Researchers at Washington University in St. Louis have developed a new method, using machine learning techniques, to more accurately predict recovery from spine surgery. By collaborating with experts in neurosurgery, they have created a model that outperforms previous methods in predicting outcomes for patients undergoing lumbar spine surgery. The diversity of outcomes in orthopedic surgeries depends not only on the patient’s structural disease but also on various physical and mental health characteristics. Understanding these factors better can lead to more personalized treatment plans and early interventions that can improve patient outcomes.
The new model takes into account preoperative physical and mental health factors that influence surgical recovery. By identifying potential pitfalls and risk factors for each patient before surgery, physicians can tailor their treatment plans more effectively. Previous methods of predicting surgery outcomes relied on patient questionnaires given at specific times and did not capture the dynamic physical and psychological patterns of recovery. Machine learning algorithms have traditionally focused on one aspect of surgery outcomes, but this new approach acknowledges the multidimensional nature of recovery, providing a more comprehensive understanding of the patient’s progress.
Utilizing mobile health data from Fitbit devices, researchers were able to monitor and compare activity levels over time to better predict post-surgery outcomes. By combining activity data with longitudinal assessments, the model can provide a more accurate prediction of the patient’s recovery. The team has laid out a statistical protocol to ensure that the AI is fed a balanced diet of data, improving the accuracy of its predictions. Previous research has already shown that patient-reported and objective wearable measurements can enhance predictions of early recovery, making Fitbit data a valuable tool in assessing post-operative progress.
The researchers also collected data on patients’ social and emotional states using ecological momentary assessments (EMAs). By combining wearables, EMAs, and clinical records, they were able to capture a broad range of information about the patients, including physical activities, pain levels, mental health, and clinical characteristics. State-of-the-art statistical tools such as Dynamic Structural Equation Modeling were essential in analyzing this complex, longitudinal data. The new machine learning technique, Multi-Modal Multi-Task Learning (M3TL), effectively combines these different types of data to predict multiple recovery outcomes, taking into account the relatedness and differences among the outcomes.
The M3TL approach considers shared information on interrelated tasks and leverages this information to make accurate predictions. By predicting post-operative pain interference and physical function scores for each patient, the model can provide valuable insights into potential long-term outcomes. The study is ongoing, with researchers continuing to refine their models to provide more detailed assessments and identify factors that can be modified to improve outcomes. This research was made possible through funding from various organizations, including AO Spine North America, the Cervical Spine Research Society, the Scoliosis Research Society, and the National Institute of Mental Health, demonstrating the importance of multidisciplinary collaboration and innovative approaches to predicting surgical outcomes.