Applying Machine Learning to Consumer Wearable Data to Predict Complications After Pediatric Appendectomy

Abstract When children are discharged from the hospital after surgery, caregivers rely mainly on subjective assessments (e.g., appetite, fatigue) to identify abnormal recovery symptoms since objective monitoring tools (e.g., thermometer) are very limited at home. Relying on such tools alone has resulted in unwarranted emergency department visits and delayed care. This study evaluated the ability of data from consumer-grade wearable devices, the Fitbit Inspire HR and Inspire 2, to predict abnormal symptoms and complications in children recovering after appendectomy. One hundred and sixty-two children, ages 3–17 years old, who underwent an appendectomy (76 simple and 86 complicated cases of appendicitis) wore a Fitbit device on their wrist for 21 days postoperatively. Symptoms and complications that arose during this monitoring period were gathered from medical records and patient report and used to label each postoperative day as either “abnormal recovery” or “normal recovery.” Fitbit-derived physical activity, heart rate, and sleep features and demographic and clinical characteristics were used to train balanced random forest classifiers to predict abnormal recovery days, separately for patients undergoing appendectomy for simple and complicated appendicitis. The classifiers accurately predicted 85% of abnormal recovery days up to the two days prior to the onset of a reported symptom/complication in complicated appendectomy patients and 70% of abnormal recovery days up to the two days prior in simple appendectomy patients. These results support the development of machine learning algorithms to predict onset of complications in children undergoing surgery and the role of the Fitbit as a monitoring tool for early detection of events.