Effects of exercise intensity on circadian rhythm and glucose regulation in adults with obesity: a randomized crossover trial

Abstract

[Purpose]

Circadian rhythm disruption is associated with metabolic dysfunction and obesity. While exercise can serve as a non-photic zeitgeber, the optimal intensity for evening exercise sessions in adults with obesity remains unclear. This pilot crossover trial compared the acute effects of evening high-intensity interval training (HIIT) versus moderate-intensity interval training (MIIT) on circadian core body temperature rhythm, 24-hour glucose regulation, and sleep quality in adults with obesity.

[Methods]

Nine adult males with obesity (BMI ≥ 25 kg/m², Asia-Pacific criterion) completed a randomised crossover trial of HIIT (~30 min at 90% heart rate reserve [HRR]) and MIIT (~50 min at 60% HRR) on a cycle ergometer between 19:00 and 21:00, separated by a 7–14-day washout. Twenty-four-hour core body temperature (ingestible telemetric capsule), interstitial glucose (FreeStyle Libre 1), and wrist actigraphy were recorded for 24 h before and 24 h after each exercise session under standardised dietary intake. The primary outcome was the change in circadian temperature amplitude (Δ amplitude) derived from a 24-hour fixed-period cosine fit with non-negative amplitude bound; mesor and acrophase were co-primary. Effect sizes (Cohen’s d_z) are reported throughout.

[Results]

Δ amplitude differed in direction between conditions (MIIT −0.01 ± 0.19 °C; HIIT +0.12 ± 0.24 °C) but did not reach conventional statistical significance (paired t(8) = −2.19, p = 0.060, d_z = 0.73). Mesor (Δ contrast p = 0.49, d_z = 0.24) and circular acrophase (Δ contrast p = 0.67, d_z = 0.15) did not differ between conditions, with both protocols producing approximately one-hour phase advances. Twenty-four-hour mean glucose post-exercise trended lower following MIIT (95.9 ± 6.1 mg/dL) than HIIT (98.7 ± 5.5 mg/dL; p = 0.081, d_z = −0.72); coefficient of variation, time in range, and mean amplitude of glycaemic excursions did not differ. Post-exercise sleep efficiency, total sleep time, and wake after sleep onset did not differ between conditions (all p > 0.50; |d_z| ≤ 0.24).

[Conclusion]

In this single-session pilot crossover trial in nine adult males with obesity, evening HIIT and MIIT produced acute changes of similar magnitude in mesor and phase, with a borderline non-significant between-condition difference in amplitude change. Twenty-four-hour mean glucose showed a comparable non-significant trend toward lower values following MIIT, whereas glucose variability and sleep markers did not differ between conditions. Larger work-matched trials are required to determine whether evening exercise intensity produces clinically meaningful differences in circadian or metabolic outcomes in adults with obesity.

INTRODUCTION

Obesity is a global health crisis affecting individuals across all demographics and contributing to numerous chronic conditions including type 2 diabetes, cardiovascular disease, and certain cancers [1,2]. Its etiology is multifactorial, involving complex interactions between genetic susceptibility and environmental factors such as diet and physical inactivity [3,4]. Epidemiological projections estimate that by 2030, approximately 58% of the global adult population will be classified as overweight or obese [3]. In South Korea, where rapid economic development has coincided with lifestyle changes, approximately 40% of adults were classified as overweight or obese as of 2020, with obesity prevalence continuing to rise [5,6].

The circadian rhythm is an endogenous 24-hour cycle that regulates physiological processes including metabolism, hormone secretion, and sleep–wake patterns [7,8]. Disruption of circadian rhythms—increasingly common in modern society due to shift work, social jet lag, and irregular lifestyle patterns—has been linked to metabolic dysfunction, altered appetite regulation, mental health disturbances, and increased obesity risk [9–11]. Core body temperature follows a circadian pattern and serves as a reliable marker of circadian phase, with its rhythm closely regulated by the suprachiasmatic nucleus [12]. The amplitude of circadian temperature rhythm has been proposed as a marker of circadian function, with circadian disruption associated with metabolic dysfunction [13].

