Real-Time Symptom and Wearable Monitoring May Improve Prediction of Postoperative Recovery in Lumbar Spine Surgery Beyond Standard Measures
Salim Yakdan, Ziqi Xu, Jingwen Zhang, Faraz Arkam, Thomas L. Rodebaugh, Burel R. Goodin, Brian J. Neuman, Michael P. Steinmetz, Michael P. Kelly, Shay Bess, Jang Yoon, Wilson Z. Ray, Chenyang Lu, Madelyn R. Frumkin, Jacob K. GreenbergBACKGROUND AND OBJECTIVES:
Accurately identifying candidates likely to benefit from surgery and addressing modifiable preoperative risk factors are central to optimizing outcomes. Current prediction tools often rely on static clinical data and patient-reported measures, which lack granularity for precision risk stratification. This study aims to determine whether preoperative mobile health (mHealth) assessments combining ecological momentary assessment (EMA) and wearable biometric monitoring improve prediction of postoperative outcomes after lumbar spine surgery.
METHODS:
Patients aged 21 to 85 years undergoing elective lumbar surgery for degenerative disease were enrolled up to 30 days before surgery. Participants completed EMA surveys up to 5 times daily for pain, disability, depression, and catastrophizing. Fitbit devices recorded activity and physiological data continuously. Participants also completed one-time retrospective self-report measures. Primary outcomes were achievement of substantial clinical benefit (SCB) at 12 months in Patient-Reported Outcomes Measurement Information System Pain Interference and disability scores. Secondary outcomes included SCB in Patient-Reported Outcomes Measurement Information System Physical Function and Numeric Rating Scale for leg and back pain. Predictive machine learning models were developed and evaluated using area under the receiver operating characteristic curve and precision-recall curve. Models using only one-time retrospective self-report data were compared with models incorporating mHealth features.
RESULTS:
Of 184 enrolled patients, 138 had sufficient data and were included in the analysis. Participants [median age, 62.4 years (IQR, 13.9); 55% female] completed 10 387 EMA surveys with a median of 78 responses per participant (IQR, 33.5). At 12-month, SCB was achieved by 66 patients for pain interference, 80 for disability, 81 for physical function, 81 for leg pain, and 89 for back pain. Compared with traditional models, mHealth models improved area under the receiver operating characteristic curve by 18.3% to 32.1% across outcomes.
CONCLUSION:
Preoperative mHealth assessments improve prediction of surgical outcomes compared with traditional assessments. If validated further, this workflow could enhance patient selection and outcome prediction.