Neurocognitive and Neuropsychiatric Trajectories in a Post-COVID Cohort: A Descriptive Longitudinal Study

1. Introduction

Post-acute COVID-19 syndrome (PACS), commonly referred to as long COVID, is characterized by the persistence of symptoms following the acute phase of SARS-CoV-2 infection. These symptoms may affect multiple organ systems and frequently include fatigue, cognitive difficulties, mood disturbances and sleep disorders, representing an increasingly recognized challenge for healthcare systems worldwide [1,2]. Although early case definitions (e.g., NICE 2021) considered symptoms persisting beyond four weeks from the acute infection, more recent and conservative definitions endorsed by the World Health Organization (WHO), the National Academies of Sciences, Engineering, and Medicine (NASEM) and the Centers for Disease Control and Prevention (CDC) require symptom persistence for at least three months, a criterion that improves specificity and reduces the inclusion of patients with self-limiting subacute symptoms [2,3,4].

Among the neurological manifestations of PACS, cognitive dysfunction—often described by patients as “brain fog”—is one of the most commonly reported symptoms and may substantially impair daily functioning and quality of life [5]. Cognitive disturbances may arise during the subacute phase of the infection or shortly after hospital discharge and can persist for several months following recovery from acute COVID-19 [6,7,8].

The pathophysiological mechanisms underlying cognitive impairment in patients recovering from COVID-19 are likely multifactorial. Proposed mechanisms include neuroinflammation, immune dysregulation, endothelial dysfunction and microvascular injury, as well as indirect effects of systemic illness such as hypoxia or widespread cytokine release [9,10,11,12,13]. Previous studies have identified impairments across multiple cognitive domains, particularly executive functions, attention, memory and processing speed [14].

Although cross-sectional investigations have consistently documented cognitive deficits in individuals recovering from COVID-19, longitudinal evidence describing the trajectory of cognitive recovery remains limited and heterogeneous. Some studies have shown that a substantial proportion of patients continue to exhibit pathological performance in at least one cognitive domain up to one year after infection [15,16]. Moreover, persistent cognitive slowing and mental fatigue have been reported even in individuals who experienced mild forms of the disease [17]

Cognitive impairment in PACS is also frequently associated with neuropsychiatric symptoms, including anxiety, depression and sleep disturbances, which may further contribute to reduced quality of life and functional limitations [18,19]. Understanding the interaction between neurocognitive deficits and psychiatric symptoms may therefore be crucial for improving long-term management strategies in these patients.

The aim of this study was to investigate longitudinal changes in neurocognitive performance and neuropsychiatric symptoms in a cohort of patients with PACS undergoing comprehensive neuropsychological assessment during follow-up.

2. Materials and Methods

2.1. Study design and setting

The Neuro-COVID Study is an observational, longitudinal, monocentric study conducted at the Post-COVID outpatient service of the National Institute for Infectious Diseases Lazzaro Spallanzani (IRCCS), Rome, Italy.

The study was approved by the Ethics Committee of INMI Lazzaro Spallanzani (approval number 119/2020, dated 20 May 2020) and conducted in accordance with the Declaration of Helsinki. All participants provided written informed consent prior to participation.

2.2. Participants

Patients with a documented history of SARS-CoV-2 infection and persistent neurological or cognitive symptoms were eligible for inclusion. Participants were either referred by clinicians or self-referred to the Post-COVID outpatient service for symptoms occurring at least four weeks after the acute infection, consistent with the definition of post-acute COVID-19 syndrome. We acknowledge that this four-week threshold reflects the earlier NICE definition and is broader than the more recent three-month criterion currently recommended by the WHO, NASEM and CDC; the implications of this inclusive window are addressed in the Discussion.

Patients who underwent neurocognitive assessment within seven months after resolution of the acute infection were considered eligible for the present analysis. The study period extended from September 2020 to January 2023. Because the baseline (t0) assessment could be performed at any time up to seven months after the resolution of the acute infection rather than at a fixed interval, the timing of the first evaluation varied across participants. This variability in the baseline window represents a potential source of heterogeneity, as patients assessed earlier and those assessed later may have been at different stages of spontaneous recovery at study entry.

A flow diagram illustrating the study population and patient selection process is shown in Figure 1.

