Predicting the outcome of AIS brace treatment using expert judgement and a fuzzy model

Abstract

Methods Data was obtained retrospectively from 28 AIS patients who had finished treatment (27 girls, 1 boy, aged 11-15 (mean 13), Cobb angles 20-44 degrees (mean 31), 21 daytime and 7 nighttime braces.) Patients were labelled ‘progressed’ if their Cobb angle had increased more than 5 degrees by the end of treatment, and ‘non-progressed’ otherwise. A fuzzy model was developed to predict treatment outcome for each patient using clinical measurements taken at the first in-brace clinic. The model considers patient age, Cobb angle, Scoliometer measurement at the apex level, and in-brace Cobb angle correction. For each patient it calculated a probability-like score for each of three possible outcomes: ‘progression’ (Cobb angle increase > 5 degrees), ‘neutral’ (Cobb angle change of 0-5 degrees), and ‘improvement’ (Cobb angle decrease). For this study, the patient was predicted to progress if the ‘progression’ score was the highest. Five AIS experts also participated: two orthopaedic surgeons, two orthotists, and one nurse practitioner. Participants were supplied with all available start-oftreatment clinical measurements for each patient, and asked to predict whether or not each patient would progress by the end of brace treatment. The multi-rater kappa was calculated to measure agreement between experts’ predictions. The correlation of each expert’s predictions and the model’s predictions with the actual treatment outcome was measured.