Prevalence of mental health conditions and brain fog in people with long COVID: A systematic review and meta-analysis

A B S T R A C T Objective: Long COVID can include impaired cognition (‘brain fog ’ ; a term encompassing multiple symptoms) and mental health conditions. We performed a systematic review and meta-analysis to estimate their prevalence and to explore relevant factors associated with the incidence of impaired cognition and mental health conditions. Methods: Searches were conducted in Medline and PsycINFO to cover the start of the pandemic until August 2023. Included studies reported prevalence of mental health conditions and brain fog in adults with long COVID after clinically-diagnosed or PCR-confirmed SARS-CoV-2 infection. Findings: 17 studies were included, reporting 41,249 long COVID patients. Across all timepoints (3 – 24 months), the combined prevalence of mental health conditions and brain fog was 20 ⋅ 4% (95% CI 11 ⋅ 1%-34 ⋅ 4%), being lower among those previously hospitalised than in community-managed patients(19 ⋅ 5 vs 29 ⋅ 7% respectively; p = 0 ⋅ 047). The odds of mental health conditions and brain fog increased over time and when validated instruments were used. Odds of brain fog significantly decreased with increasing vaccination rates (p = ⋅ 000). Conclusions: Given the increasing prevalence of mental health conditions and brain fog over time, preventive interventions and treatments are needed. Research is needed to explore underlying mechanisms that could inform further research in development of effective treatments. The reduced risk of brain fog associated with vaccination emphasizes the need for ongoing vaccination programs.

Objective: Long COVID can include impaired cognition ('brain fog'; a term encompassing multiple symptoms) and mental health conditions.We performed a systematic review and meta-analysis to estimate their prevalence and to explore relevant factors associated with the incidence of impaired cognition and mental health conditions.Methods: Searches were conducted in Medline and PsycINFO to cover the start of the pandemic until August 2023.Included studies reported prevalence of mental health conditions and brain fog in adults with long COVID after clinically-diagnosed or PCR-confirmed SARS-CoV-2 infection.Findings: 17 studies were included, reporting 41,249 long COVID patients.Across all timepoints (3-24 months), the combined prevalence of mental health conditions and brain fog was 20⋅4% (95% CI 11⋅1%-34⋅4%), being lower among those previously hospitalised than in community-managed patients(19⋅5 vs 29⋅7% respectively; p = 0⋅047).The odds of mental health conditions and brain fog increased over time and when validated instruments were used.Odds of brain fog significantly decreased with increasing vaccination rates (p = ⋅000).Conclusions: Given the increasing prevalence of mental health conditions and brain fog over time, preventive interventions and treatments are needed.Research is needed to explore underlying mechanisms that could inform further research in development of effective treatments.The reduced risk of brain fog associated with vaccination emphasizes the need for ongoing vaccination programs.
The prevalence of mental health conditions (International Classification of Diseases 11th Revision (ICD-11) [6] has been explored in studies on COVID and some studies presented such symptoms in COVID over time, without taking into account when they occur specifically in long COVID [7].The impact of hospitalization rather than community care for acute COVID-19 or of prior vaccination on the prevalence of mental health conditions in long COVID is unknown, as well as whether their prevalence varies over time after acute COVID-19 disease.
For this review we used the World Health Organisation (WHO) definition of a mental health condition as a mental disorder: 'characterized by a clinically significant disturbance in an individual's cognition, emotional regulation, or behaviour.It is usually associated with distress or impairment in important areas of functioning.'Mental disorders are listed as mental, behavioural or neurodevelopmental disorders in Chapter 6 of the International Classification of Diseases 11th Revision (ICD-11) developed by the WHO.'Mental health conditions is a broader term covering mental disorders, psychosocial disabilities and (other) mental states associated with significant distress, impairment in functioning, or risk of selfharm.'Such subthreshold symptoms are associated with distress or impairment in functioning without reaching the full criteria for a disorder.They are listed as mental or behavioural symptoms and signs in Chapter 21, symptoms, signs or clinical findings, not elsewhere classified, in the ICD-11 [6].This definition seems the most suitable, as it allows for exploring the prevalence of mental disorders as diagnosed medical conditions, but also subthreshold symptoms such as anxiety or low mood, associated with distress or impairment in functioning or future incidence of full disorder without reaching the full criteria for a mental disorder at that time [8].
'Brain fog' is not a medical term but is a term used to encompass a range of symptoms including poor concentration, feeling confused, thinking more slowly than usual, fuzzy thoughts, forgetfulness, lost words and mental fatigue [9].It can occur in many medical conditions; but, to date has been mentioned especially in the context of long COVID.It can lead to impaired functioning or distress [10].However, so far no research has explored the full range of its comprised elements and their prevalence, whereas indications are that they impact severely on work functioning.Such knowledge is important, given the high incidence of long COVID [3] and its impact on healthcare costs and workforce [11,12].
To date, there have been no systematic reviews exploring prevalence rates of mental health conditions in long COVID.Also, no studies addressed prevalence of brain fog as a composite measure of cognitive symptoms in long COVID so far.We thus sought to estimate the prevalence of any mental health condition or brain fog in long COVID, and to explore potential risk factors for their presence, covering studies published over the first 2.5 years of the pandemic.

