Predicting Declines in Attention and Memory During a Prolonged Mental Task

COG-PRED is a prospective, single-group laboratory study of cognitive performance during sustained mental effort. Subjective fatigue, autonomic responses, and task performance may change at different rates. The study will examine whether resting autonomic activity and early physiological responses provide information about subsequent performance beyond that available from sleep duration, sleepiness, and initial task accuracy. All participants will follow the same sequence of procedures, without randomization or a separate control group.

Cognitive load will be induced using a study-specific implementation of the Time Load Dual-back (TloadDback) task. Letters and digits will alternate on the screen. Participants will respond when a letter matches the preceding letter and withhold a response when it does not. For digits, participants will classify each number as odd or even. Responses will be made with the dominant hand, while the hand carrying the skin conductance electrodes remains supported.

Following familiarization, a shortened adaptive calibration will identify an individual presentation interval using a criterion of at least 85% weighted accuracy. Search blocks will contain 30 stimuli, and the selected interval will be checked in a separate 40-stimulus verification block. Calibration will last no longer than five minutes. Participants who do not obtain a verified interval will not proceed to the main task. Once verified, the interval will remain fixed throughout the main task and the post-rest task. Individual calibration adjusts task pace but does not assume identical subjective workload across participants.

Physiological recording will cover the resting measurement, task phases, and recovery period. Cardiac and skin conductance signals will be synchronized with stimulus and response events. Resting features will be derived from the final four minutes of the five-minute baseline. Early task responses will be characterized by changes from this baseline during the first four minutes of the main task.

Prespecified linear regression models will assess later task accuracy while accounting for initial accuracy. A reference model containing previous-night sleep duration, pretask sleepiness, and initial accuracy will be compared with an expanded model that additionally includes resting heart rate, heart rate variability, skin conductance, and their early task changes. Predictive performance will be assessed using 10-fold cross-validation at the participant level. Physiological information recorded after the prediction window will not be used as an early predictor.

Immediately after the main task, participants will rate mental fatigue and complete the six dimensions of the National Aeronautics and Space Administration Task Load Index (NASA-TLX), scored as an unweighted mean. These assessments will precede the full five-minute seated rest. Participants will then rate fatigue and perform a four-minute task with a new stimulus sequence at the same calibrated speed. The total interval between the main task and the post-rest task will be recorded.

Secondary analyses will describe performance and physiological changes over time and compare the final four minutes of the main task with the four-minute post-rest task. Relationships between autonomic recovery and changes in performance will be explored separately. Because there is no no-rest control condition, changes after the break will not be interpreted as effects attributable solely to rest.

• Individually Calibrated Time Load Dual-back Task With Seated Rest (Behavioral): Participants perform the Time Load Dual-back task, alternating letter 1-back matching with odd/even digit classification. A study-specific adaptive calibration starts at 1400 ms and tests presentation intervals between 400 and 1600 ms. Search blocks contain 30 stimuli; the selected interval must achieve at least 85% weighted accuracy in a separate 40-stimulus verification block. Calibration is limited to five minutes; participants without a verified interval do not proceed to the main task. The verified interval is used for a 16-minute main task and a four-minute post-rest task. Fatigue and workload ratings follow the main task, before five full minutes of quiet seated rest. A fatigue rating precedes the post-rest task. Task pace remains fixed after calibration.

Primary outcomes

• Weighted Accuracy During the Final Four Minutes of the Main Cognitive Task [During seconds 720-960 (the final four minutes) of the 16-minute main task, within a single study visit.]: Weighted accuracy (WAcc) on the Time Load Dual-back task is calculated as 0.65 x letter accuracy + 0.35 x digit accuracy, with both component accuracies expressed as percentages. Letter accuracy includes correct responses to matching letters and correct withholding of responses to nonmatching letters. Digit accuracy is the percentage of correct odd/even classifications. Errors and required-response omissions are scored as incorrect. The first letter of each new sequence and technically excluded trials are removed from the relevant denominators; analytical windows do not reset the letter reference. Scores range from 0 to 100%, with higher scores indicating better performance. The primary endpoint is the late-task score; initial accuracy is included as a predictor, rather than subtracted to define this endpoint.

