Test–retest reliability and practical suitability of cognitive outcomes for monitoring change in Parkinson’s disease
Аннотация
Cognitive tests are commonly used to evaluate the effects of cognitive and exercise interventions in Parkinson’s disease, but their ability to distinguish genuine change from measurement variability remains insufficiently characterized. This study examined the test–retest reliability, measurement error, and practical suitability of executive and related cognitive outcomes in people with Parkinson’s disease. Participants completed the same computerized and verbal cognitive battery on two occasions separated by approximately five to six weeks. Relative reliability was evaluated using intraclass correlation coefficients, while absolute reliability was examined using the standard error of measurement, minimal detectable change at the 95% confidence level, systematic between-session change, and Bland–Altman analysis. Phonemic fluency showed the most favorable relative reliability, whereas forward Corsi span demonstrated a consistent agreement profile without marked systematic bias or heteroscedasticity. Trail Making Test Part B accuracy and a parsimonious global cognitive composite showed moderate reliability but required caution because of heteroscedasticity or relatively large detectable-change thresholds. Several working-memory, action-fluency, planning, and card-sorting outcomes showed limited reliability or substantial measurement error. Completion time for the Tower of London improved systematically at retest, indicating a marked familiarization effect. These findings provide preliminary evidence that cognitive measures commonly used in Parkinson’s disease differ substantially in their short-term measurement profiles. Given the modest sample and the resulting uncertainty around several intraclass correlation coefficient estimates, the practical classifications should be interpreted as provisional and confirmed in larger independent cohorts. Outcome selection should consider absolute measurement error, practice effects, and intended group- or individual-level interpretation rather than relying on intraclass correlation coefficients alone.