Increased incidence of mild cognitive impairment in long COVID patients

Abstract

INTRODUCTION

Though brain fog is common in Long‐coronavirus disease 2019 (Long‐COVID), the incidence of mild cognitive impairment (MCI) is unknown.

METHODS

In an observational cohort study, recovered COVID‐positive, Long‐COVID, and COVID‐negative subjects underwent blinded evaluation using National Alzheimer's Coordinating Center (NACC) and National Institute on Aging (NIA) ‐Alzheimer's Association diagnostic criteria for dementia and MCI. The cumulative incidence of MCI was calculated for each group, and the hazard of MCI was compared between groups.

RESULTS

Among 260 subjects, the cumulative incidence of MCI over 4.4 years was higher with Long‐COVID (27%) versus recovered‐COVID (5%) or COVID‐negative status (1%). There was a higher hazard of MCI for patients with Long‐COVID compared to those without (hazard ratio [HR] 3.93, 95% confidence interval [CI] 1.86–8.31, p < 0.001), and specifically for the Alzheimer's disease (AD) ‐related MCI subtype (HR 3.20, 95% confidence interval [CI] 1.14–9.00, p = 0.027).

DISCUSSION

The cumulative incidence and adjusted hazard of MCI (and specifically AD‐related MCI) at 4.4 years was significantly higher among Long‐COVID patients compared to recovered‐COVID and COVID‐negative controls.

Highlights

• The cumulative incidence of mild cognitive impairment (MCI) was higher in Long‐coronavirus disease 2019 (Long‐COVID) compared to other groups.

• The hazard of MCI was four‐fold higher for Long‐COVID subjects.

• The hazard of Alzheimer's disease related MCI was significantly higher with Long‐COVID.

• The hazard of MCI due to psychiatric illness did not differ between groups.

1. BACKGROUND

Several studies have identified an epidemiological link between coronavirus disease 2019 (COVID‐19) and increased diagnoses of Alzheimer's disease (AD) and other dementias. ¹ , ² , ³ Indeed, Mendelian randomization, genomic, severe acute respiratory syndrome coronavirus‐2 (SARS‐CoV‐2) and neuropathological studies, as well as measures of neurodegenerative biomarkers among COVID‐19 patients have supported such an association. ⁴ , ⁵ , ⁶ Similarly, investigations utilizing a variety of neuropsychological batteries have reported poor performance among COVID‐19 subjects across several cognitive domains. ⁷ , ⁸ , ⁹ , ¹⁰ , ¹¹ However, major limitations of these reports include lack of adequate controls groups, short follow‐up periods, and ascertainment/selection biases, wherein patients with undiagnosed underlying cognitive impairment are more likely to present for evaluation following COVID‐19. ¹² Furthermore, it is possible that COVID‐19 related medical comorbidities (e.g., depression, anxiety, sleep disorders), medication side effects, and substance abuse may generate cognitive symptoms, and attribution biases may occur due to the prevalence of SARS‐CoV‐2 infection and the profound societal impact of the COVID‐19 pandemic.

To disentangle the relationship between SARS‐CoV‐2 infection and cognitive sequelae we conducted neuropsychological testing and blinded expert clinician evaluation following National Alzheimer's Coordinating Center (NACC) and National Institute on Aging (NIA) ‐Alzheimer's Association criteria ¹³ , ¹⁴ , ¹⁵ , ¹⁶ in a cohort of subjects with no history of cognitive impairment pre‐pandemic or pre‐index SARS‐CoV‐2 infection who were: 1. Fully recovered, laboratory documented COVID‐positive with no symptoms of Long‐COVID (recovered‐COVID); 2. COVID positive with Long‐COVID, or; 3. COVID‐negative (no clinical or laboratory history of SARS‐CoV‐2 infection and negative SARS‐CoV‐2 nucleocapsid antibodies). We preferentially recruited healthy, cognitively normal subjects followed at our Alzheimer's Disease Research Center (ADRC) with documented pre‐pandemic (pre‐January 1, 2020) and pre‐COVID normal neuropsychological testing and normal expert clinician cognitive evaluation, and normal functional status across multiple settings (Clinical Dementia Rating [CDR] score of 0). To our knowledge, this is the first study to quantify the incidence of MCI among Long‐COVID patients compared to controls using standardized NACC and NIA‐Alzheimer's Association diagnostic criteria. ¹³ , ¹⁴ , ¹⁵ , ¹⁶

