Predictive Accuracy of a Clinical Model for Carriage of Pathogenic/Likely Pathogenic Variants in Patients with Dementia and a Positive Family History at PUMCH.

Identifying carriers of Pathogenic/Likely Pathogenic Variants in patients with dementia is crucial for risk stratification, particularly in individuals with a family history. This study developed and validated a clinical prediction model using whole-exome sequencing-confirmed cohorts. A total of 601 Chinese patients with dementia and a family history were enrolled at Peking Union Medical College Hospital, with 476 in a retrospective derivation cohort and 125 in a temporal validation cohort. Predictive factors included age at onset, ε4 status, and family history characteristics. Model performance was assessed using discrimination and calibration metrics. In the derivation cohort (median age at onset 66 years), 10.3% carried Pathogenic/Likely Pathogenic Variants. Among patients with dementia, those with age at onset < 55 years (OR 2.56, = 0.0098), more than two affected relatives (OR 3.32, = 0.0039), parental disease history (OR 4.72, = 0.015), and early-onset cases in the family (OR 2.61, = 0.0096) were positively associated with Pathogenic/Likely Pathogenic Variant carriage, whereas ε4 carriage was inversely associated (OR 0.36, = 0.0041). The model achieved an area under the curve of 0.776 (95% CI, 0.701-0.853) in the derivation cohort and 0.781 (95% CI, 0.647-0.914) in the validation cohort (median age at onset 58 years), with adequate calibration. This model demonstrated strong predictive performance for Pathogenic/Likely Pathogenic Variant carriage, supporting its clinical utility in guiding genetic testing. Further research is needed to refine the model.