https://doi.org/10.

1093/braincomms/fcab295 BRAIN COMMUNICATIONS 2022: Page 1 of 19 | 1

BRAIN
AIN COMMUNICATIONS

Rapid vigilance and episodic memory

decrements in COVID-19 survivors
Sijia Zhao,1 Kengo Shibata,2 Peter J. Hellyer,3,4 William Trender,3 Sanjay Manohar,1,2
Adam Hampshire3 and Masud Husain1,2

Recent studies indicate that COVID-19 infection can lead to serious neurological consequences in a small percentage of individuals.
However, in the months following acute illness, many more suffer from fatigue, low motivation, disturbed mood, poor sleep and
cognitive symptoms, colloquially referred to as ‘brain fog’. But what about individuals who had asymptomatic to moderate
COVID-19 and reported no concerns after recovering from COVID-19? Here, we examined a wide range of cognitive functions crit-
ical for daily life (including sustained attention, memory, motor control, planning, semantic reasoning, mental rotation and spatial–
visual attention) in people who had previously suffered from COVID-19 but were not significantly different from a control group on
self-reported fatigue, forgetfulness, sleep abnormality, motivation, depression, anxiety and personality profile. Reassuringly,
COVID-19 survivors performed well in most abilities tested, including working memory, executive function, planning and mental
rotation. However, they displayed significantly worse episodic memory (up to 6 months post-infection) and greater decline in vigi-
lance with time on task (for up to 9 months). Overall, the results show that specific chronic cognitive changes following COVID-19
are evident on objective testing even amongst those who do not report a greater symptom burden. Importantly, in the sample tested
here, these were not significantly different from normal after 6–9 months, demonstrating evidence of recovery over time.

1 Department of Experimental Psychology, University of Oxford, Oxford OX1 3PH, UK

2 Nuffield Department of Clinical Neurosciences, University of Oxford, Oxford OX3 9DU, UK
3 Department of Brain Sciences, Imperial College London, 926, Sir Michael Uren Hub, 86 Wood Lane, London W12 0BZ, UK
4 Institute of Psychiatry, Psychology and Neuroscience, King’s College London, London, UK
Correspondence to: Sijia Zhao
Department of Experimental Psychology
University of Oxford
Oxford OX1 3PH, UK
E-mail: [email protected]
Keywords: COVID-19; long-COVID; cognitive deficits; sustained attention; memory
Abbreviations: AMI = Apathy Motivation Index; BF = Bayesian Factor; BFI-S = Short Big Five Inventory; CFQ = Cognitive
Failures Questionnaire; COVID-19 = Coronavirus disease 2019 caused by the Coronavirus SARS-CoV-2; GLM = generalized
linear model; GRIT-S = Short Grit Scale; HADS = Hospital Anxiety and Depression Scale; LMM = linear mixed-effect model;
NFI = Neurological Fatigue Index; PCA = Principal Component Analysis; SD = standard deviation; SES = socioeconomic status

Received July 6, 2021. Revised October 8, 2021. Accepted December 20, 2021
© The Author(s) (2022). Published by Oxford University Press on behalf of the Guarantors of Brain.
This is an Open Access article distributed under the terms of the Creative Commons Attribution License (https://creativecommons.org/licenses/by/4.0/), which permits unrestricted reuse,
distribution, and reproduction in any medium, provided the original work is properly cited.
2 | BRAIN COMMUNICATIONS 2022: Page 2 of 19 S. Zhao et al.

Graphical Abstract

task, there was a significantly larger decline in perceptual

Introduction sensitivity, despite reporting the same levels of fatigue com-
pared to healthy controls. Subsequent testing also revealed a
People who survive COVID-19 infection present a signifi-
significantly larger episodic memory decrement. These re-
cantly higher risk of major neurological and psychiatric con-
sults demonstrated that chronic cognitive reductions follow-
ditions, particularly if they were hospitalized.1–3 These
ing COVID-19 are evident upon objective testing even in
include acute cerebrovascular events such as ischaemic
people who do not report long-COVID symptoms.
stroke and intracerebral haemorrhage.4 In addition to severe
neurological conditions, there can also be more chronic,
longer-term consequences such as fatigue, low motivation,
disturbed mood and poor sleep—all commonly reported Materials and methods
symptoms amongst survivors, the so-called long-COVID
(see recent review5). Participants
Whilst many studies employ questionnaires reliant on One hundred and fifty-five participants {59 females, mean
patients’ subjective self-reports, recent investigations have age 28.6 years [standard deviation (SD) 9.7]} were recruited
begun to document a wide range of post-COVID-19 cogni- from the Prolific online recruitment platform (http://www.
tive deficits using objective cognitive testing of inpatients. prolific.co). All participants were naïve about the aim of
Particular impairments were found in attention,6–12 this study; the study was advertised as ‘a brain game’ testing
memory6–8,10–13 and executive functions.10–12,14–16 More how well people could perform. Sixty-four people had con-
recently, using 18F-FDG PET, it has been demonstrated tracted COVID-19, whilst 91 reported that they had not.
that in the most severely affected patients, the degree of cog- Although none had received treatment in an intensive care
nitive impairment was accompanied by frontoparietal unit or were seeking post-COVID care at the time the study
hypometabolism.10,11 was conducted, three participants had been hospitalized for
Understandably, most of these small-scale reports have, to COVID-19, seven participants had severe COVID-19 symp-
date, predominantly focused on symptomatic, hospitalized toms (i.e. their COVID-19 symptoms largely reduced their
patients (however see17). But what about individuals who ability to carry-out day-to-day activities) and two partici-
have not been hospitalized and do not report any ongoing pants had severe long-COVID symptoms. In total, 11 parti-
symptoms after recovery? Do they suffer cognitive deficits cipants were excluded. In addition, a follow-up survey
that they are not aware of? Here, we examined people showed that eight control participants later discovered that
who had previously contracted COVID-19 but were not they contracted COVID-19 before via antigen detection
significantly different from a control group with respect to test; although we did not find any difference in behavioural
self-reported fatigue, forgetfulness, sleep abnormality, moti- performance between them the rest of the control group,
vation, depression, anxiety or personality profile. In total, they were also excluded from further analysis.
135 participants were tested on a series of 12 online cogni- In all, 136 participants were included in further analysis
tive tasks encompassing a wide range of cognitive functions (COVID n = 53, Control n = 83, see Fig. 1 for study popula-
critical for daily life, including the ability to sustain atten- tion flow and Table 1 for demographics). There was no signif-
tion, memory, motor control, speed of response, planning, icant group difference in gender [χ2(1,N = 54) = 0.5, P = 0.5)
semantic reasoning, mental rotation and spatial–visual at- or age t(134) = –0.6, P = 0.5, Bayesian Factor (BF) = 4.4].
tention. In all of these functions, COVID-19 survivors All participants were required to complete the experi-
showed the same initial baseline performance. However, ments on the Chrome browser on a desktop with a keyboard
after only two minutes on an attentionally demanding and mouse.
Post-COVID-19 attention and memory decline BRAIN COMMUNICATIONS 2022: Page 3 of 19 | 3

Figure 1 Study population flow chart. Number of participants eligible for each experimental session. Note that as some participants might
fall into multiple exclusion criteria, the totals may not add up exactly.

Table 1 Self-reported participant demographics and questionnaire-derived measures

Measure All (n = 136) COVID (n = 53) Control (n = 83) Statistic

Age, mean years (SD) 28.6 (9.7) 28.0 (8.6) 29.0 (10.3) t(134) = −0.6, P = 0.5, BF = 4.4
Gender, female (%) 54 (39.7) 23 (43.4) 31 (37.3) χ2(1,N = 54) = 0.5, P = 0.5
COVID-19 Time from COVID-19 diagnosis, 163.0 (128.1)
mean days (SD)
COVID-19 test type: 40/4/2/7
PCR/lateral flow/unknown/not
tested
Stayed at hospital overnight for 0 (0.0)
COVID-19 treatment, yes (%)
Stayed at ICU for COVID-19 0 (0.0)
treatment, yes (%)
Questionnaires NFI Physical fatigue (SD) 8.9 (4.7) 8.4 (4.4) 9.2 (4.9) t(134) = −1.0, P = 0.3, BF = 3.4
Cognitive fatigue (SD) 4.2 (2.8) 4.0 (2.8) 4.4 (2.8) t(134) = −0.7, P = 0.5, BF = 4.1
Sleep relief (SD) 7.8 (3.4) 7.7 (3.3) 7.8 (3.5) t(134) = −0.1, P = 0.9, BF = 5.3
Sleep abnormality (SD) 6.9 (2.8) 7.2 (2.9) 6.7 (2.7) t(134) = 0.9, P = 0.4, BF = 3.8
CFQ Forgetfulness (SD) 12.7 (5.0) 13.0 (4.8) 12.4 (5.2) t(86) = 0.6, P = 0.5, BF = 4.4
False triggering (SD) 8.8 (4.7) 9.3 (4.7) 8.3 (4.7) t(86) = 1.0, P = 0.3, BF = 3.3
Distractibility (SD) 12.3 (4.5) 12.4 (4.2) 12.2 (4.9) t(86) = 0.1, P = 0.9, BF = 5.3
BFI-S Conscientiousness (SD) 10.5 (2.5) 10.4 (2.3) 10.7 (2.7) t(77) = −0.6, P = 0.6, BF = 4.6
GRIT-S Grit scale (SD) 3.2 (0.7) 3.1 (0.6) 3.3 (0.8) t(77) = −0.9, P = 0.4, BF = 3.7
AMI Behavioural apathy (SD) 10.3 (4.4) 10.9 (4.5) 9.8 (4.2) t(77) = 1.1, P = 0.3, BF = 3.0
Social apathy (SD) 12.9 (5.0) 12.5 (5.4) 13.4 (4.4) t(77) = −0.8, P = 0.4, BF = 4.1
Emotional apathy (SD) 8.0 (4.2) 8.6 (4.4) 7.3 (3.9) t(77) = 1.4, P = 0.2, BF = 2.2
HADS Depression (SD) 5.4 (3.6) 5.4 (3.9) 5.4 (3.3) t(77) = 0.0, P = 1.0, BF = 5.3
Anxiety (SD) 8.2 (3.8) 8.0 (4.0) 8.3 (3.7) t(77) = −0.4, P = 0.7, BF = 4.9
T- and χ2-tests were used to assess between-group differences, with Bayes Factor (BF) reported. The questionnaires included are Neurological Fatigue Index (NFI), Cognitive Failures
Questionnaire (CFQ), Short Big Five Inventory (BFI-S), Short Grit Scale (GRIT-S), AMI and the Hospital Anxiety and Depression Scale (HADS). For all the questionnaire-derived
indices, the mean score is shown with 1SD in the bracket.