Exercise has emerged as a non-photic zeitgeber capable of resynchronising disrupted circadian rhythms [14,15], and the field of chrono-exercise—the study of how the timing of exercise within the 24-hour day interacts with circadian physiology and health outcomes [16]—has attracted growing attention. The timing of exercise differentially affects circadian phase shifts and metabolic responses [17,18]. High-intensity interval training (HIIT) has gained attention for its time efficiency and metabolic benefits, including improved insulin sensitivity and body composition [19–21]. However, evening high-intensity exercise may disrupt sleep architecture and circadian rhythms through elevated sympathetic activity and prolonged core temperature elevation [22,23]. Recent evidence suggests that moderate-intensity interval training (MIIT) may provide comparable metabolic benefits to HIIT [24], while plausibly causing less circadian and sleep disruption given its lower sympathetic and thermogenic load.

Given that modern work schedules often necessitate evening exercise—particularly in countries like South Korea where long working hours are common—determining the optimal exercise intensity for evening sessions is clinically relevant for metabolic health optimisation. Despite growing interest in chrono-exercise, few studies have directly compared how different exercise intensities performed in the evening affect circadian rhythm markers and continuous glucose profiles in individuals with obesity.

Therefore, this study aimed to compare the acute effects of evening HIIT versus MIIT on circadian rhythm markers (core body temperature amplitude, mesor, and acrophase) and 24-hour glucose profiles in adults with obesity. We hypothesised that MIIT would better preserve circadian rhythm amplitude and produce more favourable glucose outcomes compared with HIIT when performed in the evening.

METHODS

Participants

Nine adult males aged 19–59 years with obesity (body mass index [BMI] ≥ 25 kg/m², Asia-Pacific WHO classification) who had not engaged in regular exercise for the preceding 6 months were recruited. The sample size was determined pragmatically based on previous crossover studies examining acute exercise effects on glucose metabolism and circadian markers, in which significant effects were detected with 8–12 participants [18,25]; a formal post-hoc power calculation is provided in Statistical Analysis. Participants were screened with the Morningness–Eveningness Questionnaire (MEQ); only participants scoring within the intermediate chronotype range (42–58 points) [26] were enrolled to minimise confounding from extreme chronotype on circadian responses. Exclusion criteria included orthopaedic conditions limiting exercise capacity, cancer, smoking, psychiatric disorders, use of medications affecting glucose metabolism, eating disorders, asthma, diabetes, or hypertension. Participant characteristics are presented in Table 1.

Table 1.

Participant baseline characteristics (n = 9)

Variable	Mean ± SD
Age (years)	31.2 ± 7.8
Height (cm)	177.5 ± 4.0
Body weight (kg)	89.9 ± 10.6
BMI (kg/m²)	27.6 ± 3.0
Body fat percentage (%)	28.9 ± 4.7
Fasting glucose (mg/dL)	106.6 ± 6.0
Systolic BP (mmHg)	125.4 ± 9.9
Diastolic BP (mmHg)	81.3 ± 7.9
VO2peak (mL/kg/min)	32.4 ± 3.8
Peak heart rate (bpm)	193.1 ± 11.0

BMI, body mass index; BP, blood pressure; VO2peak, peak oxygen consumption. Obesity defined as BMI ≥ 25 kg/m² per Asia-Pacific WHO classification.

Study Design

This study employed a randomised controlled crossover design. Participants were randomly allocated to perform either HIIT or MIIT first using a computer-generated random number sequence. Allocation was performed by a researcher not involved in data collection. Following completion of the first condition, participants underwent a washout period of at least 7 days (up to 14 days depending on individual scheduling) before crossing over to the alternative intervention. This minimum 7-day washout period was determined based on the acute single-session nature of the exercise interventions, allowing sufficient time for physiological parameters to return to baseline. The study was performed in accordance with the Declaration of Helsinki and was approved by the Institutional Review Board of Korea University (KUIRB-2024-0255-01). All participants provided written informed consent prior to participation.

