JAKCalc: A machine-learning approach to rationalized JAK2 testing in patients with elevated hemoglobin levels

Abstract

The demand for Janus Kinase-2 (JAK2) testing has been disproportionate to the low yield of positive results, which highlights the need for more discerning test strategies. The aim of this study is to introduce an artificial intelligence application as a more rational approach for testing JAK2 mutations in cases of erythrocytosis. Test results were sourced from samples sent to a tertiary hospital’s genetic laboratory between 2017 and 2023, meeting 2016 World Health Organization criteria for JAK2V617F mutation testing. The JAK2 Somatic Mutation Screening Kit was used for genetic testing. Machine learning models were trained and tested using Python programming language. Out of 458 cases, JAK2V617F mutation was identified in 13.3%. There were significant differences in complete blood count parameters between mutation carriers and non-carriers. Various models were trained with data, with the random forest (RF) model demonstrating superior precision, recall, F1-score, accuracy, and area under the receiver operating characteristic, all reaching 100%. Gradient boosting (GB) model also showed high scores. When compared with existing algorithms, the RF and GB models displayed superior performance. The RF and GB models outperformed other methods in accurately identifying and classifying erythrocytosis cases, offering potential reductions in unnecessary testing and costs.