EARLY AND ACCURATE DIAGNOSIS OF PATIENT WITH AXIAL SPONDYLOARTHRITIS USING MACHINE LEARNING: A PREDICTIVE ANALYSIS FROM ELECTRONIC HEALTH RECORDS IN UNITED KINGDOM
Authors: Raj Sengupta et al.
There is a long delay between symptom onset and diagnosis of axSpA, which has been estimated to be more than 6 years. Attempts to decrease the diagnostic delay by addressing general practioners or patients have had limited success. This study reports the use of a machine learning algorhythm developed with electronic health records to identify possible axSpA cases. Data from 3902 axSpA patients and 3911 matched controls were used to train the algorhythm and then test its performance. The algorhythm identified 89 best clinical predictors that differentiated between patients and healthy controls. The sensitivity of the model was 75%, the specificity 96%, the positive predictive value 81% and the negative predictive value 83%.
This study shows a machine learning approach to identify possible axSpA cases from electronic health records with a high accuracy. These findings together with other similar studies demonstrate the utility of automated analysis of health data to reduce diagnostic delay in axSpA. Of course, such a system depends on the implementation of an electronic patient dossier, which is still missing in Switzerland.