Secondary outcomes

• Weighted Accuracy Across the Main Cognitive Task [During four consecutive 4-minute windows (0-240, 240-480, 480-720, and 720-960 seconds) of the 16-minute main task.]: Weighted accuracy is calculated separately for each consecutive four-minute window as 0.65 x percentage letter accuracy + 0.35 x percentage digit accuracy. Correct letter responses include both target detections and correct withholding on non-target trials. The scoring and exclusion rules are identical to those for the primary outcome. Scores range from 0 to 100%, with higher scores indicating better performance. The repeated scores describe the course of performance during sustained cognitive load.
• Change in Weighted Accuracy From the End of the Main Task to the Post-Rest Task [During the entire 4-minute task after five minutes of seated rest; compared with seconds 720-960 of the 16-minute main task.]: Weighted accuracy during the entire four-minute post-rest task minus weighted accuracy during the final four minutes of the main task, expressed in percentage points. Both scores use the same weighting (0.65 x letter accuracy + 0.35 x digit accuracy), scoring rules, and individually calibrated presentation interval. The difference ranges from -100 to +100 percentage points. Positive values indicate better accuracy after rest. This comparison describes change after the break without attributing the change solely to rest.
• Median Correct-Response Reaction Time Across Task Phases [During each of four consecutive 4-minute windows of the 16-minute main task and during the entire 4-minute post-rest task, within one visit.]: Within-participant median time from stimulus onset to the first keypress on a correctly scored response-required trial, in milliseconds, calculated separately for each assessment window. Only correct trials requiring a response are included; correct withholding on non-target letter trials has no reaction time. Responses earlier than 100 ms and technically excluded trials are not included. Response times must fall within the stimulus response window. Main-task trajectories and post-rest versus late-main-task values will be compared. Shorter times indicate faster responses and will be interpreted together with accuracy.
• Standard Deviation of Correct-Response Reaction Time Across Task Phases [During each of four consecutive 4-minute windows of the 16-minute main task and during the entire 4-minute post-rest task, within one visit.]: Within-participant sample standard deviation of correct-response reaction times, in milliseconds, calculated separately for each assessment window. The same eligible responses are used as for median correct-response reaction time: correct keypresses from 100 ms after stimulus onset to the end of the response window. Correct withholding, anticipations, incorrect responses, and technically excluded trials are not included. Higher values indicate greater variability in response speed. Main-task trajectories and post-rest versus late-main-task values will be compared.
• Number of Incorrect Responses Across Task Phases [During each of four consecutive 4-minute windows of the 16-minute main task and during the entire 4-minute post-rest task, within one visit.]: Number of trials with an incorrect keypress in each assessment window, including responses to non-target letters, incorrect response keys on letter-target or digit trials, and anticipatory responses earlier than 100 ms. A trial is scored once according to the first-keypress rule. Required-response trials without a keypress are counted separately as omissions; correct withholding on non-target letters is not an error. Technically excluded trials and the first reference-setting letter of each sequence are omitted. Counts will be interpreted with the number of eligible trials, which varies with the calibrated presentation interval. Lower counts indicate fewer incorrect responses.
• Number of Required-Response Omissions Across Task Phases [During each of four consecutive 4-minute windows of the 16-minute main task and during the entire 4-minute post-rest task, within one visit.]: Number of letter-target and digit trials in which no keypress is recorded within the response window. Correct withholding on non-target letter trials is not an omission. Trials with anticipatory or incorrect keypresses are classified as incorrect responses rather than omissions. Technically excluded trials and the first reference-setting letter of each sequence are omitted. Counts will be interpreted with the number of eligible response-required trials. Main-task trajectories and post-rest versus late-main-task values will be compared. Lower counts indicate fewer omissions.
• Self-Reported Mental Fatigue Across the Study Visit [Before baseline; after calibration/before the main task; immediately after the 16-minute main task; after five minutes of rest/before the post-rest task; and immediately after the 4-minute post-rest task, within one visit.]: Single-item rating of current mental fatigue from 0 (no mental fatigue) to 100 (extreme mental fatigue). Ratings are obtained before baseline, after calibration immediately before the main task, immediately after the main task, after the full five-minute rest before the post-rest task, and after the post-rest task. Repeated ratings describe fatigue during preparation, sustained task performance, rest, and task resumption. Higher scores indicate greater fatigue.
• Perceived Workload on the Unweighted NASA Task Load Index [Immediately after the 16-minute main task and before the start of the five-minute seated rest.]: Unweighted mean of the six ratings on the National Aeronautics and Space Administration Task Load Index (NASA-TLX): mental demand, physical demand, temporal demand, performance, effort, and frustration. Each dimension is scored from 0 to 100. The performance dimension is oriented from success to failure so that higher values on every dimension indicate greater workload or a less favorable assessment. The overall score ranges from 0 to 100, with higher scores indicating greater perceived workload. Ratings refer only to the 16-minute main task. Pairwise weighting is not used.
• Heart Rate Variability During Baseline, Cognitive Load, and Rest [During seconds 60-300 of the 5-minute baseline, each consecutive 4-minute window of the 16-minute main task, and seconds 60-300 of the 5-minute seated rest.]: Root mean square of successive differences between normal-to-normal heartbeat intervals (RMSSD), in milliseconds, calculated from quality-controlled electrocardiographic recordings using continuous valid segments. Values are calculated separately for each assessment window. The natural logarithm of RMSSD is used in the predictive model; early reactivity is the first main-task window minus baseline on that logarithmic scale. Resting recovery compares the final four minutes of seated rest with the final four minutes of the main task and baseline. Higher RMSSD indicates greater successive-interval variability, without implying better cognitive performance.
• Mean Skin Conductance During Baseline, Cognitive Load, and Rest [During seconds 60-300 of baseline, each consecutive 4-minute window of the 16-minute main task, and the 5-minute rest (initial 60 seconds and final 240 seconds).]: Mean skin conductance, in microsiemens, calculated from valid electrodermal samples in each assessment window. This measure includes both tonic level and contributions from phasic responses; it is not a decomposed tonic-only measure. The predictive model uses ln(1 + the numerical skin conductance value in microsiemens), with early reactivity calculated as the first main-task window minus baseline on this scale. Resting recovery compares the final four minutes of seated rest with the final four minutes of the main task and baseline. Higher values indicate greater skin conductance, without implying better or worse cognitive performance. The first minute of rest is additionally examined to describe the initial transition.

