Scoliosis Research Society-22r and Ceiling Effects: Limited Capabilities for Precision Medicine With Adolescent Idiopathic Scoliosis.

Abstract

Study design: Retrospective registry analysis.

Objective: To examine predictions of individual Scoliosis Research Society-22r (SRS-22r) questions one year after surgery for adolescent idiopathic scoliosis (AIS).

Summary of background data: A precision-medicine approach to AIS surgery will inform patients of the likelihood of achieving particular results from surgery, specifically individual responses to the SRS-22r questionnaire.

Methods: A multicenter AIS registry was queried for surgical AIS patients treated between 2002 and 2020. Preoperative data collected included standard demographic data, deformity descriptive data, and SRS-22r scores. Postoperative one-year SRS-22r scores were modeled using ordinal logistic regression. The highest probability was the most likely response. Model performance was examined by c-statistics, where c >0.8 was considered excellent. Ceiling effects were measured by the proportion of patients reporting "5" to each question.

Results: A total of 3251 patients contributed data to the study; mean age 14.4 (±2.2) years, female 2631 (81%), major thoracic coronal curve 53°, mean lumbar 41°. C-statistic values ranged from 0.6 (poor) to 0.8 (excellent), evidence of varied predictive capabilities. Q17 ("days off work/school," c = 0.84, ceiling achieved 75%) and Q15 ("financial difficulties," c = 0.86, ceiling achieved 82%) had the greatest predictive capabilities while Q11 ("pain medication," c=0.73, ceiling achieved 67%), Q10 ("appearance," c=0.72, ceiling achieved 35%), and Q19 ("attractive," c=0.69, ceiling achieved 37%) performed poorly.

Conclusions: Prediction of individual SRS-22r item responses, perhaps most germane to AIS treatment, was poor. The prediction of less relevant outcomes, where ceiling effects are present, was greater as the models chose "5" for all responses. These ceiling effects may limit discrimination and hamper efforts at personalized outcome predictions.

Level of evidence: Level 3.