Exercise Interventions

Prior to the intervention, peak oxygen consumption (VO₂peak) was measured using a K-5 portable metabolic system (Cosmed, Rome, Italy) during an incremental cycling test to establish individualised exercise intensities based on heart rate reserve (HRR). Mean VO₂peak was 32.4 ± 3.8 mL/kg/min, and mean peak heart rate was 193.1 ± 11.0 bpm. Exercise sessions were performed on a cycle ergometer between 19:00 and 21:00, with the specific timing adjusted according to each participant’s schedule while maintaining within-individual consistency across conditions. Participants were instructed to finish exercise at least 3 hours before their habitual bedtime to minimise acute effects on sleep initiation.

The HIIT protocol consisted of approximately 30 minutes of interval cycling at 90% HRR (work-to-rest ratio and interval structure individualised to maintain target intensity throughout). The MIIT protocol involved approximately 50 minutes of interval cycling at 60% HRR (similarly individualised). Heart rate was continuously monitored using a Polar H10 chest strap sensor paired with a Polar Vantage monitor (Polar Electro, Kempele, Finland) to ensure target intensity was maintained.

Dietary Control

To minimise dietary confounding on glucose and metabolic outcomes, each participant’s estimated energy requirement (EER) was calculated based on age, body weight, height, and physical activity level using the Harris-Benedict equation with an activity factor of 1.55 (corresponding to a moderately active lifestyle), applied uniformly across participants and conditions. Standardised meals (breakfast, lunch, and dinner) and snacks matching individual EER were provided throughout both intervention periods. Participants were instructed to consume only the provided foods and to maintain consistent meal timing across conditions; dietary compliance was verified through daily food logs.

Outcome Measures

Outcomes were assessed across two 24-hour periods per condition: a 24-h pre-exercise baseline and a 24-h post-exercise period beginning at exercise commencement.

Core Body Temperature. Core body temperature was measured using ingestible telemetric capsules (e-Celsius Performance, Bodycap, Caen, France). For each condition, participants ingested a capsule at approximately 19:00 on the baseline day and a second capsule at approximately 19:00 on the exercise day; temperature was recorded at five-minute intervals. Temperature data were filtered to retain physiological values (35.8–38.2 °C) and fitted to a 24-hour-period cosine model T(t) = M + A·cos[2π/24·(t − φ)], where M is the mesor, A is the amplitude (bounded to [0, 2] °C to ensure positive amplitude and unambiguous acrophase interpretation), and φ is the acrophase (clock time of peak). Fits were performed with scipy.optimize.curve_fit; for each fit the initial phase guess was scanned across 13 values uniformly spaced over the 24-h domain, and the parameter set yielding the lowest residual sum of squares was retained. The fixed 24-h period was selected because the single 24-h measurement window per condition and the entrained nature of the population render free-period estimation unstable; period changes are therefore not evaluated in the present design.

Continuous Glucose Monitoring. Interstitial glucose was measured continuously using the FreeStyle Libre 1 flash glucose monitoring system (Abbott Diabetes Care, Oslo, Norway). The sensor was applied to the posterior upper arm at the start of each intervention period and remained in place throughout. Sensor accuracy was verified against capillary blood glucose obtained via finger-prick sampling at standardised time points within each condition. Twenty-four-hour CGM-derived metrics—mean glucose, coefficient of variation (CV), time in range (TIR; 70–180 mg/dL), and mean amplitude of glycaemic excursions (MAGE)—were calculated over the 24-h post-exercise window. Pre/post-exercise capillary spot glucose measurements (single finger-prick samples obtained immediately before and after the exercise session) were also collected for protocol verification; these single-time-point values are reported separately as a manipulation check and are not directly comparable to 24-h CGM-derived metrics. One participant was unable to wear the CGM sensor during the MIIT condition due to personal circumstances unrelated to the exercise intervention and was therefore excluded from CGM paired analyses, yielding n = 8 paired observations for glucose outcomes.

