Leveraging Multimodal Digital Health Data to Improve Understanding, Prediction, and Treatment of Postoperative Pain - Over 300 million surgeries are performed worldwide each year. Approximately 10-35% of surgical patients experience persistent post-surgical pain (PPSP), defined as new or worsening pain at the site of surgery lasting for 3 or more months. As the number of surgeries increases both nationally and globally, PPSP is becoming a major public health problem associated with poor quality of life, inability to return to work, and significant healthcare costs. Despite recent progress in mechanistic understanding of pain, prevention of PPSP remains difficult. A key challenge is the current lack of validated biomarkers that can reliably predict risk, guide treatment selection, or monitor recovery of postoperative pain. My program of research focuses on developing and validating novel digital biomarkers of perioperative pain derived from widely available digital health technologies (DHTs). Among spine surgery patients, I demonstrated that digital biomarkers derived from smartphones and wearable devices led to a 34% improvement in prediction of PPSP. This work was featured as an invited talk at the NIH’s 18th Annual Symposium on Advances in Pain Research. Over the next 5 years, my lab at Dartmouth will leverage novel artificial intelligence and machine learning (AI/ML) approaches for extracting features from DHTs and evaluate the utility of these features for predicting PPSP, including among patients without pre-existing pain. We will further leverage multimodal data collected postoperatively to identify behavioral (e.g., sleep, physical activity) and psychosocial (e.g., mood, anxiety, social connection) factors that drive postoperative pain fluctuations. Based on my prior NIH-funded research, I expect dynamic risk factors to vary across individuals, highlighting opportunities for precision prevention of PPSP. Together, these research areas will facilitate more accurate identification of patients at risk for PPSP and greater understanding of modifiable biopsychosocial mechanisms. Accurate prediction models can be used to enrich clinical trial samples when testing new intervention strategies, resulting in improved statistical power and more efficient trials. My long-term goal is to integrate DHTs into a larger digital ecosystem that identifies patients at risk of PPSP, provides data-driven insights into mechanisms of risk, and links patients with effective preventative measures. There are numerous areas of future research, including largescale validation of PPSP prediction models. Importantly, I envision a digital ecosystem that is bidirectional, allowing patients to gain insights into their own risk factors and postoperative recovery. In future research, we will evaluate the explainability and interpretability of novel digital biomarkers with patient and clinician stakeholders. Given rapid development of DHTs and AI/ML methods for analyzing multimodal digital health data, the flexibility afforded by this R35 is critical. As a Clinical Pain Psychologist and tenure-track faculty member in the Department of Biomedical Data Science at Dartmouth, I will continue to foster a multidisciplinary research team with expertise in digital technology, computer science, and novel AI/ML methods for extracting insights into perioperative pain from multimodal digital health data.