To compare the episodic memory deficit observed in the recruited from the local community and have been involved
COVID group (see below), we ran the same memory test in the lab’s previous studies. All participants provided writ-
on healthy elderly people (.50 years old). Sixty-four parti- ten informed consent and the study was approved by the
cipants, who reported no neurological conditions, were University of Oxford ethics committee. Participants were
4 | BRAIN COMMUNICATIONS 2022: Page 4 of 19 S. Zhao et al.

initially contacted by emailed, followed up by calls and com- target, presented randomly with a probability of 25%); no
pleted the task on their own computer devices. Fifty-three response was required for other digits. A semi-transparent
participants completed the task and one participant was jittered checkerboard pattern masked the digits, with the le-
excluded due to COVID-19 history. The results of 52 vel of difficulty manipulated by adjusting the opacity of the
participants [30 females, mean age 67.4 years (SD 7.2)] mask. Pilot testing allowed us to obtain an accuracy level of
are reported below. 80% in the first minute of the experiment. The practice
phase consisted of 90 trials, equivalent to 90 s. The difficulty
was gradually increased and feedback was provided after
Questionnaires each trial. The first 12 practice trials were highly visible sti-
The questionnaires included were: muli and participants were required to get 100% accuracy to
proceed. Subsequent non-practice trials and blocks con-
• Fatigue and sleep: Neurological Fatigue Index (NFI), which has tained no feedback. In total, each participant completed
been used to assess fatigue in persons with multiple sclerosis.18 540 trials, divided into nine blocks, each containing 60 trials
Our motivation for including this index was driven by the fact and lasting 1 min.
that it probes the interactions between sleep and fatigue. The experiment was implemented using PsychoPy
Specifically, it asks questions related to abnormal nocturnal v2021.1.2.25 To minimize the variance in latency caused
sleep, sleepiness and the need for diurnal sleep/rest. by different browsers, all participants were required to use
• Distractibility and forgetfulness: Cognitive Failures the Chrome browser on a desktop computer, although the
Questionnaire (CFQ),19 a 25-item questionnaire about minor operating system was not restricted. To minimize potential
mistakes in daily life over the last 2 weeks. For example: ‘Do instability, the use of custom-written code was deliberately
you find you forget appointments?’ ‘Do you fail to notice sign- avoided and all functions used were as provided by
posts on the road?’ ‘Do you find you forget which way to turn PsychoPy. Only the questionnaires were implemented using
on a road you know well but rarely use?’. in-house code.
• Personality: A Short 15-item Big Five Inventory (BFI-S)20 and the
Short Grit Scale (GRIT-S).21 One of BFI-S components—con-
scientiousness—provides information about the conscientious- Motivation and fatigue ratings with time on the task
ness personality trait, describing an individual’s perseverance After each minute during testing, participants were asked to
of effort combined with passions for a particular goal. To ensure report their level of fatigue (‘How tired do you feel now?’)
that this is well-captured, we also included GRIT-S: this esti- and motivation (‘How motivated do you feel?’) using a vi-
mates the same trait but with different questions. We indeed sual analogue scale. Responses were registered by clicking
found that these two scales were strongly positively correlated on the appropriate position on each scale. After completing
(Spearman’s ρ = 0.65, P , 0.0001). all ratings, a ‘confirm’ button appeared at the bottom of the
• Motivation: Apathy Motivation Index (AMI), an 18-item ques- screen, allowing participants to validate their ratings and
tionnaire, sub-divided into three subscales of apathy: emotional, start the next block. To control the time between blocks
behavioural and social apathy.22 and to reduce variance between participants, a 15 s count-
• Mood: Hospital Anxiety and Depression Scale (HADS), a down timer was displayed at the top of the screen, and the
14-item questionnaire, sub-divided into depression and next block would begin automatically once the timer lapsed.
anxiety.23 There were seven participants who had missing ratings. The
exclusion of these participants did not affect the rating results
All 155 participants completed the NFI questionnaire, of or behavioural data and there were no group differences in
which 103 also completed the CFQ. 93 of these 103 partici- age, gender or any questionnaire measures (P . 0.05).
pants completed the full set of questionnaires which were
NFI, CFQ, BFI-S, GRIT-S, AMI and HADS.
The session duration
The session of the questionnaires and vigilance test took
Vigilance test controls 22.4 min (SD 6.7) on average and COVID survivors
All participants were first tested on a version of an 23.8 min (SD 6.4). There was no statistical difference in the
established, sustained visual attention task24 adapted into session length between two groups [t(135) = 1.3, P = 0.2].
a modern online version hosted on the Pavlovia platform Although there was a relatively large variance in session dur-
(pavlovia.org). An online demo is available at https://run. ation amongst the participants, the duration of the main vig-
pavlovia.org/sijiazhao/vigilance_english_demo (open it ilance test was fixed at 11 min 15 s because each block was
with the Chrome Internet browser on a desktop computer). exactly 1 min plus a fixed break of 15 s (a countdown was
This task is designed to assess the performance decrement shown on the screen). The participants could take breaks
during sustained visual attention. A single digit (0–9) was during the questionnaire and the practice, the length of
presented at the centre of the screen for 50 ms every second which was not recorded. Importantly, the vigilance decre-
(Fig. 2 Vigilance). Participants were instructed to press the ment (see below) did not correlate with the session duration
spacebar on their keyboard as soon as they saw ‘0’ (the (Spearman’s ρ = –0.06, P = 0.5).
Post-COVID-19 attention and memory decline BRAIN COMMUNICATIONS 2022: Page 5 of 19 | 5

Figure 2 Twelve cognitive tasks measured distinct aspects of human cognition, memory, attention, motor control, planning
and verbal reasoning abilities. The vigilance task was tested through the online experiment hosting server Pavlovia and conducted first. The
rest of the 11 tasks were provided by Cognitron and ran in the following order: Motor control, Object Memory (immediate), Word memory
(immediate), Simple reaction task, Choice reaction time, 2D mental rotations, 3D mental rotations, Spatial span, Target detection, Tower of
London, Verbal analogies, Object Memory (delayed) and Word memory (delayed). The Object Memory and word memory tasks were both
tested twice: once at the beginning (‘immediate’) and again at the end of the experiment (‘delayed’), with an interval of about 30 min. The delayed
task was solely testing memory of the stimuli displayed in the first instance of the task so the memory probes were not displayed before the
delayed task.
6 | BRAIN COMMUNICATIONS 2022: Page 6 of 19 S. Zhao et al.