Other outcomes

• Heart Rate During the Post-Rest Cognitive Task [During the entire 4-minute task after five minutes of seated rest; compared with seconds 720-960 of the 16-minute main task.]: Mean heart rate, in beats per minute, derived from quality-controlled electrocardiographic data during the four-minute post-rest task. The value is compared with the final four minutes of the main task to explore the autonomic response during task resumption at the same calibrated presentation interval.
• Heart Rate Variability During the Post-Rest Cognitive Task [During the entire 4-minute task after five minutes of seated rest; compared with seconds 720-960 of the 16-minute main task.]: Root mean square of successive differences between normal-to-normal heartbeat intervals (RMSSD), in milliseconds, calculated from continuous valid segments of the electrocardiographic recording during the four-minute post-rest task. The value is compared with the final four minutes of the main task to explore the autonomic response during task resumption.
• Skin Conductance During the Post-Rest Cognitive Task [During the entire 4-minute task after five minutes of seated rest; compared with seconds 720-960 of the 16-minute main task.]: Mean skin conductance, in microsiemens, calculated from valid electrodermal samples during the four-minute post-rest task. The value is compared with the final four minutes of the main task to explore the autonomic response during task resumption. The measure includes tonic level and contributions from phasic responses.
• Individually Verified Stimulus Presentation Interval [At successful completion of the adaptive calibration, which lasts no longer than five minutes and precedes the 16-minute main task.]: Stimulus presentation interval accepted after successful calibration of the Time Load Dual-back task, in milliseconds. Permitted values range from 400 to 1600 ms in 100-ms increments. The interval must satisfy the weighted-accuracy criterion in a separate verification block and is then fixed for both the main and post-rest tasks. Lower values indicate a faster task pace, not an intelligence score. This measure is used in supplementary analyses; no interval is assigned if calibration is unsuccessful.
• Standard Deviation of Normal-to-Normal Heartbeat Intervals [During seconds 60-300 of baseline, each consecutive 4-minute window of the 16-minute main task, seconds 60-300 of seated rest, and the entire 4-minute post-rest task.]: Standard deviation of normal-to-normal heartbeat intervals (SDNN), in milliseconds, derived from quality-controlled electrocardiographic data. This supplementary measure is calculated separately within the four-minute physiological assessment windows. Higher values indicate greater overall heartbeat-interval variability within the window. SDNN is supplementary to the prespecified RMSSD measure and is not substituted for RMSSD in the primary predictive model.

