Appendix of How Can Well-Being Be Measured? Identification of Physiological Metrics for Estimating Cognitive Workload in Office Work Environments

Appendix of How Can Well-Being Be Measured? Identification of Physiological Metrics for Estimating Cognitive Workload in Office Work Environments M. García-Romero1[0000−0003−2113−3497], A. Martínez-Rojas1[0000−0002−2782−9893], A. Jiménez-Ramírez1[0000−0001−8657−992X], and J.G. Enríquez1[0000−0002−2631−5890] Languages and Informatic Systems Department, University of of Seville, Avenida Reina Mercedes, s/n. 41012, Seville, Spain. {mgarcia44,amrojas,ajramirez,jgenriquez}@us.es This appendix corresponds to the tertiary review executed to address the research question RQ1. What physiological metrics are identified in the literature to estimate the level of CWL? related to the paper "How Can Well-Being Be Measured? Identification of Physiological Metrics for Estimating Cognitive Workload in Office Work Environments". The extracted physiological metrics were grouped into 11 categories, which were developed in the following sections and presented in their corresponding tables. To support reader understanding, each table is accompanied by an explanation of what the categories and metrics represent and how these metrics typically behave under varying cognitive-workload conditions, according to prior research. 1 Electrocardiography (ECG) and Cardiovascular Electrocardiography (ECG) records the heart’s electrical activity, enabling derivation of heart rate (HR) and heart rate variability (HRV) as markers of sympathetic and parasympathetic balance under changing cognitive demands. Other Cardiovascular signals such as photoplethysmography (PPG) and blood pressure capture peripheral hemodynamics and vascular tone, extending the analysis from cardiac rhythm to beat-to-beat perfusion dynamics. In table 1, we identified 37 cardiovascular metrics, organized by physiological source and analytical domain: (I) heart rate and heart rate variability and basic statistics (ECG01–ECG06), (II) frequency–domain HRV (ECG07–ECG14), (III) time–domain HRV (ECG15–ECG25), (IV) non–linear/geometrical HRV (ECG26–ECG30), (V) Photoplethysmogram (ECG31-ECG34) and (VI) blood pressure (ECG35–ECG37). HR (ECG01) represents the number of cardiac contractions, or heartbeats, over a period of one minute (beats per minute - bpm). HR increases as CWL rises, reflecting sympathetic activation and vagal withdrawal [28,16]. The mean HR 2 M. García-Romero et al. (ECG02), minimum and maximum HR (ECG03/ECG04) values shift upward under sustained task demand and the standard deviation of HR (ECG05) over short windows typically contracts with sustained load [27], but they are not used or just mentioned in the literature. HRV (ECG06) is defined as the variation in time intervals between successive heartbeats, known as Interbeat Intervals (IBI) or R-R intervals [28]. It is an index of autonomic nervous system function, where higher variability reflects greater parasympathetic (vagal) activity and better physiological adaptability to stress [28]. Therefore, HRV decreases and estabilizes as tasks become more demanding, capturing reduced parasympathetic situation; this pattern is consistently displayed across applied domains (aviation, surgery, control rooms) to index mental effort [10,16]. 1.1 HRV Frequency domain. The majority of studies reported that the high frequency (HF; ECG07), low frequency (LF; ECG08) and LF/HF ratio (ECG09) corresponding to HRV were sensitive to differentiate CWL [28,10,4,7,31,23]. A decreased HF and LF indicate signs of increased CWL. LF/HF ratio increases as CWL becomes larger [28,5,4,31]. Very-low-frequency (VLF; ECG10) and ultra-low-frequency (ULF; ECG11) components contribute little in short task windows and are influenced by longer-term regulatory processes; when available from longer recordings, they tend to contract with the global variability reduction accompanying higher CWL [28,5,6]. However, the mid frequency (ECG12) is only mentioned in one study, but it is unclear what kind of relationship it has with CWL [28]. HRV Total power (ECG13), which indicates that it decreases as demands increase due to the parasympathetic system retirement and stabilization of HRV under stress and cognition demands conditions [28]. Non-standard bands (e.g., wavelet-packet band 2; ECG14) are less established for workload inference and see limited use in practice [28]. Identification of Physiological Metrics for Estimating CWL 3 Category Subcategory Metric ID CWL Relevance Sources Electrocardiograhpy (ECG) / Cardiovascular Heart Rate (HR) ECG01 ↑↑↑ [28], [12], [1], [20], [10], [27], [31], [26], [8], [4], [32], [25], [13], [19], [2], [6], [5], [7], [22], [30], [9], [15] Mean HR ECG02 ↑↑ [27], [23], [4], [13] Min HR ECG03 ↑[27] Max HR ECG04 ↑[27] Standard Deviation HR ECG05 ↑[27] Heart Rate Variability (HRV) ECG06 ↑↑↑ [28] [12], [1], [20], [3], [10], [27], [23], [8], [4], [13], [19], [2], [6], [5], [7], [11], [22], [24], [30], [29], [9], [15] Frequency-domain HRV High Frequency (HF) ECG07 ↑↑↑ [28], [20], [10], [31], [23], [4], [25], [19], [6], [5], [7], [11], [15] Low frequency (LF) ECG08 ↑↑↑ [28], [20], [10], [23], [4], [25], [19], [6], [5], [7], [11], [15] LF/HF ratio ECG09 ↑↑↑ [28], [31], [23], [8], [4], [21], [19], [6], [5], [7], [11] Very