Introduction

As many as 41% of individuals who meet consensus criteria for a diagnosis of PPA do not meet criteria for any one variant. Our aim was to determine whether performance during cognitive-linguistic evaluation in which a patient receives an early unclassifiable PPA designation predicted the later emergence of a given variant.

Methods

Nineteen English-speaking individuals with an initially unclassifiable variant of PPA were followed until features sufficient to identify a variant emerged. Patients were evaluated using numerous word-finding tasks targeting nouns and verbs. Receiver operating characteristic curves were used to evaluate the ability of a given task to predict eventual classification as a given variant. Tasks with an area under the curve ≥ 0.75 were examined using regression analyses to determine their ability to predict variant among the three. Optimal cut-point scores were identified for any significant independent factor in predicting variant.

Results

High mean AUROC was observed for written noun naming, Boston Naming Test, Hopkins Action Naming Assessment, and Berndt verbs. The BNT was the only test that, in isolation, resulted in a significant model, χ²(32)=49.91, p=0.02, and high accuracy in classification. Optimal cut-point scores to identify each variant using the BNT were: ≥ 28/30 for nfavPPA, ≤ 18/30 for svPPA, and ≥ 20/30 for lvPPA, but did not perform as well at that cut-point.

Conclusions

Word-finding tasks performed best in parsing between eventual variants, despite this being a common deficit in PPA. The 30-item BNT emerged as a uniquely accurate basis for predicting which variant would emerge.