Supervised machine-learning algorithms model the complex interactions between clinical characteristics, laboratory test results, and immunoassay measurements and classify patients accordingly.

The algorithm was derived and intensively validated in a prospective multicenter cohort study (n=1’393). The heparin-induced platelet activation assay (HIPA) served as the reference standard.

The performance of the algorithm was tested in a designated validation dataset. The area under the ROC-curve was 0.99 for all models (95% CI: 0.97, 1.00). Compared to the currently recommended diagnostic algorithm (4Ts score, immunoassay), the number of false-negative patients was reduced by 50.0% (ELISA), 66.7% (PaGIA), and 64.3% (CLIA). False-positive individuals were reduced by 53.1% (ELISA), 72.1% (PaGIA), and increased by 29.0% for the CLIA.

The TORADI-HIT diagnostic algorithm was validated in patients with suspected HIT. It is not developed to rule-out other thrombocytopenia disorders or to be applied as a screening tool in unselected patients. Besides, one commonly used commercial ELISA assay was employed in our study (LIFECODES PF4 enhanced, Immucor), and we cannot fully exclude that other tests perform differently. Before additional verification with other assays, the algorithm can currently only be applied to this particular ELISA.

The TORADI-HIT algorithm is a non-commercial research tool that has been developed by the PCD research group at Inselspital, University Hospital, and the University of Bern. To date, external validation is pending, and scientific societies do not yet support the application of the tool. The algorithm has not been registered as a medical device with the European Medicines Agency (EMA), and it is not CE marked. The application of the algorithm does not replace the standard diagnostic workup in patients with suspected HIT. No patient data is stored with this web-based tool.

This plot shows how the different predictors influenced the model`s prediction of the likelihood of a mild bleeding disorder. The SHapley Additive exPlanations (SHAP) values were used to show the importance of each predictor. SHAP scores quantify how much each feature contributes to a particular prediction, ensuring that the contributions are fairly distributed across the features. Positive SHAP values (red) indicate that a feature increases the predicted likelihood of HIT, and negative SHAP values (blue) indicate that a feature decreases the predicted likelihood of HIT. SHAP values should be interpreted relatively to identify the most influential predictors.