Sleep Quality. Sleep parameters were assessed using triaxial accelerometry (wGT3X-BT, ActiGraph, Pensacola, FL, USA) worn on the non-dominant wrist throughout both intervention periods. Sleep/wake classification was performed using the Cole-Kripke algorithm [27]. Sleep efficiency (%), total sleep time (TST, min), and wake after sleep onset (WASO, min) were extracted for the post-exercise sleep period. The Pittsburgh Sleep Quality Index (PSQI) and Epworth Sleepiness Scale (ESS) were administered at study entry to characterise habitual sleep quality and confirm the absence of clinical sleep disorders. One participant (EI-006) had unusable actigraphy data during the HIIT post-exercise night due to device failure (zero-valued sleep efficiency, TST, and WASO records); this participant was excluded from sleep paired analyses, yielding n = 8 paired observations.

Statistical Analysis

The primary outcome was the change in circadian temperature amplitude (Δ amplitude, post − pre) between the MIIT and HIIT conditions. Mesor and acrophase were co-primary circadian endpoints. Secondary outcomes were 24-h CGM mean glucose, CV, TIR, MAGE, and sleep efficiency, TST, and WASO. Within-condition pre-to-post change scores (Δ) were calculated for each subject, and between-condition contrasts were performed using paired t-tests on Δ scores; for outcomes assessed only post-exercise (e.g., 24-h CGM mean), post-vs-post paired comparisons were used. Cohen’s d_z (mean of paired differences ÷ SD of paired differences) is reported as the effect-size estimator. Acrophase comparisons used circular distance computation: Δφ = ((φ_post − φ_pre + 12) mod 24) − 12, yielding values in (−12, +12] hours.

To assess potential carryover or period effects given the 7–14-day washout, we computed within-subject sums and differences of paired post-exercise values for the primary outcome and tested for systematic differences between treatment- order subgroups using Welch’s t-tests (Hills-Armitage approach); neither carryover (t = 0.91, p = 0.43) nor period effects (t = −0.45, p = 0.68) reached statistical significance, supporting the assumption of acute, reversible single-session effects. We acknowledge that these tests are themselves underpowered at n = 9. A post-hoc power calculation (G*Power 3.1) confirmed that with n = 9 in a paired design at α = 0.05 (two-tailed), the achieved power was 0.71 to detect the largest observed effect (d_z = 0.73, Δ amplitude) and 0.30–0.50 for typical secondary effect sizes (d_z = 0.2–0.4), indicating that the present sample is underpowered for moderate-to-small effects.

Statistical analyses were performed in Python 3.11 using scipy and statsmodels libraries. An alpha level of 0.05 (two-tailed) was used for hypothesis testing, with effect sizes reported throughout to facilitate interpretation independent of statistical significance. Given the preliminary nature of the trial and multiple secondary outcomes, no formal correction for multiple comparisons was applied; readers are alerted to family-wise error inflation in the secondary-outcome panel and are encouraged to interpret these results as hypothesis-generating.

RESULTS

Participant Characteristics

Nine male participants completed both conditions of the crossover trial (Table 1). The mean age was 31.2 ± 7.8 years, mean height 177.5 ± 4.0 cm, mean body weight 89.9 ± 10.6 kg, and mean BMI 27.6 ± 3.0 kg/m². Body fat percentage averaged 28.9 ± 4.7%, and fasting glucose was 106.6 ± 6.0 mg/dL. Mean blood pressure was 125.4 ± 9.9 / 81.3 ± 7.9 mmHg. All participants were classified as intermediate chronotype by the MEQ and reported no clinical sleep disorders based on PSQI and ESS screening. All participants successfully adhered to the exercise protocols and dietary requirements, and no adverse events were reported. Treatment order was imbalanced (3 participants performed MIIT first; 6 performed HIIT first); carryover and period effects for the primary outcome were not significant (carryover p = 0.43; period p = 0.68), although these tests were themselves underpowered at this sample size. Eight of nine participants contributed complete CGM data for both 24-h post-exercise windows; one participant was unable to wear the CGM sensor during the MIIT condition due to personal circumstances unrelated to the exercise intervention, yielding n = 8 paired observations for glucose outcomes. Sleep paired analyses are based on n = 8 due to actigraphy device failure during HIIT for one participant (EI-006).