Low Frequency (VLF) ECG10 ↑↑ [28], [20], [25], [6], [5], [7] Ultra Low Frequency (ULF) ECG11 ↑[20], [26], [6], [5] Mid Frequency ECG12 ↑[28] Total Power ECG13 ↑[28] Wavelet-packet band 2 (non-standard) ECG14 ↑[28] Time-domain HRV Interbeat interval (IBI) ECG15 ↑↑↑ [28], [20], [10], [31], [26], [4], [25], [13], [19], [5], [7], [11], [22], [30] Mean IBI ECG16 ↑↑↑ [27], [26], [23], [4], [25], [7] pNN50 ECG17 ↑↑↑ [28], [20], [10], [27], [31], [26], [23], [25], [6], [5], [7] SDNN ECG18 ↑↑↑ [28], [20], [10], [27], [31], [23], [25], [19], [6], [5], [7] SDANN ECG19 ↑↑ [20], [5], [7] RMSSD ECG20 ↑↑↑ [28], [20], [10], [27], [31], [26], [23], [8], [4], [25], [19], [5], [7] NN50 ECG21 ↑↑ [28], [23], [5], [7] TINN ECG22 ↑↑ [28], [27] SDSD ECG23 ↑↑ [28], [25], [5], [7] Min NN ECG24 ↑[28] Max NN ECG25 ↑[28] Nonlinear/Geometrical HRV SampEn (Sample Entropy) ECG26 ↑↑ [28], [7], [17] ApEn (Approximate Entropy) ECG27 ↑↑ [28], [7], [17] HRV Triangular Index (HRVTI) ECG28 ↑↑ [28], [20], [5], [7] Poincaré (SD1/SD2) ECG29 ↑↑ [28], [23], [7] DFA α1–α2ECG30 ↑[7] Photoplethysmogram (PPG) Blood Volume Pulse (BVP amplitude) ECG31 ↑↑ [2], [7], [13], [30], [15] Blood Flow Velocity ECG32 ↑[28], [4], [13], [6] Pulse Rate Variability (PRV, from PPG) ECG33 ↑[7] Blood Oxygen % Saturation (SpO2) ECG34 ↑[28], [12], [24] Blood Pressure (BP) Systolic Blood Pressure (SBP) ECG35 ↑↑ [28], [4], [5], [22] Diastolic Blood Pressure (DBP) ECG36 ↑↑ [28], [4], [5], [22] Mean Arterial Pressure (MAP) ECG37 ↑↑ [5], [22], [9] Table 1. List of electrocardiographic (ECG) and cardiovascular metrics. 4 M. García-Romero et al. 1.2 HRV Time domain. IBI (ECG15) and HR have an inverse relationship. Shorter IBI means higher HR (HR acceleration). Conversely, longer IBI means lower HR (HR deceleration). Therefore, high workload is translated into IBI and the mean IBI is reduced (ECG16) [28,10]. Vagal-sensitive indices, such as RMSSD that measures short-term beat-to-beat variation (Root Mean Square of Successive Differences; ECG20), pNN50 (Percentage of Successive Normalized R-R intervals differing by more than 50 ms; ECG17), and SDSD (ECG23) (Standard Deviation of the Successive Differences), decrease reliably with higher cognitive demand, reflecting suppression of fast parasympathetic fluctuations and they are useful and contrast estimators of CWL [28,10,4,23,7]. SDNN (Standard Deviation of N-N intervals; ECG18) and SDANN (Standard Deviation of all N-N intervals; ECG19), reflecting overall variability (SDANN reflects over longer windows) generally declines with rising cognitive demands, although sensitivity depends on window length and non-stationarity; short, task-locked epochs yield clearer discrimination than long [28,5,4]. Count and width based measures, the number of successive N-N intervals differing by more than 50 ms (NN50/ECG21); TINN (Triangular Interpolation of the NN Interval histogram; ECG22) typically compress as high-frequency variability wanes; extreme-interval markers (Min NN/Max NN) (ECG24/ECG25) likewise narrow under cognitive load but they are not used or mentioned at all in literature [28,27,7]. 1.3 Nonlinear/Geometrical HRV. Complexity oriented indices derived from heartbeat dynamics generally decrease with increasing CWL. The sample entropy (SampEn; ECG26) and the approximate entropy (ApEn; ECG27) trend downward as cardiac control becomes more regular and less complex with effort (magnitude depends on preprocessing and epoch length) [28,7,17]. Poincaré descriptors (ECG29) shrink along the short axis (SD1) and often along the long axis (SD2), representing short-term variability. SD1 shrinks because it is heavily influenced by parasympathetic activity, which is reduced during cognitive tasks. Similarly, SD2, representing long-term variability that involves both sympathetic and parasympathetic activity, also often shrinks because the overall variability of the heart rate decreases under higher cognitive demand. [28,23,7]. The HRV Triangular Index (HRVTI; ECG28), a geometric measure of overall NN-interval variability, serves as a proxy for global autonomic flexibility. Under higher CWL, HRVTI typically decreases, reflecting reduced total variability consistent with sympathetic activation and vagal withdrawal. Evidence related to HRVTI is limited but supportive [28,5,20,7]. Detrended fluctuation analysis exponents (e.g., α1; ECG30) can shift toward more persistent correlations with cognitive load, consistent with constrained adaptability of the cardiac control system, although protocol differences warrant triangulation with time/frequency measures [7]. Identification of Physiological Metrics for Estimating CWL 5 1.4 Photoplethysmogram (PPG) and blood pressure. Photoplethysmography shows sympathetically mediated vasoconstriction during mental effort: blood volume pulse (BVP) amplitude (ECG31) decreases as workload increases, providing a complementary vascular correlate of mental demand [2,5,7,19]. Pulse-rate variability (PRV; ECG32) from PPG tracks HRV patterns at rest and under moderate movement but may diverge from ECG-HRV under strong vasomotor reactivity or motion; in workload monitoring, PRV is useful when ECG is impractical but benefits from validation against ECG in the target context [5,7,19]. Blood flow velocity (ECG33) emerges mainly as a task-sensitive cerebral hemodynamic marker, but evidence specific as a standalone workload index remains sparse [28,4,6,13], Oxygen saturation (SpO2; ECG34) typically remains stable in healthy participants performing cognitive tasks and is not a sensitive index of workload per se; it is more informative as a safety/physiology control alongside other measures [12,24,28]. Corresponding to blood pressure, systolic, diastolic, and mean arterial pressure (SBP/DBP/MAP) (ECG35/ECG37) tend to increase with CWL owing to sympathetic vasoconstriction and cardiac output changes; however, effect sizes vary with task type, posture, and measurement modality [5,7,4]. 