For a study, researchers sought to build an algorithm to identify potential pSS patients in primary care to improve early-stage disease identification. As the initial step toward a clinical decision support system, they developed a machine learning algorithm based on aggregated healthcare data. The hospital claims data (HCD) from routine healthcare data and the primary care electronic health records (EHRs), which included 1,411 pSS and 929,179 non-pSS patients, were linked at the patient level. Age, gender, diseases and symptoms, medications, and GP visits were utilized to categorize persons using the logistic regression (LR) and random forest (RF) models. The AUCs for the LR and RF models were 0.82 and 0.84, respectively. Many real pSS patients were discovered (sensitivity LR=72.3%, RF=70.1%), specificity LR and RF were 74.0% and 77.9%, and both models had a negative predictive value of 99.9%. However, most patients identified as having pSS did not receive a secondary care diagnosis for their illness (positive predictive value LR=0.4%, RF=0.5%). The study used machine learning and GP EHR data to identify individuals with pSS in primary care for the first time. Their strategy has to be validated and enhanced in clinical practice in order to help with the early diagnosis of pSS in primary care. They suggested collaborating with skilled clinicians to improve the algorithm’s capacity to detect pSS in primary care.

Source: bmcprimcare.biomedcentral.com/articles/10.1186/s12875-022-01804-w