The 2020 novel SARS-CoV-2 (COVID-19) pandemic left substantial healthcare challenges in its wake, the impact of which is still being felt today. Lifetime incidence rates of sustained symptomatology, known as long-COVID or post-acute COVID-19 syndrome (PACS), ranges from 7%-40% (1–3) with estimates reaching 34% in non-hospitalized patients (3). PACS is defined as the continuation or development of new symptoms beyond 4 weeks of an initial COVID-19 infection (4). The cluster of disorders can include numerous symptoms from all organ classes (4) and can be incredibly debilitating (5). Notably, alterations in central nervous system function, including cognitive complaints, fatigue, and psychiatric symptoms, are some of the most commonly reported features of PACS (3, 6).

Several provisional mechanisms have been put forth to explain central nervous system changes associated with PACS. Angiotensin-converting enzyme 2 (ACE2), an enzyme expressed in many regions of the brain (7, 8), has been identified as a primary point of cell entry for SARS-CoV-2 (9). Cellular damage from increased neuroinflammatory processes akin to those seen in neurodegenerative disease (10) and neurovascular disease (11), or from direct viral activity in the brain have also been proposed mechanisms of PACS-associated central nervous system changes. While no single or specific process has yet been established, independent lines of investigation provide preliminary evidence linking PACS symptoms to pathophysiologic changes that can profoundly impact brain function (12, 13). For example, data from the UK Biobank study compared pre- and post-SARS-CoV-2 infection changes in brain structure and discovered reductions in grey matter and total brain volume accompanied by significant impairments in performance-based tests of cognitive processing speed and executive function (14). In other work, brain alterations characterized by hypo-connectivity affecting hippocampal and fronto-cerebellar pathways and reduction in grey matter in cortical, limbic, and cerebellar structures were found to correlate with cognitive dysfunction as late as 11 months following SARS-CoV-2 infection (15). A recent study by our group found that over two-thirds of a neurology clinic sample who identified cognitive difficulty as a primary symptom exhibited reductions in executive function when tested at least 3 months post-SARS-CoV-2 infection, despite generally above-average performance on measures of premorbid cognitive status (16). Subsequent testing at 6 months reflected minimal recovery in performance-based measures, consistent with the persistence of subjective cognitive complaints.

While the term brain fog has been widely adopted colloquially in reference to a myriad of cognitive changes described by PACS patients, the scientific literature has been mixed regarding interpretations of severity, associated functional impact, and underlying etiology. Whereas some studies have reported moderate to large deficits in select domains of cognitive testing (17), others find impairment to be mild and restricted to only a subset of individuals who endorse new cognitive complaints following COVID-19 (18). Congruence between subjective symptom endorsement and impairment measured by performance-based neuropsychological tests can be difficult to establish for many reasons. For example, because 30-40% of PACS patients develop psychiatric symptoms in the months following their COVID-19 illness (6, 19), associated disturbances in sleep, energy, and volition could contribute to the experience of decline in cognitive status. Indeed, prolonged cognitive symptoms following SARS-CoV-2 infection have been described alongside psychiatric symptoms as sequelae of somatic, functional, and psychosocial consequences of the disease (20). Impressions of cognitive symptoms and their correlates may also be influenced by the manner in which key cognitive domains, such as attention, memory, and executive function, are assessed clinically, and the extent to which a patient’s experience and description of symptoms align with objective measures of cognitive test performance. Patient ratings of the Cognitive Failures Questionnaire were found to be uncorrelated with objective measures of cognitive status in one study of PACS patients, however, testing was completed using a web-based platform rather than by standard clinical assessment (21). It is also possible that subjective and objective measures of cognitive status are differentially impacted by additional aspects of the illness experience, wherein subjective appraisal may be influenced to a greater extent by common comorbidities of PACS (e.g., fatigue, psychiatric symptoms) (18) than are objective measures of neuropsychological test performance (22). With this distinction in mind, it is reasonable to question whether subjective appraisal of cognitive symptoms and tested ability should be expected to correlate, and how discrepant results between methods should be interpreted. Clarification of this issue could better prepare examiners to consider multiple sources of information in the evaluation of functional status and recovery through rehabilitation.