Method
This systematic review and meta-analysis followed a preregistered protocol (PROSPERO registration: CRD42023394105) [13].Results are reported following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statement (Supplementary Table 1) [14].

Search strategy (See supplementary table 2)
Medline and PsycINFO were searched to cover the start of the pandemic until August 31st 2023, to identify cross-sectional or longitudinal adult human studies relating to long COVID.No date or language restrictions were applied.Title and abstract screening used the Rayyan platform [15].Observational studies were included that reported prevalence rates, odds ratios, or hazard ratios for any mental health conditions, cognitive symptoms or brain fog ≥12 weeks after acute SARS-CoV-2 infection.We excluded abstracts, conference reports, or letters to editors; case studies, case series, qualitative studies, surveys, and intervention trials; and studies with <50 participants (to avoid small study effects) [16].

Quality assessment
Criteria for low risk-of-bias (assessed independently by two reviewers [FT, JSw] using cohort, cross-sectional and prevalence checklists from the Joanna Briggs Institute as appropriate) [17] included: • Long COVID and control groups comparable and drawn from the same population • COVID-19 measured consistently in individuals in exposed/unexposed groups • Valid instruments to assess COVID-19, long COVID and mental health conditions or brain fog, such as the medical diagnosis of COVID infection [18], a positive PCR test [19], and clear definition of long COVID in the manuscript; use of the PHQ9 [20], GAD7 [21] or another validated questionnaire for establishing depressive or anxiety symptoms; ICD codes for mental disorders [22]; and MoCA [23] or other cognitive tests to establish cognitive symptoms.• Identification of confounding factors and addressing their management within the study • Completion of follow-up, exploration of reasons for attrition, addressing incomplete follow-up within the study • Utilization of appropriate statistical analysis techniques • Participants being free of long COVID at onset of the study to only report incident cases.
Discrepancies in appraisal were resolved through discussion with a third reviewer [CFC].

Stage 1
Search results were uploaded to Rayyan [13] and after removing duplicates, titles and abstracts were screened against the concrete, predefined inclusion and exclusion criteria described in 2.1 independently by three reviewers [JSw, JSh, FT].In several stages, inclusion and exclusion criteria were discussed for clarification with an independent reviewer [CFC].Clear, pre-defined criteria, inter-rater reliability checks and regular discussions about any uncertainties in screening were incorporated into this process to decrease the risks of mistakes in inclusion [24] Duplicate blind screening of a random 10% of references was undertaken [from pairs of JSw, JSh, FT] to confirm inter-reviewer consistency.The level of agreement, rated as include or exclude, ranged from 97% indicating at least moderate levels of screening reliability [25].As agreement and consistency were high, it was considered acceptable to single-screen remaining references [26].Any uncertainties were discussed and adjudicated by an independent reviewer [CFC].Where insufficient information was available to make a clear decision, references were retained for further screening.

