Participants

The link to the SosciSurvey¹²⁶ (https://www.soscisurvey.de) online survey was distributed via social media, press releases, and the German Health Departments, which shared the link via their department-specific websites, social media, and flyers attached to the standard mail and/or e-mail to infected individuals. All participants gave written informed consent prior to participation. The study was approved by the local ethics committee of SRH University Heidelberg and Heidelberg University of Education [EV2021-09]. All methods were performed in accordance with the relevant guidelines and regulations. Participation was possible between July 15, 2021, and July 15, 2022.

Overall, N = 1405 adult participants (inclusion criterion was age > = 18 years; age range in the remaining sample: 18–90 years) were identified by their individual code, which they had generated at the very beginning of the survey, ensuring anonymity. While demographic data, information on infection status, health conditions, and participants’ responses to questionnaires were collected via SosciSurvey, another part of the study consisted of cognitive tasks that were run on the Pavlovia platform (https://pavlovia.org/) and were connected to the survey via external links. For the cognitive tasks, the sample characteristics differed from the overall sample described, as only subgroups of participants completed the various tasks, possibly due to motivational factors or because participants started the study using mobile devices without a physical keyboard. If participants took part in the survey multiple times in a row, only their first participation was considered for sample description and all analyses. According to their self-reported infection status, n = 495 participants had never had a positive COVID-19 PCR test result and n = 910 participants had received a positive test result in the past, irrespective of the time elapsed between the PCR test and their participation (M = 11.15 months, SD = 6.97, range = 0–29; 0 months indicates a positive PCR test within the last four weeks prior to participation). Infection groups (previously infected, previously not infected) did not differ regarding age, t < 1, and gender, χ² < 1 (see Table 5 for descriptives). In contrast, there was a significant association between infection status and level of education, χ² (1) = 13.56, p < .001. Specifically, individuals who had attained post-secondary education displayed a diminished prevalence of infections compared to their counterparts without completion of post-secondary education.

When participants self-reported a positive result on a PCR test, they were subsequently asked about their vaccination status preceding the positive test result (vaccinated, not vaccinated). This question was not mandatory and added to the survey only in November 2021. Furthermore, participants affirming a positive test were further prompted to delineate their acute COVID-19 symptoms at the time of infection, enduring COVID-19 symptoms, and long COVID symptoms.

Measures

Participants reached the study via a hyperlink to the SosciSurvey platform. After giving written informed consent and completion of the questions regarding demographic data, infection status, vaccination status, and both acute and post-acute COVID-19 symptoms, we assessed depressiveness, stress, anxiety, and well-being (always in this order) using the Patient Health Questionnaire (PHQ-9), the Perceived Stress Questionnaire (PSQ-20), the Generalized Anxiety Disorder Scale (GAD-7), and the WHO well-being (WHO-5) questionnaires, respectively. Next, participants were asked for information on their health condition to exclude participants with known physical or mental disorders that might affect performance in the cognitive tasks (see Analyses section). For the same reason, regular medication intake was assessed through an open-format question in which participants were asked to provide the names of any medications they were taking regularly. This approach allowed for the detailed and unrestricted reporting of medication use, ensuring comprehensive data collection for this variable.

The survey part of the study was followed by the cognitive test battery (see Fig. 1). All cognitive tasks were programmed in PsychoPy¹²⁹ (https://www.psychopy.org/) and run on the Pavlovia platform (https://pavlovia.org/). To match the survey datasets with the cognitive task data, participants generated an individual code at the beginning of the study, which they were prompted to fill in at the beginning of every cognitive task. The overall eight cognitive tasks automatically started in full-screen mode by clicking on the respective Pavlovia link. Cognitive tasks were administered in random order, with the constraints that the Corsi block-tapping backward task was always preceded by the Corsi block-tapping forward task and that the alertness task was always used as a distractor task between the study and the test phase of the MST.

Questionnaires

Depressiveness

To assess depressive symptoms within the previous two weeks, we administered the Patient Health Questionnaire (PHQ-9⁷⁹). Participants were asked to indicate the frequency of their experience with each of nine depressive symptoms (four-point scale: “not at all”, “several days”, “more than half the days”, “nearly every day”, scored as 0 to 3, respectively). This scoring system yielded a potential total score ranging from 0 to 27. PHQ-9 scores allow for a stratification of depression severity, with classifications as follows: minimal: 0–4, mild: 5–9, moderate: 10–14, moderately severe: 15–19, severe: ≥ 20).

