Postoperative VTE prediction models show substantial heterogeneity and high bias risk in lung cancer surgery patients undergoing systematic review analysis

BackgroundRisk prediction models for venous thromboembolism (VTE) in lung cancer patients undergoing surgery have increased substantially in recent years. However, the methodological quality, predictive performance, and clinical applicability of these models have yet to be systematically assessed.ObjectiveThis study aimed to systematically evaluate the published literature on the development and validation of postoperative VTE risk prediction models for patients with lung cancer.DesignA systematic review and meta-analysis of observational studies was conducted.MethodsA comprehensive search of CNKI, Wanfang, VIP, PubMed, Web of Science, The Cochrane Library, CINAHL, and Embase was conducted from inception to November 22, 2025. The data extracted from the included studies encompassed a range of characteristics, including design elements, predictors, model development strategies, validation approaches, and performance metrics. The Prediction Model Risk of Bias Assessment Tool (PROBAST) was utilized to evaluate the risk of bias and applicability. A meta-analysis of area under the receiver operating characteristic curve (AUC) values from validated models was performed using random-effects methods.ResultsA total of 4,700 records were identified, and after screening, twenty studies involving twenty prediction models were included. The majority of the studies were retrospective and single-center, and all were adjudged to have a high risk of bias according to PROBAST. Logistic regression emerged as the predominant modeling approach, while a limited number of studies adopted machine learning methods, including XGBoost and stacked models. The most frequently utilized predictors were D-dimer and age. The extent of reported model discrimination exhibited significant variability, with AUC values ranging from 0.66 to 0.99. A total of eight models that had undergone validation were deemed eligible for the quantitative synthesis, resulting in a pooled AUC of 0.85 (95% confidence interval [CI]: 0.78–0.93). However, substantial heterogeneity was observed (I² = 89.1%).ConclusionWhile several models showed some discriminatory ability, all included studies demonstrated a high risk of bias and limitations in applicability. The extant evidence does not support the routine clinical use of existing postoperative VTE prediction models in lung cancer patients. Future studies should adopt rigorous methodological frameworks, ensure adequate sample sizes, apply standardized predictor handling, and conduct multicenter external validation to improve the reliability and clinical utility of prediction models.Systematic review registrationhttps://www.crd.york.ac.uk/prospero/, identifier CRD420251232098.