Reliability of Regression-Based Normative Data for the Oral Symbol Digit Modalities Test: An Evaluation of Demographic Influences, Construct Validity, and Impairment Classification Rates in Multiple Sclerosis Samples

Abstract

The oral Symbol Digit Modalities Test (SDMT) has been recommended to assess cognition for multiple sclerosis (MS) patients. However, the lack of adequate normative data has limited its clinical utility. Recently published regression-based norms may resolve this limitation but, because these norms were derived from a relatively small sample, their stability is unclear. We aimed to evaluate the stability of regression-based SDMT norms by comparing existing norms to a cross-validation dataset. First, regression-based normative data were created from a similarly-sized, independent, control sample (n = 94). Next the original and cross-validation norms were compared for equivalency, management of demographic influences, construct validity, and impairment classification rates in a mildly affected MS sample (n = 70). Lastly, similar comparisons were made for a large, representative MS clinic sample (n = 354). We found construct validity and management of demographic influences were equivalent for the two sets of regression-based norms but lower T-scores were obtained using the original dataset, resulting in discrepancies in impairment classification. In conclusion, regression-based norms for the oral SDMT attenuate demographic influences and possess adequate construct validity. However, norms generated using small samples may yield unreliable classification of cognitive impairment. Larger, representative databases will be necessary to improve the clinical utility of regression-based norms.

Keywords:

• Symbol Digit Modalities Test
• Normative data
• Psychometrics
• Multiple sclerosis
• Neuropsychology.

Notes

¹ Given the established relation between age and processing speed (Salthouse, 1996), it is unexpected that age was not a significant predictor of performance for the original sample. This might have occurred due to issues relating to multicollinearity as age also did not emerge as a significant predictor for the cross-validation sample when the age variable was not centred before conducting the regression analysis.