What question did this study set out to answer?

The study aims to develop and validate a machine learning tool for predicting mild bleeding disorders preoperatively.

January 17, 20260 citationsOpen Access

MBD-Check: An Interpretable Machine Learning Decision Support Tool for the Preoperative Prediction of Mild Bleeding Disorders—Model Development, External Validation, and User-Centric Evaluation

HNHenning Nilius FMFabrizio Minervini MAMarcel Adler

Key Points

The study aims to develop and validate a machine learning tool for predicting mild bleeding disorders preoperatively.
Collected clinical and laboratory data from two independent cross-sectional studies
Trained multiple machine learning algorithms on training cohort of 555 patients
Conducted external validation on 217 patients
Evaluated user-friendliness using System Usability Scale (SUS)
87.5% sensitivity in predicting MBD in external validation cohort
54.3% specificity in correctly excluding patients without MBD
AUROC of 0.86 indicating good model performance
Median SUS score of 82.5 indicating excellent usability

Abstract

Background: Mild bleeding disorders (MBD) are the most common inherited bleeding disorders, often leading to perioperative hemorrhages. Preoperative screening for MBD remains challenging due to a lack of effective tools, resulting in a significant proportion of patients being referred for preoperative work-up. This study aimed to develop, externally validate, and implement an easy-to-use, explainable machine learning-based decision support tool.sMethods: Clinical and laboratory data were collected in two independent cross-sectional studies, including consecutive patients referred for suspected MBD (n = 555, training cohort; n = 217, external validation cohort). Diagnostic workup followed current guidelines, with final diagnoses established by an expert panel. Multiple machine learning algorithms were trained, and the best-performing model underwent external validation. To evaluate user-friendliness, we created a survey platform incorporating four case vignettes and the System Usability Scale (SUS), a validated software usability questionnaire.Findings: The following predictors were selected: (a) activated partial thromboplastin time, (b) PFA-200 closure time (epinephrine/collagen cartridge), (c) sex, and (d) a streamlined bleeding history. In the external validation cohort, 87.5% of patients with MBD were correctly predicted (sensitivity; 95% confidence interval CI, 79.9–93.0), while 54.3% of patients without MBD were correctly excluded from further work-up (specificity; 95% CI, 44.3–64.0). The area under the receiver operating characteristic curve (AUROC) was 0.86 (95% CI, 0.81–0.90). The final decision support tool (available at https://toradi-hit.dbmr.unibe.ch/mbdcheck/) was assessed by 33 surgeons, 29 anesthesiologists, and 24 hematologists. The median time to complete the tool was 72 seconds (interquartile range IQR, 49.0–79.5). The median System Usability Scale (SUS) score was 82.5 (IQR, 72.5–90), indicating excellent usability (SUS > 80).Interpretation: MBD-Check is an interpretable machine learning solution that may simplify the preoperative prediction of mild bleeding disorders, potentially reducing unnecessary referrals and improving patient care.

AI에게 질문

Bookmark

View Full Paper