Statistical analysis on the number of participants who had COVID-19 within

2 months) were selected with replacement from each group
Measuring of vigilance decrement and a difference between means was computed. This process
The accuracy of each minute was measured as an F1 score, was then repeated 10 000 times. Group difference was
the harmonic mean of precision and sensitivity, given by deemed significant if the proportion of bootstrap interac-
tions that fell above or below zero was more than 95%
2 × TP (i.e. P , 0.05).
F1 =
2 × TP + FP + FN
Correlation
where TP is the true positive rate, FP is the false positive rate
To control for outlier effects, all reported bivariate and par-
and FN is the false-negative rate. F1 score is a particularly
tial correlations were performed using the conservative
suitable accuracy measure for this task because it emphasizes
Spearman’s rank correlation method. Additionally, we con-
true positives (responses to the rare target) and disregards
ducted partial correlations after controlling the effect of the
true negatives (non-responses to non-targets). For reaction
grit scale of personality.
time (RT), only trials with correct responses were consid-
ered. In each block, trials with outliers (two standard devia-
tions away from mean) were excluded, representing 3.7% of Cognitive test battery
trials overall. On a subsequent occasion, a subset of participants also com-
To investigate the effect of group and time on minute-wise pleted a sequence of 11 cognitive tasks to measure distinct
performance and the three ratings, all the values were aspects of human cognition, episodic memory, attention,
z-scored across participants and mixed-effects generalized motor control, planning, mental rotation and verbal reason-
linear models (GLM) were conducted using the MATLAB ing abilities (Fig. 1). These 11 tasks can be viewed at https://
function fitglme with Laplace approximation as the method oxmh1.cognitron.co.uk. They were adapted from estab-
for estimating model parameters. The models included fixed lished behavioural paradigms in a manner that is sensitive
effects of group and time and random effects of participants to population variables of interest whilst being robust
and their age. against the type of device that a person is tested on. The
The vigilance decrement for each individual was com- tasks were conceived, designed and programmed in
puted by subtracting the average accuracy over the first HTML5 with JavaScript by A.H. and W.T., and hosted on
3 min from the average over the last 3 min and then normal- a custom server system ‘Cognitron’ developed by P.J.H. on
ized by dividing the average over the first three blocks. The the Amazon EC2 platform.
reason for using the average over the first three blocks as a This study was conducted independently to the vigilance
performance baseline is because, in COVID-19 participants, task. All 155 participants were invited to attend this experi-
the vigilance decrement happened relatively quickly, at the ment and 81 completed this part of the study between 6 and
group level the accuracy was significantly lower than the first 20 May 2021, with 36 from the COVID group [mean time
minute from the fourth minute (Fig. 3A). since COVID-19 diagnosis: 233.8 days (129.1), age 27.4
(8.6), 14 females] and 44 from the control group [age 26.3
Time-series analysis (8.0), 17 females]. No significant difference in age [t(80) =
To identify time intervals in which groups exhibit accuracy 1.8, P = 0.08, BF = 1.1] or gender [χ2(1,N = 31) = 0.0005,
rate differences, a non-parametric bootstrap-based statisti- P = 1.0] or any questionnaire-derived measures (all P .
cal analysis was used.26 In each iteration, N data sets (N = 0.1) were observed in this subset of participants.
53 here; based on the number of participants in the
COVID group) were selected with replacement from each Object episodic memory
group and a difference between means was computed. Participants were asked to memorize 20 images, all everyday
These time series were subjected to bootstrap re-sampling objects, depicted in black and white. The images were pre-
(10 000 iterations). At each time point, differences were sented sequentially in a random order, each for 2000 ms
deemed significant if the proportion of bootstrap iterations with an inter-image interval of 500 ms. Immediately after
that fell above or below zero was more than 95% (i.e. the presentation, participants’ memory of the 20 images
P , 0.05). This analysis was applied to Fig. 3A–C. was tested in 20 randomly ordered trials. Each trial required
participants to select a previously displayed image from a set
Group comparison of eight images; incorrect images differed in the object itself,
T-test statistics and BF are reported. All P-values reported look or orientation (see the example in Fig. 2’s Object
are two-tailed. When the sample sizes are not balanced, a Memory) in order to measure not only whether the correct
bootstrap-based permutation test was used to confirm the target was identified, but also at higher precision the similar-
reliability of the significance. For example, when comparing ity of selected objects to the original target when errors were
the vigilance decrement in participants who had COVID made. A delayed memory recall test was repeated at the end
within 2 months (N = 13) versus the control participants of the experiment, i.e. without additional sequential encod-
(N = 73), in each iteration, N data sets (N = 13 here; based ing of images; on average the time between the first and
Post-COVID-19 attention and memory decline BRAIN COMMUNICATIONS 2022: Page 7 of 19 | 7

Figure 3 COVID group showed a larger and faster vigilance decline on the task. (A) Accuracy rate was computed for every minute
(i.e. every block) of the vigilance test and plotted against the time. The y-value at t = 0 corresponds to the accuracy rate over the 1 min-long
practice block. The shaded area shows +1 SEM and the black horizontal line at the bottom indicates time intervals where bootstrap statistics
confirmed significant differences between the two groups (P , 0.05, details see ‘Time-series analysis’ in ‘Vigilance test’); the divergence was
significant from the fourth minute to the eighth minute. (B) Group average of self-reported ratings of motivation against time (shaded area shows
+1 SEM). The rating at t = 0 corresponds to the rating after the practice block. No group difference was found in the motivation rating over time.
(C) COVID-19 survivors felt more tired from the beginning (shaded area shows +1 SEM). (D) However, the fatigue rating (averaged over all 10
ratings) did not correlate with the size of the vigilance decrement in neither the COVID group nor the control group. The Spearman’s correlation
coefficient and their two-tailed P-values was shown for each group. (E) Vigilance decrement showed a significant correlation with the time from
COVID-19 diagnosis. Both Spearman’s and Pearson’s correlation coefficients and their two-tailed P-values are shown above the plot. (F)
Participants who had COVID-19 within the last 9 months displayed significantly larger vigilance decrements than the controls. The number of
participants for each bin was labelled above each bar. Each grey dot represents individual data and the error bar indicates 1 SE. Group comparison
performed by permutation test (with 10 000 iterations). *P , 0.05, m (months).

second test was 28.8 min. Population mean of the immediate vegetables or clothes. The words were presented sequentially
accuracy was 60.1% (20.9) compared with the delayed ac- in a pseudo-random order, each presented for 1000 ms and
curacy of 57.6% (21.0), both well above chance level with an inter-image interval of 200 ms. Immediately post-
(12.5%). The memory decrement was computed by sub- presentation, word memory was tested with 24 words, includ-
tracting the accuracy in the immediate memory test from ing 50% non-targets, of which half were foils selected as being
that in the delayed memory test for each individual. semantically similar to the targets [population mean of accur-
acy 90.9% (6.8)]. After an average of 27.8 min later, partici-
Word memory pants were tested again to assess the episodic memory [mean
This task involved memorizing a sequence of English words accuracy 86.3% (9.1)]. The memory decrement was com-
(e.g. Peas, Monkey, Dress, Aubergine, Mouse, etc.). Words puted by subtracting the accuracy in the immediate memory
could be randomly drawn from three categories: animals, test from that in the delayed memory test for each individual.
8 | BRAIN COMMUNICATIONS 2022: Page 8 of 19 S. Zhao et al.

Spatial span number of target shapes clicked. Population mean was

This task measured spatial short-term memory capacity. It 60.2 (SD 10.4).
was a variation of the Corsi Block Tapping Test27,28 partici-
pants were presented with a 4 × 4 grid, where each indivi- Tower of London
dual cell can light up sequentially. Participants were This task measured spatial planning and was a variant of the
required to memorize the sequence and replicate it by click- original Tower of London task.29 The participant was
ing the appropriate squares in the order they lit up. The dif- shown two sets of three prongs with coloured beads on
ficulty was incremented using a ratchet system, every time a them. The first set was the initial state and the second set
sequence was recalled correctly, the length of the subsequent was the target state. The task was to work out the lowest
sequence was incremented by one. The test was terminated number of moves it would take to transition from the initial
when three consecutive mistakes were made on a particular state to the target state, then input this number using an
sequence length. The outcome measure was the maximum on-screen number pad. The test consisted of 10 trials of vari-
sequence length correctly recalled. Minimum level = 2, max- able difficulty, scaled using the number of beads and the con-
imum level = 16, ISI = 0 ms, encoding time = 1500 ms. The volutedness, defined as the number of moves that must be
mean sequence length successfully recalled was 6.6 (SD 1.3). made that do not place a bead in its final target position.
The outcome measure was the total number of correct trials.
Simple reaction task Unlike in the original task,29 the pegs were of equal height
This task measured the basic cognitive process of perception and the task was done mentally—the beads could not be
and response execution. The participants were instructed to moved. The total number of correct trials was recorded
click on a red circle as soon as it appeared at the centre of the [population mean was 5.8 (SD 2.7)].
screen. They were presented with 60 stimuli, the ISI was jit-
tered using a uniform random distribution between 0.5 and Verbal analogies
2 s. The dependent measure was the speed of response This task examined semantic reasoning. Participants were
[population mean was 0.3 s (SD 0.1)]. presented with two written relationships that they must
decide if they had the same type of association or not (e.g.
Motor control ‘Lion is to feline as cabbage is to vegetable’). Participants in-
This task was identical to the reaction time task above, dicated their decision by selecting the True or False buttons
except that the location of the red circle was different each presented below the written analogies. Analogies were
time. Participants had to move their cursor to the target varied across semantic distance to modulate difficulty and
and respond as fast as possible. The participants were pre- association types switch throughout the sequence of trials.
sented with 30 stimuli, the next stimulus was presented To obtain maximum points, participants must solve as
with a 0 ms delay after the response to the previous stimulus. many problems as possible within 3 min. For every correct
Here, the population mean was 0.7 s (SD 0.2). response, the total score increased by one. For every incor-
rect response, the total score decreased by one. The outcome
Choice reaction time measure was the total score [population mean total point
A black arrow either pointing left or right appeared score was 19.9 (SD 12.9)].
on-screen, indicating the side of the screen the participant
needed to click as fast as possible. The participants were pre- Two-dimensional mental rotation
sented with 60 stimuli with a 50% chance of each stimulus The 2D mental rotation test measured the ability to spatially
pointing left. The ISI was jittered using a uniform random manipulate objects in the mind.30 In this version, we used
distribution between 0.5 and 2 s. The mean RT over button 6 × 6 grids with various arrangements of coloured cells.
presses for both sides was computed [population mean was Provided with a target grid and a set of four further grids,
0.5 s (SD 0.1)]. participants had to identify which of the four was the target
but merely rotated by 90°, 180° or 270° (incorrect grids had
Target detection five incorrectly coloured cells). Participants were given 3 min
This spatio-visual attention task involved identifying and to answer as many as possible, with correct trials being
clicking on a target shape amongst a field of distractor awarded with one point to their score [population mean
shapes. The participant was presented with a target shape was 34.5 (SD 8.5), technical issues beyond our control
on the left of the screen and a probe area on the right side caused the loss of five participants’ records for this task].
of the screen. After 3 s, the probe area began to fill with
shapes, the participant must identify and click the target Three-dimensional mental rotation
shape whilst ignoring the distractor shapes. Shapes were Akin to the 2D version above, participants were tasked with
added every 1 s and a subset of the shapes in the recognizing a rotated version of the target from four options.
probe area is removed every 1 s. The trial ran for a total In this 3D version, the grids were instead 3D scenes of build-
of 120 addition/removal cycles. The target shape was in- ings arranged upon a green surface. Again, the correct an-
cluded in the added shapes pseudo-randomly, at a frequency swer was identical to the target but from a different
of 12 in 20 cycles. The outcome measure was the total viewpoint (rotated by 90°, 180° or 270°), whilst the others
Post-COVID-19 attention and memory decline BRAIN COMMUNICATIONS 2022: Page 9 of 19 | 9