Stage 2
Full texts of included articles were reviewed independently [JSw, JSh, FT].A random 10% of total papers were checked by a second reviewer to ensure reliability.Disagreements were discussed and arbitrated by an alternative reviewer [CFC].To ensure transparency, all C. van der Feltz-Cornelis et al.
papers excluded during full-text screening and the associated reasons for exclusion are listed in Supplementary List 1.

Data extraction
A data extraction format was designed to extract (independently by each) (1) general study information (e.g.author, year of publication, title, country) (2) study aim (3) design (e.g.control group, timeframe, sample size, sampling method and description, COVID-19 assessment method) (4) reporting of comorbidities (5) prevalence rates of mental health and cognitive symptoms or conditions (6) long COVID definition and diagnostic method (7) method of mental health assessment (e.g., interview, validated or non-validated questionnaire, routine medical data) (8) cut-off criteria for mental health assessments to meet the criteria for a mental health condition or cognitive symptoms of brain fog; and (9) main findings.Uncertainties were decided by an independent reviewer [CFC].Where additional information was required, corresponding authors were contacted.Where authors did not respond or did not have relevant data, studies were excluded from analysis (supplementary list 1).
In general, we assumed that if a study reported cognitive symptoms separately from anxiety/depression, those cognitive symptoms stood alone and were not reported as part of the mental health conditions themselves.However, as overlap between mental health conditions and cognitive symptoms might exist (e.g.concentration problems and anxiety), we showed the amount of overlap (Results, Fig. 1).We assessed mental and physical fatigue where differentiated, or if fatigue was listed as a physical or general symptom.
Country and time vaccination and COVID-19 period data were identified using Our World in Data [27] and the Johns Hopkins University Coronavirus Resource Centre [28]; virus strain from the Nextstrain project [29]; and variants of concern from the European Centre for Disease Prevention and Control [30,31].

Data analysis
Pooled prevalence rates of mental health conditions and brain fog were assessed by random effects meta-analysis (Comprehensive Meta-Analysis Version 2) [32].The effect size was the event rate, reported as prevalence rates in percentages with 95% Confidence Interval (CI) and weights provided.Between-study variability was examined for heterogeneity, using the Q statistic for quantifying inconsistency.We planned a moderator analysis exploring the influence of several potential factors on prevalence (e.g.post-hospitalization versus neverhospitalised (community-managed) COVID-19 patients; presence/absences of pre-COVID-19 mental health conditions and brain fog).Overlapping symptoms such as fatigue or fatiguability can occur in mental health and other medical conditions as they are intertwined, and attention has been drawn to the diagnostic issues with that [33].To deal with this issue, we planned to run the analysis for mental health conditions with a random model without fatigue, 'Brain Fog' was assessed as the combined presence of cognitive symptoms and of mental fatigue [9,34] where data concerning mental fatigue were available, or as the presence of one or more cognitive symptom if not.
We conducted meta-regression to estimate changes in the prevalence of mental health conditions and brain fog over time after acute COVID-19 disease, and their association with geographically-specific vaccination rates at the time of the study, and with diagnosis from medical files Fig. 1.Forest plot event rates of combined mental health conditions and brain fog cognitive symptoms at all timepoints.
C. van der Feltz-Cornelis et al.
versus that derived from validated instruments.Tau-squared was calculated to establish the variance of true effect sizes in logit units.A sensitivity analysis excluding studies with high risk-of-bias was planned.

Role of the funding source
The funder of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report.