2. METHODS

2.1. Study population

In this observational cohort study, subjects who met the following inclusion criteria were enrolled between January 18, 2023 and January 14, 2025: aged ≥18 years, consent to participate, available study partner/informant, able to complete neuropsychological testing and clinician interview, and no history of cognitive impairment, MCI, AD, or dementia pre‐ January 1, 2020 (for COVID‐negative controls) and pre‐index SARS‐CoV‐2 infection (for COVID‐positive subjects). Exclusion criteria were: presence or history of structural brain lesions such as ischemic stroke, intracranial hemorrhage, traumatic brain injury, multiple sclerosis/demyelinating disease, brain tumor, hydrocephalus, central nervous system infection, amyotrophic lateral sclerosis, history of a neurosurgical procedure, life expectancy < 6 months, participation in a COVID‐19 related clinical trial, and pre‐pandemic/pre‐COVID modified Rankin Score (mRS) > 0.

COVID‐positive subjects were required to have laboratory evidence of SARS‐CoV‐2 infection (either reverse transcriptase‐polymerase chain reaction [RT‐PCR] or antigen testing). Long‐COVID was defined according to the World Health Organization definition as the continuation or development of new symptoms (any symptom from the WHO Long‐COVID symptom checklist ¹⁷ ) 3 months after initial SARS‐CoV‐2 infection. ¹⁸ Recovered‐COVID was defined as full recovery from SARS‐CoV‐2 symptoms within < 3 months. COVID‐19 negative controls were required to have an absence of any positive SARS‐CoV‐2 RT‐PCR or antigen test, no clinical episode that resembled acute COVID‐19, and negative SARS‐CoV‐2 nucleocapsid IgG testing performed at the time of evaluation. Subjects who reported no clinical history of COVID‐19, and no history of positive SARS‐CoV‐2 testing, but for whom SARS‐CoV‐2 nucleocapsid antibodies were positive (based on laboratory thresholds) were coded as recovered‐COVID; 26/212 (12%) COVID‐positive patients fell into this category. SARS‐CoV‐2 strain was assigned based on the month, year and location of infection (New York City metropolitan area) using data from the Centers for Disease Control and Prevention (CDC), ¹⁹ World Health Organization, ²⁰ , ²¹ and Global Initiative on Sharing All Influenza Data (GISAID). ²² , ²³

2.2. Recruitment

Participants were preferentially recruited from an extant cognitively normal control population enrolled in the New York University (NYU) ADRC, all of whom had normal Uniform Data Set (UDS version 3) neuropsychological testing and clinical evaluations at visits prior to January 1, 2020 (pre‐pandemic in New York City) and/or prior to index SARS‐CoV‐2 infection (if COVID‐positive). The NYU ADRC subjects all had baseline CDR scores of 0, reflecting normal cognitive performance and memory, as well as normal functional performance in hobbies, professional activities, community affairs, personal care, and household activities. The remaining subjects were recruited from an existing longitudinal COVID‐19 cohort, ²⁴ , ²⁵ , ²⁶ from outpatient general medicine and neurology clinics, and as self‐referrals via a research portal accessible to the general public. All subjects had no documented history of prior cognitive impairment or dementia, which was confirmed by the electronic health record, prior ADRC testing and clinical evaluation (when available), a close contact informant, and self‐report. This study was approved by the NYU Grossman School of Medicine Institutional Review Board. All patients or their surrogates provided written consent to participate.

2.3. Evaluation

All subjects underwent an augmented in‐person NACC UDS version 3 ¹⁵ , ¹⁶ neuropsychological battery administered by trained clinicians (Supplemental Table 1). The UDS batteries are normed for age, sex, and education level for individual participants.

Subjects and informants were then interviewed by neurologists specializing in cognition and dementia care who were blinded to the COVID and Long‐COVID status of the participants. The diagnoses of mild cognitive impairment (MCI) or dementia were conferred by this consensus team of expert clinicians utilizing NACC diagnostic criteria ¹⁶ and the 2011 and 2018 NIA‐Alzheimer's Association guidelines for the diagnosis of MCI, and subtypes of MCI. ¹³ , ¹⁴ The clinical diagnosis of MCI was based on review of neuropsychological testing, patient and informant interview, review of education level, pertinent medical history and co‐morbidities, neuroimaging, evaluation of plasma and positron emission tomography (PET) biomarkers (when available), and genetic factors (apolipoprotein E4 [APOE4] status). ¹³ , ¹⁴ , ¹⁵ , ¹⁶

RESEARCH IN CONTEXT

1. Systematic review: A systematic review using PubMed yielded several reports which support a link between coronavirus disease 2019 (COVID‐19) and increased risk of Alzheimer's disease (AD) and other dementias. However, limitations of the current literature include: reliance on International Classification of Diseases, Revision 10 (ICD‐10) codes for dementia diagnoses; limited cognitive testing which does not parse etiology; inadequate control groups; and lack of rigorous, blinded clinician evaluation using standardized criteria to specifically diagnose mild cognitive impairment (MCI) etiology accounting for relevant confounders.

