Acute blood biomarker profiles predict cognitive deficits 6 and 12 months after COVID-19 hospitalization

Abstract

Post-COVID cognitive deficits, including 'brain fog', are clinically complex, with both objective and subjective components. They are common and debilitating, and can affect the ability to work, yet their biological underpinnings remain unknown. In this prospective cohort study of 1,837 adults hospitalized with COVID-19, we identified two distinct biomarker profiles measured during the acute admission, which predict cognitive outcomes 6 and 12 months after COVID-19. A first profile links elevated fibrinogen relative to C-reactive protein with both objective and subjective cognitive deficits. A second profile links elevated D-dimer relative to C-reactive protein with subjective cognitive deficits and occupational impact. This second profile was mediated by fatigue and shortness of breath. Neither profile was significantly mediated by depression or anxiety. Results were robust across secondary analyses. They were replicated, and their specificity to COVID-19 tested, in a large-scale electronic health records dataset. These findings provide insights into the heterogeneous biology of post-COVID cognitive deficits.

Fig. 1. Factors associated with post-COVID cognitive deficits.

Association between baseline characteristics and MoCA at 6 months (measuring objective cognitive deficits, lower indicate more deficits) and C-PSQ at 6 months (measuring subjective cognitive deficits, higher means more deficits). Age was z-transformed in this analysis which means that the coefficient corresponds to a difference in MoCA/C-PSQ corresponding to a difference of 1 × s.d. in age. Only one level of education and one income level are presented (with no education and income <£19,000 being taken as references respectively). The same graphs with all education and income levels, as well as ethnicity, are presented in Supplementary Figs. 1 and 2. n = 1,837 individual participants. Dots indicate point estimates and horizontal lines indicate 95% CI. P values were estimated as part of a generalized linear model and are two-sided and not adjusted for multiple comparisons: ^oP < 0.1, *P < 0.05, **P < 0.01, ***P < 0.001. NVQ, national vocational qualification.

Fig. 2. High fibrinogen is linked with objective and subjective cognitive deficits.

a,b, A first dimension of covariation links high fibrinogen with relatively low CRP to higher C-PSQ at 6 and 12 months (signs of subjective cognitive deficits) and lower MoCA at 6 and 12 months (signs of objective cognitive deficits). P values are derived from permutation tests, two-sided and not corrected for multiple comparisons. c–h, Distribution of different variables between the top half and the bottom half of the cohort along this first dimension (n = 768, 1,777, 1,837, 626, 1,502 and 584 individual participants, respectively). The center of the boxes represents the median, their bounds represent the 25th and 75th centiles and the lower and upper ends of whiskers represent the smallest/largest values, no further than 1.5 × IQR from the box-plot respective end. Distribution of all variables investigated is found in Supplementary Figs. 4–6. NS, P > 0.05.

Fig. 3. High D-dimer is linked with subjective but not objective cognitive deficits.

a,b, A second dimension of covariation links high D-dimer with relatively low CRP to higher C-PSQ at 6 and 12 months (signs of subjective cognitive deficits) and signs of occupational impact (in terms of affected ability to work at 6 and 12 months and occupational changes at 6 months) but little difference in MoCA. P values are derived from permutation tests, two-sided and not corrected for multiple comparisons. c–h, Distribution of different variables between the top half and the bottom half of the cohort along this second dimension (n = 977, 1,777, 1,502, 584, 233 and 252 individual participants, respectively). The center of the boxes represents the median, their bounds represent the 25th and 75th centile and the lower and upper ends of whiskers represent the smallest/largest value, no further than 1.5 × IQR from the box-plot respective end. Distribution of all variables investigated can be found in Supplementary Figs. 8–10. NS, P > 0.05.

Fig. 4. Mediation by other clinical features.

Mediation of the associations captured in the first and second dimensions of covariation by scales representing other aspects of health at 6 months after COVID-19. The names of the scales are reported in brackets. P values were estimated using nonparametric bootstrap with 1,000 repetitions and are two-sided and not adjusted for multiple comparisons: ***P < 0.001. For the second dimension, the P value for fatigue was 0.004 and for shortness of breath it was <0.001 (below the minimum threshold detectable with 1,000 repetitions). FACIT, Functional Assessment of Chronic Illness Therapy; BPI, Brief Pain Inventory; SARC-F, Sarcopenia screen; ISWT, Incremental Shuttle Walk Test; GPPAQ, General Practice Physical Activity Questionnaire; SPPB, Short Physical Performance Battery; PHQ-9, Patient Health Questionnaire; PTSD, post-traumatic stress disorder; PCL-5, PTSD Checklist; GAD-7, Generalized Anxiety Disorder scale; PAI, Physical Activity Index.

