Development of the OncoScore: A Diagnostic Prediction Score for Previously Unrecognized Malignancy in Patients with Acute Pulmonary Embolism

Background/Objectives: Pulmonary embolism (PE) may represent the first clinical manifestation of a previously unrecognized malignancy. We aimed to develop the OncoScore, a diagnostic prediction score for estimating the probability of previously unrecognized malignancy being diagnosed during the index hospitalization for acute PE. Methods: We conducted a retrospective diagnostic prediction model derivation study using prospectively collected data from the multicenter Regional Pulmonary Embolism Registry (REPER). Candidate predictors were assessed using logistic regression and Chi-square Automatic Interaction Detection (CHAID) analysis, and independent predictors were incorporated into the OncoScore. Discrimination was assessed by receiver operating characteristic (ROC) analysis, calibration by calibration slope and a calibration plot, and internal validation by bootstrap resampling. Diagnostic performance measures were calculated using the cut-off value determined by the Youden index. Results: Among 2615 patients with acute PE, 91 had newly diagnosed malignancy, 265 had previously known malignancy, and 2259 had no malignancy. The final OncoScore included hemoglobin, C-reactive protein, platelet count, and leukocyte count, with a total score ranging from 0 to 7 points. The OncoScore yielded an area under the curve (AUC) of 0.702 (95% confidence interval [CI] 0.653–0.751; p < 0.001). Bootstrap internal validation of the fixed score yielded an optimism-corrected AUC of 0.7013 and an optimism-corrected calibration slope of 1.015. When predictor selection and CHAID cut-point determination were repeated within the bootstrap procedure, the optimism-corrected AUC was 0.6560. Patients were stratified as low (≤2), intermediate (3–4), and high risk (≥5), with observed malignancy prevalences of 1.1%, 5.5%, and 10.8%, respectively. At a cut-off of ≥3 points, sensitivity was 87.5%, specificity 46.5%, positive predictive value 6.4%, and negative predictive value 98.9%. Conclusions: The OncoScore is a simple laboratory-based prediction tool that showed moderate discrimination and identified clinically distinct risk strata for malignancy diagnosed during the index hospitalization for acute PE. The score should be considered a preliminary derivation-stage model, and independent external validation is required before clinical implementation.