2 Electroencephalography (EEG) Electroencephalography (EEG) provides a millisecond-resolved window into cortical dynamics during task performance. By capturing both oscillatory activity and time-locked responses, EEG indexes neural activation, inhibition, and information-processing efficiency that covary with CWL. Spectral band power and band-ratio metrics summarize shifts in rhythmic activity; event-related potentials (ERPs) reflect resource allocation and processing speed through amplitude and latency; and entropy/complexity measures quantify signal irregularity. When paired with rigorous preprocessing (artifact control, referencing) and standardized windowing, these complementary indices make EEG a sensitive probe of CWL fluctuations. In table 2, we identified 40 electroencephalographic metrics, grouped by analytical focus into three main domains: (I) spectral band power (EEG01–EEG05), which quantifies the absolute or relative power within canonical frequency bands (δ,θ,α,β,γ); (II) spectral ratios (EEG06–EEG13), representing composite indices of cortical activation or inhibition through the proportion between bands; (III) event-related potentials (ERP) (EEG14–EEG30), comprising amplitude and latency-based components; and (IV) entropy measurements (EEG31-EEG40), which quantify the irregularity of ECG signals. 2.1 Spectral band power. Spectral band power refers to the amount of power within a specific range of frequencies or wavelengths. It quantifies the strength of the signal within that 6 M. García-Romero et al. chosen band, providing a way to analyze how power is distributed across different parts of the spectrum. Power in the canonical EEG bands is sensitive to changes in CWL. Frontal midline θpower spectral density power (EEG02) (4-8 Hz) typically increases as task demands increase, especially for workingmemory, arithmetic and continuous monitoring tasks [28,17,26,10]. In contrast, parietal–occipital αpower (EEG03) (8-13 Hz) generally decreases with higher workload, consistent with cortical activation and inhibition release in areas relevant to the task [28,5,26,10]. βpower (EEG04) (13-30 Hz) often shows moderate increases with focused, sensorimotor, and vigilance demands, but its directionality is less consistent across paradigms and recording sites [28,17]. Evidence for γ power (ECG05) ( >30 Hz) indicates localized increases during intensive perceptual binding or memory operations, yet findings are heterogeneous and sensitive to noise and muscle artifacts and careful preprocessing is required [5,17]. Finally, δpower (EEG01) (<4Hz) is less reliable as a workload marker; when it does vary, it more often indexes drowsiness or low arousal than increased cognitive demand, so it should be used cautiously to measure CWL [5,2,22]. 2.2 Spectral ratios. Composite spectral ratios summarize concurrent band shifts and often enhance robustness. Ratios that place θin the numerator or αin the denominator, e.g θ/β (EEG07) and θ/α (EEG13) tend to increase with higher cognitive control demands because θrises while βholds or modestly increases [28,26,20]. Conversely, ratios dominated by αin the numerator, e.g., (α/β,α/θ; EEG06/EEG08) usually decrease with rising workload due to αsuppression [28,17]. The engagement-style index β/(α+θ)(EEG09) increases with sustained, focused engagement and is frequently used as a continuous proxy for task involvement; its inverse (α+θ)/β (EEG12) moves in the opposite direction and is sometimes framed as a workload index in settings where θrises strongly and βremains stable [28,17,4]. Ratios involving δor γ-e.g., δ/γ,γ/θ (EEG10/EEG11)- are less standardized for occupational workload monitoring and show task-dependent behavior; they are better treated as exploratory or adjunct features rather than primary markers [7,32,17]. Identification of Physiological Metrics for Estimating CWL 7 Category Subcategory Metric ID CWL Relevance Sources Electroencephalography (EEG) Spectral band power PSD δ(power) EEG01 ↑↑ [28], [2], [5], [7], [22], [9], [15], [17] PSD θ(power) EEG02 ↑↑↑ [28], [20], [10], [26], [23], [4], [2], [6], [5], [7], [22], [9], [15], [17] PSD α(power) EEG03 ↑↑↑ [28], [20], [10], [26], [4], [2], [6], [5], [7], [22], [9], [15], [17] PSD β(power) EEG04 ↑↑↑ [28], [20], [10], [26], [4], [2], [6], [5], [7], [22], [9], [15], [17] PSD γ(power) EEG05 ↑↑↑ [28], [20], [10], [26], [4], [2], [5], [7], [22], [9], [15], [17] Spectral ratios α/βRatio EEG06 ↑↑↑ [20], [26], [4], [32], [7], [17] θ/βRatio EEG07 ↑↑↑ [28], [20], [26], [4], [7] α/θRatio EEG08 ↑↑↑ [28], [20], [26], [4], [7] β/(α+θ) Ratio EEG09 ↑↑↑ [28], [20], [4], [17] δ/γRatio EEG10 ↑[32], [7] γ/θRatio EEG11 ↑[20], [7] (α+θ)/βRatio EEG12 ↑↑ [17] θ/αRatio EEG13 ↑↑ [17] Event-Related Potentials (ERP) N1 EEG14 ↑[17] N2 EEG15 ↑[17] N100 EEG16 ↑[28], [17] P200 EEG17 ↑[28], [17] N200 EEG18 ↑[28] P300 (P3 complex) EEG19 ↑↑ [28], [5], [17] P700 EEG20 ↑[17] P2 EEG21 ↑[17] P1 EEG22 ↑[17] P3a EEG23 ↑↑ [28], [17] P3b EEG24 ↑↑ [28], [17] Late Positive Potential (LPP) amplitude EEG25 ↑[28], [17] Mismatch Negativity (MMN) EEG26 ↑[28], [17] Early Slow-Wave component (SW1) EEG27 ↑[17] Late Slow-Wave component (SW2) EEG28 ↑[17] Mean Bandpower EEG29 ↑[15] Mean Amplitude EEG30 ↑[15] Entropy measurements Wavelet Entropy (WEn) EEG31 ↑[17] Sample Entropy (SampEn) EEG 32 ↑[17] Spectrum Entropy (SPEn) EEG 33 ↑[17] Peak-to-Peak Sample Entropy (PP-SampEn) EEG 34 ↑[17] Approximate Entropy (ApEn) EEG 35 ↑[17] Shannon’s Entropy EEG 36 ↑[17] Tsallis Wavelet Entropy EEG 37 ↑[17] Log Energy Entropy EEG 38 ↑[17] Rényi’s Entropy EEG 39 ↑[17] Fuzzy Entropy EEG 40 ↑[17] Table 2. List of Electroencephalography (EEG) metrics. 8 M. García-Romero et al. 2.3 Event-Related Potentials (ERPs). Event-Related Potentials (ERPs) components index how efficiently the brain can allocate attention to incoming stimuli while performing a task. As mental effort rises, a consistent pattern emerges across components: amplitudes tend to decrease (reflecting fewer resources available for the probe stimulus), and latencies tend to increase (slower processing). Higher CWL is characterized by lower ERP amplitudes and longer latencies. In ERP terms, P1 (EEG22) is an early positive deflection around 80–120 ms indexing initial visual sensory processing [17]; N1/N100 (EEG14/EEG16) is an early negative component ( 100–180 ms) linked to selective attention and perceptual gating [17]; P2/P200 (EEG21/EEG17), a positive component ( 150–250 ms), reflects early categorization and rapid feature evaluation [28,17]; N2/N200 (EEG15/EEG18), a negative component ( 200–350 ms), indexes conflict detection and inhibitory control [28,17]; the P300 complex (EEG19), with P3a (EEG23) reflecting novelty-driven orienting and P3b (EEG24) reflecting evaluation of task-relevant events and working-memory updating, marks context updating and resource allocation [28,17,5]; the Late Positive Potential, LPP (EEG25), is a sustained late positivity associated with affective appraisal and sustained attention [28,17]; SW1/SW2 (EEG27/EEG28) are post-stimulus slow waves capturing prolonged monitoring and effort maintenance [17]; the Mismatch Negativity (MMN; EEG26) is a preattentive auditory negativity elicited by deviations from a repetitive pattern [28,17]; Mean amplitude (EEG30) is the average voltage within a component window, mean bandpower around the ERP (EEG29) and quantifies spectral energy in the ERP time range [17]; under higher CWL, all of these typically show reduced amplitudes and prolonged latencies. P300 and its components p3a and p3b contribute to context updating and the allocation of cognitive resources to respond to stimuli and are crearly associated with CWL [28,5,17]. 2.4 Entropy measurements. Entropy measures quantify the irregularity or complexity of EEG activity and, in CWL research, are used to track how cortical dynamics reorganize under demand. Entropy indices, including wavelet Entropy (WE; EEG31), spectrum entropy (SPEn; EEG33), log-energy entropy (EEG38), Shannon’s entropy (EEG36), Rényi’s entropy (EEG39), Tsallis wavelet entropy (EEG37), approximate entropy (ApEn; EEG35), sample entropy (SampEn; EEG32), peak-to-peak variants (PP-SampEn; EEG34), and Fuzzy entropy (EEG40), capture shifts in signal complexity that can rise with acute visual/cognitive load but often decrease with prolonged time-on-task, transitions toward drowsiness, or sustained fatigue [17]. EEG entropy measures are not yet widely adopted for CWL estimation. Their use remains less standardized than spectral power/ratios or ERPs, and reported effects are heterogeneous across tasks, states (acute load vs. fatigue/drowsiness), analysis windows, and parameter choices[17]. Identification of Physiological Metrics for Estimating CWL 9 3 Eye tracking Eye tracking provides information about CWL through visual attention and oculomotor control. Moment-to-moment changes in gaze position, fixation stability, saccadic dynamics, pupil diameter, and blinks index how visual information is sampled and processed, while also reflecting arousal and effort related to CWL. With proper calibration and artifact control, these signals offer complementary sensitivity to fluctuations in demand that is distinct from electrophysiological and cardiovascular markers. In table 3, we identified 19 metrics of eye tracking, which we grouped into five sub-categories based on the underlying physiological action: (I) pupillometry (EYE01–EYE05), which relate to the size of the pupils; (II) fixations (EYE06–EYE09), which are stationary periods of eye gaze; (III) saccades (EYE10-EYE14), defined as the rapid eye movements that occur between fixations; (IV) blinks (EYE15EYE17) corresponding to the close and open of the eyelids; and (V) higher-level gaze movements (EYE28/EYE19), such as dwell time and AOI revisits. 