Perceived stress

To evaluate perceived stress within the concurrent two-week timeframe, aligning with the temporal scope of the other administered questionnaires, an adapted version of the Perceived Stress Questionnaire (PSQ-20⁸⁰) was employed. This modification involved a temporal adjustment, focusing on the preceding two weeks instead of the original four-week duration. The PSQ-20 comprises four distinct subscales, namely demands, tension, joy (with reverse coding in total score), and worries, containing five items each. Participants were asked to rate how often an item applies to their experience within the last two weeks using a four-point scale (“almost never”, “sometimes”, “often”, “usually”, scored as 0 to 3, respectively). Mean scores of the subscales were divided by 3 and multiplied by 100, resulting in a mean subscale score that ranges from 0 to 100. The total score represents the mean of the transformed subscale scores.

Anxiety

The Generalized Anxiety Disorder Scale (GAD-7⁸¹) was employed to assess the frequency of seven symptoms associated with general anxiety. Participants rated their experiences on a four-point scale ranging from: “not at all”, “several days”, “more than half the days”, to “nearly every day”, scored as 0 to 3, respectively, resulting in a possible total score range between 0 and 21). GAD-7 scores allow for classification based on anxiety levels, with scores exceeding 10 falling within the clinical range (mild: 5–9, moderate: 10–14, severe: ≥ 15).

Well-being

To assess the overall well-being of our sample, we used the WHO well-being index (WHO-5⁸²; see also ⁸³). Participants were asked to report the frequency of experiencing five different states reflecting high levels of well-being within the last two weeks using a six-point scale ranging from “never” to “all the time” (scored as 0 to 5). Raw scores were then transformed into a total score (i.e., multiplied by 4) covering a range from 0 to 100. Scores ≤ 50 are considered indicators of a poor emotional state, warranting further examination¹³⁰.

Cognitive tasks

Long-term memory

The Mnemonic Similarity Task (MST⁶⁷), a modified object recognition memory task, is designed to be highly sensitive to hippocampal function, particularly by imposing robust demands on pattern separation. Widely adopted in behavioral research, the MST has consistently demonstrated sensitivity to age-related memory decline, hippocampal connectivity, and overall hippocampal function, with a specific emphasis on the DG (see ⁶⁷, for an overview). This task has proven extremely useful in identifying hippocampal dysfunction across various conditions, including healthy aging, dementia, schizophrenia, depression, and other clinical disorders^87,131. For instance, in amnesic patients with lesions limited specifically to the hippocampus, impaired lure discrimination performance has been observed, while item recognition performance remains intact¹³². Further supporting the role of the DG in mnemonic pattern separation, a rare case with damage isolated to the DG displayed selective impairments in discriminating similar lures, while recognition performance remained normal¹³³. Similar findings have been reported in patients with CA1-specific lesions, a crucial output structure of the DG¹³⁴. Extending beyond clinical populations, studies involving healthy individuals have revealed that pattern-separation performance as measured by the MST reflects hippocampal activity, distinguishing between accurately discriminated lures and instances where the lure is mistakenly identified as old^135,136. This discrimination ability can be localized to some extent to the DG/CA3 regions within the hippocampus¹³⁷.

The MST comprises an incidental picture encoding phase followed by an unexpected recognition memory assessment (see Fig. 1). All pictures used in the MST were drawn from the openly available stimulus pool provided on the Stark lab github repository (https://github.com/celstark/MST). Participants were randomly assigned to one of six stimulus sets (Set 1–6 of the stimulus pool), each containing 384 color photographs of objects organized into 192 pairs that consist of two similar items each. Before the study phase, the stimulus pairs were randomly shuffled, individually for each participant. During the study phase, participants were presented with 128 items drawn from these unique pairs. At the subsequent memory test, 64 of these objects were reintroduced as old items (targets). The remaining previously presented objects had their respective similar item presented at test (lures). In addition, the 64 objects of the stimulus pool that had not been encoded in the study phase were presented as new items (foils), resulting in 192 recognition test trials overall.

At the beginning of each of the 128 study-phase trials, a fixation cross was presented in the center of the screen for 500 ms, which was then replaced by an object picture for 2000 ms. For as long as the picture was presented, participants could indicate that they think the object typically appears indoor or outdoor by pressing the ‘f’ or ‘j’ key on their computer keyboard (counterbalanced between subjects), with key assignment reminders displayed at the bottom of the screen throughout the encoding phase. Following the study phase, a simple 5-minute stimulus-response alertness task was administered as a filler task (see below). During the subsequent recognition test phase, participants encountered the 192 pictures as described earlier (targets, lures, foils; see Fig. 1) in random order. Each trial began with the presentation of a fixation cross in the center of the screen for 500ms, followed by the presentation of the object for 2000 ms. The participants’ task was to indicate whether the objects were old, similar, or new by pressing the ‘c’, ‘b’, or ‘m’ key on the computer keyboard, respectively. Responding was possible for as long as the picture was presented.

