The risk prediction models for deep venous thrombosis in perioperative patients with lower limb fractures: a systematic review and meta-analysis

The quality of model development and its applicability are still unknown, even though numerous studies have developed or validated predictive models to estimate the risk of perioperative deep venous thrombosis (DVT) in lower limb fractures. The aim of the study was to systematically assess the DVT risk prediction model in lower limb fracture patients in the perioperative phase. PubMed, EMBASE, Web of Science, and China National Knowledge Infrastructure (CNKI) were assessed regarding studies on risk prediction models for perioperative DVT in limb fractures. The search period spanned from the database establishment to December 20, 2024. Two investigators independently screened the articles and extracted the data, and the quality of the included articles was evaluated using the PROBAST assessment tool. A total of 21 studies describing 25 predictive models were included, and the acquired data indicated that the incidence of DVT ranged from 1.2% to 72.5% in the perioperative period of lower limb fractures. The most commonly used predictors were D-dimer, age, time from injury to admission, and time from injury to surgery. The areas under the curve (AUCs) for all models ranged from 0.676 to 0.979. While twenty-three models demonstrated good predictive performance, many exhibited biases, often stemming from unreported methods for handling missing data, some models lacked external validation. The combined AUC value for the training models was 0.82 (95% confidence interval: 0.78–0.87), suggesting a moderate level of discrimination. The existing model exhibits good overall prediction performance and a low applicability risk, however the bias risk is high. To enhance the robustness of these findings, further research should encompass prospective studies, external validation of the current model, and refinement of the statistical analysis aspects.