Researchers at Washington University in St. Louis have developed a machine learning (ML) approach that uses data from wearable devices and other sources to predict recovery outcomes after lumbar spine surgery.
The research team emphasized that predicting outcomes after orthopedic surgery is key to improving clinical decision-making and personalizing treatment. The outcome of spine surgery can vary greatly depending on the patient's structural disease, as well as their physical and mental health characteristics, such as excessive stress or physiological problems leading up to the surgery.
The researchers further noted that existing models for predicting spine surgery outcomes typically rely on patient-reported outcome measures (PROMs), which are limited in their ability to capture insight into a patient's long-term condition before surgery.
Many predictive tools also focus on a single surgical outcome, but recovery is multidimensional and consists of a variety of unique yet related outcomes, including physical function, pain interference, and quality of recovery.
The advent of wearable devices and smartphones offers the opportunity to more effectively collect longitudinal patient data outside of clinical settings, and previous work by researchers in this field has demonstrated that PROMs and wearable data can improve outcome prediction after surgery.
The study correlated Fitbit data with ecological momentary assessment (EMA) data. This information, which captures participants' emotional and social states, was collected using their smartphones. To improve on this approach, the researchers turned to ML.
“We combine wearables, EMA and clinical records to collect a wide range of information about patients, from physical activity to subjective reports of pain and mental health to clinical characteristics,” Chengyang Lu, PhD, Fulgraf Professor in the university's McKelvey School of Engineering, said in a news release.
In doing so, ML can provide a more accurate “big picture” of factors that affect surgical recovery and how they relate to each other. The approach, called “multimodal multitask learning,” the researchers explained, emphasizes the relevance of outcomes to help with predictions.
This analysis provides a predicted change score for physical function and pain interference that can guide treatment. Tests showed that the ML model outperformed standard prediction approaches.
“Predicting outcomes before surgery can help us set some expectations, aid in early intervention and identify high-risk factors,” said Zhiqi Xu, a doctoral student in Lu's lab and lead author of the study.
According to Jacob Greenberg, MD, assistant professor of neurosurgery in the School of Medicine, the researchers plan to continue to fine-tune the approach to enable more accurate predictions and help clinicians “understand what factors we might be able to modify to improve long-term outcomes.”
These efforts reflect growing interest in how predictive analytics can enhance healthcare, with AI and ML increasingly being applied to improve risk stratification and patient outcomes.
