MANAGING MISSING SCORES ON THE ROLAND MORRIS DISABILITY QUESTIONNAIRE Peter Kenta and Henrik Hein Lauridsenb aBack Research Centre and bInstitute of Sports Science and Clinical Biomechanics, University of Southern Denmark Background There is no standard method to calculate Roland Morris Disability Questionnaire (RMDQ) sum scores when one or more questions have not been answered. However, missing data are common on the RMDQ and the current options are: calculate a sum score regardless of unanswered questions, reject all data containing unanswered questions, or to impute scores. Other questionnaires, such as the Oswestry Disability Index (Oswestry) convert their raw score into a standardized score out of 100. An advantage of this method is that it allows missing data to be accommodated by proportional recalculation. For example, if 17 questions had been answered ’yes’ on a RMDQ questionnaire containing 23 yes/no questions, a raw sum score of 17 would be a standardized score of 74/100 if all 23 questions were answered (17/23 x 100/23). However, the raw sum score of 17 would be proportionally recalculated to a standardized score of 85/100 if 3 questions had not been answered (17/20 x 100/20). Quantification of the measurement error introduced when using this method would allow determination of whether it is a valid strategy. The aim of this study was to quantify the measurement error in RMDQ (standardized) scores when one or more questions were unanswered and to compare this with the measurement error present when the Oswestry was scored in the same way. Methods · For each of 311 fully completed RMDQ23 questionnaires from people seeking primary or secondary care, a sum score was calculated and standardized to a 100-point scale. · Using random number generation, questions were systematically dropped from each person’s raw scores and the standardized score was proportionally recalculated. This process was repeated until 6 questions had been dropped from each person’s questionnaire. · The error (absolute and percentage) introduced by each level of dropped question was calculated for each person’s score. The absolute error was the number of standardized RMDQ points that the proportionally recalculated score varied from the original standardized score. The percentage error was the absolute error expressed as a proportion of the original standardized score. · This process was repeated using the Oswestry scores from the same people. Results · The mean standardized score for the RMDQ23 was 49/100 (SD25, range 4-91) and for the Oswestry was 37/100 (SD13, range 3-74). · The absolute error from a RMDQ23 standardized score ranged from a mean of 2 (SD2, range 0-9) when 1 question was dropped to a mean of 4 (SD3, range 0-20) when 6 questions were dropped. · The proportional error from a RMDQ23 standardized score ranged from a mean of 6.6% (SD5.8%, range 0.0%-38.3%) when 1 question was dropped to a mean of 13.9% (SD15.9%, range 0.4%-100.0%) when 6 questions were dropped. · These results were equal to or less than those found when dropping the same number of questions from the Oswestry. Conclusions The practice of expressing RMDQ scores as a standardized score allows missing data to be accommodated by proportional recalculation and is as valid for the RMDQ23 as it is for the Oswestry.