Alertness

The alertness task closely followed the procedure of the TAP (Testbatterie zur Aufmerksamkeitsüberprüfung¹³⁸) Alertness test part assessing intrinsic alertness. In each trial, participants were presented with a black fixation cross against a white background in the center of the screen for 1000 ms (see Fig. 1). Their task was to press the space key as fast as possible as soon as the fixation cross appeared. Inter-stimulus intervals were jittered ranging from 700 to 3500 ms in steps of 100 ms. These intervals were randomly selected from a pool comprising all 29 jitter durations. The task was organized into four blocks of 20 trials each. The blocks were separated by 8-second breaks, resulting in an overall duration of approximately 5 min.

Executive functions

Working Memory

Corsi block tapping

We assessed visuo-spatial working memory utilizing an online adaptation of the Corsi block-tapping task (inspired by the digital version outlined in Arce & McMullen¹³⁹), primarily derived from the online Corsi block-tapping demo accessible via https://gitlab.pavlovia.org/demos/corsi). Each participant engaged in both a forward and a backward version of the Corsi block-tapping task, consistently following this order. In both versions, participants were presented with nine randomly positioned, non-overlapping white squares on grey background. Their task was to memorize a random sequence of squares, visually indicated by sequentially highlighting the blocks. Specifically, a square changed its color to red for one second before reverting to white, and the subsequent square in the sequence turned red (see Fig. 1). After completion of a sequence, participants were instructed to replicate the sequence by clicking on the blocks either in the original order (forward version) or in reverse order, starting with the last highlighted block (backward version), without any time constraints. A tapped square transitioned to light grey and remained so until the participant’s number of clicks matched the length of the sequence. After completing the required number of taps of the respective sequence (regardless of success), a 500 ms inter-trial interval was inserted, in which a fixation cross was presented, before the next trial started. In both the forward and backward version, the length of the block sequence progressively increased during the task. Starting with a sequence length of three blocks, participants completed five trials at each length before the sequence extended by one block. This process continued until reaching a maximum sequence length of nine blocks. If a participant failed to reproduce a sequence correctly in at least three out of the five trials, subsequent sequence lengths were presented only once. This approach ensured comparable task familiarization and interference levels for all participants, irrespective of their success, particularly as the performance in the backward version immediately following should not be influenced by varying degrees of task exposure in the forward version.

N-back task

We assessed visual working memory using a 2-back version of the well-established n-back task¹⁴⁰. In this task, participants were sequentially presented with single letters and were instructed to press the space key whenever the current letter matched the one shown two trials before (i.e., n-back trials). A set of twenty-four letters was randomly selected from a pool of 16 unique letters (A, B, C, D, F, G, H, J, K, L, M, N, P, S, T, and U). These letters were displayed in black Arial font against a white background (see Fig. 1). Each trial started with a fixation cross that was presented at the center of the screen for 2000 ms. Subsequently, the fixation cross was replaced by the randomly chosen letter, which remained on the screen until a response was made or for a maximum of 500 ms. Participants could respond throughout letter presentation plus an additional 500 ms after the letter was no longer visible. The entire task consisted of 96 trials, organized into four blocks separated by a 10-second break. Each block comprised 24 trials, including six n-back trials.