This study sought to better understand the clinical underpinnings of reported cognitive symptomatology in a sample of help-seeking individuals who presented to a neurology clinic post-acute COVID-19 infection with cognitive complaints. The current analysis was designed to incorporate not only an extensive battery of clinical neuropsychological tests in domains relevant to daily function, including working memory, processing speed, and executive functions, but also an estimate of premorbid cognitive status and normatively-referenced scores to aid in interpretability of test scores in the absence of a pre-COVID-19 baseline. Questionnaire-based measurement of cognitive symptoms included both subjective ratings provided by patients, as well as observer ratings provided by an informant that was used in a parallel set of analyses for comparison with patient data. Finally, appreciating that the experience and expression of cognitive symptoms may differentially relate to underlying disease processes, statistical models incorporated measures of fatigue, psychiatric symptoms, and serum markers of inflammatory activity in order to account for the impact of non-cognitive aspects of PACS on the experience of cognitive symptomatology.

This analysis was conducted in a clinical sample of patients who presented to an academic medical center’s neurology clinic with complaints of cognitive symptoms three or more months following SARS-CoV-2 infection (16). Subjective complaints were highly endorsed (98% in the clinically significant range) and, for the majority of participants, also recognized as a change from pre-COVID-19 status by an observer (68% in the clinically significant range). Cognitive complaints were accompanied by psychiatric disturbance characterized by depression, anxiety, and fatigue symptoms. However, when evaluated with a comprehensive battery of neuropsychological tests, rates of objectively-defined cognitive impairment appeared much lower, with less than 50% of participants performing neuropsychological testing in a range expected to warrant clinical attention (Figure 1B). A discrepancy between subjective and objectively measured cognitive symptoms appears throughout the COVID-19 literature and, at first glance, the present findings would appear consistent given the high endorsement of symptoms in the presence of generally intact impressions of neuropsychological test performance. In line with the literature, correlations between cognitive symptoms and other potential explanatory factors, including anxiety (19), depression (51, 52), fatigue symptoms (53), and inflammatory profiles, were generally modest (54, 55). Of these factors, anxiety was identified as the independent measure most highly correlated with subjectively-rated cognitive symptom severity, with a smaller but appreciable correlation also observed with depression ratings (Table 2). From this perspective, results are consistent with prior reports suggesting that subjective cognitive complaints of PACS are associated with aspects of psychological distress, independent of neuropsychological status (56). Nevertheless, when predictors of subjective cognitive complaints were evaluated using a multivariate modeling approach, independent and additive contributions of objective neuropsychological testing, specifically in domains of memory and pre-morbid cognitive status, also emerged. The most parsimonious model, based on anxiety and pre-morbid status, accounted for 21% of variance in MyCog score. Interpreted directionally, the combination of higher anxiety and lower pre-morbid intellectual status predicted higher severity MyCog scores. Together with these predictors, a more comprehensive model, accounting for 36% of total variance, further suggested that lower verbal memory recall in the presence of higher short-term working memory contributed to higher endorsement of cognitive complaints.

In contrast, the optimal model predicting observer-rated cognitive symptoms (TheirCog) was composed entirely of objective neuropsychological tests, involving domains of memory, visuoconstructional processing, and executive function, as well as vocabulary. Exploratory analyses of anxiety scores found no direct or interactive relationships in prediction of TheirCog. A second exploratory analysis was conducted to differentiate contributions of current neuropsychological function from estimated pre-morbid cognitive capacity by removing Vocabulary, addressing the question of whether both aspects of ability must be represented for models to capture discrepancy suggestive of decline. In this case, tests of current function explained the most variance (24%), and the contribution from Vocabulary was additive, but not required. In light of the finding that anxiety was contributory to subjective, but not observer, ratings of cognitive symptoms, input from a knowledgeable informant would appear less influenced by aspects of distress associated with symptomatology and perhaps offers a more direct appraisal of behavioral change associated with cognitive function.

