A machine learning (ML) algorithm was developed to predict bleeding risk in venous thromboembolism (VTE) patients receiving anticoagulant therapy. British Journal of Hematology.
The ML algorithm was generated using 55 baseline variable predictors using data from 49,587 patients from the Registro Informatizado de Engermedad TroomboEmbólica (RIETE) and the final model was compared with the RIETE and VTE-BLEED scores. I was. RITE is an ongoing international registry of her VTE patients. Learning was prospective using data from patients recruited from the RIETE registry.
An internal prospective validation was performed in a new cohort of 10,337 patients and an external validation was performed using data from 3027 patients from the COMMAND-VTE database.
keep reading
Five ML methods were used for training. XGBoost was the best performing method. In the prospective validation cohort, the rate of major bleeding was 2.2%.The algorithm correctly identified 8.7% of major bleeding episodes and failed to identify 1.6% (odds ratio [OR], 5.9; 95% CI, 4.4–7.8). The sensitivity and specificity of the algorithm in the prospective validation cohort were 33.2% and 93%, respectively. The positive predictive value was 10%.
The F1 score was 15.4%, higher than the RIETE score of 8.6% and the VTE-BLEED score of 6.4%.
“The XGBoost algorithm performed poorly in our externally validated cohort,” the authors write in their report. They pointed out that this was because his COMMAND-VTE dataset “lacked 14 predictors in the algorithm.”
In the externally validated cohort, sensitivity and specificity were 10.3% and 87.6%, respectively. The positive predictive value was 3.5%. The F1 score was 5.2%, while the RIETE score was 17.3% and the VTE-BLEED score was 9.75%.
The authors conclude that the algorithm’s performance “outperformed RIETE and VTE-BLEED scores only in the prospective validation cohort, but not in the external validation cohort.”
Disclosure: Some study authors have declared affiliations with biotechnology, pharmaceutical, or device companies. See the original reference for the full list of disclosures.
reference
Mora D, Mateo J, Nieto JA, et al. Machine learning for predicting hemorrhage during anticoagulation for venous thromboembolism: possibilities and limitations. Br J Hematol. Published online on March 21, 2023. doi: 10.1111/bjh.18737