2.3. Neurocognitive assessment

Neurocognitive assessment (NCA) was performed at three time points during follow-up: approximately three months (t0), six months (t1) and twelve months (t2) after the resolution of the acute infection.

Participants underwent a comprehensive standardized neuropsychological battery including 20 tests covering five cognitive domains: memory, attention, language, executive functions and psychomotor processing speed.

The following tests were administered:

Mini-Mental State Examination (MMSE) for global cognitive screening [20]

Rey Auditory Verbal Learning Test (RAVLT-ST, RAVLT-DR, RAVLT-REC) for verbal memory [21]

Digit Span Forward and Backward (DSF, DSB) for verbal short-term and working memory [22]

Corsi Span Forward and Backward (CSF, CSB) for visuospatial memory [22]

Rey-Osterrieth Complex Figure (ROCF-Copy and ROCF-Delayed Recall) for visuospatial constructional abilities and visual memory [23]

Trail Making Test A and B (TMTA, TMTB) for processing speed and executive functioning [24]

WAIS-R Digit Symbol (DS) for psychomotor processing speed [25]

Stroop Test (ST) for inhibitory control and cognitive flexibility [26]

Multiple Features Target Cancellation (MFTC) for attention [27]

Phonological and categorical verbal fluency tests (PVF, CVF) for language production [28]

Functional autonomy was evaluated using the Instrumental Activities of Daily Living scale (IADL) [29].

Raw test scores were adjusted for age, sex and educational level and converted into equivalent scores according to Italian normative standards [30,31]. Equivalent scores range from 0 to 4, where 0 indicates pathological performance and 4 indicates performance within the upper normal range. Cognitive impairment was defined as performance below the normative cut-off (equivalent score = 0).

2.4. Neuropsychiatric assessment

The following questionnaires were administered to assess the presence of neuropsychiatric symptoms and their possible influence on neurocognitive performance: the Beck Anxiety Inventory (BAI) [32] which assess the cognitive and physiological symptoms of anxiety [range = 0–63 were classified as in: 0–7 = no symptoms, 8–15 = mild symptoms, 16–25 = moderate symptoms; 26–63 = severe symptoms]; the Beck Depression Inventory (BDI-II) [33], which assess cognitive, affective, and physiological symptoms of depression (Somatic-affective SA/ Cognitive C dimensions) [range = 0–63 were classified as in: 0–13 = no symptoms, 14–19 = mild symptoms, 20–28 = moderate symptoms; 29–63 = severe symptoms]; the Pittsburgh Sleep Quality Index (PSQI) [34] for sleep quality assessment (if score >5 indicates the presence of poor sleep quality). It should also be noted that several BDI-II items (e.g., fatigue, loss of energy, changes in sleep, appetite or weight, difficulty concentrating and reduced interest in activities) overlap substantially with the somatic and cognitive manifestations of PACS itself. Consequently, BDI-II scores in this population may partly capture the somatic burden of long COVID rather than depression per se, and depression severity may be overestimated. The use of the separate somatic-affective (SA) and cognitive (C) dimensions was intended to partially mitigate this issue.

2.5. Statistical analysis

Data were summarized using descriptive statistics. Continuous variables are presented as medians and interquartile ranges (IQR), while categorical variables are expressed as frequencies and percentages.

Comparisons between categorical variables were performed using the Chi-square test or Fisher’s exact test when appropriate. Longitudinal changes in cognitive and psychiatric variables across the three time points were analysed using the non-parametric Friedman test.

Statistical significance was defined as p < 0.05. To account for the multiple comparisons inherent in the cognitive–neuropsychiatric association analyses, p-values were adjusted using the Benjamini–Hochberg false discovery rate (FDR) procedure, with a significance threshold set at FDR < 0.05. Effect sizes for longitudinal changes were quantified using Kendall’s W (interpreted as small ≥0.10, medium ≥0.30, large ≥0.50). All analyses were conducted using SPSS version 29.0 (IBM Corp., Chicago, IL, USA).

To minimize practice (test–retest) effects arising from repeated assessment, alternate (parallel) forms of the neuropsychological tests were administered across the three time points whenever available. The use of alternate forms reduces the influence of task-specific learning on serial performance, so that the longitudinal changes observed are less likely to be attributable to mere re-exposure to the same test material.