Core Body Temperature Rhythms

Cosine fits with 24-h fixed period and non-negative amplitude bound (0–2 °C) were obtained for all 9 × 2 × 2 datasets using multi-start optimisation (13 phase initialisations uniformly spaced over 24 h, lowest residual sum of squares retained; Table 2, Figure 1).

Table 2.

Core body temperature circadian rhythm parameters

Parameter	MIIT Δ	HIIT Δ	Between-condition difference (MIIT − HIIT)	t	p	Cohen’s d_z
Amplitude (°C)	−0.01 ± 0.19	+0.12 ± 0.24	−0.13	−2.19	0.060	−0.73
Mesor (°C)	+0.19 ± 0.38	+0.12 ± 0.20	+0.07	+0.72	0.491	+0.24
Acrophase (h, circular)	−0.91 ± 1.41	−1.04 ± 1.74	+0.13	+0.45	0.667	+0.15

Δ, change from pre- to post-exercise; values are mean ± SD. Cosine fit performed on 24-h pre- and post-exercise core body temperature data with fixed 24-h period and non-negative amplitude bound (0–2 °C); multi-start optimisation with 13 phase initialisations uniformly spaced over 24 h. Acrophase comparisons used circular distance computation: Δφ = ((φ_post − φ_pre + 12) mod 24) − 12, yielding values in (−12, +12] hours. d_z, Cohen’s d for paired design (mean of paired differences ÷ SD of paired differences). MIIT, moderate-intensity interval training; HIIT, high-intensity interval training. n = 9 paired. No comparison reached α = 0.05.

Figure 1. Representative example of cosine fits applied to core body temperature data for participant EI-009.

(A) HIIT condition: pre-exercise (blue) and post-exercise (red) raw observations with corresponding cosine fits. (B) MIIT condition: pre-exercise (blue) and post-exercise (red) raw observations with cosine fits. Dashed horizontal lines indicate fitted mesor values; vertical annotations indicate fitted amplitude and acrophase. Cosine fits shown are obtained under the non-negative-amplitude parametrisation used for all quantitative analyses (see Methods).

Amplitude. Following MIIT, fitted amplitude was essentially unchanged (pre 0.66 ± 0.28 °C, post 0.67 ± 0.14 °C; Δ = −0.01 ± 0.19 °C); following HIIT, fitted amplitude increased (pre 0.47 ± 0.18 °C, post 0.61 ± 0.20 °C; Δ = +0.12 ± 0.24 °C). The between-condition Δ contrast did not reach statistical significance (paired t(8) = −2.19, p = 0.060, d_z = 0.73). The directional increase under HIIT may partly reflect contamination of the 24-h fit by the immediate post-exercise thermogenic peak (descriptive 0–1 h post-exercise temperature: HIIT 37.6 ± 0.2 °C, MIIT 37.4 ± 0.3 °C); cosine modelling of the nocturnal-only window is precluded by the limited data span.

Mesor. Both conditions produced small post-exercise increases in mesor (MIIT Δ = +0.19 ± 0.38 °C; HIIT Δ = +0.12 ± 0.20 °C; between-condition Δ contrast t(8) = 0.72, p = 0.49, d_z = 0.24).

Acrophase. Acrophase advanced by approximately one hour in both conditions (MIIT Δ = −0.91 ± 1.41 h; HIIT Δ = −1.04 ± 1.74 h; circular between-condition Δ contrast t(8) = 0.45, p = 0.67, d_z = 0.15).

Glucose Regulation

Twenty-four-hour CGM-derived metrics from the post-exercise window are summarised in Table 3 and Figure 2 (n = 8 paired). Twenty-four-hour mean glucose post-exercise trended lower following MIIT (95.9 ± 6.1 mg/dL) than HIIT (98.7 ± 5.5 mg/dL), without reaching statistical significance (paired t(7) = −2.04, p = 0.081, d_z = −0.72). Pre-exercise 24-h CGM means were 101.2 ± 10.1 mg/dL (MIIT) and 101.7 ± 7.3 mg/dL (HIIT), differing by approximately 0.5 mg/dL—well within day-to-day intra-individual variability characteristic of single-session CGM.

Table 3.