Flexibility

To assess the costs associated with switching between two different tasks in comparison to conditions in which only one task was performed¹⁴¹, the task-switching test encompassed three different phases (primarily derived from the online task-switching experiment accessible via https://gitlab.pavlovia.org/c8b/task-switching). The first two phases were task-homogeneous, meaning participants engaged in a single task, while the third phase was task-heterogeneous, necessitating participants to flexibly switch between two tasks. In both tasks, a grid of black lines dividing a square into four white quarters was centrally presented on the screen (see Fig. 1). Each trial involved the appearance of either a square or diamond shape outlined by a green frame, filled with green dots. A total of four combinations of shapes and fillings could occur in one quarter at a time (i.e., square comprising two dots, square comprising three dots, diamond comprising two dots, diamond comprising three dots). In the task-homogeneous shape task, diamonds and squares filled with two or three dots exclusively appeared in the upper half of the grid. Participants were instructed to respond with respect to the shape (i.e., press the “x” key for diamonds, press the “m” key for squares) and ignore the filling. Conversely, in the task-homogeneous filling task, stimuli consistently appeared in the lower half of the grid. Participants were instructed to respond with respect to the filling (i.e., press the “x” key for two dots, press the “m” key for three dots) and ignore the shape. During the task-heterogeneous phase, both tasks were integrated: Participants were instructed to respond with respect to shape when a stimulus appeared in the top half of the grid and to decide based on the filling when a stimulus appeared in the bottom half. Half of the task-heterogenous trials were repeat trials (e.g., a “filling” trial preceded by another “filling” trial) and the other half were switch trials (e.g., a “filling” trial preceded by a “shape” trial). The experimental phase comprised 28 task-homogeneous shape-only trials, 28 task-homogeneous filling-only trials, and 52 task-heterogeneous trials (26 switch trials, 26 repeat trials). Four buffer trials at the beginning of each phase were not included in the analyses, resulting in 24, 24, and 48 valid trials, respectively. In a preceding practice phase, participants completed 8 trials in the shape-only condition, 8 trials in the filling-only condition, and 16 trials in the task-heterogeneous condition (8 repeat trials, 8 switch trials).

Inhibition

Go/no-go task

The go/no-go task serves as a measure of inhibition, assessing the ability to withhold responses to a specific stimulus (no-go stimulus), while responding to other, more frequent stimuli (go stimuli). All stimuli in this task were black letters presented in Arial font against a white background in the center of the screen (see Fig. 1). In the first of two phases, the letters E, F, H, K, M, N, T, V, W, and Y were used as no-go stimuli, each presented twice, resulting in 20 no-go trials. The go stimulus X was presented 80 times. To intensify inhibition demands and elevate between-task interference (as in¹⁴²), the second phase reversed the response instructions, with X serving as the no-go stimulus and the letters A, E, F, H, I, K, M, N, R, T, V, W, Y, and Z as go stimuli. Ten of these letters were presented six times and five letters were presented four times. This resulted in a total number of 80 go trials, aligning with the go/no-go trial ratio from Phase 1. In both phases, participants were instructed to respond to the go stimuli by pressing the space key and to refrain from responding in no-go trials. Letters were sequentially presented for 300 ms, following the procedure used by Redick et al.¹⁴². Responses could be made throughout stimulus presentation plus 700 ms after the letter had disappeared, resulting in a total response window of 1000 ms. Participants did not receive feedback. Participants could familiarize with the task in a practice phase consisting of 12 go trials and three no-go trials. In each practice trial, a letter was presented for 400 ms and responses could be made throughout the presentation plus further 1000 ms. Feedback on response accuracy was provided for 400 ms after each practice trial.

Stop-signal task

In the stop-signal task, participants were required to stop an already initiated response. Each trial started with the presentation of a black fixation cross against a white background in the center of the screen. After 250 ms, the fixation cross was replaced by either a black square or circle (see Fig. 1). Participants were asked to indicate which shape they saw as fast as possible by keypress (“f” for circles, “j” for squares). Each shape was presented in 50% of the overall 192 trials. Trials were organized in three blocks, each consisting of 64 trials (48 go trials, 16 stop trials; see¹⁴³), with a 10-second break between blocks. In go trials, the stimulus remained on the screen for 1250 ms and the default inter-stimulus interval was fixed at 2000 ms, regardless of participants’ response latencies. In stop trials, the black shape turned red, signaling participants to inhibit their response. During these stop trials, participants were required to inhibit their initial go reaction and refrain from keypress. In stop-signal tasks, the delay between the onset of the go signal and the onset of the stop signal (stop-signal delay, SSD) determines the difficulty level of inhibiting an initiated response since inhibition becomes easier with shorter SSDs. The stop-signal task was programmed in an adaptive manner with a varying SSD, employing the adaptive staircase tracking algorithm of the STOP-IT software¹⁴⁴. In particular, the SSD began at 250 ms in the first stop trial and continuously adjusted, based on inhibition success in the previous stop trial. Specifically, the SSD increased by 50 ms after successful inhibition and decreased by 50 ms after inhibition failure¹⁴⁴. Consequently, the probability of a response given a stop signal was expected to be at 50%. In that way, the strategy to intentionally prolong one’s response latencies to facilitate inhibition would not be successful.

