Comparison of computational methods for the evaluation of learning potential in schizophrenia

Learning potential (LP) refers to the ability to improve cognitive performance as a result of training. It is typically assessed by test-train-test administrations of a task, wherein changes in pre-post performance are an index of LP. In schizophrenia research, LP has been suggested as a mediator of the relationship between static neurocognition and functional outcome. While a number of studies do indicate that LP assessment improves prediction of functioning beyond standard administrations of the same task, multiple approaches of computing LP indices have been used in this work. Multiple psychometric issues have been raised with respect to computation of change scores, but have not been widely recognized in LP assessment. To address this issue, the current study aimed to evaluate the test-retest reliability, interrelatedness, construct, and criterion validity of several conventional indices of LP, obtained from a test-train-test version of a list-learning task administered to 43 individuals with chronic schizophrenia. Overall, test-retest and intercorrelation coefficients indicated variable reliability and convergence across methods. While LP indices generally correlated more highly with independent measures of neurocognition and community functioning than pretraining list learning scores, coefficients were comparably small. Recommendations and measurement issues inherent to the LP construct are discussed.