3. Results

3.1. Participant characteristics

A total of 42 patients were included in the analysis. The median age was 57 years (IQR 38–81), and 35.7% of participants were female. The median educational level was 13 years (IQR 8–18). Most patients had been previously hospitalized during the acute phase of COVID-19 (81%), and 66.7% required oxygen therapy. Of the 967 patients initially enrolled at the Post-COVID outpatient service, only 42 (4.3%) completed all three neurocognitive assessments and were included in the present longitudinal analysis (see Figure 1). Non-completion of follow-up assessments was attributed to multiple factors, including perceived cognitive recovery leading to voluntary dropout, lack of motivation to return in the absence of ongoing symptoms, severe psychological distress precluding continued participation, and work or logistical constraints limiting availability for repeated testing.

To characterise potential selection bias, baseline demographic and clinical characteristics of study completers were formally compared with those of non-completers (n=544; Table S1). Age (completers: median 57 years, IQR 51–60; non-completers: median 54 years, IQR 47–61; p=0.273) and educational level (both groups: median 13 years; p=0.359) did not differ significantly between groups. Notably, however, completers had a significantly higher prevalence of acute-phase respiratory support (66.7% vs 33.6%; p<0.001), pulmonary embolism (16.7% vs 2.9%; p<0.001), corticosteroid use (66.7% vs 47.4%; p=0.016), and tocilizumab/sarilumab treatment (7.1% vs 1.1%; p=0.021), and were more frequently female (64.3% vs 44.7%; p=0.014). These findings indicate that completers were, if anything, more severely ill during the acute phase than non-completers, rather than representing a healthier or more motivated subgroup. The direction of any resulting selection bias is therefore likely to be towards overrepresentation of severe acute disease and its sequelae, which should be considered when interpreting the observed cognitive trajectories.

Baseline demographic and clinical characteristics of the study population are summarized in Table 1.

3.2. Longitudinal neurocognitive outcomes

Longitudinal analyses revealed significant changes in several cognitive domains across the three assessments.

Improvements in verbal short-term learning (RAVLT-ST) were observed over time, with the proportion of impaired patients decreasing from 23.8% at baseline (t0) to 11.9% at t1 and 9.5% at t2 (p = 0.032).

Similarly, visuospatial short-term memory (CSF) improved significantly, with impairment rates decreasing from 31% at t0 to 2.4% at t1 and 0% at t2 (p < 0.0001). Improvements were also observed in visuospatial working memory (CSB) (p = 0.002).

Significant improvements were additionally observed in constructional praxis (ROCF-C; p = 0.018), phonological verbal fluency (PVF; p = 0.05) and psychomotor processing speed (Digit Symbol; p = 0.002).

Sleep quality also significantly improved across follow-up evaluations, with the proportion of patients reporting poor sleep quality decreasing from 31% at baseline to 7.1% at t1 and 2.4% at t2 (p < 0.0001).

Detailed results of the longitudinal neurocognitive and neuropsychiatric analyses are presented in Table 2.

3.3. Association between neurocognitive performance and neuropsychiatric symptoms

The following analyses of associations between neurocognitive performance and neuropsychiatric symptoms are considered exploratory, given the multiple comparisons and small sample size. P-values were adjusted using the Benjamini–Hochberg FDR procedure; all reported associations survived FDR correction (FDR < 0.05).

At baseline (t0), increased anxiety symptoms were significantly associated with poorer performance in verbal short-term memory (Digit Span Forward; p = 0.040) and psychomotor processing speed (Digit Symbol; p = 0.020). Sleep disturbances were also associated with reduced processing speed (p = 0.018).

At the second assessment (t1), anxiety symptoms were associated with poorer verbal learning performance (RAVLT-ST; p = 0.035), while sleep disturbances were associated with poorer performance in both verbal learning (RAVLT-ST; p = 0.033) and verbal long-term memory (RAVLT-DR; p = 0.010).

At the final assessment (t2), depressive symptoms were associated with poorer performance in verbal long-term memory (RAVLT-DR).

A detailed representation of these associations is provided in Figure 2A–C.