Analyses

All analyses were conducted using R¹⁴⁵. Regardless of the specific tasks, participants under the age of 18 were excluded from all analyses. In cases where participants engaged in the survey or completed cognitive tasks multiple times consecutively, we ensured that only the initial datasets were included in the analyses. In particular, only their initial dataset from the respective cognitive task was considered for analysis and this was only the case if this dataset could be uniquely matched with their first participation in the survey. For the analyses of the cognitive tasks, individuals lacking normal or corrected-to-normal vision were excluded. Additionally, participants with a history of critical health conditions, such as a stroke, cardiac arrest, cancer, or a brain surgery, as well as those who underwent major surgery within the last month or experienced a loss of consciousness for more than five minutes in the past, were excluded from the analyses. Further exclusions encompassed participants reporting diseases that could impact cognitive functioning, including psychiatric or neurologic disorders (e.g., narcolepsy, schizophrenia, epilepsy, diagnosed depression, migraine), untreated hypertension, and those under medication that could influence cognitive functioning (e.g., neuroleptics, antidepressants).

For all cognitive task analyses, a linear regression model was employed. Infection status (previously infected, previously not infected) served as fixed effect, while controlling for the covariates age (mean-centered), gender (male, female), level of education (binary as “high” and “low”, i.e., completed vs. no completed post-secondary education; see¹⁴⁶, depressiveness (PHQ-9), anxiety (GAD-7), and stress (PSQ-20). Gender was coded as binary variable in the analyses of all cognitive tasks, as the three individuals with diverse gender did not participate in any of the tasks. All p values are reported two-tailed. Effect size d for independent t-tests comparing groups with and without a history of infection was calculated as the difference between group means of the dependent variable, divided by the pooled standard deviation. As the absence of evidence does not constitute evidence for the absence of an effect (see, e.g^146,147,148), we additionally calculated Bayes Factors (BF) to better quantify the strength of evidence for or against the null hypothesis. For the regression analyses, we report BF₁₀ values, representing the ratio of the likelihood of the data under the full model (including the effect of interest) to that under a reduced model without the effect. These were computed using the BayesFactor package for R¹⁴⁹. For t-tests, BF₁₀ were calculated using a Cauchy prior on effect size scaled at r = \(\:\sqrt{2}/2\), indicating the degree of support for the alternative hypothesis (H₁) over the null hypothesis (H₀).

To interpret the resulting Bayes factors, we followed established classification schemes^150,151. Specifically, BF₁₀ values greater than 10 were considered to reflect strong to very strong evidence for H₁, values between 3 and 10 indicated moderate evidence for H₁, and values between 1 and 3 were taken as anecdotal evidence for H₁. A BF₁₀ around 1 suggested no diagnostic value in favor of either hypothesis. Conversely, values between 1/3 and 1 were interpreted as anecdotal evidence for H₀, values between 1/10 and 1/3 as moderate evidence for H₀, and values smaller than 1/10 were considered strong to very strong evidence for H₀.

Long-term memory

Mnemonic similarity task

After excluding participants who met the general exclusion criteria outlined above, we further excluded participants who did not press each key in the recognition test at least once, as this suggests a potential misunderstanding of task instructions. The final datasets all contained a minimum of 10 trials per item category (target, lure, foil), resulting in a sample size of N = 151 (n = 86 previously infected, n = 65 previously not infected; age: M = 43.36, SD = 13.83, range = 18–69 years; 101 female, 50 male; n = 68 with completed post-secondary education, n = 83 without completed post-secondary education). On average, participants in this sample did not respond in 10.53 trials within the given response-time window. Infection groups (previously infected, previously not infected) did not differ in terms of age, t < 1, gender, χ² (1) = 1.97, p = .160, or level of education, χ² (1) = 1.95, p = .163. Within the subgroup of previously infected participants, the average time between the positive PCR test and study participation was 9.27 months (SD = 6.36, range: 0–25 months), with 13 participants having participated within three months after testing positive. Among the remaining 73 participants with positive test results of more than three months ago, 45 reported experiencing at least one post-acute/long-COVID symptom.

To assess mnemonic pattern separation, we computed each participant’s lure discrimination ability, quantified by the Lure Discrimination Index (LDI). The LDI indicates the ability to correctly identify lure items as lures, or, in other words, to recognize that the lures are similar (but not identical) to a studied item, corrected for a response bias towards accepting foils as lures. Thus, the LDI is calculated as the ratio of “similar” responses to lures minus the ratio of “similar” responses to foils, that is, P(“similar” | lure) – P(“similar” | foil). As an indicator of item recognition memory, we calculated the pr score as recognition index (REC). The REC index represents the ability to recognize target items as old while correcting for a bias towards endorsing foils as targets, that is, P(“old” | target) – P(“old” | foil). For all analyses of the MST data, participants with LDI or REC values beyond two standard deviations from the mean were excluded (see also⁸⁷), separately for the previously infected and previously not infected group (n = 10 previously infected; n = 3 previously not infected). This resulted in a final sample size of N = 138 participants (n = 76 previously infected, n = 62 previously not infected). On a trial level, responses faster than 100 ms were excluded to prevent the results from being biased by accidental key presses (5.31% of all trials; see¹⁵²). Additionally, a pattern completion (PC) bias reflecting the tendency to respond “old” to lure items was calculated as the ratio of “old” responses to lures minus the ratio of “similar” responses to lures, that is, P(“old” | lure) – P(“similar” | lure).