Results
Searches yielded 7541 studies: 7415 were excluded (582 duplicates; 6833 not meeting criteria for progression to stage 2), leaving 126 studies of which 109 were excluded by full-text screening (reasons in Supplementary Fig. 1, Supplementary List 1).Seventeen studies with data allowing the estimation of prevalence rates were thus included in the meta-analysis (9 prospective observational cohort studies, 5 observational, 3 cross-sectional).Three were Spanish, 3 British, 3 German, 2 French and 2 Chinese, with one study each from South Korea, Switzerland, Luxembourg, and India.Full details of included studies are shown in Supplementary Figure 1 [14].
The number of COVID-19 patients per study ranged from 72 to 86,157, with 41,249 of the total 146,231 (28%) suffering from long COVID.Seven studies included long COVID patients never hospitalised for SARS-CoV-2 infection (n = 38,774, 94%).Ten included those who had been hospitalised (2475, 6%).Twelve studies (n = 4609) reported the gender of long COVID sufferers (female 2660: 58%).Subramanian and colleagues [35] failed to respond to our inquiries regarding healthy control data, making it impossible to compare the prevalence of mental health conditions and brain fog between long COVID and non-COVID-19 control subjects.There were insufficient data to include mental fatigue in a composite measure of brain fog, so we included only cognitive symptoms encompassing concentration difficulties, memory impairment and mental confusion.Characteristics of included studies are shown in Table 1.
Although cut scores for clinical levels of depression (PHQ9) and anxiety disorder (GAD7), (HADS) exist, studies reported depression and anxiety at symptom level, not as clinical diagnosis, except in case of one study using ICD codes [36].Only two studies [37,38] used a validated test for cognitive symptoms (the MOCA).We have included concentration difficulties, memory loss and mental confusion as cognitive symptoms of brain fog.
The combined prevalence of all mental health conditions and brain fog reported in all studies over all follow-up periods was 20⋅4% (95% CI 11⋅1%-34⋅4%; 17 studies).There was significant and large heterogeneity (99%, p < ⋅001) between studies.When differentiating between specific mental health conditions and brain fog; however, there was no significant heterogeneity (p = ⋅92).
Meta-regression showed that brain fog was twice as likely to be reported when validated assessment instruments were used (p < ⋅001), compared to being diagnosed from medical files (Fig. 4).For mental health conditions, probability was higher if validated measures were used, however to a smaller degree (p < .001)(Fig. 5).
By meta-regression analysis, the probability of brain fog, but not of mental health conditions, decreased with increasing vaccination rates by up to 3.7 times (p < ⋅001) (Supplemental Fig. 2, and Supplemental Fig. 3).
Meta-regression indicated brain fog was more prevalent with longer duration after acute SARS-CoV-2 infection (p ≤⋅001), with the probability rising 2.5 from onset of long COVID to 24 months after infection.(Supplemental Fig. 4).For mental health conditions, the increase was approximately 1.2 times (p < .001).(Supplemental Fig. 5).
The majority of included papers (n = 15; 88%) reported the information expected for the study design-type and were rated to have low risk-of-bias.Eight studies had unclear reporting regarding follow-up and missing data.Two [1,25] had moderate risk, with unclear reporting of three and four out of ten criteria respectively.None were rated high-risk, precluding the need to perform a sensitivity analysis.Full quality assessment results in Supplementary Table 3.
As there was no control group without COVID-19, we could not run a publication bias analysis.However, the estimate of the prediction interval for true effects was 0⋅02 to 0⋅72 with the true effect size in 95% of all comparable populations falling within this interval [56].This supports the validity and generalizability of our findings as there are no indications of publication bias.