4. Discussion

To our knowledge, this study represents one of the few longitudinal investigations providing detailed neuropsychological monitoring across multiple cognitive domains in patients with PACS using in-person standardized assessment.

This longitudinal study investigated the evolution of neurocognitive performance and neuropsychiatric symptoms in patients with post-acute COVID-19 syndrome (PACS) undergoing comprehensive neuropsychological assessment over a one-year follow-up period. Our findings indicate that cognitive performance may gradually improve over time in several domains, particularly verbal short-term learning, visuospatial memory, working memory, constructional praxis, phonological verbal fluency and psychomotor processing speed. However, recovery was heterogeneous across cognitive domains, and a subset of individuals continued to show impaired or borderline performance during follow-up.

These findings are consistent with previous studies reporting partial recovery of cognitive function after COVID-19, in which deficits persisting for months tend to improve progressively in a proportion of patients [35,36,37,38], particularly in attentional and executive functions, while residual impairments remain in some individuals [8,16]. These improvements should nonetheless be interpreted with caution. The use of alternate (parallel) test forms across assessments makes test–retest learning an unlikely explanation, supporting a genuine change in performance. However, most participants had been hospitalized, frequently with oxygen and steroid treatment, and substantial cognitive improvement during the year after hospital discharge is well documented across critically ill populations independently of SARS-CoV-2; as detailed in the Limitations, the attrition pattern in our cohort (Table S1) selectively overrepresents patients with more severe acute illness. The early entry criterion (from four weeks after infection) further increases the likelihood that part of the improvement reflects natural resolution of acute illness rather than a chronic PACS-specific process, contributions our design cannot fully disentangle. Recent large longitudinal studies are consistent with this pattern: in a prospective in-person cohort, most cognitive domains improved progressively while processing speed and executive functioning remained below the normative mean [39], and serial assessments in the COVID and Cognition Study similarly described gradual symptom and cognitive change over time [40].

The mechanisms underlying cognitive dysfunction in PACS are likely multifactorial. Neuroinflammatory responses triggered by SARS-CoV-2 have been proposed as a major contributor to persistent neurological symptoms [13], potentially affecting neuronal function, synaptic transmission and cerebral microcirculation, while systemic factors such as hypoxia, endothelial dysfunction and metabolic disturbances during acute infection may also contribute to neural injury [9,10,11,12]. However, as the present study did not include neuroimaging or inflammatory biomarkers, these pathways were not measured and no causal link can be established. The gradual improvement seen in our cohort is compatible with, but cannot be attributed to, the progressive resolution of inflammatory processes; alternative explanations, including expected post-hospitalization recovery and practice effects, are at least equally plausible.

Improvements were not uniform across domains: while several functions recovered, others remained stable and a small proportion of individuals showed worsening of previously normal scores. Similar heterogeneity has been reported previously, with cognitive trajectories varying substantially between individuals [8,40]. A recent 36-month study of hospitalized COVID-19 patients identified four distinct trajectories—stable normal function, recovery, persistent impairment and delayed decline [41]—while other long-term data reported essentially negligible change in most patients [42], confirming that the magnitude and direction of cognitive change remain heterogeneous across studies. These patterns may reflect differences in disease severity, individual vulnerability, pre-existing conditions or psychological factors.

Sleep quality improved significantly over time. Sleep disturbances are among the most prevalent symptoms in PACS [43,44] and may contribute to cognitive complaints by affecting attention, memory consolidation and executive functioning; the observed improvement may therefore have played a role in the recovery of certain cognitive functions, as sleep disturbances and mental fatigue are closely related to performance in tasks involving processing speed and attentional resources [17,45].

Our results also highlight the complex relationship between neurocognitive performance and neuropsychiatric symptoms in patients with PACS. Anxiety symptoms were associated with poorer performance in verbal short-term memory and psychomotor processing speed during the first assessment. In addition, sleep disturbances were associated with reduced performance in processing speed and verbal memory tasks during follow-up. These findings are consistent with previous research indicating that psychological distress may contribute to cognitive impairment in patients recovering from COVID-19 [40,46,47,48].