Alertness

Following the application of the exclusion criteria specified in the general Analyses section, we further excluded participants who failed to respond in at least 50% of the trials (n = 4, with a mean percentage of 7.50% trials with a response), resulting in a sample size of N = 155 (n = 92 previously infected, n = 63 previously not infected; age: M = 43.60, SD = 14.37, range = 18–76 years; 105 female, 50 male; n = 67 with completed post-secondary education, n = 88 without completed post-secondary education). Infection groups did not differ in terms of age, t < 1, gender, χ² < 1, and level of education, χ² = 1.16, p = .281. The primary outcome measure of this task were mean response latencies, with accuracy (i.e., hit rate, reflecting the proportion of correct responses on the signal relative to the overall trial number) also reported. In all analyses of the alertness data, participants whose dependent variables of interest (i.e., response latencies, accuracy) deviated by more than two standard deviations from the mean were excluded, separately for the previously infected and previously not infected group. For the response latency analyses, the removal of eight outlier participants (n = 4 previously infected; n = 4 previously not infected) resulted in a final sample size of N = 147 participants (n = 88 previously infected, n = 57 previously not infected). In accuracy analyses, six outlier participants were excluded (n = 4 previously infected; n = 2 previously not infected), resulting in a final sample size of N = 149 participants (n = 88 previously infected, n = 61 previously not infected).

Executive functions

Working memory

Corsi block tapping

In addition to the general exclusion criteria (see above), we excluded participants who were unable to accurately repeat even the initial sequence level (i.e., sequences of three blocks in at least three out of five trials). This precaution was taken to address any potential comprehension issues with the task. Overall, N = 250 participants were included in the analyses of the forward version (n = 192 previously infected, n = 58 previously not infected; age: M = 43.80, SD = 13.65, range = 18–72 years; 166 female, 84 male; n = 101 with completed post-secondary education, n = 149 without completed post-secondary education) and N = 224 participants in the analyses of the backward version (n = 176 previously infected, n = 48 previously not infected; age: M = 43.71, SD = 13.64, range = 18–72 years; 150 female, 74 male; n = 95 with completed post-secondary education, n = 129 without completed post-secondary education). Infection groups did not differ regarding age, t < 1, or gender, χ² < 1, in either task version. However, there was an uneven distribution of education levels across infection groups, χ² (1) = 4.66, p = .031, with a greater proportion of individuals with lower educational level in the group previously infected (n = 70 with completed post-secondary education, n = 122 without completed post-secondary education) compared with the group previously not infected (n = 31 with completed post-secondary education, n = 27 without completed post-secondary education). In the backward version, a similar pattern was observed, reaching marginal significance, χ² (1) = 2.87, p = .090, again with more individuals with a lower educational level in the group previously infected (n = 69 with completed post-secondary education, n = 107 without completed post-secondary education) than in the group previously not infected (n = 26 with completed post-secondary education, n = 22 without completed post-secondary education). Within the previously infected sample, the time between the positive PCR test and participation in the forward Corsi task was 11.34 months on average (SD = 7.10, range: 0–29 months) with 25 participants participating within three months of testing positive. Among the remaining 167 participants who had previously tested positive and whose infection was diagnosed more than three months ago, 112 reported experiencing at least one post-acute/long-COVID symptom. For the regression analyses, an additional participant (59 years old, previously infected) was excluded from the analyses of both task versions as he had not completed the questionnaires on depressiveness, anxiety, and stress.

The primary outcome measure for both Corsi versions was the maximum sequence length (i.e., block span) that could be reproduced correctly in at least three out of five trials, calculated separately for the forward and backward version.