Discussion
This systematic review and meta-analysis of 17 studies and 41,249 long COVID patients infected in the first 30 months of the pandemic across three continents is the first to specifically assess the prevalence of any mental health condition or brain fog in long COVID, and to compare between the two when exploring factors that may be relevant to their manifestation.
To summarize, mental health conditions and brain fog both occurred in around one in five of patients between three months and two years after COVID-19 infection.This is a major public health problem.Given the high percentage of people developing long COVID in most studies, this finding is concerning and should have implications for provision of care, that seems to be stretched currently [57].The findings may also be important for recovery as comorbid mental health conditions in chronic conditions are known to impair both recovery and participation in rehab programmes.The increase of prevalence over time after acute infection occurs both in brain fog and in mental health conditions and would potentially allow preventive approaches.Given that vaccination appears to be protective for brain fog, this calls for sustained vaccination programs.
Regarding the interpretation of the findings, it is known that depression or cognitive symptoms can occur following other infective conditions such as pneumonia [58] and stroke in similar percentages and with a similar increase over time as in long COVID [59].Alternatively, mental health conditions in long COVID might be reactive to the presence of long-term illness.In addition, long COVID may be a chronic condition that clusters with depression and anxiety as it does for other disease states like diabetes, COPD and cardiovascular disorder [60][61][62].
Brain fog is not listed as a mental disorder in the ICD-11, but as 'clouding of consciousness' (with brain fog being listed as a synonym) in the ICD-11 chapter 'Symptoms, signs or clinical findings' [ICD-11] C. van der Feltz-Cornelis et al. [6,22].Also, brain fog is not listed as a medical condition but as a set of symptoms that can occur in long COVID [9].Nevertheless, overlap among mental health conditions and brain fog might occur as cognitive symptoms are part of depressive and anxiety disorder and this could affect the findings.We assumed that if a study reported anxiety or depression and also cognitive symptoms, those cognitive symptoms were independent of the mental health conditions, as this is common diagnostic and classification procedure in clinical and research assessments.However, as this was not always clear from the studies included, such overlap might still have occurred, potentially leading to underreporting of cognitive symptoms in the context of brain fog.We therefore showed the n of the relevant symptom compared to the N of the sample per study in Fig. 1, and this shows that in four of the 17 studies there are indications that overlap may have occurred [38,41,45,47].In those studies, the overall prevalence rates should not be interpreted as mutually exclusive.As the relative weight of those studies was similar to the other studies, this may however have had a small influence in the analysis.Furthermore, the strong association with validated assessment methods for brain fog indicate that its diagnosis requires specific and valid neuropsychological testing.Prevalence rates were higher in studies that used validated diagnostic instruments versus routine medical data or databases which might (at the beginning of the pandemic) have suffered from underreporting.Taking 'brain fog' as a catch-all term to include mental health conditions would not be supported by our findings.Formal diagnostic criteria for the cognitive symptoms in brain fog are needed, which should separate them from mental health conditions such as anxiety, depression and psychological distress.Research is therefore needed to address the characteristics, and pathogenesis of brain fog in long COVID.As it seems that fatigue and cognitive aspects may be strongly intertwined in brain fog in long COVID; both should be addressed.
Prevalence rates were lower in patients who had been admitted to hospital with acute COVID-19.The potential influence of bias seems low as the use of validated instruments for classifying mental health conditions and brain fog did not differ between these study populations, which were seemingly screened and reported in the same way.Survivor bias might help explain our findings.Likewise, those hospitalised might have received medication [63] that reduced acute symptoms and hence later prevalence rates.Hospitalised patients may also have been 'grateful to be alive', with those ill in the community being more socially isolated.Finally, those who stayed in the community may have had less access to medical care [64,65].

Strengths of the review
Study strengths include the fact that both mental health conditions and brain fog were simultaneously studied and that factors associated with their presence in long COVID were compared.The most common approach for assessment was the use of standardized assessments such as GAD-7 [21] and PHQ-9 [20] in an interview [38,40,43,45,48] or for selfreporting [41,46,50], for mental health conditions, and the use of valid tests such as the MOCA for cognitive symptoms.The size, timing, geographical spread and methodology of included studies makes the risk of bias low to moderate.Heterogeneity was low 12 months after infection and when type of mental health condition or brain fog were taken into account.We explored the association between prevalence rates and follow-up time and vaccination, and hospital admission versus community management in the acute phase.We also explored potential factors affecting prevalence rates separately for brain fog and mental health conditions.

