To assess whether linguistic features in paragraph-recall responses can identify cognitive impairment and provide information beyond conventional memory scores.
Approach:
Participants and measures: Researchers analyzed immediate and delayed Logical Memory recall recordings from 710 Long Life Family Study participants: 598 with normal cognition and 112 with cognitive impairment.
Analysis: Language features were measured with Linguistic Inquiry and Word Count software. Features associated with cognitive status were combined into immediate- and delayed-recall polyfeature scores and compared with traditional Logical Memory scoring. Follow-up TICS-…
Key Findings:
Eight immediate-recall and seven delayed-recall linguistic features were associated with cognitive impairment. Patterns included greater use of negations and fewer words in several categories, with differences varying by recall timing.
With demographic characteristics, the delayed-recall polyfeature score had a precision-recall AUC of 0.77, compared with 0.81 for traditional delayed Logical Memory scoring; the difference was not statistically significant. Combining both linguistic scores wit…
Linguistic scores were moderately correlated with traditional Logical Memory scores, suggesting overlapping but nonidentical information.
Among 514 participants with follow-up assessments, higher baseline delayed-recall scores were associated with lower TICS-m performance over follow-up, including after adjustment for traditional Logical Memory scores. The score did not predict a faster rate of …
In participants with normal cognition at baseline, the association between delayed-recall score and subsequent TICS-m performance was only marginal. Exploratory subtype findings were based on small groups.
Interpretation:
The authors reported that subtle features of verbal responses on a brief episodic-memory assessment were informative for detecting cognitive impairment and predicting cognitive performance over time.
Limitations:
Traditional Logical Memory scores contributed to cognitive-status classification, while linguistic measures came from the same recall responses, creating potential circularity that may have inflated predictive accuracy.
Feature selection used the full sample rather than an independent training sample, potentially producing optimistic estimates.
The language software was not designed for paragraph recall and did not assess syntax, word relationships, or acoustic and temporal speech features.
Subtype analyses were underpowered; the authors called for replication and validation in more diverse cohorts and clinical samples.
These 10 states make it more practical for physicians to participate in hospital ownership by aligning statutory structure, corporate practice of medicine rules, and population trends.
