"See What I Mean?" Linguistic Digital Markers of Cognitive Decline in Nursing Home Residents

Cognitive decline in nursing homes is often under-recognized. Access to specialized neuropsychological assessments, which involve detailed evaluations of how the brain influences behavior and thinking, is often limited. In this study, we examined the feasibility of administering an ecological reading task, involving reading activities that imitate real-life situations, in a real-world nursing home setting. We also explored whether reading-derived linguistic metrics and measures of language use during reading, along with an eye-tracking component that monitored participants' eye movements, were associated with cognitive impairment.

This cross-sectional observational pilot study included 60 nursing home residents aged 65 years or older, classified as either cognitively impaired (CI, n = 30) or healthy control (HC, n = 30) based on neuropsychological profiles and clinical evaluations. All participants completed the Mini-Mental State Examination (MMSE), Repeatable Battery for the Assessment of Neuropsychological Status (RBANS), Frontal Assessment Battery (FAB), Geriatric Depression Scale (GDS), Barthel Index, and Cognitive Reserve Index questionnaire (CRIq). The reading task involved reading aloud a 177-word Italian passage followed by eight comprehension questions. Primary outcomes were total reading errors (TRE), total comprehension errors (TCE), and words per minute (WPM). Eye-tracking data (Gazepoint GP3, 60 Hz) were available for a subsample with usable calibration and validation data (CI: n = 26; HC: n = 24). Analyses included descriptive comparisons, covariate-adjusted generalized linear models, receiver operating characteristic (ROC) analyses, and partial Spearman correlations.

All 60 participants completed the task. Usable audio recordings were obtained for everyone, and usable eye-tracking data were available for 50 participants. The CI group showed higher TRE and TCE, lower WPM, and longer reading times compared with the HC group. In models adjusted for gender, age, education, GDS, Barthel Index, and CRIq, TCE showed the strongest association with CI status (odds ratio [OR] = 11.108, 95% CI 2.662-246.186, p = 0.024), and TRE was also associated with CI status (OR = 1.238, 95% confidence interval [CI] 1.054-1.629, p = 0.045). ROC analyses showed high areas under the curve (AUCs) for TCE (0.982), TRE (≈0.95), and audio-based WPM (0.936). In the eye-tracking subsample, timing-related measures also showed good discrimination (AUC = 0.960 for eye tracking [ET]-WPM; 0.942 for ET-Total Reading Time), whereas conventional first-order oculomotor metrics did not differ significantly between groups.

An ecological reading task was feasible within a nursing home setting. It generated reading-derived linguistic metrics linked to cognitive impairment and broader cognitive-functional status. Measures related to comprehension, overall reading error load, and reading speed were shown to be useful as digital linguistic markers.