had their buildings in the wrong locations. Each participant attending this study. Compared to the age-matched control
was presented with 12 trials scored in the same manner as group who did not report contracting COVID-19 before,
before [population mean was 4.1 (SD 6.2)]. there were no difference in gender or baseline group differ-
ences in a wide range of measures including fatigue/sleep ab-
The session duration normality (NFI), motivation (AMI), distractibility/
This session took controls 37.4 min (SD 13.6) on average forgetfulness (CFQ), mood (HADS) and personality
and COVID survivors 35.3 min (SD 11.7). There was no sta- (GRIT-S and BFI-S) (Table 1).
tistical difference in the session length between two groups Compared with the controls, COVID-19 survivors
[t(94) = –0.8, P = 0.4]. The memory decrement (see below) showed a significantly larger decline in performance in
did not correlate with the session duration (Spearman’s a 9 min-long sustained visual attention task (Fig. 1
ρ = 0.1, P = 0.4). Vigilance). There was no significant group difference in ac-
curacy over the baseline [i.e. the first 3 min, t(135) = –0.9,
Statistical analysis P = 0.4, BF = 3.8]. However, COVID-19 participants
showed a greater and more rapid decline in accuracy over
For each cognitive function, a principal component analysis time [significant interaction between group and time using
(PCA) was performed to derive a global index (PCA scores) a mixed-effects GLM [F(1,1346) = 18.8, P , 0.0001] along
quantifying its level across all relevant tasks used to assess it: with main effects of time [F(1,1346) = 15.1, P = 0.0001]
and group [F(1,1346) = 8.1, P = 0.005)]. Comparing the
• Short-term memory score: PCA of full recall correct rate of minute-by-minute accuracy rate (Fig. 3A), the group differ-
Object Memory, full recall correct rate of Word Memory and ence began to emerge by the start of the fourth minute and
Spatial Span memory capacity. ended after the eighth minute.
• Executive and attentional function score: PCA of reaction times On average, control participants’ accuracy dropped from
in Simple Reaction Task, Motor Control and Choice Reaction 78.5% (SD 20.2, average accuracy over the first 3 min, see
Time and the overall score in Target Detection. Table 2 for more details) to 75.4% (SD 20.9, average
• Mental rotation score: PCA of scores in 2D and 3D mental rota- accuracy over the last 3 min), whilst COVID survivors
tion. This quantifies each individual’s mental ability to spatially started with a similar baseline at 75.5% (19.2), reducing
manipulate objects. to 67.8% (23.0) by the end of the ninth minute.
Comparing the absolute change in performance over time,
Group comparisons and correlations were performed as COVID-19 survivors showed a significantly larger decline
in Vigilance. [t(135) = –2.7, P = 0.008, BF = 5.0]. Importantly, we at-
As the cognitive battery test was used as a broad-brush way tained a similar result after normalizing the change by indi-
of assessing a wide range of cognitive functions, Bonferroni vidual’s baseline performance [COVID −12.3% (17.4)
correction was applied to the significant P-value(s). versus Control −0.9% (29.1), t(135) = −2.6, P = 0.01,
Unadjusted estimates would be provided if stated, along BF = 3.8] suggesting that this larger vigilance decrement
with adjusted estimates reported in the Results section. amongst the COVID group was regardless of the indivi-
dual’s baseline performance.
Ethics statement Across both groups, lower motivation (Fig. 3B) and great-
er fatigue (Fig. 3C) were reported as the experiment
All participants gave electronic informed consent prior to
progressed (mixed-effects GLM, main effect of time P ,
the experiment. The study was approved by the local ethics
0.001). Crucially, the COVID group showed a significantly
committee and carried out in accordance with the relevant
faster fatigue accumulation [interaction of group × time,
guidelines and regulations.
F(1,1347) = 8.8, P = 0.003] and significantly larger
fatigue rating over time [main effect of group, F(1,1347) =
Data availability 5.7, P = 0.02]. Although the COVID group had a statisti-
The data supporting the findings of this study are available cally similar fatigue rating before the experiment started
from the corresponding author upon reasonable request. [t(131) = 1.4, P = 0.2, BF = 2.3, COVID: 53.6%, SD 29.5,
A demo of the vigilance task is available at https://run. Control: 46.6%, SD 28.7], they started to report a signifi-
pavlovia.org/sijiazhao/vigilance_english_demo (Use the cantly higher fatigue rating after completing the first minute
Chrome internet browser on a desktop computer). The ex- of the test (Fig. 3C), but these were not significant for moti-
perimental code is available at https://gitlab.pavlovia.org/ vation ratings (Fig. 3B). Noticeably, the fatigue rating—
sijiazhao/vigilance_english_demo. neither the average over all 10 ratings, the baseline rating,
nor the change in the rating in the first minute—correlated
with the vigilance decrement (Fig. 3D). This suggests that
on an individual level, the vigilance decrement does not
Results merely reflect the subjective feeling of being tired.
COVID-19 survivors in this study were young, with many There was a significant positive correlation between indi-
testing positive for COVID-19 several months before vidual vigilance decrement (normalized difference in average
10 | BRAIN COMMUNICATIONS 2022: Page 10 of 19 S. Zhao et al.

Table 2 Results of vigilance test

Measures All COVID Control Statistic

(n = 136) (n = 53) (n = 83)
Accuracy Baseline accuracy (averaged over the first 3 min), % (SD) 77.3 (19.8) 75.5 (19.2) 78.5 (20.2) t(135) = −0.9, P = 0.4, BF =
3.8
Final accuracy (averaged over the first 3 min), % (SD) 72.4 (22.0) 67.8 (23.0) 75.4 (20.9) t(135) = −2.0, P = 0.05,
BF = 1.1
Accuracy over 9 min, % (SD) 74.7 (20.0) 71.4 (19.8) 76.8 (20.0) t(135) = −1.5, P = 0.1, BF =
1.9
Absolute change in accuracy (final minus baseline), % (SD) −4.9 (10.1) −7.7 (10.3) −3.0 (9.5) t(135) = − 2.7, P = 0.008,
BF = 5.0
Vigilance decrement, i.e. change in accuracy normalized by −5.4 (25.7) −12.3 (17.4) −0.9 (29.1) t(135) = − 2.6, P = 0.01,
individual baseline, mean % (SD) BF = 3.8
RT RT over 9 min, s (SD) 0.5 (0.1) 0.5 (0.1) 0.5 (0.1) t(134) = −0.2, P = 0.8, BF =
5.2
Absolute change in RT (final minus baseline), s (SD) 0.02 (0.1) 0.01 (0.1) 0.02 (0.1) t(130) = −0.8, P = 0.4, BF =
4.0
Change in RT normalized by individual baseline, % (SD) 4.0 (10.3) 3.0 (12.2) 4.6 (8.9) t(130) = −0.9, P = 0.4, BF =
3.8
Ratings Baseline fatigue rating (average over the first 3 min), % (SD) 57.7 (28.6) 65.1 (27.2) 52.8 (28.6) t(135) = 2.5, P = 0.01, BF
= 3.1
Fatigue rating, average over 9 min% (SD) 63.3 (26.2) 68.2 (25.7) 60.1 (26.1) t(135) = 1.8, P = 0.08, BF
= 1.3
Change in fatigue rating (last 3 min minus first 3 min), % (SD) 11.6 (17.5) 7.9 (17.4) 14.1 (17.2) t(135) = −2.1, P = 0.04, BF
= 1.3
Baseline motivation rating, % (SD) 54.6 (24.6) 52.9 (26.4) 55.7 (23.4) t(135) = −0.7, P = 0.5, BF =
4.4
Motivation rating, average over all ratings% (SD) 46.3 (24.5) 45.2 (26.4) 47.0 (23.3) t(135) = −0.4, P = 0.7, BF =
4.9
Change in motivation rating (last 3 min minus first 3 min), % (SD) −16.3 (17.8) −15.7 (17.0) −16.7 (18.4) t(135) = 0.3, P = 0.8, BF
= 5.1
T-tests used to assess between-group differences, with Bayes Factor (BF) reported. The significant t-tests are highlighted in bold.