Sex: All, ages: 18 Years to 30 Years, accepts healthy volunteers

Inclusion Criteria:

* Age 18-30 years.
* Good general health according to self-report.
* Normal or corrected-to-normal vision and ability to read the letters and digits presented on the screen.
* Ability to understand instructions in Polish and perform the computer task.
* Provision of voluntary informed consent.

Exclusion Criteria:

* Neurological, psychiatric, or cardiovascular conditions that significantly affect participant safety, task performance, or heart rhythm.
* Use of medications or substances that significantly affect heart rhythm, arousal, or cognitive performance.
* Acute infection, fever, feeling unwell, or skin damage that makes sensor placement difficult.
* Failure to understand the task after repetition of the instructions and practice, or unsuccessful calibration.
* Caffeine or nicotine use within four hours, alcohol consumption within 24 hours, or intense physical exercise within 12 hours before the visit.
* Absence or withdrawal of informed consent.

• Laborde S, Mosley E, Thayer JF. Heart Rate Variability and Cardiac Vagal Tone in Psychophysiological Research - Recommendations for Experiment Planning, Data Analysis, and Data Reporting. Front Psychol. 2017 Feb 20;8:213. doi: 10.3389/fpsyg.2017.00213. eCollection 2017. PMID 28265249
• Lim J, Kwok K. The Effects of Varying Break Length on Attention and Time on Task. Hum Factors. 2016 May;58(3):472-81. doi: 10.1177/0018720815617395. Epub 2015 Dec 29. PMID 26715686
• Matuz A, van der Linden D, Kisander Z, Hernadi I, Kazmer K, Csatho A. Enhanced cardiac vagal tone in mental fatigue: Analysis of heart rate variability in Time-on-Task, recovery, and reactivity. PLoS One. 2021 Mar 3;16(3):e0238670. doi: 10.1371/journal.pone.0238670. eCollection 2021. PMID 33657124
• Matuz A, van der Linden D, Darnai G, Csatho A. Generalisable machine learning models trained on heart rate variability data to predict mental fatigue. Sci Rep. 2022 Nov 21;12(1):20023. doi: 10.1038/s41598-022-24415-y. PMID 36414673
• O'Keeffe K, Hodder S, Lloyd A. A comparison of methods used for inducing mental fatigue in performance research: individualised, dual-task and short duration cognitive tests are most effective. Ergonomics. 2020 Jan;63(1):1-12. doi: 10.1080/00140139.2019.1687940. Epub 2019 Nov 18. PMID 31680632
• Hrabovecky JM, De Tiege X, Samyn C, Borragan G, Peigneux P, Strauss M, Slama H. Is a single calibration for the TloadDback cognitive fatigue induction task reliable? Front Psychol. 2025 Jul 24;16:1561819. doi: 10.3389/fpsyg.2025.1561819. eCollection 2025. PMID 40777221
• Borragan G, Slama H, Bartolomei M, Peigneux P. Cognitive fatigue: A Time-based Resource-sharing account. Cortex. 2017 Apr;89:71-84. doi: 10.1016/j.cortex.2017.01.023. Epub 2017 Feb 3. PMID 28237888
• Borragan G, Slama H, Destrebecqz A, Peigneux P. Cognitive Fatigue Facilitates Procedural Sequence Learning. Front Hum Neurosci. 2016 Mar 3;10:86. doi: 10.3389/fnhum.2016.00086. eCollection 2016. PMID 26973501