N-back task

After applying the exclusion criteria outlined in the general Analyses section, we further excluded one participant who did not press the space key at least once, suggesting a potential misunderstanding of the task. In total, N = 328 participants were included in the analyses (n = 208 previously infected, n = 120 previously not infected; age: M = 41.94, SD = 13.92, range = 18–80 years; 223 female, 105 male; n = 146 with completed post-secondary education, n = 182 without completed post-secondary education). No significant differences emerged between infection groups in terms of age, t < 1, or gender, χ² < 1. However, there was an uneven distribution of education levels across infection groups, χ² (1) = 4.39, p = .036, with a higher proportion of individuals exhibiting lower educational levels in the group previously infected (n = 83 with completed post-secondary education, n = 125 without completed post-secondary education) compared with the group previously not infected (n = 63 with completed post-secondary education, n = 57 without completed post-secondary education). The primary outcome measure for assessing working memory performance was accuracy, quantified by the sensitivity measure d’ (i.e., z(Hit rate) – z(False Alarm rate)). To avoid d’ being undetermined in subjects with hit rates or false alarm rates equal to 0 or 1, we replaced scores equal to 0 by 0.5/n and scores equal to 1 by (n-0.5)/n, with n representing the total number of trials of the respective target category (i.e., target trials or foil trials; see also^153,154,155). Participants identified as outliers within their group based on accuracy (i.e., above or below two standard deviations of the mean) were excluded from all analyses (n = 10 previously infected, n = 7 previously not infected). In addition to a linear regression model testing the impact of infection status on accuracy, the same model was also applied to mean response latencies for correct target responses (i.e., hits). For the regression analyses, an additional participant (59 years old, previously infected) was excluded from the analyses as he had not completed the questionnaires on depressiveness, anxiety, and stress. On the trial level, we excluded trials with response latencies less than 100 ms (0.04% of all trials).

Flexibility

In addition to the exclusion criteria outlined in the general Analyses section, seven participants were excluded due to incomplete participation in the initial two (task-homogeneous) experimental phases, as well as failing to complete a minimum of 10 switch trials and 10 repeat trials within the subsequent third (task-heterogeneous) phase. Overall, N = 325 participants could be included in the analyses (n = 216 previously infected, n = 109 previously not infected; age: M = 42.54, SD = 13.72, range = 18–69 years; 222 female, 103 male; n = 138 with completed post-secondary education, n = 187 without completed post-secondary education). Demographic comparisons indicated no significant differences in age between infection groups, t < 1, or gender distribution, χ² < 1. However, an uneven distribution was observed for educational level across infection groups, χ² (1) = 7.11, p = .008, with a notably higher proportion of individuals with lower educational levels in the group previously infected (n = 80 with completed post-secondary education, n = 136 without completed post-secondary education) compared with the group previously not infected (n = 58 with completed post-secondary education, n = 51 without completed post-secondary education). On average, participants in this sample did not respond in 4.51 trials within the given response-time window. Cognitive flexibility was assessed through two types of costs, namely (1) slowed responses and (2) increased error rates. Higher costs in both aspects indicate reduced flexibility. Specific switch costs were computed by subtracting repeat-trial response latencies (or error rates) in the task-heterogeneous block from switch-trial response latencies (or error rates) in the same block. Mixing costs were determined by subtracting response latencies (or error rates) in the task-homogeneous blocks from non-switch trial response latencies (or error rates) in the task-heterogeneous block. Error rates represented the proportion of incorrect responses within a trial category. Response latencies were log-transformed for the regression analyses to meet the requirement of normally distributed residuals. On the subject level, we further excluded participants who were identified as outliers within their group regarding the respective cost effect (i.e., above or below two standard deviations of the mean). For the costs in response latencies, we excluded 16 outliers for the specific switch-cost analyses (n = 11 previously infected, n = 5 previously not infected) and 15 outliers in the mixing-cost analyses (n = 10 previously infected, n = 5 previously not infected). For the analyses of error costs, we excluded 15 outliers (n = 10 previously infected, n = 5 previously not infected) for the specific switch-cost analyses and 19 outliers (n = 13 previously infected, n = 6 previously not infected) for the mixing-cost analyses. On the trial level, trials with response latencies of < 100 ms were excluded from all analyses (1.44% of all trials). For regression analyses, two participants (20 years old, previously not infected, and 59 years old, previously infected) were excluded due to incomplete questionnaires on depressiveness, anxiety, and stress.

