serum uric acid, platelet-to-lymphocyte ratio (PLR), BMI, systemic inflammation (SII), and platelet counts were associated with development of preeclampsia.The interaction of NLR with other variables (OR 0.566, 95% CI 0.245-1.309, p = 0.184) was also not statistically significant. Age was statistically significant (OR 1.036, 95% CI 1.000-1.074, p = 0.053), but on the edge of significance. The apparent area under the ROC curve was 0.854. The 10-fold cross-validated area under the curve was 0.850. The optimism corrected area under the curve was 0.838. The Brier score was 0.095 (apparent) and 0.097 (cross-validated). Based on the evaluation of the model in the simulated dataset, the interaction of the variables is steady and the model displays acceptable levels of overall performance. However, the results shown here should not be interpreted as evidence of the model’s validity or its application to real-world situations. Prior to application to patients, extensive evaluation for validity and reliability including several different patient populations should be conducted.