2. Interpretation: Using National Alzheimer's Coordinating Center (NACC) and National Institute on Aging (NIA‐Alzheimer's Association) criteria we found a significantly increased cumulative incidence and hazard of MCI (and specifically AD subtype of MCI) in Long‐COVID patients, compared to recovered‐COVID and COVID‐negative controls.

3. Future directions: Our study supports ongoing investigations into the pathophysiological effects of severe acute respiratory syndrome coronavirus‐2 (SARS‐CoV‐2) on cognition and neurodegeneration, and initiatives aimed at identifying therapeutic targets for post‐COVID cognitive impairment.

Of note, the UDS battery does not include systematic performance validation tests. However, there are certain components of the NACC/UDS exam that are used to support performance validity. First, the standard operating procedure for the UDS battery requires that exams be performed in a fixed order with verbatim instructions in a controlled environment, to minimize testing variances. Second, the examiner administering the battery is required to rate the exam validity during the testing session. If validity is questioned, the examiner must code this, as well as an explanation for invalidity. Last, the primary clinician and the adjudication panel of clinicians evaluate discrepancies in test performance, error counts, and domain‐specific composite scores and flag inconsistent performances. The final adjudicated diagnosis (cognitively normal, MCI, or dementia) accounts for concomitant medical and psychiatric conditions along with cognitive testing results and estimations of testing validity. Data suggest that, among older adults in their 60s and 70s undergoing dementia evaluation, performance validation tests have a low likelihood of identifying invalid results that were initially coded as valid (0.7–4.3%). Conversely, performance validation tests more often generate false positive results (e.g. identifying valid results as invalid) in this population, limiting their utility. ²⁷

MCI core clinical criteria include: (1) clinical concern about decline in cognition compared to baseline cognitive function (based on participant, informant, and clinician judgement), (2) impairment in one or more cognitive domains compared to the patient's estimated baseline, or supported by objective longitudinal neuropsychological evidence of decline, and (3) largely preserved functional independence or functional dependence that is not related to cognitive impairment. ¹³ , ¹⁴ The etiologies of MCI were coded as primary or contributing. Primary MCI related to AD (MCI‐AD) was coded in patients with single or multi‐domain amnestic MCI after excluding other possible primary diagnoses including: MCI due to vascular contributions to cognitive impairment and dementia (MCI‐VCID), MCI due to psychiatric illness ([MCI‐psychiatric illness] including anxiety, depression, bipolar disorder, post‐traumatic stress disorder based on Diagnosis and Statistical Manual of Mental Disorders, Fifth Edition {DSM‐5} criteria), MCI due to systemic/medical illness (MCI‐systemic illness), or MCI due to other factors (e.g., medications, Lewy body disease, Parkinson's, multi‐system atrophy, corticobasal degeneration, progressive supranuclear palsy, primary progressive aphasia, frontal temporal lobe degeneration, LATE [TDP‐43], Huntington disease, prion disease, alcohol or drug abuse, hydrocephalus, traumatic brain injury, central nervous system neoplasm, multiple sclerosis, or amyotrophic lateral sclerosis). ¹³ , ¹⁴ , ¹⁵

2.4. Outcomes

The primary outcome was the cumulative incidence of MCI in Long‐COVID patients compared to those without Long‐COVID (including COVID‐negative and recovered COVID‐positive subjects). We performed subgroup analyses comparing Long‐COVID patients to COVID‐positive only, as well as COVID‐negative status only. The association of Long‐COVID with MCI subtypes by etiology, including MCI‐AD, MCI‐VCID, MCI‐systemic illness, and MCI‐psychiatric illness, was similarly explored. Two sensitivity analyses were performed: 1) evaluating only subjects who were previously enrolled at the NYU ADRC, and hence have rigorous pre‐COVID and pre‐pandemic normal UDS neuropsychiatric testing and were diagnosed as cognitively normal (CDR 0) by a consensus panel of dementia experts; and 2) an analysis excluding patients hospitalized for acute COVID‐19, to mitigate against the confounding impact that hypoxia may have on neurodegeneration.

