Ranking qualitative and timing‐based verbal fluency scores as predictors of incident cognitive impairment

Background Verbal fluency tasks are a valuable tool for neuropsychological assessment of cognitive impairment. In recent decades, researchers have proposed qualitative scores, such as clustering and switching , to index cognitive processes underlying fluency. However, the best approach to computing these scores is not clear, and no study addresses the handling of erroneous utterances. Method We transcribed animals and letter F fluency recordings on 703 cases of ICI and matched controls from the REGARDS study, amending each transcription with word timings. We then calculated clustering and switching scores with three methods described in the literature and new scores indexing speed of switch and edge transitions. After fitting a base model with only counts of valid words, perseverations, and intrusions, we fit a series of logistic regression models adding either clustering and switching scores or switch and edge speed scores, ranking the models with Watanabe‐Akaike Information Criterion. Results For animal fluency, we found that a model with switch‐edge speeds dominated the other models. Clustering and switching, when defined by the slope difference algorithm, offered a slim improvement over the base model. For letter fluency, none of the novel scores we computed yielded an improvement over the base model. A model with animal and letter fluency variables offered further improvement. Conclusion Switch‐edge speed scores improve on a base model predicting ICI from animal fluency. Animals and letter F fluency provide complementary prognostic information.