Your Heart’s Rhythm Reveals When Stress Helps or Hurts Your Performance

When a pilot fights turbulence, a surgeon operates through a laparoscope, or an air traffic controller tracks a dozen aircraft at once, the same question haunts safety scientists: how can we know, in real time, whether the person at the controls is coping, compensating, or quietly coming apart? A new study from the University of Chieti-Pescara suggests the answer may not lie in a single physiological number at all, but in recognizing which of several distinct autonomic states the heart happens to be in at any given moment. The research, published in Frontiers in Neuroergonomics, shows that the relationship between heart rate variability and task performance changes shape depending on both the level of physiological activation and the kind of pressure the operator is under.

Heart rate variability, or HRV, refers to the tiny beat-to-beat fluctuations in the timing of heartbeats, and it has long been a favorite window into the autonomic nervous system. When the parasympathetic, or vagal, branch dominates, the heart’s rhythm becomes more variable; when the sympathetic branch takes over under stress or effort, that variability tends to shrink. Decades of research have produced a whole toolbox of HRV metrics: time-domain measures like SDNN, which captures overall variability, and RMSSD, which reflects short-term vagal modulation; frequency-domain measures like high-frequency power, tied to vagal tone, and low-frequency power, which mixes autonomic and baroreflex influences; and non-linear complexity measures such as Sample Entropy, Fuzzy Entropy, Permutation Entropy, and the correlation dimension D2, which quantify the irregularity and fractal structure of the heartbeat series. The problem is that studies have often treated these metrics as if one of them should be the universal performance marker, and the aviation literature in particular has been frustratingly inconsistent about which one wins.

The Italian team, led by Ludovica Di Pompeo and Arcangelo Merla, suspected that the inconsistency itself was the clue. Rather than assuming a single linear relationship between physiological activation and performance across the whole arousal spectrum, they proposed that the heart operates in distinct regimes, and that different HRV indices might carry different predictive value in each. Their framework draws on the century-old Yerkes-Dodson law, which describes an inverted U-shaped relationship between arousal and performance: too little activation and attention drifts, too much and it collapses, with an optimal middle zone in between. If that is true, then pooling data across all arousal levels might wash out the very signals that matter most.

To test the idea, forty-two healthy young adults, averaging twenty-five years old, flew a demanding multi-task flight simulation built on the AeroStim platform, an evolution of NASA’s Multi-Attribute Task Battery II. Participants were randomly assigned to one of two conditions. In the Workload-Only condition, twenty participants performed two primary tasks: a Tracking task requiring them to keep a moving crosshair inside a target area with a joystick, and a System Monitoring task requiring continuous vigilance for visual anomalies. In the Stress-Oriented condition, twenty-two participants faced the same primary tasks but with additional concurrent secondary tasks layered on top, including Stroop interference tests, radio communications requiring selective attention, and resource management puzzles, all designed to introduce divided attention, time pressure, and stress. An energy bar that drained irreversibly with every error added a relentless countdown, and its depletion could terminate an exercise early.

Throughout the roughly thirty-minute session, every heartbeat was captured by an EqVital sensorized T-shirt, a wearable garment with textile electrodes that records a two-lead ECG at 256 Hz. The researchers extracted a comprehensive battery of linear and non-linear HRV indices from each exercise epoch, then normalized every metric within each participant using Min-Max scaling so that individual differences in absolute heart rhythm would not swamp the analysis. The crucial methodological move came next: instead of dividing the data by task difficulty, they fed the normalized SDNN values into a Gaussian Mixture Model, which partitioned each person’s own physiological range into three relative activation zones labeled Easy, Medium, and Hard. The three-regime structure was specified in advance on theoretical grounds, but the model provided empirical support, and sensitivity checks with four different normalization procedures showed 98.4 to 100 percent agreement in classification, indicating the tripartite split was robust rather than an artifact of one particular scaling choice.

What emerged was a strikingly context-dependent picture. In the Workload-Only condition, the HRV indices that correlated with performance shifted systematically across regimes: low-frequency power was the only nominal correlate in the Easy range, global variability measures like SDNN and SD2 joined LF in the Medium range, and Sample Entropy appeared in the Hard range. In the Stress-Oriented condition, the pattern changed again. The Medium range now involved vagal indices such as RMSSD, pNN50, and SD1, along with the non-linear complexity measures Fuzzy Entropy and Permutation Entropy, while the Hard range was dominated by vagal and LF-related measures tied to error rates. In other words, when stressors overlapped with workload, the physiological signature of good performance migrated toward parasympathetic and complexity-based metrics, a pattern consistent with the idea that stress reshapes the autonomic processes supporting task execution rather than simply degrading them.

