Predicting self-image satisfaction after adult spinal deformity surgery: a machine learning approach using patient phenotypes

Abstract

PurposeTo characterize the evolution of the Self-Image (SI) domain of the Scoliosis Research Society-22 item questionnaire (SRS-22) in adult spinal deformity (ASD) patients undergoing surgical correction and to identify predictors of achieving the patient-acceptable symptom state (PASS) both preoperatively and at 2-year follow-up.MethodsThis retrospective analysis of a prospective, multicenter registry included 710 surgically treated ASD patients with a minimum 2-year follow-up. Patients were stratified into three validated phenotypic clusters: Young Coronal (YC), Old Moderate (OM), and Old Severe (OS). Random Forest models were developed to predict achieving the PASS for the SRS-22 SI domain (score >= 3.3).ResultsPreoperatively, only 13% of patients were SI PASS + (YC: 28%; OM: 23%; OS: 1%). Postoperatively, this proportion increased to 65% (YC: 85%; OM: 68%; OS: 52%). Functional disability (Oswestry Disability Index), pain (Visual Analog Scale), and physical/social functioning (SF-36) were the strongest predictors of SI, consistently outperforming radiographic parameters and complications across all models. Predictive accuracy for achieving SI PASS + at 2 years was high, with Area Under the Curve (AUC) values ranging from 0.73 to 0.94 depending on the cluster.ConclusionsSI is severely impaired in most ASD patients but improves substantially after surgery. Functional status and pain are more influential than spinal alignment or complications in determining SI satisfaction. These findings challenge the traditional radiograph-centric evaluation of surgical success and underscore the importance of SI as a key indicator of the holistic patient experience in ASD.