2.5. Statistical analyses

Demographics, details of SARS‐CoV‐2 infections, and subtypes of MCI diagnoses were compared between subjects with and without Long‐COVID using Mann–Whitney U, Kruskal–Wallis, chi‐squared, and Fisher's exact tests, as appropriate. Because only one subject was diagnosed with dementia, this patient was analyzed with the MCI group. Individual neuropsychological tests were dichotomized as abnormal (versus normal) per published literature or at the median. Cumulative incidence function curves for the occurrence of MCI and associated confidence intervals were constructed using the etmcif function in the ETM package in R. Cox proportional hazards models for MCI (and MCI subtypes) were fit treating index COVID and Long‐COVID diagnoses as time‐dependent variables at which the hazard for MCI (and its subtypes) might change. All time analyses were performed using January 1, 2020 as time zero, prior to the first SARS‐CoV‐2 diagnosis in the United States and New York City. This time point was selected to control for pandemic related stressors that may contribute to cognitive impairment. ²⁸ Times to MCI for participants who were enrolled with prior SARS‐CoV‐2 infection were treated as left‐truncated by the time of the index infection, as they were required to be cognitively unimpaired prior to that time. An analysis adjusting hazard ratios for race was performed. Since UDS 3 tests are normed for age, sex, and education level, we did not perform a duplicated adjustment for these factors in multivariable analyses. All analyses were performed using SPSS (IBM version 29) and RStudio (2023.06.1+524).

3. RESULTS

We included 260 subjects from the New York City area: N = 212 COVID‐19 patients (59% female, median age 70 years, median time from index SARS‐CoV‐2 infection to evaluation 2.0 years [interquartile range {IQR} 1.2–3.2 years], median time from pandemic onset to evaluation 3.9 years [IQR 3.6–4.2]) and N = 48 COVID‐negative controls (73% female, median age 71 years, median 3.9 years [IQR 3.5–4.2] from pandemic onset to evaluation, Table 1, Figure 1). Of the COVID‐positive subjects, 113 had Long‐COVID (median age 66, 57% female, median 2.7 years from index SARS‐CoV‐2 infection to evaluation), and 99 were recovered‐COVID (median age 72, 61% female, median 1.5 years from index SARS‐CoV‐2 infection to evaluation). Dates of index SARS‐CoV‐2 infection ranged from March 1, 2020 through February 21, 2024. The majority of participants (189/260, 73%) were recruited from the extant cognitively normal population enrolled in the NYU ADRC, (48/48 [100%] COVID‐negative and 141/212 [67%] of COVID‐positive patients), all of whom had normal UDS neuropsychological testing, CDR = 0, and normal clinical evaluations at visits prior to January 1, 2020 (pre‐pandemic in New York City) and/or pre‐index SARS‐CoV‐2.

TABLE 1.

Demographics of Long‐COVID patients compared to those without Long‐COVID (including recovered COVID patients and COVID‐negative controls).

Parameter	Long‐COVID (N = 113)	Recovered‐COVID (N = 99)	COVID‐negative (N = 48)	p‐Value
Age (years, median, IQR) *	66 (50–72)	72 (67–77)	71 (67–77)	< 0.001
Sex (female)	64/113 (57%)	60/99 (61%)	35/48 (73%)	0.151
Race (N, %)
White *	87/112 (78%)	53/98 (54%)	22/48 (46%)	< 0.001
Black *	25/112 (22%)	43/98 (44%)	35/48 (52%)	< 0.001
Asian	0/112 (0%)	2/98 (2%)	1/48 (2%)	0.312
Hispanic (N, %)	17/111 (15%)	7/97 (7%)	6/47 (13%)	0.190
Education (years, median, IQR)	16 (16–18)	17 (16–18)	16 (14–18)	0.169
APOE4 status (N, %)
Heterozygous	14/61 (23%)	31/78 (40%)	13/40 (33%)	0.110
Homozygous	2/61 (3%)	5/78 (6%)	1/40 (3%)	0.534
Past medical history (N, %)
Hypertension	47/113 (42%)	47/99 (48%)	23/48 (48%)	0.784
Diabetes	19/112 (17%)	9/98 (9%)	4/48 (8%)	0.205
Hypercholesterolemia	52/113 (46%)	52/99 (53%)	23/48 (48%)	0.903
Cancer	16/113 (14%)	15/99 (15%)	8/48 (17%)	0.919
Thyroid disease	15/113 (13%)	16/99 (16%)	7/47 (15%)	0.146
Vitamin B12 deficiency	6/112 (5%)	4/99 (4%)	4/48 (8%)	0.490
Tobacco Use * (pack years, median, IQR)	0 (0–7)	0 (0–10)	1 (0–30)	0.041
Alcohol use (drink frequency, median, IQR)	1 drink/month (< 1/month to 1/week)	1 drink/week (< 1/month to a few/week)	1 drink/month (< 1/month to a few/week)	0.116
Date of index SARS‐CoV‐2 infection * (median, IQR)	2/1/2021 (3/1/2020–5/1/2022)	7/1/2022 (10/16/2021–2/14/2023)	–	< 0.001
Hospitalized for COVID * (N, %)	22/113 (20%)	1/99 (1%)	0/48 (0%)	< 0.001
Index SARS‐CoV‐2 variant † (N, %)
Original/wild‐type * (3/2020‐9/2020)
Alpha * (10/2020‐4/2021)	37/112 (33%)	10/88 (11%)	–	0.001
Delta (5/2021‐11/2021)	20/112 (18%)	8/88 (9%)	–	0.011
Omicron * (12/2021‐8/2024)	7/112 (6%)	4/88 (5%)	–	0.846
Time from COVID to evaluation * (years, median, IQR)	2.7 (1.8–3.3)	1.5 (0.8–2.1)	–	< 0.001
Time from January 1, 2020 to evaluation (years, median, IQR)	3.9 (3.6–4.2)	4.0 (3.6–4.3)	3.9 (3.5–4.2)	0.472
# Episodes of COVID * (years, median, IQR)	1 (1–2)	1 (1–2)	0 (0–0)	< 0.001
SARS‐CoV‐2 vaccination (N, %)	110/113 (97%)	99/99 (100%)	46/48 (96%)	0.170