The behavioral results added their own twist. Performance across activation ranges only partially supported the classic inverted-U. In the Workload-Only condition, omission rates on the monitoring task were significantly lower in the Medium activation range than in the Easy range, offering partial support for an intermediate-activation advantage. But no such optimum appeared in the Stress-Oriented condition, where performance was statistically flat across regimes, suggesting that compensatory effort and motivation may have buffered behavior even as the underlying physiological cost climbed. The most robust finding in the entire study was a speed-accuracy trade-off between conditions: Workload-Only participants responded faster but made more errors, while Stress-Oriented participants slowed down and became more accurate, a pattern that survived false-discovery-rate correction and is consistent with the risk-averse response strategies documented under stress.

The authors are careful, and appropriately so, about how far these conclusions can be pushed. None of the HRV-performance correlations survived false-discovery-rate correction, so all of them must be treated as exploratory, and the per-range samples were small, with only fourteen to nineteen aggregated observations and correspondingly wide confidence intervals. There is also a methodological circularity to contend with: because SDNN was used both to define the activation regimes and as a predictor within them, some apparent range-specific associations may reflect statistical range restriction rather than genuine physiology. A sensitivity analysis confirmed that stratifying by task difficulty instead of SDNN produced weaker or even opposite correlations for the same metrics in the Workload-Only condition. Early termination by the energy bar shortened roughly a third of the exercise epochs, and a window-length sensitivity analysis showed that frequency-domain measures like LF and HF were noticeably less stable than time-domain and non-linear indices when the recording window shrank to 180 seconds, underscoring the need for standardized recording durations in future work.

Those caveats do not diminish the study’s central contribution, which is conceptual as much as statistical. The findings argue that there is no single universal HRV marker of performance, and that the physiological correlates of task execution depend jointly on the activation state and the operational context. This resonates with recent reviews of pilot workload detection, which report that different HRV domains show sensitivity across tasks and flight conditions but that no single metric consistently dominates, and with real-flight evidence showing that workload classifiers perform well within individual pilots but degrade substantially when generalized across them. The regime-based framework offers a way to reconcile that heterogeneity: instead of asking which HRV index predicts performance, adaptive systems could first identify which activation regime an operator is in, then apply the index profile appropriate to that regime and context.

The practical implications reach well beyond the laboratory. Adaptive human-machine systems, from cockpit aids that reallocate attention to warning systems that detect overload before errors accumulate, need workload models that can distinguish not only performance degradation but also the physiological cost of maintaining performance, since two operators can look behaviorally identical while burning very different autonomic reserves. A regime-based, multimodal HRV monitor, calibrated briefly to each individual and then classifying activation states in real time, could provide exactly that capability. Before that vision becomes reality, the authors note, the field needs standardized recording durations, independent autonomic markers to validate the regime boundaries, respiratory monitoring to disentangle breathing influences from true autonomic shifts, and validation across larger, more diverse samples and genuine operational environments. But the direction is clear: the heart does not speak a single language of stress and workload, and learning to listen in the right regime at the right moment may be the key to keeping humans safe in the most demanding jobs on Earth.

Subject of Research: Heart rate variability and autonomic regulation of performance under cognitive workload and stress

Article Title: Physiological activation regimes reveal distinct autonomic mechanisms underlying operational performance under cognitive workload and stress

Article References: Di Pompeo, L., Tritto, M., Romano, F., Perpetuni, D., Cardone, D., & Merla, A. (2026). Physiological activation regimes reveal distinct autonomic mechanisms underlying operational performance under cognitive workload and stress. Frontiers in Neuroergonomics, 7, Article 1946611. https://doi.org/10.3389/fnrgo.2026.1946611

Image Credits: AI Generated

DOI: 10.3389/fnrgo.2026.1946611

Keywords: heart rate variability, autonomic nervous system, cognitive workload, stress, flight simulation, neuroergonomics, performance monitoring, wearable ECG, Yerkes-Dodson law, adaptive human-machine systems, Gaussian mixture model, speed-accuracy trade-off