Using Wearable Technology to Assess Recovery and Detect Post-Operative Complications Following Cardiothoracic Surgery

Project Summary. Every year, more than 500,000 patients undergo operations for heart and lung disease. After surgery, patients often experience pain, fatigue, and disturbed sleep that can persist for weeks to months. In addition, up to 32% of patients develop postoperative complications, which often occur after discharge from

the hospital and may lead to readmission. Complications are costly and can be deadly; they are associated with a 200-300% increase in healthcare costs and a 6-fold increase in 90-day postoperative mortality. Currently, after surgery, when a patient is discharged from the hospital, the patient and their family members

are responsible for monitoring the patient’s health status. Patients are usually not seen by a doctor for 2-4 weeks after discharge. Attempts to improve postoperative monitoring include home health visits and telemedicine approaches. However, these methods have been shown to be ineffective, costly, and allow for

only vague and intermittent assessments of recovery. They do not detect complications until they are at a more severe stage. As such, accurate, easy-to-implement and inexpensive methods to assess postoperative recovery and to detect complications at their earliest stage—before symptom onset—are urgently needed.

We previously showed that machine learning analysis of biometrics collected by wearables could detect Lyme Disease and Covid-19. We then, in a pilot study, applied our algorithm, previously developed to identify Covid- 19, to patients undergoing thoracic surgery and showed that this algorithm could detect 89% of complications a

median of 3 days before symptom onset. When we evaluated the postoperative recovery of cardiothoracic patients, we showed that machine learning analysis of biometrics could classify patients into distinct recovery groups. Thus, wearables and machine learning algorithms could lead to a highly accurate and accessible

method to predict complications early and improve assessments of recovery. Our overall objective is to optimize and validate our machine learning algorithm—previously developed for the early detection of Covid-19—for the detection of postoperative complications prior to symptom onset and to use machine learning analysis to predict the quality of a patient’s recovery using pre- and intraoperative data.

Our project aims to first use wearables to collect high-resolution physiologic data of cardiothoracic surgical patients. We will then extend our previously developed algorithm for early detection of postoperative complications and develop an algorithm to predict the quality of a patient’s postoperative recovery.

The proposed project will develop an innovative method to detect postoperative complications prior to symptom onset and predict the quality of a patient’s postoperative recovery using pre- and intraoperative data. Importantly, our proposed method could be scaled to not only improve outcomes for cardiothoracic surgical

patients, but for patients undergoing other types of surgery. The results of this study will enable a future randomized trial that evaluates whether real-time postoperative monitoring with machine learning algorithms and wearables can lead to 1) earlier detection of complications, 2) earlier outpatient interventions that improve

recovery and/or reduce severity of complications, and 3) decreases in unplanned hospital readmissions.