Direct relationships between subjective appraisal of cognitive symptoms and objective test performance have traditionally been difficult to establish in other clinical populations. For instance, patients with persistent infection with the bacterium Coxiella burnetiid who report more severe cognitive complaints do not exhibit corresponding deficits on objective measures of cognitive dysfunction (57). In another example, reports of cognitive complaints in patients after a stay in the intensive care unit were found to be uncorrelated with performance on a computer-administered battery of cognitive tests (58). While some prior research has found no relationship between subjective cognitive symptoms and objective test performance in PACS (21) there are several explanations for why this relationship may vary from study to study, including the cognitive domains represented in the neuropsychological battery and individual variance in performance across the tests included. In the current study, performance on any individual neuropsychological test ranged considerably, including scores well within a range typically associated with impairment (scaled score < 3) as well as the high end of superior (scaled score ‗16) (Table 1). Heterogeneity in performance profiles can also be appreciated between and within the domains of neuropsychological tests (Figure 1A), with rates of below-average performance reaching only 35% for select measures of executive function and processing speed (TMT-B, SCWT Word Reading) and larger spread of low and intact performance noted across other cognitive domains.

The PACS participants included in the current analysis represent a healthy segment of the general population, who were screened thoroughly for pre-morbid conditions that could account for the presence of cognitive symptoms. Premorbid intelligence was estimated to be at least in the average range based on several indicators, including the Word Accentuation Test (WAT), which has been shown to correlate highly with full-scale performance measures of intellectual capacity and is resistant to mental deterioration (29), and thus useful for estimation of pre-morbid ability in a sample without measured cognitive baseline. No participant within this analysis scored below 16 on the WAT, the threshold for below average intelligence, while the mean WAT score in this sample was approximately 1 standard deviation above that associated with the midpoint of the average range of the estimated full-scale IQ equivalence (29), (Supplementary Figure 1). However, nearly every participant in this sample endorsed significant cognitive complaints, captured on the MyCog portion of the SCD-Q. TheirCog was also rated as elevated for most participants, with a moderately strong correlation found between these two scores; thus a basic level of consistency between sources can be assumed. When considering severity and congruence in endorsement of cognitive symptoms from patient and observer perspectives, we are inclined to interpret these data as a reliable indication that participants experienced change in cognitive status following SARS-CoV-2 infection. Although symptom validity measures were not collected in this study, the fact that this sample was recruited from a neurology clinic population would suggest that cognitive symptoms were sufficiently concerning to cause these individuals to seek evaluation and care.

The current analysis was motivated by questions regarding the extent to which subjective cognitive symptomatology in the context of PACS aligns with objective clinical indicators of impairment. Using a liberal threshold of 1 standard deviation below the population average (16^th percentile) to identify neuropsychological test scores that could warrant clinical attention, many participants in the current sample would likely avoid detection as impaired based on neuropsychological testing alone. While 80% of the sample was identified by below average performance on at least 1 test, only 50% met this threshold for performance on three or more tests, which extended to only 30% when the criterion for detection was set at six tests or more (Figure 1B). In comparison with other PACS samples described in the literature, we surmise that participants included in the current analysis are reasonably representative. In one study, performance in the range of 1–2 standard deviations below average was observed in 3 or more neuropsychological tests in approximately 60% and in 6 or more tests in approximately 40% of PACS patients (N = 49) (59), suggesting only a slightly higher rate of possible impairment in a comparably sized sample.