accuracy between first and last 3 min) and time from about the preciseness of the memory recalled. One novel fea-
COVID diagnosis (Fig. 3E, ρ = 0.3, P = 0.04, two-tailed; ture of this test is that Object Memory was tested twice, once
partial correlation after controlling age and personality of immediately after presentation and again around 30 min la-
grit scale: ρ = 0.4, P = 0.03, two-tailed). Furthermore, parti- ter, which provides a measure of memory decrement over
cipants who had COVID-19 in the last 9 months (Fig. 3F) time.
showed a significantly larger vigilance decrement than the On average, the controls achieved 59.7% accuracy in the
control group [n = 37, t(118) = −2.7, P = 0.009, BF = 4.7, immediate memory test. COVID-19 survivors showed a si-
bootstrap-based P = 0.001], with this difference no longer milar short-term memory performance to the control group
apparent in participants who had COVID-19 more than 9 [t(78) = −0.02, P = 1.0, BF = 4.3; Fig. 4A and see Table 3
months previously [n = 16, t(97) = −0.8, P = 0.4, BF = 2.7, for more details]. However, a difference emerged in the later
bootstrap-based P = 0.2]. (delayed) test; whilst the controls displayed no memory dec-
In order to obtain a more comprehensive examination of rement [no difference from zero, one sample t-test: t(43) =
our COVID-19 survivors’ cognitive profiles, we subse- 0.9, P = 0.4, BF = 5.8], COVID-19 survivors showed a
quently invited all participants to complete a cognitive significant memory decrement [t(35) = −4.1, P = 0.0003,
battery comprising 11 cognitive tests (all bar Vigilance in BF = 107.5], which was larger than in controls by 9.2%
Fig. 2) 2 months after the Vigilance test. Thirty-six [t(78) = 3.3, P = 0.001, BF = 23.2; Fig. 4B]. This group dif-
COVID-19 survivors had been diagnosed on average ference would survive through Bonferroni correction (29
233.8 days previously (129.1) and 44 age-matched controls statistical tests were done for all 11 cognitive battery tests,
attended the battery (Table 3 for demographics). see Tables 3 and 4, adjusted P = 0.029). Like the vigilance
Memory was a key cognitive function measured in this decrement, the memory decrement is computed as a
battery. In one of the memory tests—Object Memory— normalized change in performance (i.e. the difference be-
participants were shown 20 images of everyday objects to tween delayed and immediate memory test, divided by the
memorize. Specifically, participants needed to remember immediate memory test). Importantly, this difference in epi-
not only the object (e.g. a spoon), but also its look (e.g. a sodic memory decrement was not due to the variance in
spoon with a long handle) as well as its orientation (e.g. memory maintenance duration, because there was no differ-
handle pointing towards top-right). This provides us details ence in the duration between the immediate and delayed
Post-COVID-19 attention and memory decline BRAIN COMMUNICATIONS 2022: Page 11 of 19 | 11

Table 3 Demographics and the results of the Object Memory task in the battery test

Measure All COVID Control Statistic

(n = 80) (n = 36) (n = 44)
Age, years, mean (SD) 26.8 (8.2) 27.4 (8.6) 26.3 (8.0) t(78) = 0.6, P = 0.6,
BF = 3.7
Gender, female, n (%) 31 (38.8) 14 (38.9) 17 (38.6) χ2(1,N = 31) = 0.0,
P = 1.0
COVID-19 Time from COVID-19 diagnosis, mean days (SD) 233.8 (129.1)
COVID-19 test type: PCR/lateral flow/unknown/not tested 26/2/3/5
Stayed at hospital overnight for COVID-19 treatment, yes (%) 0 (0.0)
Stayed at ICU for COVID-19 treatment, yes (%) 0 (0.0)
Object Immediate memory test Accuracy, mean % (SD) 60.1 (20.9) 60.0 (15.9) 60.1 (24.4) t(78) = −0.02, P = 1.0,
Memory BF = 4.3
RT in s, mean (SD) 3.3 (1.8) 3.2 (1.6) 3.4 (2.0) t(78) = −0.4, P = 0.7,
BF = 4.0
Delayed memory test Accuracy, mean % (SD) 57.6 (21.0) 52.5 (20.6) 61.8 (20.7) t(78) = − 2.0, P =
0.05, BF = 1.3
RT in s, mean (SD) 2.3 (0.8) 2.2 (0.6) 2.4 (1.0) t(78) = −1.3, P = 0.2,
BF = 2.1
Memory decrement (normalized Percentage change in 3.0 (46.8) −13.6 (21.4) 16.6 (56.9) t(78) = − 3.0, P =
by immediate accuracy) accuracy, mean % (SD) 0.003, BF = 10.7
Time between immediate and Duration in minutes, mean 28.8 (9.4) 27.9 (8.7) 29.6 (9.9) t(78) = −0.8, P = 0.4,
delayed tests (SD) BF = 3.2
The time from COVID-19 diagnosis was computed as the days between the date they attended this battery test and the self-reported date of their positive COVID-19 test. T- and χ2
tests used to assess between-group differences, with Bayes Factor (BF) reported. The significant t-tests are highlighted in bold.

memory tests across the two groups [COVID: 27.9 min (8.7), task was much simpler than Object Memory which also
control: 29.6 min (9.9), t(78) = –0.8, P = 0.4, BF = 3.2]. had precision manipulation.
Furthermore, the larger episodic memory decrement As part of a separate study, we ran the same memory tests
amongst COVID-19 survivors was driven by errors in which on 52 healthy elderly participants [Elderly Control group:
the wrong orientation was chosen for a correct item. In the 53–82 years old, mean 67.4 (7.2), 30 females, no self-report
immediate memory test, COVID-19 survivors had a 30.6% COVID-19]. Amongst controls (N = 96, including all
(13.3) false alarm rate where they chose the right object but from Control and Elderly Control groups), there was a
wrong orientation, misbinding object identity with object weak correlation between memory decrement and age
orientation. This was not significantly different from controls (Pearson r = −0.2, P = 0.03; Spearman’s ρ = −0.2, P =
[28.6 (18.3), t(80) = 0.5, P = 0.6, BF = 3.8]. However, this 0.07). Compared with young controls, elderly participants
orientation-specific false alarm rate increased to 35.3% had a significantly larger memory decrement [mean (SD):
(15.7) 30 min later in the COVID group, which was signifi- –7.1 (21.8), t(94) = 2.8, P = 0.007, BF = 6.0, Fig. 4B], but
cantly higher than the controls [27.3% (14.7), t(80) = 2.4, importantly elderly participants’ episodic memory decre-
P = 0.02, BF = 2.6]. This difference suggests that the deficit ment was not statistically different from COVID-19 survi-
in episodic memory in the COVID group might be associated vors [t(86) = –1.4, P = 0.2, BF = 1.9, Fig. 4B], indicating
with a deficit in binding information in memory. that COVID-19 survivors performed as if they were older.
In another memory test (Fig. 2 Word Memory), partici- However, the elderly controls spent a longer time com-
pants were instructed to memorize 24 simple English pleting other tasks between the immediate and delayed
words. Although both COVID and control groups showed memory tests [mean 39.2 (6.1) minutes] than the COVID
significant memory decrement [COVID: t(35) = −4.5, P = group [t(86) = −7.2, P , 109, BF . 107]. The requirement
0.00008, BF = 305.0; control: t(43) = −4.4, P = 0.00007, to maintain memory for a longer period was associated
BF = 314.5], they did not differ significantly from each other with greater memory decrements (partial correlation of
[t(78) = −0.04, P = 1.0, BF = 4.3]. Because our participants time between memory tests and memory decrement amongst
reported having different first languages, we also ran a 2 young and elderly controls after controlling the effect of age:
(group: COVID versus control) × 2 (first language: English Pearson r = 0.2, P = 0.03; Spearman’s ρ = 0.05, P = 0.6).
or non-English) ANOVA on the memory decrement. There Therefore, we regressed out the effect of the memory main-
were no main effects of first language [F(1,76) = 1.1, P = tenance time from each individual’s memory decrement, but
0.3] or group [F(1,76) = 0.1, P = 0.7] and no language– the pattern remained unchanged.
group interaction [F(1,76) = 0.1, P = 0.7], indicating that Amongst survivors who contracted COVID-19 within the
this null effect was not due to differences in first language last year, the size of memory decrement was weakly but sig-
amongst participants. This null effect in word memory nificantly correlated with the time from diagnosis (Fig. 4C,
decrement might be due to the fact that the word memory Pearson r = 0.6, P = 0.001; Spearman ρ = 0.5, P = 0.008),
12 | BRAIN COMMUNICATIONS 2022: Page 12 of 19 S. Zhao et al.

Figure 4 COVID group showed a mild episodic memory deficit compared with age-matched controls. (A) The distribution of the
short-term memory, measured as the correct percent in the memory test immediately after viewing the sequence of objects, is plotted as a violin
for COVID (n = 36), Control (n = 44) and Elderly Control (n = 52, all above 50 years old, data collected separately). Group comparison
performed by t-test. There were no statistical differences between groups in the short-term memory [COVID versus Control: t(78) = −0.02, P =
1.0, BF = 4.3; Control versus Elderly Control: t(94) = −0.4, P = 0.7, BF = 4.4; COVID versus Elderly Control: t(86) = −0.5, P = 0.6, BF = 4.0].
(B) Approximately 30 min later, their memory was tested again. COVID and Elderly controls showed significantly larger memory decrements
than the younger controls [COVID versus Control: t(78) = −3.0, P = 0.004, BF = 10.7; Control versus Elderly Control: t(94) = 2.8, P = 0.007, BF
= 6.0; COVID versus Elderly Control: t(86) = −1.4, P = 0.2, BF = 1.9]. (C) In COVID-19 survivors who contracted COVID-19 within 1 year, the
size of memory decrement was positively correlated with the time from COVID-19 diagnosis. Both Spearman’s and Pearson’s correlation
coefficients and their two-tailed P-values are shown above the plot. (D) Participants who had COVID-19 within the last 6 months showed
significantly larger memory decrement than the age-matched controls. The number of participants for each bin was labelled above each bar. Each
grey dot represents individual data and the error bar indicates 1 SE. Group comparison performed by permutation test (with 10 000 iterations).
*P , 0.05, **P , 0.01, m (months).