Limitations
Our prevalence estimates and rates should be interpreted with caution.The lack of matched studies reporting the prevalence rate of   23⋅2-28⋅0) in cases of anxiety disorders worldwide in 2020 [66,67], how many of these patients suffered long COVID is unknown.Another limitation is that COVID disease affected some groups disproportionately but we could not get data by ethnicity, age, socio-economic group   to explore this.In addition, regarding the negative association between vaccination rates and prevalence of brain fog, as these analyses are based on approximations at country level, we could not explore an association between virus strain and the prevalence of mental health conditions or brain fog as virus strain data for included study samples were not reported, and most studies in this review were performed during waves with several strains.So this finding is exploratory and should be confirmed in further research.In addition, although overlap between brain fog and mental health conditions seems to have been limited, nevertheless its potential occurrence is a limitation of the study.
Because no studies reported mental fatigue explicitly, we could only provide an estimate of brain fog prevalence based on a composite of cognitive measures.We lacked information on a history of previous COVID-19, brain fog or of mental health conditions.
It should be noted that the research summarised in this review is fundamentally descriptive, so causal attribution to acute COVID-19 disease is not possible.However, our findings that approximately one in five to one in four of long COVID cases experience mental health conditions or brain fog suggest there is a high demand for effective and accessible treatments.Funding for research into effective treatments for mental health conditions and brain fog in people with long COVID is urgently needed.This is the first systematic review to determine the prevalence rates of both mental health conditions and brain fog in long COVID, and to compare factors that may be associated with their presence.Both are common, with prevalence rising with time after SARS-CoV-2 infection.The prevalence of brain fog, but not of mental health conditions, is inversely related to vaccination rates, suggesting some degree of  pathogenic independence.Research is required to better characterise the neurocognitive features of brain fog, their pathogenesis, and brain fog's relationship (if any) to other mental health conditions, such that therapeutic targets might be identified.As our findings emphasize the value of vaccination in preventing brain fog as a core symptom of long COVID, ongoing vaccination programmes should be encouraged and reinforce the need for effective treatments to manage mental health conditions and brain fog in long COVID.Furthermore, given the increasing prevalence of mental health conditions over time, and their potential negative impact on recovery, preventive treatment for mental health conditions in long COVID may be helpful.

Fig. 2 .
Fig. 2. Forest plot event rates grouped by mental health conditions and cognitive symptoms at 12 months.

Fig. 3 .
Fig. 3. Forest plot event rates of brain fog as a composite measure of cognitive symptoms.

C
. van der Feltz-Cornelis et al.

Fig. 4 .
Fig. 4. Meta-regression of probability that brain fog will occur in relation to diagnostic method. 1 = routine medical file/database; 2 = non-validated questionnaire; 3 = validated instruments.Area proportional to study weight.

Fig. 5 .
Fig. 5. Prevalence of mental health conditions association with diagnostic methods.Area proportional to study weight.

Table 1
Characteristics of included studies.

Table 1
(continued ) Studies reported depression and anxiety at symptom level, not as clinical diagnosis, except in case of studies using ICD codes.
*Original physician diagnosis texts ('long COVID syndrome,' 'post COVID syndrome,' 'post COVID complications').ICD-10 diagnoses like chronic fatigue, breathing abnormalities, smell and taste disturbances, malaise and fatigue, and attention disturbances were surrogate markers for LCS.†Control group included in this study: People in primary care without known SARS-COV-2 infection (n = 110,737).No prevalence data were provided in that group.C. van der Feltz-Cornelis et al.mental health conditions in long COVID with non-COVID-19 subjects is a limitation.Although the COVID-19 pandemic seems associated with a 27⋅6% increase (95% uncertainty interval: 25.1-30.3) in cases of major depressive disorder and a 25⋅6% increase (95% uncertainty interval: