Machine Learning Model to Aid Diagnosis of Heparin-Induced Thrombocytopenia - EMJ

Machine Learning Model to Aid Diagnosis of Heparin-Induced Thrombocytopenia

2 Mins

A NOVEL, accurate, and user-friendly machine learning model has been developed to diagnose heparin-induced thrombocytopenia (HIT), which may reduce misdiagnosis.

A team of researchers conducted a prospective, multicentre, observational study including 1,393 patients who were suspected to have HIT between 2018 and 2021. They collected detailed laboratory data and clinical information, which included results from immunoassays of the participants. The washed platelet heparin-induced platelet activation assay (HIPA) was used as the reference standard, and for each immunoassay, five separate machine learning prediction models were evaluated. Of the 1,393 patients, 1,274 were HIPA negative and 119 were HIPA positive, resulting in an HIT prevalence of 8.5%.

The team then completed a selection process to balance trade-offs between the model’s usability and accuracy, using a training data set that included 75% of the cohort. They identified causes of thrombocytopenia such as platelet nadir, identified immunoassay test result, C-reactive protein, unfractionated heparin use, and timing of thrombocytopenia as predictor variables. A collection of models, including enzyme-linked immunosorbent assay (ELISA), chemiluminescent immunoassay (CLIA), and particle-gel immunoassay (PaGIA), formed the TORADI-HIT algorithm, a multivariable diagnostic prediction tool. The support vector machine algorithms for ELISA and CLIA were found to perform the best, as well as a gradient-boosting machine algorithm for PaGIA. Researchers noted an area under the receiver-operating characteristic curve of 0.99 for each of these in the validation dataset, which included 25% of participants.

Furthermore, the number of false-positives and false-negatives was compared with the diagnostic algorithm that is currently recommended, and this showed a reduction in numbers of false-negatives of 50.0% for ELISA, 64.3% for CLIA, and 66.7% for PaGIA. The number of false-positives reduced by 29.8% for ELISA and 68.5% for PaGIA; however, it increased by 29.0% for CLIA.

“The TORADI-HIT algorithm has the potential to reduce delayed diagnosis and overtreatment in clinical practice. Future studies shall assess usability and performance in other patient populations and healthcare systems,” the researchers concluded in their report. They have made the tool accessible online to facilitate its use. Limitations of the study included the use of results from commonly employed immunoassays only, and the fact that most patients included were from Switzerland.

Keep your finger on the pulse

Join Now

Elevating the Quality of Healthcare Globally