suggesting that people who had COVID-19 more recently Do these cognitive differences relate to the symptom se-
tended to forget more over the 30-min interval. This signifi- verity experienced during COVID-19 illness or the post-
cant memory decrement could be observed up to 6 months illness long-COVID? In a follow-up survey, we asked the
[N = 15, t-test: t(57) = −2.4, P = 0.02; Fig. 4D]. These ana- participants for their experience during and after their
lyses provide preliminary evidence that COVID-associated COVID-19 illness (Fig. 5A and C). The questions were mod-
reductions in sustained attention and episodic memory ified from Office for National Statistics—Coronavirus
may persist for months, but may normalize subsequently, al- (COVID-19) Infection Survey; for example, the question
though the findings have to be taken with caution given the about the COVID-19 symptom severity was ‘Do any of
sample size. the COVID-19 symptoms reduce your ability to carry out
Post-COVID-19 attention and memory decline BRAIN COMMUNICATIONS 2022: Page 13 of 19 | 13

Table 4 Results of the 10 cognitive tasks in the battery test

Measure All COVID Control Statistic

(n = 80) (n = 36) (n = 44)
Word memory Immediate memory test Accuracy, mean % (SD) 90.9 (6.8) 91.1 (7.1) 90.7 (6.7) t(78) = 0.2, P = 0.8,
BF = 4.2
RT in s, mean (SD) 1.0 (0.3) 1.0 (0.3) 1.0 (0.2) t(78) = 0.2, P = 0.8,
BF = 4.2
Delayed memory test Accuracy, mean % (SD) 86.3 (9.1) 86.3 (8.6) 86.2 (9.5) t(78) = 0.1, P = 0.9,
BF = 4.3
RT in s, mean (SD) 0.9 (0.2) 0.8 (0.2) 0.9 (0.2) t(78) = −1.1,
P = 0.3, BF = 2.6
Memory decrement (normalized Percentage change in −5.1 (46.8) −5.1 (21.4) −5.1 (56.9) t(78) = −0.04,
by immediate accuracy) accuracy, mean % (SD) P = 1.0, BF = 4.3
Time between immediate and Duration in minutes, mean 27.8 (8.4) 26.7 (6.8) 28.8 (9.5) t(78) = −1.1,
delayed tests (SD) P = 0.3, BF = 2.6
Spatial short-term Spatial span n, mean (SD) 6.6 (1.3) 6.4 (1.4) 6.8 (1.3) t(78) = −1.2,
memory capacity P = 0.2, BF = 2.3
Motor control Simple reaction time RT in s, mean (SD) 0.3 (0.1) 0.3 (0.1) 0.3 (0.1) t(78) = −1.2,
P = 0.2, BF = 2.2
Motor control RT in s, mean (SD) 0.7 (0.2) 0.7 (0.2) 0.7 (0.2) t(78) = 0.5, P = 0.6,
BF = 3.9
Choice reaction time RT in s, mean (SD) 0.5 (0.1) 0.5 (0.1) 0.5 (0.1) t(78) = −0.2,
P = 0.9, BF = 4.2
Spatial–visual Target detection Score, mean (SD) 60.2 (10.4) 60.4 (11.0) 60.1 (10.0) t(78) = 0.1, P = 0.9,
attention BF = 4.3
Spatial planning Tower of London Score, mean (SD) 5.8 (2.7) 5.8 (2.7) 5.9 (2.7) t(78) = −0.2,
P = 0.8, BF = 4.2
Semantic reasoning Verbal analogies Score, mean (SD) 19.9 (12.9) 18.3 (11.2) 21.2 (14.2) t(78) = −1.0,
P = 0.3, BF = 2.8
Mental rotation 2D mental rotation Score, mean (SD) 34.5 (8.5) 34.7 (7.4) 34.4 (9.3) t(73) = 0.2, P = 0.9,
BF = 4.2
3D mental rotation Score, mean (SD) 4.1 (6.2) 4.3 (6.1) 4.0 (6.3) t(78) = 0.3, P = 0.8,
BF = 4.2
Results from the same 80 participants reported in Table 3 are shown. T- and χ2-tests used to assess between-group differences, with Bayes Factor (BF) reported. Note that due to the
technical issue, we lost five participants’ data for the 2D mental rotation task, thus resulting in a smaller degree of freedom in the t-test for that task.

day-to-day activities?’ with four options: Yes, a lot/Yes, a lit- confirmation of these relations is out of scope of the present
tle/Not at all/No symptom, which in turn corresponds to study and should be addressed in patient studies amongst
Severe/Moderate/Mild/Asymptomatic. To assess the relation COVID-19 inpatients and/or long-COVID patients.
between the post-COVID cognitive decrements observed Although the correlation between cognitive decrement
here and the COVID-19/long-COVID severity, a linear and time since COVID-19 diagnosis (Figs 3E and 4D) pro-
mixed-effect model (LMM) with COVID-19 severity level vides strong evidence linking cognitive differences observed
and long-COVID severity level as fixed effects, and partici- in the present with COVID-19 infection, we additionally
pant as a random effect was applied. Amongst the considered additional non-infectious factors. First, we
COVID-19 survivors who had any COVID-19 or ruled out basic factors including age [Fig. 6A, t(134) =
long-COVID symptoms, larger vigilance and memory decre- −0.6, P = 0.5], gender [Fig. 6B, χ2(1,N = 54) = 0.5, P =
ments were associated with more severe COVID-19 symp- 0.5], first language [Fig. 6C, χ2(1,N = 50) = 0.03, P = 0.9],
toms [Fig. 5B, main effect on vigilance decrement: F(1,8) = country of current residence [Fig. 6D, χ2(1,N = 36) = 0.2,
6.8, P = 0.03; main effect on memory decrement: F(1,6) = P = 0.7] and ethnicity (Fig. 6E, all categories P . 0.1), as
15.3, P = 0.008] but not with long-COVID symptoms none showed any difference between the two groups.
[Fig. 5D, main effect on vigilance decrement: F(1,8) = 4.5, Secondly, we pondered whether the more significant decline
P = 0.07; main effect on memory decrement: F(1,6) = 0.09, in attention and memory of the COVID group could be
P = 0.8]. However, two caveats require attention: first, the attributed to a lower socioeconomic status (SES). This is
positive relation between the COVID-19 symptom severity highly possible: a recent study31 suggested strong associa-
and the cognitive decrements must be taken with caution be- tions between low SES and high probability of COVID-19
cause it disappears if taking asymptotic participants into ac- infection, along with higher infection fatality rate.
count. Secondly, the null effect of long-COVID severity on Meanwhile, lower SES has established negative pressures
the cognitive decrements might be specific to the present study on cognition including attention and memory (see review32).
as most of COVID-19 survivors in the present study did not To address this concern, all participants received a follow-up
have any long-COVID symptoms. Nevertheless, further survey covering a wide range of demographic and
14 | BRAIN COMMUNICATIONS 2022: Page 14 of 19 S. Zhao et al.

Figure 5 Cognitive decrements sorted by COVID-19 symptom and long-COVID symptom severity. Fifty-one out of 64 COVID-19
survivors (including the three participants who stayed hospital overnight for COVID-19) reported their COVID-19 symptom severity (A) and
long-COVID symptom severity (C). In both (A) and (C), the number of participants for each severity level is labelled above the corresponding
bar. (B) The vigilance (left) and memory (right) decrements binned by COVID-19 symptom severity. Each grey dot represents individual data and
the error bar indicates 1 SE. An LMM with participant as a random effect showed that COVID-19 severity level had main effect on vigilance
decrement [F(1,8) = 6.8, P = 0.03] and main effect on memory decmrenet [F(1,6) = 15.3, P = 0.008]. Similarly, (D) shows the cognitive
decrements for each long-COVID symptom severity level. An LMM with participant as a random effect showed that long-COVID symptoms had
no effect on vigilance decrement [F(1,8) = 4.5, P = 0.07] or memory decrement [F(1,6) = 0.09, P = 0.8].

socioeconomic measures, encompassing education, income, experience of cognitive testing than the controls. This hy-
occupation, work sector as well as subjective SES measured pothesis can be tested by comparing the number of studies
by the MacArthur Scale of Subjective Social Status.33 One that the participants attended on Prolific. In contrast, the
hundred and seventeen responded (75% of 155 participants, COVID group had greater experience of online experiments
COVID N = 51, control N = 66), with no statistical differ- [t(134) = 4.4, P , 10−4]. The average number of studies at-
ence in education level (Fig. 6F), annual income (Fig. 6G) tended was 248 studies (SD 358) in the COVID group and
or employment status (Fig. 6H). In fact, the COVID group 67 (SD 82) in the control group. Moreover, the number of
reported a slightly higher subjective SES [Fig. 6I, COVID attended studies does not correlate with vigilance decrement
subjective SES = 6.1 (1.2), control subjective SES = 5.6 (Spearman’s ρ = −0.09, P = 0.3) or memory decrement
(1.5), COVID versus control: t(96) = 2.5, P = 0.02]. (Spearman’s ρ = −0.05, P = 0.7). These confirm that cogni-
Moreover, we found no difference in the proportion of es- tive differences observed in the present study were not
sential workers amongst the groups (Fig. 6J, χ2(1,N = 51) caused by the effect of test familiarity. Although it is unclear
= 0.8, P = 0.4) or the method of the commute during the if smoking history could affect the cognitive ability, we also
pandemic (Fig. 7A, all categories P . 0.1). These suggest checked for this. Reassuringly, there was no difference in
that SES cannot fully explain the cognitive difference be- the proportion of past smokers [Fig. 7B, χ2(1,N = 29) =
tween COVID and control groups in the present study. 0.9, P = 0.3] or present smokers [Fig. 7C, χ2(1,N = 14) =
Another potential confound is testing experience; partici- 0.9, P = 0.4] between the two groups. Finally, we asked
pants from the COVID group might, by chance, have less for COVID-19 vaccination history: unsurprisingly, there
Post-COVID-19 attention and memory decline BRAIN COMMUNICATIONS 2022: Page 15 of 19 | 15

Figure 6 Demographics and socioeconomics profile of the participants. T-test was used to assess between-group difference in age (A)
and subjective SES (I). For the measures with binary outcomes, including gender (B), first language (C), country of current residence (D), and
essential workers (J), χ2-test was used to assess between-group differences. Their P-values were unadjusted for multiple comparison. For the
measures with multiple categories—ethnicity (E), education (F), annual income (G) and employment status (H), χ2-test was run for each
category of each measure and P-values were adjusted using the Bonferroni method (i.e. multiplying the number of categories in that measure).
Amongst all measures, only one measure showed significant difference: the COVID group showed a significantly higher subjective SES [H, t(96) =
2.5, P = 0.02]. No difference was found in other measures and annotated as n.s. (not significant).
16 | BRAIN COMMUNICATIONS 2022: Page 16 of 19 S. Zhao et al.

