Predicting the appropriate timing of vascular access intervention therapy using machine learning

Abstract

[Background] Ultrasonography is useful in vascular access (VA) surveillance. However, there are no criteria to predict the appropriate timing of intervention. AI technology has been advancing rapidly in recent years. Its application in the medical field has also been progressing. In this study, we aimed to predict the appropriate timing of VA intervention therapy using machine learning with Python. [Method] We targeted 1862 patients who underwent ultrasonography. All the data were learned and predicted using machine learning. We examined the accuracy of the obtained data after the area under the curve (AUC) was calculated from the receiver operating characteristic (ROC) curve of each algorithm. We compared the rates of accuracy between the machine learning algorithms and the criteria of guidelines. [Results] We determined that the best algorithm was logistic regression, given that it yielded the following values: AUC (0.88), sensitivity (0.85), specificity (0.71), and accuracy rate (0.83). The guidelines yielded the following values: sensitivity (0.69), specificity (0.86), and accuracy rate (0.72). [Conclusion] Machine learning predictions exceeded the Guideline accuracy score. These results suggest that machine learning may be used to predict the appropriate timing of VA treatment.