From this perspective, how do we reconcile the high rate of endorsed cognitive symptomatology with a neuropsychological profile that, depending in part on the selection of tests administered, could appear fundamentally intact? One possible explanation is the inherent challenge in interpreting test scores falling below expectation in the absence of a comparable baseline, from which interval changes are measured for more direct interpretation of change. The fact that high-performing individuals can remain capable of performing neuropsychological tests at or above the average range despite concerns of decline (39) has been offered elsewhere as an explanation for the observed discrepancy in correlation between subjective and objective measures of cognitive status in PACS (60). This explanation is offered as one of many alternative solutions, which may include the impact of various sources of error variance that differentially impact questionnaire- and performance-based measures as well as matters yet to be fully understood regarding the duration of PACS and stability of neuropsychological performance in appreciation of a relapsing versus remitting course of symptomatology (61). In consideration, the correlational results of the present analysis are informative in promoting an integrated model that supports a direct relationship of cognitive symptoms to objectively measured function, while also accounting for the impact of anxiety on the reporting of symptoms.

This study has several limitations, some of which are detailed in the parent study report (16). As previously described, the current analyses include a small sample size from one clinical site, and all participants were referred by healthcare providers to partake in the neuropsychological battery, which may present a referral bias. This study lacked a healthy control population due to limits imposed by the local Ethics Committee. Our results failed to replicate findings of heightened inflammatory activity, reported both in SARS-CoV-2 infection and PACS patients (62), which may be in part attributed to the limited subset of cytokines available for the current analysis. These serum cytokine measures failed to enter models based on MyCog and TheirCog; however, this outcome is consistent with a previous study that found no association between peripheral inflammation and cognitive symptoms in PACS patients (54). Notably, cytokine concentrations are only one part of the inflammatory response that may alter over time during chronic disease. Changes in neuroinflammation ascertained via PET imaging have been reported in PACS patients (55), and it is widely accepted that neuroinflammation can result in neuropsychiatric changes (63–65). Thus, it may be important to explore these relationships further in PACS patients using alternative modalities.

In summary, the current analyses led to several conclusions regarding the interpretation of subjective cognitive symptoms in PACS. First, our analyses are novel in the inclusion of both patient and observer-rated symptom scales, which were found to be correlated significantly, albeit modestly. Importantly, observer data, collected from an individual well known to the patient, lends confidence to the impression that participants experienced detectable cognitive sequelae of PACS. Using a comprehensive battery of neuropsychological tests, psychiatric scales relevant to PACS, and inflammatory cytokine measures, we next sought to identify underlying clinical features that could account for the experienced and observed cognitive symptoms. In this case, a significant portion of variance in SCD-Q scores, rated both by patient and observer, was explained by neuropsychological test performance. Given that the optimal models in each case explained roughly 35% of the variance, it is evident that our predictor set is incomplete, and substantial variance remains to be explained by factors presently unaccounted for. Nevertheless, as a second conclusion, we posit that cognitive symptoms associated with PACS are more likely attributable to changes in brain function that are detectable using routine neuropsychological testing than to epiphenomena of psychiatric symptoms, fatigue, or other COVID-19-related comorbidities. Although anxiety did enter the MyCog model, its contribution was additive to that provided by neuropsychological testing, which explained a near equal amount of variance independently. Moreover, anxiety did not appear to have any measurable influence on the relationship between objective testing and TheirCog ratings; neuropsychological testing was the only significant predictor. Remarkably, multivariate models found no influence of fatigue or inflammatory markers on cognitive symptom ratings in either the MyCog or TheirCog models, further highlighting the relative strength of the observed relationships to neuropsychological test performance. In this regard, the current results lend support to the validity of patient-reported outcome instruments, such as the SCD-Q, in the assessment of PACS-associated cognitive symptoms. Finally, although statistical analyses did uncover direct relationships between cognitive symptoms and neuropsychological test performance, suggesting that testing does capture aspects of cognitive symptomatology experienced by patients, questions remain about how valid and objective test results can be best used to inform the detection of cognitive impairment associated with PACS when differences from expected performance appear subtle or isolated to a few domains. In these cases, questionnaire-based measures, particularly those capturing input from informed observers, may be particularly valuable for identification of functional changes useful for contextualizing test results in areas suspected to be most impacted.