Invited review: Noninvasive and minimally invasive approaches for acute and chronic stress assessment in dairy cattle

In modern dairy production, cattle are routinely exposed to a wide range of management-related, environmental, and biological stressors, all of which can compromise welfare, reproductive efficiency, immunocompetence, and lactation performance when the animal's adaptive capacity is exceeded. Accurate and timely stress quantification is therefore essential for welfare monitoring, research design, and evidence-based herd management decisions. However, many conventional approaches involve handling or restraint that can confound most of the parameters being measured, underscoring the need for noninvasive or minimally invasive methods that allow repeated or continuous sampling without materially altering the animal's physiological or behavioral baseline. Heart rate variability reflects beat-to-beat fluctuations in cardiac autonomic control and provides the highest temporal resolution among available stress indicators. Consistent with Porges' polyvagal framework, stress is operationally characterized as parasympathetic withdrawal, making time and frequency domain heart rate variability parameters sensitive early indicators of both acute and subacute stress responses. Validated telemetric Holter systems permit continuous, unrestrained monitoring over multiple days, detecting autonomic shifts that precede overt behavioral or cortisol responses during milking transitions, veterinary procedures, or parturition. Cortisol and its metabolites provide complementary temporal coverage across sampling matrices: salivary cortisol reflects circulating concentrations with an approximately 20-min lag and is suitable for acute event detection; milk cortisol is obtainable at routine milking without additional disturbance; fecal glucocorticoid metabolites integrate hypothalamic-pituitary-adrenal axis activity over a 9- to 15-h window, enabling restraint-free chronic stress profiling; and hair cortisol retrospectively captures cumulative adrenocortical activation over weeks to months. Each matrix introduces specific analytical challenges, including diurnal rhythmicity, matrix-specific immunoreactive metabolite profiles, lactation stage effects, and coat-related confounders. Infrared thermography detects sympathetically mediated surface temperature changes without physical contact, but requires standardized environmental conditions and region-specific calibration to yield valid results. Behavioral indicators represent the most accessible and continuously obtainable dimension of stress assessment. Machine learning algorithms applied to accelerometer, acoustic, and video data now achieve behavioral classification accuracies of 85% to 96%, enabling population-level, real-time welfare monitoring with minimal observer interference. Critically, no single indicator provides a complete or unambiguous stress assessment: different stressors generate distinct physiological and behavioral signatures, substantial inter-individual variation in coping style modulates all measured parameters, and each method carries inherent limitations in specificity, invasiveness, or practical applicability. The central argument of this review is therefore that validated, integrated multi-indicator approaches are not merely preferable but necessary to achieve both the sensitivity and specificity required for reliable welfare assessment in commercial dairy systems. Future research priorities include characterizing chronic-acute stress interactions and translating validated models into practical precision livestock farming tools compatible with on-farm sensor infrastructure.