Item reduction of the patient-rated wrist evaluation using decision tree modelling

Abstract

Background: The aim of this study is to assess the viability of a decision tree version of an often used questionnaire to measure wrist pain and disability, the Patient Rated Wrist Evaluation.Methods: Patient Rated Wrist Evaluation scores were collected from a cohort of 10394 patients who are part of a routine outcome measurement system. A decision tree version of the Patient Rated Wrist Evaluation (PRWE) was created. The intraclass correlation was used to evaluate the inter-version reliability between the original PRWE and the decision tree version.Results: The decision tree reduced the number of questions from 5 to 3 for the pain subscale, and from 10 to 3 for the disability subscale. The intraclass correlation between the original PRWE and the decision tree version was 0.97. The mean difference between the Patient Rated Wrist Evaluation and the decision tree Patient Rated Wrist Evaluation total sumscore was 0.35 (95% CI −9.92–10.62).Conclusions: We found that the decision tree was successful at reducing the items of the Patient Rated Wrist Evaluation from fifteen to only six questions with very high similarity to the scores of the full questionnaire.Implications for rehabilitationThe Patient Rated Wrist Evaluation can reliably be used with 6 instead of 15 questions.Decision trees are useful statistical tools to shorten lengthy questionnaires, especially when large amounts of data are available.Having a shortened Patient Rated Wrist Evaluation saves patients and clinicians time in answering this specific questionnaire.