Secondary outcomes: 24-h continuous glucose monitoring and sleep quality (post-exercise)

Parameter	MIIT post	HIIT post	t	p	Cohen’s d_z
Glucose regulation (24-h post-exercise CGM, n = 8 paired)
Mean glucose (mg/dL)	95.9 ± 6.1	98.7 ± 5.5	−2.04	0.081	−0.72
CV (%)	18.2 ± 6.6	19.3 ± 8.6	−0.70	0.509	−0.25
TIR 70–180 (%)	96.4 ± 5.3	95.1 ± 10.7	+0.57	0.586	+0.20
MAGE (mg/dL)	27.3 ± 8.2	27.4 ± 7.9	−0.03	0.979	−0.01
Sleep quality (post-exercise night, n = 8 paired)
Sleep efficiency (%)	90.0 ± 14.1	87.5 ± 13.7	+0.48	0.648	+0.17
Total sleep time (min)	351 ± 76	321 ± 113	+0.67	0.526	+0.24
WASO (min)	31.8 ± 41.0	42.3 ± 44.8	−0.62	0.557	−0.22

Values are mean ± SD. Between-condition contrasts performed by paired t-tests on post-exercise values (post-vs-post paired comparison) within each subject. CV, coefficient of variation; TIR, time in range; MAGE, mean amplitude of glycaemic excursions; WASO, wake after sleep onset. d_z, Cohen’s d for paired design. One participant was unable to wear the CGM sensor during the MIIT condition due to personal circumstances unrelated to the exercise intervention (yielding n = 8 paired for glucose); one participant (EI-006) was excluded from sleep paired analysis due to actigraphy device failure during HIIT (yielding n = 8 paired for sleep). No comparison reached α = 0.05.

Figure 2. Comparison of 24-hour post-exercise continuous glucose monitoring profiles following HIIT (red) and MIIT (blue).

Glucose values (mg/dL) are plotted against time (hours) from exercise commencement. The HIIT condition shows more pronounced postprandial glucose excursions, particularly at meal times (approximately 10 and 17 hours post-exercise).

Glucose variability (CV: MIIT 18.2 ± 6.6%, HIIT 19.3 ± 8.6%; t = −0.70, p = 0.51), TIR 70–180 mg/dL (96.4 ± 5.3% vs 95.1 ± 10.7%; p = 0.59), and MAGE (27.3 ± 8.2 vs 27.4 ± 7.9 mg/dL; p = 0.98) did not differ between conditions.

Pre/post-exercise capillary spot glucose measurements (single finger-prick samples obtained immediately before and after the exercise session) yielded mean pre-to-post differences of approximately −22 mg/dL (MIIT) and −16 mg/dL (HIIT); these single-time-point values reflect acute exercise-induced glucose decline at the moment of measurement and are reported here only as a manipulation check, not for between-condition inference.

Sleep Quality

Post-exercise sleep parameters (n = 8 paired, after exclusion of EI-006 due to actigraphy device failure during HIIT) did not differ significantly between conditions: sleep efficiency MIIT 90.0 ± 14.1% vs HIIT 87.5 ± 13.7% (paired t(7) = 0.48, p = 0.65, d_z = 0.17); total sleep time 351 ± 76 vs 321 ± 113 min (t = 0.67, p = 0.53, d_z = 0.24); WASO 32 ± 41 vs 42 ± 45 min (t = −0.62, p = 0.56, d_z = −0.22). All effect sizes were small and no comparison reached statistical significance. Baseline night-to-night variability was substantial (e.g., pre-MIIT efficiency range 0–100%), reflecting the limitations of single-session sleep assessment.

DISCUSSION

This preliminary single-session randomised crossover trial compared the acute effects of evening HIIT versus MIIT on circadian core body temperature parameters, 24-h glucose regulation, and sleep quality in adults with obesity. None of the primary or secondary outcomes reached conventional statistical significance with the present sample of nine participants. The largest observed effect was the between-condition difference in Δ circadian temperature amplitude (p = 0.060, d_z = 0.73), in which HIIT was associated with a directional increase in fitted amplitude and MIIT with no change. Twenty-four-hour CGM mean glucose trended lower following MIIT (p = 0.081, d_z = −0.72), and sleep markers did not differ between conditions.