Abbreviations: APOE4, apolipoprotein E4; COVID, coronavirus disease 2019; IQR, interquartile range; SARS‐CoV‐2, severe acute respiratory syndrome coronavirus‐2;.

^*There were significant differences (p < 0.05) between groups.

^† SARS‐CoV‐2 variants were classified according to the most prevalent strain circulating in the New York City area at the time.

FIGURE 1.

Enrollment diagram with the frequency of diagnoses of mild cognitive impairment (MCI), dementia, and MCI related to Alzheimer's disease (MCI‐AD) among Long‐coronavirus disease 2019 (Long‐COVID), recovered COVID, and COVID‐negative controls (COVID‐).

Among Long‐COVID subjects, 63% had symptoms that were persistent at the time of evaluation, and 76% reported brain fog (Supplemental Figure 1). Brain fog was significantly correlated with a diagnosis of MCI (Spearman's rho 0.256, p < 0.001). Compared to subjects without Long‐COVID (N = 147: including recovered COVID‐positive [N = 99] and COVID‐negative controls [N = 48]), those with Long‐COVID tended to be younger, White, experienced more episodes of COVID‐19, and had index COVID‐19 earlier in the pandemic than those without Long‐COVID. Long‐COVID patients were also hospitalized for COVID‐19 at a higher rate than recovered COV+ participants (20% versus 1%, p < 0.001, Table 1). Education level, APOE‐4 genetic status, and relevant past medical history did not differ between groups. Long‐COVID patients had significantly worse neuropsychometric test scores in measures of global function and cognition, depression, stress, and subjective cognitive decline compared to recovered COVID‐positive and COVID‐negative controls (Supplemental Figure 2).

Overall, MCI was diagnosed in 37 patients (14%) and dementia (AD type) was diagnosed in one (0.4%) Long‐COVID patient. Primary and contributing MCI diagnoses among patients with Long‐COVID, recovered‐COVID, and COVID‐negative controls are shown in Table 2. Four primary subtypes of MCI were diagnosed including MCI‐AD (N = 18), MCI‐VCID (N = 4), MCI‐psychiatric illness (N = 6), and MCI‐systemic illness (N = 10). Psychiatric illness was a contributing factor in 7 (6%) MCI diagnoses, all among those with Long‐COVID. The most common pattern of cognitive impairment was multi‐domain amnestic, consistent with Alzheimer's‐type cognitive deficits.

TABLE 2.

Primary and contributing MCI diagnoses and subtypes among patients with long‐COVID, recovered COVID, and controls (COVID‐negative).