3.1 Pupillometry The pupil diameter (EYE01) and its derivatives -maximum, mean, percent dilation (EYE02/EYE03/EYE05)- are sensitive to CWL. As task difficulty and working-memory demands rise, mean and pupil diameter increase, and pupil dilation (∆PD%) becomes larger, reflecting heightened central arousal and attentional involvement under relatively constant luminance [8,1,4]. In parallel, the complexity of the pupil entropy (EYE04) can change with sustained effort, capturing fluctuations in moment-to-moment resource allocation; however, direction and magnitude depend on time windowing and control of confounding factors (e.g., illumination or accommodation) [1,10,30]. 3.2 Fixations. The fixations behavior reflects how visual attention is allocated as CWL increases. Higher cognitive load is commonly accompanied by a longer average fixation duration (EYE06) (more processing time per glance), while fixation count (EYE07) and fixation rate (EYE08) can either decrease (fewer, deeper fixations) or increase (more intensive scanning) depending on task structure and information layout [28,20,20,8]. Spatial dispersion or entropy of fixations (fixation spread; EYE09) often grows when search is broad/inefficient, but can shrink when attention is narrowly focused under high CWL; interpretation should therefore be tied to task goals and display complexity [28,10,8,29]. 16 M. García-Romero et al. Category Subcategory Metric ID CWL Relevance Sources Neuroendocrine Plasma/Blood Plasma Cortisol NE01 ↑↑ [28], [7] Plasma Adrenaline NE02 ↑[28] Plasma Noradrenaline NE03 ↑[28], [7] Plasma Adrenocorticotropic Hormone (ACTH) NE04 ↑[28] Plasma Prolactin NE05 ↑[28] Beta-endorphin NE06 ↑[28] Dopamine NE07 ↑[28], [7] Salivary Salivary Cortisol NE08 ↑↑↑ [28], [12], [1], [10], [6], [5], [7] Salivary α-amylase (sAA) NE09 ↑↑ [12], [10], [7] Salivary Immunoglobulin A (sIgA) NE10 ↑↑ [12], [6], [7] Salivary Testosterone NE11 ↑[12], [1] Salivary NO− 3NE12 ↑[10] Metabolic Blood Glucose NE13 ↑[7] Table 5. List of neuroendocrine metrics. Identification of Physiological Metrics for Estimating CWL 17 6 Respiratory and Voice/speech Respiratory and voice/speech signals offer indices of autonomic regulation and cognitive–affective load during task performance. Breathing adaptively modulates its rate, depth, variability, and inspiratory–expiratory timing as demands change, while gas-exchange measures capture ventilatory efficiency and CO2/O2 dynamics. Acoustic features of speech reflect concurrent cognitive effort through prosody (F0and range), intensity/voice quality (e.g., SPL, cepstral and MFCC descriptors), and temporal organization (speech rate). In Table 6, we identified 30 metrics spanning respiratory physiology and acoustic speech behavior. They are organized into five groups: (I) breathing dynamics (RP01–RP11), capturing rate, amplitude/variability, and inspiratory timing/ratios; (II) gas exchange (RP12–RP18), indexing end-tidal CO2, oxygen uptake, CO2production, and respiratory efficiency; (III) prosodic features (V01–V07), summarizing pitch and intensity patterns (e.g., F0, range, normalized variation); (IV) voice intensity and quality (V08–V11), including SPL, cepstral/relative fundamental frequency and MFCC descriptors; and (V) temporal features (V12), centered on speech rate. 6.1 Breathing dynamics. Across controlled tasks, CWL reliably alters breathing patterns. The most consistent change is an increase in respiration rate (RP01) [28,12,1,10], often accompanied by higher minute ventilation (RP05) [14,7]; by contrast, tidal volume (RP04) tends to remain stable or shows mixed effects, indicating that people usually cope with greater mental demand by breathing faster rather than deeper [14,7,15]. In time–based parameters, inspiratory time (RP08) often shortens and derived ratios such as mean inspiratory flow (TV/Ti; RP09) [14,7] and the inspiratory duty cycle (Ti/Ttot; RP10) can increase, reflecting a shift toward quicker, more effortful inspirations [14]. Regarding variability, total breath–to–breath variability in rate (RP03) is not consistently affected, but the correlated (structured) fraction decreases, suggesting less regularity in the breathing rhythm under load [14]. Sigh rate (RP11) shows heterogeneous behavior (increasing or unchanged), likely depending on task design and speaking requirements [14]. Finally, cardiorespiratory coupling indexed by respiratory sinus arrhythmia (RP06) typically diminishes with higher CWL (vagal withdrawal), a pattern frequently reported alongside other autonomic indicators in applied settings [7,4]. 18 M. García-Romero et al. Category Subcategory Metric ID CWL Relevance Sources Respiratory Breathing dynamics Respiration Rate RP01 ↑↑↑ [28], [12], [1], [10], [4], [32], [25], [19], [2], [5], [7], [14], [30], [15] Respiratory Amplitude RP02 ↑↑ [13], [14] Standard Deviation Respiration RP03 ↑[27], [14] Tidal Volume RP04 ↑↑ [7], [14], [15] Minute Ventilation RP05 ↑↑ [7], [14] Respiratory Sinus Arrhythmia (RSA) RP06 ↑↑ [7], [14] Inspiratory/Expiratory time ratio (Ti/Te) RP07 ↑↑ [7], [14] Inspiratory Time (Ti) RP08 ↑[14] Mean Inspiratory Flow Rate (TV/Ti) RP09 ↑[14] Inspiratory Duty Cycle (Ti/Ttot) RP10 ↑[14] Sigh Rate (SR) RP11 ↑[14] Gas exchange End-tidal CO2(petCO2) RP12 ↑↑ [7], [14] Oxygen Consumption (VO2) RP13 ↑↑ [14], [15] Mean Oxygen Consumption (VO2) RP14 ↑[15] Standard Deviation Oxygen Consumption (VO2) RP15 ↑[15] Peaks Oxygen Consumption (VO2) RP16 ↑[15] Carbon Dioxide Production (VCO2) RP17 ↑[14] Respiratory Exchange Ratio (RER) RP18 ↑[14] Voice and Speech Features Prosodic Features Mean Vocal Intensity V01 ↑↑ [10], [4] Median Vocal Intensity V02 ↑[10] Fundamental Frequency (F0) V03 ↑↑ [10], [4], [21] Mean F0 V04 ↑[21] F0 Variability V05 ↑[21] Normalized F0 Variation V06 ↑[21] Pitch Range V07 ↑[21] Voice Intensity and quality Sound Pressure Level (SPL) V08 ↑[21] Cepstral Peak Prominence V09 ↑[21] Relative Fundamental Frequency (RFF) V10 ↑[21] Mel-frequency Cepstral Coefficients (MFCC) V11 ↑[10] Temporal Features Speech Rate V12 ↑↑ [10], [4] Table 6. List of respiratory and voice/speech metrics. Identification of Physiological Metrics for Estimating CWL 19 6.2 Gas exchange. Gas–exchange measures provide convergent evidence that breathing is upregulated relative to metabolic need during demanding cognition. End–tidal CO2 (petCO2; RP12) often falls (consistent with overbreathing) [7,14], while oxygen consumption (VO2; RP13/RP16) and carbon–dioxide production (VCO2; RP17) [15] rise from baseline to task; the respiratory exchange ratio (RER; RP18) shows mixed responses across studies and difficulty levels [14]. 6.3 Prosodic features. Prosodic voice markers show sensitivity, in a exploratory way, to CWL. Effects are heterogeneous and strongly dependent on speaking and dual-task context. Fundamental frequency (F0; V03) and its derivatives—F0mean (V04), F0variability (V05), and normalized F0variation (V06)—often show little change in single-task lab settings, whereas dual-task/free-speech paradigms sometimes report decreases in F0level and dispersion [21,10,4]. Pitch range (V07) and vocal intensity—mean (V01) and median (V02) likewise exhibit mixed, task-contingent shifts. In applied control-room contexts, however, bundles of prosodic features (V01–V07) have contributed to workload prediction within multimodal models despite laboratory heterogeneity [21]. 6.4 Voice Intensity and quality. Measures of intensity and voice quality show task-contingent but informative trends. Sound pressure level (SPL; V08) can decrease in free-speech dual-task paradigms yet increase when concurrent motor demands raise articulatory effort [21]. Cepstral peak prominence (CPP; V09) and relative fundamental frequency (RFF; V10) exhibit sensitivity that depends on task constraints (and, in some reports, age), consistent with changes in vocal effort [21]. Mel-frequency cepstral coefficients (MFCCs; V11) are typically used as features within machine-learning pipelines for workload inference rather than as standalone mechanistic markers [21]. Overall, intensity/quality metrics appear sensitive but are highly taskand population-dependent, warranting careful protocol control and covariate reporting (e.g., age, vocal health). 6.5 Temporal features. Speech rate (V12) can shift with CWL, but the direction and magnitude depend on linguistic demands and dual-task configuration; scoping and field reviews highlight speech rate as a useful predictor when combined with other acoustic or physiological channels, despite single-study heterogeneity [10,4]. 20 M. García-Romero et al. 7 Motion patterns and EMG Somatomotor signals provide a complementary view of workload by indexing how attention modulates motor tone, micro-movements, and posture during seated information work. Subtle changes in facial actions, head pose and movement rates, and postural sway reflect shifts in visual search and control strategies, while surface EMG captures task-related muscle activation (e.g., neck–shoulder co-contraction). In Table 7, we identified 8 somatomotor metrics organized into two categories: (I) motion pattern features (MOT01–MOT03) and (II) electromyography (EMG; EMG01–EMG05). 7.1 Motion patterns. Facial action units/expressions (MOT01) capture micro-movements linked to attention, affect, and stress. They are useful alongside physiological signals but the evidence is heterogeneous, so treat them as supportive rather than standalone CWL indicators [3,32,9]. Head pose/movement (MOT02) reflects visual search versus focused processing, but the direction of change is task/interface dependent, so interpret with task context and concurrent signals [10,4,7,9]. Approximate entropy (ApEn) of postural control (MOT03) generally increases as postural and cognitive demands grow, indicating less regular (more complex) control when confounders are controlled. Collectively, these metrics are informative but not primary markers of CWL in the office environment [18]. 7.2 Electromyography (EMG). EMG amplitude (EMG01) indexes muscle activation that often rises with mental workload, particularly in neck–shoulder muscles under seated, informationprocessing tasks [28,10,26]. In field-like operations (e.g., road traffic control), increasing task demand was accompanied by higher trapezius EMG amplitude [10]. Reviews concur that trapezius EMG is a practical indicator of stress/tension related to workload; nevertheless, EMG is also sensitive to physical load and posture, so ergonomic control and task-matched baselines are essential to avoid misattribution [5,7,10]. Mean and standard deviation summaries as well as extrema (min, max) (EMG02–EMG05) of EMG amplitude are widely used features that improve discrimination of load levels over time windows and help smooth transient artifacts; these features are commonly reported in VR/ergonomic studies but they are not used as indicators of CWL levels [15]. Identification of Physiological Metrics for Estimating CWL 21 Category Metric ID CWL Relevance Sources Motion Patterns Facial Action Units (AUs)/Expressions MOT01 ↑↑ [3], [32], [9] Head Pose/Movement (Rot./Trans. Rates) MOT02 ↑↑ [10], [4], [7], [9] Postural Control ApEn MOT03 ↑[18] Electromyography (EMG) Amplitude EMG EMG01 ↑↑ [28], [10], [26], [23], [13], [5], [7], [30] Mean EMG EMG02 ↑[15] Standard Deviation EMG EMG03 ↑[28], [15] Min EMG EMG04 ↑[15] Max EMG EMG05 ↑[15] Table 7. List of somatomotor (Motion patterns and Electromyography) metrics. 