Inter-university Neuroscience & Mental Health Conference
Statistical modelling of post-stroke aphasia recovery
Sydney, NSW, Australia
/University of Technology Sydney
/University of Technology Sydney/DVC (Research)
/University of Technology Sydney/DVC (Teaching and Learning)
Background: Aphasia is a communication disorder caused by damage to the language centres of the brain due to Traumatic Brain Injury or Stroke. The Western Aphasia Battery Aphasia Quotient (AQ) is a standardised test battery with high testretest reliability but is negatively affected by a ceiling effect. Maximal Potential Recovery (MPR) is calculated from the AQ by dividing the improvement in AQ (from baseline) by the maximum achievable improvement. The MPR was introduced to overcome the ceiling effect of the AQ. This work assesses the use of MPR to model post-stroke aphasia recovery to determine the most statistically appropriate model. Methods: Secondary analysis of data from two Australian randomised controlled trials was used to compare competing statistical models. Data from study 1 (N=59) was used to develop the models with different outcome measures: a) AQ and b) MPR. Data from study 2 (N=20) was used to validate and compare the models. Quasi Information Criterion (QIC), Residual plots and Residual Sum of Squares (RSS) were compared across the two models to determine the optimal aphasia measure. Results: The MPR model demonstrates substantially lower values for QIC and RSS. It also demonstrates considerable improvement in residual scores compared to the AQ model. The lower values of QIC together with the residual plot indicate a better statistical model fit; the lower value of RSS indicates greater accuracy in predicting recovery. Conclusions: Modelling aphasia recovery on MPR provides a better fit than AQ. This provides a more accurate prediction of recovery.