The relationship between depressive symptoms and cognition changed across assessments: at baseline, depressive symptoms were unexpectedly associated with better visuospatial memory (CSF/CSB), whereas later evaluations showed an association with poorer verbal long-term memory. The counter-intuitive baseline association is most likely spurious, given the small sample, the many cognitive–neuropsychiatric pairs tested without correction for multiple comparisons, and dichotomized categories with few impaired cases. All associations were tested using Chi-square or Fisher’s exact test (expected counts <5) on dichotomised variables, with n=42 at each time point, and should be regarded as hypothesis-generating only. These fluctuations may reflect the bidirectional interaction between mood and cognition, whereby psychological distress influences cognition through reduced motivation, impaired attentional control or mental fatigue, while persistent cognitive difficulties may maintain anxiety and depressive symptoms.

From a clinical perspective, these findings emphasize the importance of comprehensive neuropsychological assessment in PACS, as cognitive complaints may reflect both objective alterations and psychological factors. Long-term monitoring may help identify individuals at risk of persistent impairment and guide targeted rehabilitation, and integrated approaches addressing both cognitive and emotional symptoms may improve quality of life.

Several limitations should be considered. First, the study did not include pre-infection neurocognitive assessments, limiting estimation of cognitive decline attributable to SARS-CoV-2. Second, the absence of a matched control group prevents comparison with individuals without prior COVID-19. Third, the small sample size limits statistical power and generalizability: only 42 of 967 enrolled patients (4.3%) completed all three assessments. As shown in Table S1, completers did not differ from non-completers in age or education but were more severely ill acutely (respiratory support 66.7% vs 33.6%, p<0.001; pulmonary embolism 16.7% vs 2.9%, p<0.001; corticosteroids 66.7% vs 47.4%, p=0.016), so the sample is enriched for hypoxia-related and post-critical-illness trajectories rather than healthier individuals. Fourth, the cohort was older than the general PACS population and predominantly hospitalized, two-thirds receiving oxygen and steroids; cognitive dysfunction and its improvement may be driven largely by hypoxia-related injury, critical illness and expected post-hospitalization recovery common across non-COVID populations, rather than PACS-specific mechanisms, while the high prevalence of hypertension (42.9%) and vascular risk factors may have contributed through a vascular pathway. These features limit generalizability to the broader, mostly non-hospitalized long-COVID population with brain fog. Fifth, no neuroimaging or inflammatory biomarker data were available, so mechanistic interpretations remain speculative. Finally, the inclusion criterion of symptoms from four weeks after infection reflects an earlier, broader definition of PACS than the three-month criterion now recommended by the WHO, NASEM and CDC, and the variable baseline window (up to seven months) adds heterogeneity in the timing of first evaluation.

Despite these limitations, the present study provides valuable longitudinal data based on detailed in-person neuropsychological assessments across multiple cognitive domains. The use of a comprehensive neuropsychological battery and repeated evaluations allowed a more nuanced characterization of cognitive trajectories in patients with PACS.

Future studies should aim to include larger multicentre cohorts and appropriate control groups in order to better clarify the mechanisms and long-term trajectory of cognitive impairment following COVID-19. In addition, further research integrating neuroimaging, inflammatory biomarkers and neuropsychological assessment may help elucidate the biological processes underlying cognitive dysfunction in PACS.

5. Conclusions

Our results provide descriptive longitudinal evidence suggesting that, in a small and predominantly hospitalized cohort with post-acute COVID-19 syndrome, performance on several cognitive measures tended to improve over the first year of follow-up, while deficits in a subset of cognitive functions persisted. Critically, a formal comparison showed that study completers were more severely ill during the acute phase than non-completers, indicating that the sample selectively represents patients with more severe acute disease rather than those who recovered most easily; the observed improvements may therefore partly reflect the expected post-hospitalisation recovery trajectory rather than PACS-specific processes. The use of alternate test forms across assessments makes a substantial contribution of practice effects unlikely. Exploratory analyses indicated that anxiety, depressive symptoms and sleep disturbances were associated with poorer performance across several cognitive domains; these findings should nonetheless be regarded as hypothesis-generating only. Integrated neuropsychological and psychiatric monitoring appears warranted in this population. Larger controlled studies with non-hospitalised comparison groups are needed to clarify the specific contribution of PACS to the observed cognitive trajectories.