Work

Early Detection of Maternal Supraphysiological Hypercholesterolemia: Development and Internal Validation of a Machine Learning Model and Characterization of First- Trimester Atherogenic Biomarkers

Year: 2026
Type: preprint
Study design: Observational Study
Abstract: Background Maternal supraphysiological hypercholesterolemia (MSPH), defined by total cholesterol ≥ 290 mg/dL at term in women without pre-existing dyslipidemia, is associated with a more atherogenic profile and adverse cardiovascular consequences for mother and offspring. Because MSPH is recognized only late in pregnancy, early markers identifying at-risk women are lacking. We aimed to develop and internally validate supervised machine learning models using first-trimester clinical and biochemical measurements to predict MSPH at term, and to characterize early atherosclerosis-related biomarkers in women who developed MSPH. Methods This prospective observational cohort included 160 women with singleton, uncomplicated pregnancies recruited at two prenatal centers in Santiago, Chile. Predictors comprised 11 first-trimester anthropometric, metabolic, hemodynamic, and lipid-related variables. ApoB, ApoA1, oxidized LDL, and malondialdehyde were measured in a subgroup (n = 78). Third-trimester lipid measurements, used to define MSPH, were excluded as predictors. A complete-case dataset (n = 148) and a median-imputed full-cohort (n = 160) were evaluated. Multiple classifier families, preprocessing strategies, and internal-validation schemes, including stratified train-test splits, k-fold cross-validation, and leave-one-out validation, were benchmarked within leakage-controlled pipelines. Stochastic partitioning strategies were evaluated across fixed random seeds, and model selection prioritized predictive performance, stability, and parsimony. Interpretability was assessed using SHAP and permutation importance. Results Of 160 women, 44 (27.5%) developed MSPH. First-trimester total and LDL cholesterol showed the strongest between-group differences, while multivariable exploratory analyses indicated substantial overlap between groups. The selected Linear Discriminant Analysis pipeline achieved a ROC-AUC of 0.77 and a PR-AUC of 0.52. At the fixed operating threshold of 0.60, sensitivity was 0.62, specificity 0.79, accuracy 0.75, and MCC 0.40. Outputs were interpreted as discrimination scores rather than calibrated absolute risk estimates. SHAP and permutation importance identified first-trimester total cholesterol as the dominant predictor. Women who subsequently developed MSPH also showed higher first-trimester ApoB and malondialdehyde. Conclusions First-trimester clinical and lipid measurements contain discriminatory information for subsequent MSPH, with total cholesterol providing the strongest contribution. Atherogenic and oxidative alterations are also detectable in the first trimester of pregnancies that later develop MSPH. These findings support further evaluation of interpretable first-trimester risk assessment approaches; external validation is required before clinical application. Trial registration : Not applicable. Analytical code: https://github.com/kren-ai-lab/MSPH_classification_models.
Language: English
Field: Medicine
Open Access status: green
OpenAlex ID: W7236174358
Cites: 46
Cited by:
Related to: 10