IDENTIFYING PATIENTS AT RISK FOR PERSISTENT MEDICALLY UNEXPLAINED PHYSICAL SYMPTOMS USING DATA MINING TECHNIQUES IN PRIMARY CARE ELECTRONIC MEDICAL RE

Medically unexplained physical symptoms (MUPS) have major impact on the quality of life of patients and lead to high health care costs. Early identification of patients at risk could lead to early intervention in order to prevent persistence. We aimed to develop and validate a predictive model that can help identifying these patients using data mining techniques applied to primary care electronic medical records (EMRs). Methods: We used all coded anonymised EMR data from 22 Dutch general practices (156.176 patients) over a five-year period (2007-2011). We defined MUPS as having a diagnosis of irritable bowel syndrome, fibromyalgia, chronic fatigue syndrome or low back pain without radiation. We balanced the dataset by including all patients with these codings (n=7840) and a randomly selected sample of non-MUPS patients (n=7988). We analysed EMR data from the year prior to the first MUPS diagnoses, or from a random year in the control group for the detection of predictors. We applied three different data mining techniques and evaluated the models by means of 10-fold cross-validation. By calculating the area under the curve (AUC) we determined performance measures. The models were compared by using the Student’s t-test. Results and Conclusions: We were able to classify patients at risk for persistent MUPS with an UAC of 0.796 (95% CI 0.792-0.801) by using a well performing algorithm including 408 variables. When implemented in advanced use of EMR data, early identification of patients at risk could support general practitioners with proactive care for MUPS.