Development and Validation of a Prediction Model for Pulmonary Arterial Hypertension in Diabetic Patients: A Retrospective Study of 57,375 Patients

Pulmonary arterial hypertension (PAH) is an important cardiopulmonary complication in patients with type 2 diabetes mellitus (T2DM), yet practical models integrating routinely available clinical, laboratory, and echocardiographic information remain limited. We conducted a single-center retrospective electronic medical record cohort study with a prespecified temporal split. Patients admitted from January 1, 2018 to December 31, 2021 formed the training cohort, and those admitted from January 1, 2022 to July 31, 2025 formed the temporal validation cohort. Among 47,042 pre-2022 records screened, 10,458 eligible patients were included in the training cohort, including 423 patients with PAH (4.0%); among 60,069 later records screened, 14,060 patients were included in the validation cohort, including 600 patients with PAH (4.3%). Nine prespecified clinical domains were initially evaluated. Because tricuspid regurgitation velocity/gradient contributed to the echocardiographic PAH classification, a sensitivity analysis excluded the PAH-related pressure-gradient domain from the exploratory combined model to reduce incorporation bias. In the training-cohort sensitivity analysis, cardiac cavity and vascular dilatation was the strongest remaining individual discriminator (area under the curve [AUC], 0.819), followed by coagulation dysfunction (AUC, 0.697) and cardiac dysfunction (AUC, 0.627); the pressure-gradient-excluded combined model achieved an AUC of 0.823. When applied without refitting to the temporal validation cohort, this sensitivity model achieved an AUC of 0.780. The eight-variable echocardiography-informed nomogram had a C-index of 0.666 (95% confidence interval [CI], 0.631-0.699), with a Hosmer-Lemeshow p value of 0.453. Because the nomogram includes echocardiographic predictors, including tricuspid regurgitation velocity and pulmonary valve velocity, it is intended for risk stratification after echocardiography rather than pre-echocardiographic screening or prediction of incident PAH. Diabetes-specific variables such as HbA1c, diabetes duration, treatment exposure, insulin resistance indices, and microvascular complications were not consistently available in the historical records and were not modeled. Prospective multicenter studies with systematic collection of diabetes-specific variables are needed.