Posthospitalization COVID-19 cognitive deficits at 1 year are global and associated with elevated brain injury markers and gray matter volume reduction

Abstract

The spectrum, pathophysiology and recovery trajectory of persistent post-COVID-19 cognitive deficits are unknown, limiting our ability to develop prevention and treatment strategies. We report the 1-year cognitive, serum biomarker and neuroimaging findings from a prospective, national study of cognition in 351 COVID-19 patients who required hospitalization, compared with 2,927 normative matched controls. Cognitive deficits were global, associated with elevated brain injury markers and reduced anterior cingulate cortex volume 1 year after COVID-19. Severity of the initial infective insult, postacute psychiatric symptoms and a history of encephalopathy were associated with the greatest deficits. There was strong concordance between subjective and objective cognitive deficits. Longitudinal follow-up in 106 patients demonstrated a trend toward recovery. Together, these findings support the hypothesis that brain injury in moderate to severe COVID-19 may be immune-mediated, and should guide the development of therapeutic strategies.

Fig. 1. Flow diagram of patients included from the COVID-CNS.

Nationally at least 16,279 patients were screened of whom at least 2,712 were eligible. Matched community data were collected separately and held in a large normative database. ‘Other’ includes autonomic dysfunction, cerebral hypoxic injury, headache, headache and fatigue, hyperkinetic movement disorder, Parkinsonian movement disorder, seizures and speech and sensory. ^aSix patients with ‘anosmia/ageusia’ reclassified as COVID from NeuroCOVID. The parentheses show n with biomarkers, n with neuroimaging. MOCA, Montreal Cognitive Assessment.

Fig. 2. Cognitive scores and recovery trajectories.

a, Violin plot of DfE cognition scores by diagnostic group including median (IQR) (black). Statistics compare each group with normative data, n = normative (2,927), cerebrovascular, encephalopathy, inflammatory, neuropsychiatric, peripheral and other. Exact P values are listed in Supplementary Table 8. b, Pattern of deficits in clinical groups by median DfE accuracy and responsive time minus matched community controls across six cognitive tasks. Exact effect sizes and P values in listed Supplementary Table 9. c, Recovery trajectories in NeuroCOVID and COVID patients following postacute assessment. A black dot indicates a single observation, lines connect paired observations between postacute assessment and follow-up 1, and follow-up 1 and follow-up 2. Center line, median; box limits, upper and lower quartiles; whiskers, 1.5× IQR; the dashed line shows normal cognition; the numbers under the x axis show n for each assessment. *P < 0.05, **P < 0.01, ***P < 0.001 two-sided Mann–Whitney U-test, adjusted for multiple comparisons based on the FDR approach in a and b (adjusted for n = 8 and n = 12 comparisons, respectively). 2D, two-dimensional; NS, nonsignificant.

Fig. 3. Brain injury biomarkers and neuroimaging.

a, Brain injury markers (in pg ml⁻¹) by diagnostic group. Lower limit of quantification (dashed line) if included in scale. Normative values from n = 60 healthy controls. *P < 0.05, **P < 0.01, ***P < 0.001; NS, nonsignificant, unadjusted two-sided Mann–Whitney U. b, Brain regions represented by the IDPs utilized in analyses. These regions are parcellated as per the Desikan-Killiany cortical atlas. For each region and regions combined, IDP composites for thickness and volume were utilized. ^aIDP composites that have significant correlations with overall cognition (Supplementary Table 2). Created using Matlab and BrainNet Viewer. c, Scatter plots for IDP composite z-scores against GDfE in the overall cohort, using Pearson’s correlation. The black line indicates the line of best fit (least squares method) and gray error band indicates the 95% CI. Significance persisting after adjusted for multiple comparisons based on the FDR approach (adjusted for n = 14 for each IDP composite test). Center line, median; box limits, upper and lower quartiles; whiskers, 1.5× IQR. ACC, anterior cingulate cortex; ERC, entorhinal cortex; UCH-L1, ubiquitin carboxy-terminal hydrolase L1.

Fig. 4. Heatmap and unsupervised cluster analysis.

Heatmap and unsupervised cluster analysis (Euclidean, complete) in the full cohort (n = 351) of cognitive tasks shaded by correlation (Spearman), including cognition (accuracy and inverse RT), clinical variables, biomarkers and neuroimaging. *P < 0.05, **P < 0.01, ***P < 0.001 two-tailed Spearman correlation adjusted for multiple comparisons (37 ×3 7 matrix, n = 1,369) using the FDR approach. ACB, anticholinergic burden; BIB, brain injury marker; COG, cognitive task; OFC, orbitofrontal cortex; PHG, parahippocampal gyrus; SR, self-report; STG, superior temporal gyrus.