Aims Familial hypercholesterolemia (FH) significantly increases cardiovascular risk from childhood yet remains widely underdiagnosed. This cross-sectional study aimed to evaluate existing paediatric FH diagnostic criteria in real-world cohorts and to develop two novel diagnostic tools: a semi-quantitative scoring system (FH-PeDS) and a machine learning model (ML-FH-PeDS) to enhance early FH detection. Methods and results Five established FH diagnostic criteria were assessed (Dutch Lipid Clinics Network [DLCN], Simon Broome, EAS, Simplified Canadian, and Japanese Atherosclerosis Society) in Slovenian (N = 1360) and Portuguese (N = 340) paediatric hypercholesterolemia cohorts, using FH-causing variants as the reference standard. FH-PeDS was developed from the Slovenian cohort, and ML-FH-PeDS was trained and tested using a 60%/40% split before external validation in the Portuguese cohort. Only 47.4% of genetically confirmed FH cases were identified by all established criteria, while 10.9% were missed entirely. FH-PeDS outperformed DLCN in the combined cohort (AUC 0.897 vs. 0.857; P < 0.01). ML-FH-PeDS showed superior predictive power (AUC 0.932 in training, 0.904 in testing vs. 0.852 for DLCN; P < 0.01) and performed best as a confirmatory test in the testing subgroup (39.7% sensitivity, 87.7% PPV at 98% specificity). In the Portuguese cohort, ML-FH-PeDS maintained strong predictive performance (AUC 0.867 vs. 0.815 for DLCN; P < 0.01) despite population differences. Conclusion Current FH diagnostic criteria perform sub-optimally in children. FH-PeDS and ML-FH-PeDS provide tools to improve FH detection, particularly where genetic testing is limited. They also help guide genetic testing decisions for hypercholesterolemic children. By enabling earlier diagnosis and intervention, these tools may reduce long-term cardiovascular risk and improve outcomes.
原文链接
分享
引用
