Development and validation of a new model for predicting occult cancer after unprovoked VTE and its comparative performance against riete and some scores

Abstract Background: Venous thromboembolism (VTE) can be the first clinical manifestation of occult cancer. In our prior work, we externally validated the RIETE and SOME scores in patients with unprovoked VTE and found their discriminative performance to be modest. In the same cohort, we developed a new two-variable logistic regression model—now named the Riyadh Score—using age and hemoglobin to estimate the probability of cancer within 12 months ( EHA 2025, 4159687, 4159683). However, internal validation and comparative analysis with existing scores had not yet been performed. Objective: To internally validate the Riyadh Score and compare its predictive performance against the RIETE and SOME scores in patients with unprovoked VTE. Method We conducted a retrospective cohort study of adult patients diagnosed with objectively confirmed unprovoked VTE at a tertiary academic center between January 2016 and December 2022. Unprovoked VTE was defined according to ISTH criteria and included pulmonary embolism, deep vein thrombosis, mesenteric vein thrombosis, and cerebral venous sinus thrombosis. Patients with active cancer at baseline, provoked VTE, or incomplete follow-up were excluded. Clinical and laboratory variables were obtained through manual chart review. The Riyadh Score was previously derived in this cohort using multivariable logistic regression and incorporates two continuous predictors: age and hemoglobin level. The predicted probability of cancer was calculated using the equation: Logit(p) = −3.6317 + 0.0325 × Age − 0.0088 × Hemoglobin, and converted to risk using the logistic function. For comparison, the RIETE and SOME scores were calculated based on their original point-based algorithms and analyzed as continuous variables. Model performance was assessed by evaluating discrimination using the area under the receiver operating characteristic curve (AUC) and calibration using the Brier score and calibration plots. Comparisons between models were performed using DeLong's test for correlated ROC curves. Statistical analyses were conducted using SPSS version 29.0.2. The study was approved by the Institutional Review Board, and informed consent was waived.Results: Among 1,680 patients screened, 1,230 were excluded due to provoked VTE and 23 because of loss of follow-up, leaving 427 patients with unprovoked VTE for analysis. The median age was 60 years (IQR 22), and 40% were male. Most patients (73%) had at least one comorbidity. Over 12 months, 19 patients (4.4%) were diagnosed with cancer, most commonly colorectal (31.6%), followed by lung (15.8%) and pancreatic cancer (10.5%). The majority (58%) had localized disease at diagnosis. Model Performance The Riyadh Score demonstrated good calibration, with predicted risks closely aligned with observed outcomes across deciles of risk. The Brier score was 0.043, indicating low overall prediction error, and the Hosmer–Lemeshow test showed no evidence of miscalibration (p = 0.29). In terms of discrimination, the Riyadh Score achieved an AUC of 0.72 (95% CI: 0.64–0.80). In comparison, the RIETE Score yielded an AUC of 0.60 (95% CI: 0.49–0.70) and the SOME Score an AUC of 0.60 (95% CI: 0.48–0.71). The Riyadh Score significantly outperformed both the RIETE and SOME Scores in predicting cancer risk, with p-values < 0.001 for both comparisons (DeLong's test). Conclusion: We developed a novel prediction model—the Riyadh Score—based on only two readily available variables (age and hemoglobin) to estimate the risk of occult cancer following unprovoked VTE. The score demonstrated good calibration and discrimination, outperforming the RIETE and SOME scores. Unlike prior tools, it provides a continuous risk estimate rather than relying on arbitrary dichotomized thresholds. To support future research and clinical exploration, a web-based calculator is available. Prospective external validation is warranted to confirm its generalizability and clinical impact.