Figure 7 Work, smoking history and vaccination status of the participants. For the measures with multiple categories—transport
means used to commute (A), work sector (E) and work from home status (F), χ2-test was run for each category of each measure and P-values
were adjusted using the Bonferroni method (i.e. multiplying the number of categories in that measure). For the measures with binary outcomes,
including smoking history (B and C) and COVID-19 vaccination history (D), χ2-test was used to assess between-group differences. Amongst all
measures, only one measure showed significant difference: the COVID group showed a lower rate of COVID-19 vaccination [D, χ2(1,N = 15) =
8.6, P = 0.003]. No difference was found in other measures and annotated as n.s. (not significant).

was a significantly higher vaccination rate amongst controls PCA scores between COVID-19 survivors and controls, no
than COVID-19 survivors [Fig. 7D, χ2(1,N = 15) = 8.6, P = significant difference was found in the short-term memory
0.003]. On average, the days from the last dose of vaccina- [t(78) = −1.2, P = 0.2, BF = 2.2], the executive function
tion to the date of attending the test was 63.7 (SD 37.9, [t(78) = 0.3, P = 0.7, BF = 4.1] or the mental rotation ability
minimum = 12, maximum = 139) and was not correlated [t(73) = 0.2, P = 0.8, BF = 4.2], suggesting that most of the
with the vigilance decrement (Spearman’s ρ = 0.1, P = 0.6). key cognitive functions were normal.
Taking all the evidence together, the results suggest that
the cognitive differences in attention and memory observed
here seem to be strongly associated with COVID-19 infec-
tion, rather than an outcome of a single demographic or so-
Discussion
cioeconomic metric. In the present study, we examined a wide range of cognitive
Encouragingly, apart from these two cognitive differ- abilities in COVID-19 survivors and age-matched controls.
ences, COVID-19 survivors did not show any significant dif- The COVID group did not require hospitalization and had
ference from the age-matched controls in a wide range of not sought medical help for long-COVID symptoms after re-
cognitive capabilities, including short-term memory covery. The good news is that COVID-19 survivors per-
(Object Memory, Word Memory, Spatial Span), the re- formed well in most cognitive abilities tested, including
sponse speed (Simple Reaction Task, Motor Control, working memory, executive function, planning and mental
Choice Reaction Time), spatial–visual attention (Target rotation. However, even though their questionnaire-derived
Detection), spatial planning (Tower of London), semantic measures (fatigue, forgetfulness, motivation, sleep abnor-
reasoning (Verbal Analogies) and mental rotations (2D or mality, depression and anxiety levels) were no different
3D) (see Tables 3 and 4 for details). Similarly, comparing from age-matched controls (Table 1), they showed a
Post-COVID-19 attention and memory decline BRAIN COMMUNICATIONS 2022: Page 17 of 19 | 17

significantly larger vigilance decrement along with faster fa- survivors in the present study did neither indicated any
tigue build-up over the course of a 9 min-long attentionally sign of higher fatigue, forgetfulness, apathy, anxiety, depres-
demanding task (Fig. 3A and C). They also had significantly sion or sleep abnormality (Table 1), nor felt any more tired
worse episodic memory decrement over time, comparable to than their age-matched controls over the time course of the
a healthy, elderly person in their 60s (Fig. 4B). Notably, both vigilance test, suggesting a dissociation between self-report
deficits scaled with the time from COVID-19 diagnosis sug- symptoms and objectively measured deficits. Our findings
gesting a strong relation with COVID-19 itself (Figs 3E and highlight that cognitive reductions are not limited to patients
4D). who had prolonged neurological manifestations after recov-
To our knowledge, this is the first report describing defi- ery,7,8 but might exist more widely in a sub-clinical form
cits in sustained attention and episodic memory amongst amongst COVID-19 survivors who would not consider
mildly-affected COVID-19 survivors long after the acute ill- themselves requiring any post-COVID treatment.
ness, in people who were not complaining of long-COVID At the outset of the pandemic, Hampshire et al.17 con-
symptoms (cf. Zhou et al.9 reporting sustained attention im- ducted a large-scale online test involving over 13 000 people
pairments in recently-recovered patients). Our findings are with suspected or biologically confirmed COVID-19 circa 2
consistent with the most prevalent complaints concerning months. That study shared a subset of tasks with our inves-
post-COVID cognitive issues, including poor concentration tigation, covering a wide range of cognitive functions includ-
and/or impaired memory in 18–50% of patients post- ing semantic problem solving, visual–spatial attention, speed
recovery.8,34,35 However, in contrast with previous reports of response and working memory. All functions showed
of long-COVID patients,6–8,12,13,16 here we found no differ- some degree of cognitive deficit amongst people who had
ence in short-term attention (performance over the first few contracted COVID-19, scaling with respiratory symptom se-
minutes of a vigilance test; overall performance in tasks mea- verity. In the present study, however, we did not find any
suring executive function and response speed, Table 4) or group differences in these cognitive domains (Table 4).
working memory (performance in the immediate object or This difference is likely to be attributable to the relative
word memory tests, Tables 3 and 4 and Fig. 4A). mildness of the symptoms experienced by our COVID-19
In the present study, COVID-19 survivors began with ap- survivors, combined with the length of time since infection
parently normal behavioural performance followed by a (some over 9 months ago). This suggests that these function-
gradual decline away from age-matched controls, suggesting al deficits might not be obvious in milder COVID-19 pa-
reduced ability to attentively track and maintain information tients, with recovery expected within months.
over time. The inconsistency with previous reports might be There are some limitations to our study. Although the ma-
due to the fact that patients featured in those studies had se- jority of participants from our COVID group reported a
vere COVID-19 symptoms,3,4,6,12 clinically significant cogni- PCR-confirmed COVID-19 positive result with none report-
tive impairment16 or at least reported persistent cognitive ing the need for post-COVID treatment, our study was lim-
symptoms,7,8,13 whilst our participants were mostly non- ited by our reliance on self-reports of positive/negative
hospitalized and devoid of self-reported abnormality. COVID-19 tests and timing of diagnosis, which might in-
The mechanisms underlying these cognitive deficits as yet crease or decrease our estimate of the prevalence and dur-
remains unclear. Although a direct effect of virus persisting ation of COVID-associated cognitive deficits. Our study is
in the brain cannot be excluded, the evidence from post- also constrained by a relatively small sample size (N =
mortem studies suggests there is very little presence of virus 136) with under-representation of the over 70s, thus any
within the brain in COVID-19 patients.36 Rather, there generalization should be taken carefully.
might be indirect effects of the virus on cognitive function Overall, the findings here show that COVID-19 survivors
mediated via a range of possible mechanisms, including im- showed a significant reduction in their ability to sustain at-
munological and microvascular changes (see review37). One tention on a demanding task up to 9 months after
investigation of COVID-19 survivors demonstrated that the COVID-19 infection, along with mild, but significantly
most severely cognitively affected patients demonstrated a worse, episodic memory for up to 6 months. Just as the acute
degree of cognitive impairment accompanied by hypometa- illness of COVID-19 demonstrates a wide severity spectrum
bolism in the frontoparietal regions.10 These brain regions from asymptomatic to fatal forms,45 our findings show that
are implicated in sustained attention38 as well as in episodic post-COVID cognitive deficits too can also manifest a wide
memory.39–41 Reassuringly, the follow-up study of Hosp severity spectrum. They highlight a pressing need to measure
et al.10 showed slow but evident improvement after 6 cognitive performance objectively in order to better under-
months.11 This is in line with the mildly-affected individuals stand how the brain is affected by COVID-19.
reported here: both vigilance and episodic memory decre-
ments gradually resolved over time (Figs 3E and 4C).
Episodic memory returned to normal levels after 6 months
(Fig. 4D) and those who had COVID-19 over 9 months
ago did not exhibit the vigilance decrement (Fig. 3F).
Acknowledgements
Unlike other survey-based reports focusing on self- The authors are grateful to Stephen Walsh for his involve-
reported long-COVID symptoms,35,42–44 COVID-19 ment in the early stage of the project.
18 | BRAIN COMMUNICATIONS 2022: Page 18 of 19 S. Zhao et al.