Fig. 5. Replication and generalization of the findings using electronic health records data.

Kaplan–Meier curves represent the cumulative incidence of cognitive deficits between those with high versus low fibrinogen (or D-dimer) and CRP level ≤10 mg l⁻¹. The same analysis conducted in people without COVID-19 (bottom). Curves represent the Kaplan–Meier estimates and shading around curves represents 95% CI. P values are derived from log-rank tests, are two-sided and not adjusted for multiple comparisons. The same figures within the COVID-19 cohort wherein the criterion on CRP is relaxed to include all those with a CRP ≤ 20 mg l⁻¹ or is removed altogether are presented as Extended Data Fig. 5.

Extended Data Fig. 1. Contributions of each cognitive item to the dimensions of covariation.

These report the weights of each item in the weighted combinations that represent the cognitive profile for each dimension. Note that positive weights (for MoCA items) and negative weights (for C-PSQ items) do not necessarily imply that individuals who scored high on the dimension of covariation had better cognitive outcomes for those items since these items might covary with other items with opposite weights. See Supplementary Figs. 5–6, 9 and 10 for distribution of individual items along dimensions of covariation.

Extended Data Fig. 2. Association between dimensions of covariation and clusters of post-acute impairment.

Distribution of individuals in the top and bottom half of the cohort along both dimensions of covariation in terms of predefined clusters of post-acute impairment. For both dimensions, those who scored in the top half of the cohort tended to have more severe impairment.

Extended Data Fig. 3. Change in subjective cognitive function between pre- and post-COVID.

Distribution of the change in C-PSQ-2 (assessing subjective cognitive deficits) between a pre-COVID baseline and 6 months (left) or 12 months (right) post-COVID. The dashed lines represent the mean change. In both cases, the change was, on average, significantly greater than zero indicating worsening of subjective cognitive function following COVID-19 (mean [s.e.m.] change in C-PSQ-2: 0.48 [0.04] between pre-COVID and 6 months post-COVID, p < 0.0001; and 0.40 [0.055] between pre-COVID and 12 months post-COVID-19, p < 0.0001). At six months, 43/547 participants (7.9%) had better cognition, 288 (52.7%) had no change, and 216 (39.5%) had worse cognition compared to before COVID-19. At 12 months, 12/205 participants (5.9%) had better cognition, 116 (56.6%) had no change, and 77 (37.6%) had worse cognition compared to before COVID-19.

Extended Data Fig. 4. Mediation of the associations between biomarker and cognitive profiles by markers of severity of infection.

No marker reached statistical significance. Fraction explained by the mediator are reported as negative if they are negatively associated with the cognitive profile. WHO, World Health Organization clinical progression scale; ICU, Intensive care unit; NEWS, National Early Warning Scores.

Extended Data Fig. 5. Associations between biomarkers and post-acute cognitive deficits in the EHR data with different constraints on CRP.

Kaplan–Meier curves represent the cumulative incidence of cognitive deficits between those with high versus low fibrinogen (or D-dimer) and CRP level ≤ 20 mg/L (top panels), or any CRP level (bottom panels). Curves represent the Kaplan–Meier estimates and shading around curves represents 95% confidence intervals. P-values are derived from log-rank tests, two-sided, and not adjusted for multiple comparisons.

Extended Data Fig. 6. Associations between COVID-19 status and risks of venous thromboembolism and ischemic stroke among those with raised D-dimer.

Comparison between matched cohorts of patients with high D-dimer and normal CRP with COVID-19 vs. without COVID-19 in terms of risk of venous thromboembolism [VTE] (left) and ischaemic stroke (right). Curves represent the Kaplan–Meier estimates and shading around curves represents 95% confidence intervals. P-values are derived from log-rank tests, two-sided, and not adjusted for multiple comparisons.