Study design: Analysis of Roland Morris Disability Questionnaire (RMDQ) and Oswestry Disability Index (Oswestry) responses.
Objective: To determine the prevalence of unanswered questions on the RMDQ23 (23-item RMDQ version) and Oswestry questionnaires. To determine whether managing RMDQ23 missing data using proportional recalculation is more accurate than simply ignoring missing data.
Summary of background data: It is likely that the most common method for calculating an RMDQ sum score is to simply ignore any unanswered questions. In contrast, the raw sum score on the Oswestry is converted to a 0 to 100 scale, with the advantage of allowing missing data to be accommodated by proportional recalculation.
Methods: The prevalence of unanswered RMDQ23 questions was measured in a research project and a routine care setting. The accuracy of the RMDQ23 proportional recalculation method was measured using 311 fully completed RMDQ23 and matching Oswestry questionnaire sets. Raw sum scores were calculated, and questions systematically dropped. At each stage, sum scores were converted to a score on a 0 to 100 scale and the error calculated. Wilcoxon Tests were used to compare the magnitude of the error scores.
Results: The prevalence of people who did not answer one or more questions was 29.5% (RMDQ23) in routine care, and 13.9% (Oswestry) and 20.3% (RMDQ23) in a research project. Proportional recalculation was a more accurate method to calculate RMDQ sum scores than simply ignoring missing data, when two or more questions were unanswered.
Conclusion: Because of less error when missing data are present, the most accurate method for expressing RMDQ sum scores collected using Yes/No answers is conversion to a 0 to 100 scale. This conversion method is (a) if all questions are answered or only one question is unanswered, multiply the raw sum score by 100 divided by the total number of questions, and (b) if two or more questions are unanswered, multiply the raw sum score by 100 divided by the number of answered questions.