22 M. García-Romero et al. 8 Neuroimaging Neuroimaging tracks hemodynamic and neural responses driven by neurovascular coupling. Load increases typically modulate activity within executive and salience circuits, most prominently prefrontal and cingulate regions, yielding systematic changes in oxygenated/deoxygenated hemoglobin or BOLD signal. Functional Near-Infrared Spectroscopy (fNIRS) enables portable, scalp-level monitoring suitable for semi-naturalistic tasks, while Functional Magnetic Resonance Imaging (fMRI) offers whole-brain coverage and high spatial specificity for mechanism validation. In Table 8, we identified 6 neuroimaging metrics organized into two subcategories: (I) fNIRS (NIMG01–NIMG03), comprising oxyhemoglobin (HbO2), deoxyhemoglobin (HbR), and their differential (HbO2–HbR) as cortical hemodynamic proxies; and (II) fMRI (NIMG04–NIMG07), including the blood-oxygenationlevel–dependent (BOLD) signal and region-specific activations in prefrontal (PFC) and anterior cingulate (ACC). 8.1 Functional Near-Infrared Spectroscopy (fNIRS). fNIRS indexes cortical hemodynamics through changes in oxyhemoglobin (HbO2; NIMG01) and deoxyhemoglobin (HbR; NIMG02), typically summarized either separately or as a differential oxygenation signal (HbO2–HbR; NIMG03) via the Modified Beer–Lambert Law. Increases in CWL are commonly reflected by higher HbO2and lower HbR over the prefrontal cortex (PFC), consistent with greater executive-control demands; the combined oxygenation metric (HbO2– HbR) tends to scale with task difficulty and is widely used as the dependent variable [10,6,5,7,22]. While most workload studies concentrate on the PFC (where hair-related optical attenuation is minimal), parietal contributions have also been reported, and spatial patterns can vary with task class and individual strategy [10,6]. 8.2 Functional Magnetic Resonance Imaging (fMRI). The Blood Oxygenation Level Dependent (BOLD; NIMG04) signal provides a high–spatial-resolution proxy of neural activity [22,32]. With rising workload, fMRI studies typically show increased activation in control-related regions e.g., dorsolateral PFC (NIMG05), and greater modulation of cingulo-opercular and default-mode networks e.g., increased anterior cingulate cortex ACC (NIMG06) engagement and stronger deactivation of posterior cingulate/precuneus, with effects that track task difficulty [22]. In aging and clinical samples, a compensatory pattern is often observed: over-activation of frontal or temporo-parietal areas can help maintain performance at moderate loads, whereas at higher loads under-recruitment or failure to further upregulate BOLD responses accompanies performance decline [32]. Although fMRI establishes region-specific CWL markers, it is intrusive and poorly suited to in-situ office settings. [22,32]. Identification of Physiological Metrics for Estimating CWL 23 Category Subcategory Metric ID CWL Relevance Sources Neuroimaging fNIRS Oxyhemoglobin (HbO2) Concentration NIMG01 ↑↑↑ [10], [6], [5], [7], [22] Deoxygenated Hemoglobin (HbR) concentration NIMG02 ↑↑↑ [10], [6], [5], [7], [22] Cerebral Blood Oxygenation Change (HbO2–HbR) NIMG03 ↑↑↑ [10], [6], [5], [7], [22], [9] fMRI Blood Oxygenation Level Dependent (BOLD) Signal NIMG04 ↑↑ [32], [22] PFC Region Activation NIMG05 ↑↑ [32] ACC Region Activation NIMG06 ↑↑ [32] Table 8. List of neuroimaging (fNIRS and fMRI) metrics. 24 M. García-Romero et al. References 1. Almukhtar, A., Caddick, V., Naik, R., Goble, M., Mylonas, G., Darzi, A., OrihuelaEspina, F., Leff, D.R.: Objective assessment of cognitive workload in surgery: a systematic review. Annals of surgery 281(6), 942–951 (2025) 2. Banuelos-Lozoya, E., Gonzalez-Serna, G., Gonzalez-Franco, N., Fragoso-Diaz, O., Castro-Sanchez, N.: A systematic review for cognitive state-based qoe/ux evaluation. Sensors 21(10), 3439 (2021) 3. Bassi, G., Orso, V., Salcuni, S., Gamberini, L.: Understanding workers’ well-being and cognitive load in human-cobot collaboration: Systematic review. Journal of Medical Internet Research 27, e75658 (2025) 4. Braarud, P.