Reliability of delirium rating scale (DRS) and delirium rating scale-revised-98 (DRS-R98) using variance-based multivariate modelling.

Delirium's characteristic fluctuation in symptom severity complicates the assessment of test-retest reliability of scales using classical analyses, but application of modelling to longitudinal data offers a new approach. We evaluated test-retest reliability of the delirium rating scale (DRS) and delirium rating scale-revised-98 (DRS-R98), two widely used instruments with high validity and inter-rater reliability. Two existing longitudinal datasets for each scale included DSM-IV criteria for delirium diagnosis and repeated measurements using the DRS or DRS-R98. To estimate the reliability coefficients RT and RΛ for each scale we used a macros provided by Dr. Laenen at http://www.ibiostat.be/software/measurement.asp. For each dataset a linear mixed-effects model was fitted to estimate the variance-covariance parameters. A total of 531 cases with between 4 and 9 measurement points across studies including both delirious and non-delirious patients. Comorbid dementia in the datasets varied from 27% to 55%. Overall RT for the DRS were 0.71 and 0.50 and for DRS-R98 0.75 and 0.84. RΛ values for DRS were 0.99 and 0.98 and for DRS-R98 were 0.92 and 0.96. Individual RT measures for DRS-R98 and DRS across visits within studies showed more range than overall values. Our models found high overall reliability for both scales. Multiple factors impact a scale's reliability values including sample size, repeated measurements, patient population, etc in addition to rater variability.