Researchers have found a new way to predict how patients will recover from spine surgery by combining AI and mobile health technology. The team, led by Chenyang Lu and Jacob Greenberg, used machine learning techniques from the AI for Health Institute at Washington University in St. Louis to improve the accuracy of their predictions. This method builds on their previous work using Fitbit data to anticipate surgical outcomes.The findings, which were recently featured in the journal Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies, demonstrate the superiority of their model in predicting the outcomes of spine surgery compared to previous models. This is significant because the success of lower back surgery and other orthopedic procedures can differ greatly based on the patient’s structural condition, as well as their physical and mental health attributes. It’s been established that the recovery process following surgery is impacted by the patient’s preoperative physical and mental health. Additionally, some individuals may experience catastrophic thoughts during this time.Some people experience anxiety or excessive worry when dealing with pain, which can actually make the pain and recovery process worse. Others may have physiological issues that make their pain even worse. If doctors can anticipate these challenges for each patient, it will enable them to create more personalized treatment plans.
“By predicting the outcomes before the surgery, we can help establish some expectations and help with early interventions and identify high risk factors,” said Ziqi Xu, a PhD student in Lu’s lab and first author on the paper.
Previous methods of predicting surgery outcomes typically involved using patient questionnaires given once or twice in clinics.The study found that traditional methods of measuring recovery only capture a single moment in time. Xu pointed out that these methods fail to account for the long-term physical and psychological changes in patients. Previous machine learning algorithms focused on a narrow aspect of surgery outcomes, neglecting the multidimensional nature of recovery. Researchers have utilized mobile health data from Fitbit devices to track recovery and compare activity levels over time. However, they have determined that combining activity data with longitudinal assessment data provides a more accurate prediction of a patient’s post-surgery recovery.According to Greenberg, the current research demonstrates the potential of using multimodal machine learning to give doctors a more accurate overview of the various factors that impact recovery. Before this study, the team established the statistical methods and protocol to ensure that the AI receives the appropriate data. Previously, the team also published a preliminary proof of concept in Neurosurgery, which indicated that patient-reported and objective wearable measurements can enhance predictions of early recovery compared to traditional patient assessments.In a study by Frumkin, Ray, Rodebaugh, and Lu, they found that Fitbit data can be linked to various surveys that measure an individual’s social and emotional well-being. The research also utilized ecological momentary assessments (EMAs) through smartphones to gather this data. Rodebaugh is currently based at the University of North Carolina at Chapel Hill.nts receive frequent reminders to evaluate their mood, pain levels, and behavior multiple times throughout the day.
“We use a combination of wearables, EMA, and clinical records to gather a wide range of information about the patients, including physical activities, subjective reports of pain and mental health, and clinical characteristics,” Lu explained.
Greenberg also noted that advanced statistical tools, such as “Dynamic Structural Equation Modeling,” were crucial in analyzing the complex, longitudinal EMA data, with contributions from Rodebaugh and Frumkin.
In the most recent study, they then considered all of these factors and conducted an analysis.
Researchers have developed a new machine learning technique called “Multi-Modal Multi-Task Learning (M3TL)” to effectively combine different types of data in order to predict multiple recovery outcomes.
With this approach, the AI learns to assess the relatedness among the outcomes while also capturing their differences from the multimodal data, Lu explains.
This method involves using shared information on interrelated tasks of predicting different outcomes, and then utilizing that shared information to help the model understand how to make an accurate prediction, according to Xu.
All of this comes together in the final package to produce a predicted change for each outcome.The study is still in progress as researchers work on refining their models to make more accurate predictions about post-operative pain and physical function scores. They are also trying to identify factors that can be modified to improve long-term outcomes. The study received funding from several organizations including AO Spine North America, the Cervical Spine Research Society, the Scoliosis Research Society, the Foundation for Barnes-Jewish Hospital, Washington University/BJC Healthcare Big Ideas Competition, the Fullgraf Foundation, and the National Institute of Mental Health.Greenberg,