12. Beaud V, Crottaz-Herbette S, Dunet V, et al. Pattern of cognitive

Funding deficits in severe COVID-19. J Neurol Neurosurg Psychiatry.
2021;92:567–568.
This work was supported by the Wellcome Trust and 13. Hellmuth J, Barnett TA, Asken BM, et al. Persistent
National Institute for Health Research (NIHR) Oxford COVID-19-associated neurocognitive symptoms in non-
Biomedical Research Centre. S.Z. and M.H. were funded hospitalized patients. J Neurovirol. 2021;27(1):191–195.
by the Wellcome Trust (206330/Z/17/Z). S.M. was funded 14. Raman B, Cassar MP, Tunnicliffe EM, et al. Medium-term effects
by an Medical Research Council (MRC) Clinician Scientist of SARS-CoV-2 infection on multiple vital organs, exercise cap-
acity, cognition, quality of life and mental health, post-hospital dis-
Fellowship (MR/P00878/X). K.S. was funded by the charge. EClinicalMedicine. 2021;31:100683.
Berrow Foundation. A.H. was supported by the UK 15. Ferrucci R, Dini M, Groppo E, et al. Long-lasting cognitive abnor-
Dementia Research Institute Care Research, NIHR malities after COVID-19. Brain Sci. 2021;11(2):235.
Technology Centre and Imperial Biomedical Research 16. Miskowiak K, Johnsen S, Sattler SM, et al. Cognitive impairments
Centre. P.J.H. was supported by the NIHR Biomedical four months after COVID-19 hospital discharge: Pattern, severity
and association with illness variables. Eur Neuropsychopharmacol.
Research Centre, South London and Maudsley NHS 2021;46:39–48.
Foundation Trust and King’s College London. W.T. was 17. Hampshire A, Trender W, Chamberlain SR, et al. Cognitive deficits
funded by the Engineering and Physical Sciences Research in people who have recovered from COVID-19. EClinicalMedicine.
Council (EPSRC) through the Imperial College Centre for 2021;39:101044.
Doctoral Training (CDT) for Neurotechnology. The funders 18. Mills RJ, Young CA, Pallant JF, Tennant A. Development of a pa-
tient reported outcome scale for fatigue in multiple sclerosis: the
had no role in study design, data collection and analysis, de- Neurological Fatigue Index (NFI-MS). Health Qual Life
cision to publish or preparation of the manuscript. Outcomes. 2010;8(1):22.
19. Broadbent DE, Cooper PF, FitzGerald P, Parkes KR. The Cognitive
Failures Questionnaire (CFQ) and its correlates. Br J Clin Psychol.
1982;21(1):1–16.
Competing interests 20. Lang FR, John D, Lüdtke O, Schupp J, Wagner GG. Short assess-
ment of the Big Five: robust across survey methods except telephone
The authors declare no competing financial interests. interviewing. Behav Res Methods. 2011;43(2):548–567.
21. Duckworth AL, Quinn PD. Development and validation of the
Short Grit Scale (Grit–S). J Pers Assess. 2009;91:166–174.
References 22. Ang Y-S, Lockwood P, Apps MAJ, Muhammed K, Husain M.
Distinct subtypes of apathy revealed by the apathy motivation in-
1. Taquet M, Geddes JR, Husain M, Luciano S, Harrison PJ. 6-month dex. PLoS ONE. 2017;12(1):e0169938.
neurological and psychiatric outcomes in 236 379 survivors of 23. Zigmond AS, Snaith RP. The hospital anxiety and depression scale.
COVID-19: A retrospective cohort study using electronic health re- Acta Psychiatr Scand. 1983;67(6):361–370.
cords. Lancet Psychiatry 2021;8(5):416–427. 24. Nuechterlein KH, Parasuraman R, Jiang Q. Visual sustained atten-
2. Mao L, Jin H, Wang M, et al. Neurologic manifestations of hospi- tion: image degradation produces rapid sensitivity decrement over
talized patients with coronavirus disease 2019 in Wuhan, China. time. Science. 1983;220(4594):327–329.
JAMA Neurol. 2020;77(6):683–690. 25. Peirce J, Gray JR, Simpson S, et al. PsychoPy2: Experiments in be-
3. Helms J, Kremer S, Merdji H, et al. Neurologic features in havior made easy. Behav Res Methods. 2019;51(1):195–203.
severe SARS-CoV-2 infection. N Engl J Med. 2020;382(23): 26. Efron B, Tibshirani RJ. An introduction to the bootstrap. CRC
2268–2270. Press; 1994.
4. Varatharaj A, Thomas N, Ellul MA, et al. Neurological and neu- 27. Corsi PM. Human memory and the medial temporal region of the
ropsychiatric complications of COVID-19 in 153 patients: A brain. Unpublished PhD thesis. McGill University; 1972.
UK-wide surveillance study. Lancet Psychiatry. 2020;7(10): 28. Berch DB, Krikorian R, Huha EM. The Corsi block-tapping task:
875–882. Methodological and theoretical considerations. Brain Cogn.
5. Nalbandian A, Sehgal K, Gupta A, et al. Post-acute COVID-19 syn- 1998;38(3):317–338.
drome. Nat Med. 2021;27(4):601–615. 29. Shallice T, Broadbent DE, Weiskrantz L. Specific impairments of
6. Jaywant A, Vanderlind WM, Alexopoulos GS, Fridman CB, Perlis planning. Philos Trans R Soc Lond B Biol Sci. 1982;298(1089):
RH, Gunning FM. Frequency and profile of objective cognitive def- 199–209.
icits in hospitalized patients recovering from COVID-19. 30. Silverman I, Choi J, Mackewn A, Fisher M, Moro J, Olshansky E.
Neuropsychopharmacology. 2021;46:2235–2240. Evolved mechanisms underlying wayfinding. further studies on the
7. Graham EL, Clark JR, Orban ZS, et al. Persistent neurologic symp- hunter-gatherer theory of spatial sex differences. Evol Hum Behav.
toms and cognitive dysfunction in non-hospitalized Covid-19 “long 2000;21(3):201–213.
haulers”. Ann Clin Transl Neurol. 2021;8(5):1073–1085. 31. Mena GE, Martinez PP, Mahmud AS, Marquet PA, Buckee CO,
8. Woo MS, Malsy J, Pöttgen J, et al. Frequent neurocognitive deficits Santillana M. Socioeconomic status determines COVID-19 inci-
after recovery from mild COVID-19. Brain Commun. 2020;2(2): dence and related mortality in Santiago, Chile. Science. 2021;372-
fcaa205. (6545):eabg5298.
9. Zhou H, Lu S, Chen J, et al. The landscape of cognitive function in 32. Hackman DA, Farah MJ, Meaney MJ. Socioeconomic status and
recovered COVID-19 patients. J Psychiatr Res. 2020;129:98–102. the brain: mechanistic insights from human and animal research.
10. Hosp JA, Dressing A, Blazhenets G, et al. Cognitive impairment and Nat Rev Neurosci. 2010;11(9):651–659.
altered cerebral glucose metabolism in the subacute stage of 33. Adler N, Stewart J. The MacArthur scale of subjective social status. San
COVID-19. Brain. 2021;144(4):1263–1276. Francisco: MacArthur Research Network on SES & Health. 2007.
11. Blazhenets G, Schröter N, Bormann T, et al. Slow but evident 34. Rogers JP, Chesney E, Oliver D, et al. Psychiatric and neuropsychia-
recovery from neocortical dysfunction and cognitive impairment tric presentations associated with severe coronavirus infections: a
in a series of chronic COVID-19 patients. J Nucl Med. 2021;62: systematic review and meta-analysis with comparison to the
910–915. COVID-19 pandemic. Lancet Psychiatry. 2020;7(7):611–627.
Post-COVID-19 attention and memory decline BRAIN COMMUNICATIONS 2022: Page 19 of 19 | 19

35. Blomberg B, Mohn KG-I, Brokstad KA, et al. Long COVID in a pro- 41. Dudukovic NM, Kuhl BA. Cognitive control in memory encoding
spective cohort of home-isolated patients. Nat Med. 2021;27: and retrieval. In: The Wiley handbook of cognitive control. John
1607–1613. Wiley & Sons, Ltd; 2017:355–375.
36. Lee M-H, Perl DP, Nair G, et al. Microvascular injury in the brains 42. Carfì A, Bernabei R, Landi F, for the Gemelli Against COVID-19
of patients with Covid-19. N Engl J Med. 2021;384(5):481–483. Post-Acute Care Study Group. Persistent symptoms in patients after
37. Alnefeesi Y, Siegel A, Lui LMW, et al. Impact of SARS-CoV-2 infec- acute COVID-19. JAMA. 2020;324:603–605.
tion on cognitive function: A systematic review. Front Psychiatry. 43. Huang C, Huang L, Wang Y, et al. 6-month consequences of
2021;11:621773. doi:10.3389/fpsyt.2020.621773 COVID-19 in patients discharged from hospital: a cohort study.
38. Langner R, Eickhoff SB. Sustaining attention to simple tasks: a Lancet. 2021;397:220–232.
meta-analytic review of the neural mechanisms of vigilant atten- 44. Cirulli ET, Barrett KMS, Riffle S, et al. Long-term COVID-19
tion. Psychol Bull. 2013;139(4):870–900. symptoms in a large unselected population. medRxiv. Published on-
39. Cabeza R, Ciaramelli E, Olson IR, Moscovitch M. The parietal cor- line December 1, 2020:2020.10.07.20208702. 10.1101/2020.10.
tex and episodic memory: An attentional account. Nat Rev 07.20208702
Neurosci. 2008;9(8):613–625. 45. Wiersinga WJ, Rhodes A, Cheng AC, Peacock SJ, Prescott HC.
40. Davis SW, Wing EA, Cabeza R. Contributions of the ventral parie- Pathophysiology, transmission, diagnosis, and treatment of corona-
tal cortex to declarative memory. Handb Clin Neurol. 2018;151: virus disease 2019 (COVID-19): A review. JAMA. 2020;324(8):
525–553. 782–793.