Construction of a prediction model for isolated distal deep vein thrombosis after total knee arthroplasty based on machine learning.
Source: PubMed, NCBI / U.S. National Library of Medicine
Isolated distal deep vein thrombosis (IDDVT) is a common complication after arthroplasty. In this study, we calculated the incidence of IDDVT after total knee arthroplasty (TKA), established a prediction model for the occurrence of IDDVT after arthroplasty. We collected data on 590 patients who underwent TKA from September 2021 to June 2023 from the hospital's electronic system. We performed feature variable selection and model performance evaluation by Lasso regression, random forest, XGBoost. And take the intersection variable for the above algorithm. The selected variables were incorporated into the prediction model using Logistic regression. Effects of the prediction model were evaluated by selecting cross-validation, Bootstrap validation and external validation. The incidence of IDDVT within three days after TKA was 15.3%. Our prediction model incorporated 7 variables. The training set AUC of the prediction model in cross-validation was 0.829, which is higher AUC than Random Forest and XGBoost models. The AUC for the validation set of the predictive model was 0.726 and the external validation was 0.760. The net benefit of the predictive model in DCA was positive at probability thresholds in the range of 0.15-0.25. The Bootstrap validation (B = 500 repetitions) model had an AUC mean was 0.805, suggesting that the model has a high diagnostic value. The incidence of IDDVT after TKA is high. We developed an online prediction platform for clinical applications to as
Abstract
Isolated distal deep vein thrombosis (IDDVT) is a common complication after arthroplasty. In this study, we calculated the incidence of IDDVT after total knee arthroplasty (TKA), established a prediction model for the occurrence of IDDVT after arthroplasty. We collected data on 590 patients who underwent TKA from September 2021 to June 2023 from the hospital's electronic system. We performed feature variable selection and model performance evaluation by Lasso regression, random forest, XGBoost. And take the intersection variable for the above algorithm. The selected variables were incorporated into the prediction model using Logistic regression. Effects of the prediction model were evaluated by selecting cross-validation, Bootstrap validation and external validation. The incidence of IDDVT within three days after TKA was 15.3%. Our prediction model incorporated 7 variables. The training set AUC of the prediction model in cross-validation was 0.829, which is higher AUC than Random Forest and XGBoost models. The AUC for the validation set of the predictive model was 0.726 and the external validation was 0.760. The net benefit of the predictive model in DCA was positive at probability thresholds in the range of 0.15-0.25. The Bootstrap validation (B = 500 repetitions) model had an AUC mean was 0.805, suggesting that the model has a high diagnostic value. The incidence of IDDVT after TKA is high. We developed an online prediction platform for clinical applications to assess the probability of risk of IDDVT in TKA patients, facilitating timely identification of applications and treatment.