Ø.: Measuring cognitive workload in the nuclear control room: a review. Ergonomics 67(6), 849–865 (2024) 5. Charles, R.L., Nixon, J.: Measuring mental workload using physiological measures: A systematic review. Applied ergonomics 74, 221–232 (2019) 6. Ciptomulyono, U., Dewi, R.S., et al.: Physiological and biochemical measures of mental workload of air traffic controllers: A systematic literature review. In: 2nd South American Conference on Industrial Engineering and Operations Management, IEOM 2021. pp. 2026–2037. IEOM Society (2021) 7. Costa, N., Costa, S., Pereira, E., Arezes, P.M.: Workload measures—recent trends in the driving context. Occupational and Environmental Safety and Health pp. 419–430 (2019) 8. Dai, X., Vitrano, G.: Evaluating mental workload measures in human-robot collaborative assembly. In: 2024 IEEE International Conference on Industrial Engineering and Engineering Management (IEEM). pp. 1059–1063. IEEE (2024) 9. Darvishi, A., Khosravi, H., Sadiq, S., Weber, B.: Neurophysiological measurements in higher education: A systematic literature review. International Journal of Artificial Intelligence in Education 32(2), 413–453 (2022) 10. Diarra, M., Theurel, J., Paty, B.: Systematic review of neurophysiological assessment techniques and metrics for mental workload evaluation in real-world settings. Frontiers in Neuroergonomics 6, 1584736 (2025) 11. Dias, R.D., Ngo-Howard, M.C., Boskovski, M.T., Zenati, M.A., Yule, S.J.: Systematic review of measurement tools to assess surgeons’ intraoperative cognitive workload. Journal of British Surgery 105(5), 491–501 (2018) 12. Gellisch, M., Bablok, M., Brand-Saberi, B., Schäfer, T.: Neurobiological stress markers in educational research: a systematic review of physiological insights in health science education. Trends in Neuroscience and Education 37, 100242 (2024) 13. Glasserman-Morales, L.D., Carlos-Arroyo, M., Ruiz-Ramirez, J.A., AlcantarNieblas, C.: Use of wearable devices in the teaching-learning process: a systematic review of the literature. In: Frontiers in Education. vol. 8, p. 1220688. Frontiers Media SA (2023) 14. Grassmann, M., Vlemincx, E., Von Leupoldt, A., Mittelstädt, J.M., Van den Bergh, O.: Respiratory changes in response to cognitive load: A systematic review. Neural plasticity 2016(1), 8146809 (2016) 15. Halbig, A., Latoschik, M.E.: A systematic review of physiological measurements, factors, methods, and applications in virtual reality. Frontiers in virtual reality 2, 694567 (2021) 16. Hughes, A.M., Hancock, G.M., Marlow, S.L., Stowers, K., Salas, E.: Cardiac measures of cognitive workload: a meta-analysis. Human factors 61(3), 393–414 (2019) Identification of Physiological Metrics for Estimating CWL 25 17. Ismail, L.E., Karwowski, W.: Applications of eeg indices for the quantification of human cognitive performance: A systematic review and bibliometric analysis. Plos one 15(12), e0242857 (2020) 18. Keshner, E.A., Slaboda, J.C., Day, L.L., Darvish, K.: Visual conflict and cognitive load modify postural responses to vibrotactile noise. Journal of neuroengineering and rehabilitation 11(1), 6 (2014) 19. Marchand, C., De Graaf, J.B., Jarrassé, N.: Measuring mental workload in assistive wearable devices: a review. Journal of NeuroEngineering and Rehabilitation 18(1), 160 (2021) 20. Pereira, E., Sigcha, L., Silva, E., Sampaio, A., Costa, N., Costa, N.: Capturing mental workload through physiological sensors in human–robot collaboration: A systematic literature review. Applied Sciences 15(6), 3317 (2025) 21. Pyfrom, M., Lister, J., Anand, S.: Influence of cognitive load on voice production: A scoping review. Journal of Voice (2023) 22. Ranchet, M., Morgan, J.C., Akinwuntan, A.E., Devos, H.: Cognitive workload across the spectrum of cognitive impairments: A systematic review of physiological measures. Neuroscience & Biobehavioral Reviews 80, 516–537 (2017) 23. Raza, M.S., Murtaza, M., Cheng, C.T., Muslam, M.M., Albahlal, B.M.: Systematic review of cognitive impairment in drivers through mental workload using physiological measures of heart rate variability. Frontiers in Computational Neuroscience 18, 1475530 (2024) 24. Simmons, C., Lendrum, R., Perkins, Z., Grier, G., Marsden, M.: A scoping review of cognitive load assessment tools suitable for clinicians performing reboa. Scandinavian Journal of Trauma, Resuscitation and Emergency Medicine 33(1), 121 (2025) 25. Sriranga, A.K., Lu, Q., Birrell, S.: A systematic review of in-vehicle physiological indices and sensor technology for driver mental workload monitoring. Sensors 23(4), 2214 (2023) 26. Suzuki, Y., Wild, F., Scanlon, E.: Measuring cognitive load in augmented reality with physiological methods: A systematic review. Journal of Computer Assisted Learning 40(2), 375–393 (2024) 27. Tamantini, C., Cristofanelli, M.L., Fracasso, F., Umbrico, A., Cortellessa, G., Orlandini, A., Cordella, F.: Physiological sensor technologies in workload estimation: A review. IEEE Sensors Journal (2025) 28. Tao, D., Tan, H., Wang, H., Zhang, X., Qu, X., Zhang, T.: A systematic review of physiological measures of mental workload. International journal of environmental research and public health 16(15), 2716 (2019) 29. Tokuno, J., Carver, T.E., Fried, G.M.: Measurement and management of cognitive load in surgical education: a narrative review. Journal of surgical education 80(2), 208–215 (2023) 30. Vitti, M., Padovano, A., Facchini, F.: A review on cognitive workload for industry 5.0. Computers & Industrial Engineering p. 111350 (2025) 31. Wang, P., Houghton, R., Majumdar, A.: Detecting and predicting pilot mental workload using heart rate variability: a systematic review. Sensors 24(12), 3723 (2024) 32. Yu, R., Schubert, G., Gu, N.: Biometric analysis in design cognition studies: A systematic literature review. Buildings 13(3), 630 (2023)