Optimizing Diagnostic Accuracy of Clinical Red Flags in <scp>RASopathies</scp>

RASopathies are a group of genetic disorders caused by pathogenic variants in the RAS-mitogen-activated protein kinase (RAS-MAPK) signaling pathway, often presenting with congenital heart defects, craniofacial dysmorphisms, and developmental delays. To assess the diagnostic yield of genetic testing in patients with suspected RASopathies and to identify predictive clinical "red flags" using machine learning (ML). This retrospective study included patients evaluated at our institution between 2020 and 2023. All patients underwent cardiovascular evaluation, and a subset of individuals with suspected RASopathies underwent genetic assessment including exome sequencing. Thirteen clinical "red flags" were analyzed as predictors of a molecular diagnosis within the subgroup of patients with suspected RASopathies. Diagnostic performance was assessed via sensitivity, specificity, and area under the curve (AUC). A random forest classifier identified the most predictive clinical features. Among 669 patients, 34 were clinically suspected of RASopathy, with a confirmed diagnosis in 24 cases (71%). Noonan syndrome was most frequent (18/24), and PTPN11 was the most commonly mutated gene (n = 13). Pulmonary valve stenosis (PVS) and facial dysmorphisms were the strongest individual predictors of a positive genetic test result. A threshold of ≥ 2 red flags balanced sensitivity (92%) and accuracy (73.5%). ML analysis identified PVS and facial dysmorphisms as the top predictors. A model including only these features achieved an AUC of 0.86 in the derivation cohort. PVS and facial dysmorphisms are key diagnostic indicators for RASopathies and should prompt early genetic testing. While the ML model showed high performance in this derivation cohort, external validation is needed to confirm its generalizability.