Parameter	Total N * (%)	Primary diagnosis	Contributing diagnosis	Amnestic, single domain	Amnestic, multiple domains	Non‐amnestic, single domain	Non‐amnestic, multiple domains
Long‐COVID N = 113
Normal	84 (74%)	–	–	–	–	–	–
MCI‐AD †	13 (12%)	13 (12%)	0	3 (3%)	10 (9%)	0	0
MCI‐ Psychiatric	12 (11%)	5 (4%)	7 (6%)	0	8 (7%)	1 (1%)	3 (3%)
MCI‐VCID	4 (4%)	3 (3%)	1 (1%)	0	2 (2%)	0	2 (2%)
MCI‐ systemic illness	10 (10%)	8 (7%)	2 (2%)	1 (1%)	8 (7%)	0	1 (1%)
Recovered COVID N = 99
Normal	92 (93%)	–	–	–	–	–	–
MCI‐AD	4 (4%)	4 (4%)	0	2 (2%)	2 (2%)	0	0
MCI‐Psychiatric	1 (1%)	1 (1%)	0	0	1 (1%)	0	0
MCI‐VCID	0	0	0	–	–	–	–
MCI‐systemic illness	2 (2%)	2 (2%)	0	0	0	0	2 (2%)
COVID negative N = 48
Normal	46 (96%)	–	–	–	–	–	–
MCI‐AD	1 (2%)	1 (2%)	0	0	1 (2%)	0	0
MCI‐psychiatric	0	0	–	–	–	–	–
MCI‐VCID	1 (2%)	1 (2%)	0	0	0	0	1 (2%)
MCI‐ systemic illness	0	0	–	–	–	–	–

Note: Subjects may have only one primary diagnosis, but may have multiple contributing diagnoses.

Abbreviations: AD Alzheimer's dementia; COVID, coronavirus disease 2019; MCI, mild cognitive impairment; VCID, vascular contributions to cognitive impairment and dementia.

^* Includes primary and contributing diagnoses.

^† Includes one patient with AD‐type dementia.

The cumulative incidence of MCI over 4.4 years was higher with Long‐COVID (27%, 95% confidence interval [CI] 17%–40%) versus recovered‐COVID (5%, 95% CI 2%–10%) or COVID‐negative status (1%, 95% CI 0.2%–4%, Figure 2). Long‐COVID (treated as a time‐dependent covariate) was associated with a significantly higher hazard of MCI compared to no Long‐COVID (HR 3.93, 95% CI 1.86–8.31, p < 0.001, Table 3). Similar results were found comparing Long‐COVID status to COVID‐positive only (HR 3.15, 95% CI 1.38–7.19, p = 0.007) and COVID‐negative only subgroups (HR 6.68, 95% CI 1.59–27.99, p = 0.009). Conversely, COVID‐positive status (versus COVID‐negative status) was not significantly associated with increased hazard of MCI (HR 2.2, 95% CI 0.45–10.71, p = 0.33). When adjusting for race, which differed significantly between groups, the adjusted hazard of MCI was also higher among Long‐COVID subjects (adjusted HR [aHR] 4.01, 95% CI 1.83–8.79, p = 0.001) compared to other groups (Supplemental Table 2).

FIGURE 2.

Cumulative incidence of mild cognitive impairment (MCI) among Long‐coronavirus disease 2019 (Long‐COVID), green line) patients compared to recovered COVID (covid+, red line) patients and COVID‐negative (covid‐, black line) controls. The solid lines represent the cumulative incidence and the dotted lines represent 95% confidence intervals.

TABLE 3.

Cox proportional hazards models computing the time‐dependent HR (95% CI) for MCI and primary MCI subtypes comparing those with long‐COVID to those without (inclusive of COV+ and COV− subjects), and compared to COV+ only (without long‐COVID), or COV− only groups.

Primary MCI type	Long‐COVID vs. no Long‐COVID * HR (95% CI), p	Long‐COVID vs. COV+ only HR (95% CI), p	Long‐COVID vs. COV− only HR (95% CI), p
Any MCI	3.93 (1.86–8.31) p < 0.001	3.15 (1.38–7.19) p = 0.007	6.68 (1.59–27.99) p = 0.009
MCI‐AD	3.20 (1.14–9.0) p = 0.027	2.48 (0.81–7.62) p = 0.113	6.09 (0.80–46.59) p = 0.082
MCI‐psychiatric illness	5.98 (0.70–51.23) p = 0.103	3.72 (0.43–31.88) p = 0.231	—
MCI‐systemic illness	5.05 (1.07–12.85) p = 0.041	3.18 (0.67–15.06) p = 0.114	—

Note: Models were not stable for MCI‐VCID as this occurred as a primary diagnosis in only 4 subjects.

Abbreviations: AD, Alzheimer's dementia; aHR, adjusted hazard ratio; CI, confidence interval; COV, coronavirus disease 2019; MCI, mild cognitive impairment; VCID, vascular contributions to cognitive impairment and dementia.

^*Includes both recovered COV+ and COV− controls.