Inhibition

Go/no-go task

In addition to the general exclusion criteria (see above), six participants were omitted from the analyses (n = 4 previously infected, n = 2 previously not infected) due to their failure to complete a minimum of 10 go and 10 no-go trials within each phase. Consequently, the final sample comprised N = 427 participants (n = 219 previously infected, n = 111 previously not infected; age: M = 43.45, SD = 14.14, range = 18–82 years; 237 female, 93 male; n = 136 with completed post-secondary education, n = 194 without completed post-secondary education). Comparisons between infection groups revealed no significant differences in age, t < 1, or gender distribution, χ² < 1. However, the distribution of educational levels was uneven across infection groups, χ² (1) = 4.29, p = .038, indicating a disproportionate number of individuals with lower educational levels in the group previously infected (n = 81 with completed post-secondary education, n = 138 without completed post-secondary education) compared with the group previously not infected (n = 55 with completed post-secondary education, n = 56 without completed post-secondary education). To assess the ability to inhibit responses in no-go trials we focused on the false-alarm rate (i.e., incorrect responses on no-go trials). Sensitivity, represented by d’ (i.e., z(Hit rate) – z(False Alarm rate)), was calculated to account for response bias. The rationale behind this measure posited that, in the event of inhibition impairment caused by a prior SARS-CoV-2 infection, previously infected participants would exhibit a decrease in d’ particularly driven by an increase in false alarm rate. For all participants irrespective of infection status, a decrease in d’ should be observed between the first and the second test phase. To avoid d’ being undetermined in case of hit rates or false alarm rates equal to 0 or 1, scores equal to 0 were replaced by 0.5/n and scores equal to 1 were replaced by (n-0.5)/n, with n representing the total number of trials in the respective trial category (i.e., go trials or no-go trials; see also^153,154,155). Outliers with respect to d’ (i.e., above or below two standard deviations of the mean), identified separately for their infection status groups (n = 19 previously infected, n = 9 previously not infected), were removed.

Stop-signal task

Performance in the stop-signal task was modeled as a race between two different processes: a go process initiated by the go signal and a stop process activated by a stop signal. Following the presentation of both signals, these processes operate concurrently, and the outcome depends on which process finishes first – whether responses are successfully inhibited or not. Successful inhibition is characterized by the completion of the stop process before the go process finishes. Conversely, if the go process finishes before the stop process, participants are unable to inhibit their response. Measuring response inhibition constitutes a unique challenge due to the absence of participant reactions in (correct) response-inhibition trials. However, the stop-signal reaction time (SSRT) can be mathematically estimated based on an independent race model¹⁵⁶. According to this model¹⁵⁷, SSRT is calculated as the difference between stop-signal delay (SSD; i.e., the time between the onset of the go signal and the onset of the stop signal) and the time when the internal response to the stop signal occurs. For the adaptive staircase procedure we used in this task, the internal response to the stop signal equals the mean go reaction time as participants should necessarily stop half of their responses in the stop trials¹⁵⁸. However, the proportion of commission errors (i.e., false alarms in stop trials) in participants who respond extremely rarely or extremely often can deviate from 50%. Therefore, participants with < 25% or > 75% commission errors or > 10% omission errors (i.e., no response in go trials) were excluded (n = 49 previously infected, n = 29 previously not infected), following recommendations by Verbruggen et al.¹⁴³  (see also^144,159). Moreover, remaining participants whose response latencies on unsuccessful stop trials were numerically longer than those on go trials were also excluded, as SSRT cannot be reliably estimated in such cases[143] (n = 6 previously infected, n = 1 previously not infected;). After removing outliers regarding SSRT (two standard deviations above or below the mean of the respective infection group; n = 8 previously infected, n = 5 previously not infected), the final sample consisted of N = 245 participants (n = 167 previously infected, n = 78 previously not infected), age: M = 43.09, SD = 13.93, range = 18–80 years; 167 female, 78 male; n = 117 with completed post-secondary education, n = 128 without completed post-secondary education). No significant differences between infection groups were observed regarding age, t < 1, or gender distribution, χ² < 1. However, educational levels were unevenly distributed across infection groups, χ² (1) = 5.13, p = .023, revealing a higher proportion of individuals with lower educational levels in the group previously infected (n = 71 with completed post-secondary education, n = 96 without completed post-secondary education) compared with the group previously not infected (n = 46 with completed post-secondary education, n = 32 without completed post-secondary education). For SSRT estimation, the integration method was employed, as it is more reliable and less biased than the mean method (see ^143,156,160). The integration method considers SSRT as the time point at which the integral of the response-time distribution equals the likelihood for incorrect responses in stop trials (see¹⁴³). The time point at which the stopping process finishes thereby equals the nth response time, where n is the number of all go-trial responses (including choice errors and responses made before the stop signal occurred) multiplied by the likelihood of responding in a stop trial. To account for missing response latencies in omission-error go trials, maximum response latency was assigned to these trials (see¹⁴³). SSRTs were log-transformed for regression analyses to meet the requirement of homoscedasticity.