We examined primary MCI etiologies/subtypes using similar Cox proportional hazard models and found a higher hazard of MCI‐AD subtype among those with Long‐COVID versus no Long‐COVID (HR 3.20, 95% CI 1.14–9.0, p = 0.027, Table 3). This relationship persisted after adjusting for race (aHR 3.25, 95% CI 1.10–9.60, p = 0.033, Supplemental Table 2). Additionally, plasma levels of pTau‐217 were significantly higher in subjects diagnosed with MCI‐AD compared to those who were not (median 0.22 ng/mL versus 0.13 ng/mL, p = 0.024), but were similar among MCI‐AD patients with and without Long‐COVID (median 0.28 ng/mL vs. 0.22 ng/mL, p = 0.461). Long‐COVID was also associated with a higher hazard of MCI due to systemic illness (HR 5.05, 95% CI 1.07–12.85, p = 0.041), though this relationship was not significant after adjusting for race (Supplemental Table 2). No significant relationships were identified for Long‐COVID and MCI due to psychiatric illness. Models evaluating relationships between COVID status and MCI‐VCID were unstable, as this occurred as a primary diagnosis in only four subjects.

We performed sensitivity analyses using the same adjusted Cox proportional hazard models using time‐dependent covariates in a subgroup of patients previously followed at the ADRC (N = 189 with documented pre‐pandemic normal cognitive evaluations), as well as an analysis excluding patients hospitalized for acute COVID‐19 (N = 237 non‐hospitalized COVID‐19 patients) and found consistent results (Supplemental Tables 3 and 4).

4. DISCUSSION

Overall, we found that Long‐COVID patients had a higher cumulative incidence of MCI, and particularly primary MCI‐AD, compared to those without Long‐COVID over 4.4 years. In adjusted cox analysis using both COVID and Long‐COVID status as a time‐dependent covariates, there was a four‐fold increased hazard of MCI and a three‐fold increased hazard of MCI‐AD. Furthermore, the observed incidence of MCI among Long‐COVID patients is substantially higher than the expected background population incidence rates. For example, the baseline incidence of MCI pre‐pandemic among men and women aged 70–74 years has been estimated to be 2.41% per year (24.1 per 1000‐patient years), ²⁹ which would equate to approximately 10.1% cumulative incidence of MCI over 4.4 years, assuming a constant background incidence rate. The incidence rate of MCI among sexagenarians is not well described in the literature, but is expected to be lower. Since the median age of Long‐COVID subjects at the time of evaluation was 66 years, the observed 27% cumulative incidence of MCI over 4.4 years in this group is substantially higher than the expected rate (< 10.1%). The lower observed cumulative incidence rates of MCI in recovered‐COVID and COVID‐negative subjects compared to expected rates may be due, in part, to younger age and higher levels of education than published populations. ³⁰ ,  

This is the first study, to our knowledge, to use standardized NACC and NIA‐Alzheimer's Association diagnostic process to determine the incidence of new MCI diagnoses, and ascribe MCI etiologies in Long‐COVID patients compared to controls. ¹³ , ¹⁴ , ¹⁶ Furthermore, 73% of our cohort underwent the same extensive neuropsychological battery and clinician interviews pre‐pandemic and pre‐COVID with documentation of normal cognitive and functional status at baseline. In contrast, other studies have not had extensive documentation of normal baseline cognitive status, nor have they differentiated post‐COVID cognitive deficits due to psychiatric disorders, systemic illness, or other causes. Other strengths of this study include use of a contemporaneous COVID‐negative control group, blinded expert interviews with subjects and informants, and adjustment for relevant confounders (race, age, education level, sex). Though psychiatric disease was a contributing factor in 6% of Long‐COVID subjects (Table 2), MCI‐AD was the most common primary diagnosis (12%). Co‐morbid psychiatric illness is not unexpected since, in the general population, up to 43% of patients with MCI have documented neuropsychiatric symptoms including depression, apathy, and irritability. ³¹ Our data suggest that MCI in Long‐COVID patients is primarily multi‐domain amnestic in nature, and may be related to AD‐type neurodegeneration. Further study of imaging and neurodegenerative biomarkers may help identify characteristics that differentiate post‐SARS‐CoV‐2 cognitive impairment from AD. However, only gold standard neuropathological studies can definitively confirm or refute the relationship between Long‐COVID brain fog and neurodegeneration. Proteomic analysis of amyloid beta plaques, the phosphorylated tau interactome, and even normal appearing white and gray matter can be used to identify differences in mechanistic pathways contributing to AD‐type pathology, as we have previously done to identify uniquely enriched proteins and altered signaling pathways in Down syndrome‐related AD compared to sporadic early and late onset AD. ³² , ³³ , ³⁴ , ³⁵

There are limitations to this study. First, 53% of participants reported experiencing Long‐COVID compared to CDC estimates of 30% of the U.S. adult population, ³⁶ suggesting an element of referral bias. However, there may be a higher prevalence of COVID‐positive individuals in the New York City (NYC) area. Additionally, our study included a highly educated urban population, and results may not generalize to a broader population. Second, we present a single assessment for both COVID and MCI. We are currently collecting longitudinal repeated measures of neuropsychological testing and expert interviews to corroborate cognitive trajectories. Third, though amyloid and tau PET and biomarker data were utilized as supporting evidence during the MCI diagnostic process, these were not required. Of note, revised criteria for the diagnosis and staging of Alzheimer's disease were published in August 2024, ³⁷ after enrollment in our study began. These guidelines incorporate imaging and blood biomarkers into the diagnosis of AD, though specific thresholds for abnormality in MCI and AD are not specified. Importantly, new UDS version 4 diagnostic criteria ¹⁶ still rely on clinical evaluation and are not biomarker dependent, though biomarkers can be used to support a diagnosis. Fourth, Long‐COVID subjects had a longer exposure time from index SARS‐CoV‐2 to evaluation than those with recovered‐COVID (median 2.7 years versus 1.5 years). However, both COVID‐positive and Long‐COVID status were handled as time‐dependent covariates in cox models to help accommodate differences in exposure time, along with standard right censoring at end of follow‐up. Fifth, it is difficult to definitively establish COVID negative status since SARS‐CoV‐2 nucleocapsid antibodies can wane over time and subjects can convert to seronegative status. However, in addition to negative nucleocapsid antibody testing, we only classified subjects as COVID negative if they were free from any episode of symptoms that clinically resembled COVID, and if they never had a positive SARS‐CoV‐2 PCR or antigen test. Given the prevalence SARS‐CoV‐2 screening/testing, we believe this was the most stringent possible way to define COVID negative status. Last, we cannot establish causality for SARS‐CoV‐2 leading to AD‐type neurodegeneration without a better understanding of the underlying pathobiology, though our results support continued work in this arena. While it remains possible that some patients had an undetected/preclinical cognitive disorder or subjective cognitive decline (SCD, which is a risk factor for MCI ³⁸ ) prior to contracting COVID‐19, 73% of subjects were recruited from our ADRC and had extensive neuropsychological testing and expert clinician evaluation certifying normal cognitive status pre‐pandemic and pre‐COVID. Furthermore, our sensitivity analysis including only ADRC patients recapitulated the results of the primary analysis. It is also conceivable that hypoxic brain injury among those hospitalized for COVID‐19 could contribute to cognitive impairment. However, our sensitivity analysis excluding the N = 23 subjects hospitalized for acute COVID‐19 still supported a relationship between Long‐COVID and MCI (HR 3.43, 95% CI 1.50–7.83, p = 0.004).

In conclusion, participants with Long‐COVID had a higher cumulative incidence and higher hazard of MCI, and specifically AD‐type MCI, than those without Long‐COVID, even after controlling for confounders. Further research into SARS‐CoV‐2‐related mechanisms of AD‐type neurodegeneration is warranted.

AUTHOR CONTRIBUTIONS

J.A.F. conceptualized the project, analyzed data, and drafted the manuscript. A.V.M. conceptualized, collected and interpreted data and critically revised the manuscript. All authors have approved the submitted manuscript. R.A.B. performed statistical analyses, interpretation of data and critically revised the manuscript. Z.A., A.B., J.C., S.H., J.H., L.J., M.L., J.L., K.M., H.P., F.S., S.T., D.V., and A.Y. acquired the data and critically revised the manuscript. A.V. acquired and collated the data and critically revised the manuscript. Y.G. and T.W. conceptualized and designed the project and critically revised the manuscript.

CONFLICT OF INTEREST STATEMENT

J.A.F. receives consulting fees from Serb Pharmaceuticals (not related to article content). A.V.M. serves as a section editor for Alzheimer's & Dementia. R.A.B., Z.A., A.B., J.C., S.H., J.H., L.J., M.L., J.L., K.M., H.P., F.S., S.T., D.V., A.Y., Y.G., and T.W. report no conflicts of interest. Author disclosures are available in the supporting information.

CONSENT STATEMENT

This study was approved by the NYU Grossman School of Medicine Institutional Review Board. All patients or their surrogates provided written consent to participate.

Supporting information

Supporting Information

Supporting Information