Contrary to our a priori hypothesis that HIIT would flatten the circadian temperature amplitude relative to MIIT, the present analysis suggests that fitted amplitude increased under HIIT. Physiologically, this directional increase may reflect the greater acute thermogenic and sympathetic response evoked by high-intensity exercise, whereby elevated post-exercise core temperature and catecholamine activity transiently amplify the temperature waveform. This interpretation should nonetheless be made with caution, because the apparent increase may in part be a methodological artefact: the 24-h cosine fit can be contaminated by the immediate post-exercise thermogenic peak. The 0–1 h post-exercise core temperature was descriptively higher under HIIT (37.6 ± 0.2 °C) than MIIT (37.4 ± 0.3 °C), and inclusion of this transient daytime spike within the 24-h fitting window can inflate the fitted peak-to-trough amplitude without reflecting a true change in the underlying circadian rhythm. Restricting the fit to a nocturnal-only window would isolate the circadian component but is precluded here by the limited data span; multi-day continuous measurement is required to evaluate true circadian period and amplitude changes [12]. The substantially larger between-individual variability of post-HIIT acrophase (SD 1.74 h) relative to post-MIIT (SD 1.41 h) suggests heterogeneous individual responses to high-intensity evening exercise, plausibly reflecting between-individual variability in sympathetic recovery, cortisol clearance, or heat dissipation capacity.

Several physiological pathways may underlie the differential acute effects of evening HIIT versus MIIT. First, high-intensity exercise produces a more pronounced sympathetic response and more sustained elevation of plasma catecholamines and cortisol than moderate-intensity exercise, with cortisol exhibiting an intensity-dependent threshold effect [28,29]. The 3-hour buffer between exercise completion and habitual bedtime adopted in this protocol may not be sufficient to permit full attenuation of these neuroendocrine responses; residual sympathetic and glucocorticoid activity at sleep onset can attenuate nocturnal vasodilation and the typical decline in core body temperature that accompanies sleep entry [22,23]. Second, glucocorticoids interact with central and peripheral circadian regulation [8,30]; sustained post-exercise glucocorticoid elevation could plausibly attenuate the amplitude of physiological rhythms. Third, the directional trend toward lower 24-h CGM mean glucose following MIIT is consistent with established evidence that moderate-intensity exercise enhances insulin-mediated glucose disposal without provoking the strong counter-regulatory hormone response that characterises high-intensity exercise [31,32]. The present design did not directly measure cortisol, catecholamines, or clock-gene expression; these mechanisms are therefore proposed as testable hypotheses for future work rather than as inferences from the present data.

The small absolute magnitude of between-condition differences in 24-h CGM mean (≈ 2.8 mg/dL) is comparable to the manufacturer-specified measurement uncertainty of the FreeStyle Libre 1 flash glucose monitoring system, suggesting that this metric did not reliably differentiate the conditions in our sample. Sleep markers showed no consistent intensity-dependent pattern, consistent with the literature suggesting that the 3-h exercise–bedtime buffer may be sufficient to mitigate evening high-intensity exercise effects on sleep continuity in non-clinical populations [22,23].

Several limitations should be considered. First, with n = 8–9 paired observations across outcomes, the present study was underpowered to detect moderate-to-small effects (achieved power 0.30–0.50 for d_z = 0.2–0.4); findings should be interpreted as hypothesis-generating rather than confirmatory. Second, the HIIT (~30 min) and MIIT (~50 min) protocols differed simultaneously in intensity and total duration, so the present data cannot disentangle intensity-specific effects from those attributable to volume or accumulated mechanical/metabolic load; the directional differences observed could plausibly reflect intensity-dependent catecholamine release, volume-dependent insulin-sensitising effects, or some combination. Future trials employing energy-matched protocols are required to resolve this confound. Third, only male participants were included; sex differences in circadian rhythms [33] limit generalisability. Fourth, the single-session design does not capture potential adaptations to repeated training, and the absence of a non-exercise control condition limits separation of exercise-specific effects from day-to-day variability. Fifth, the single 24-h measurement window per condition prevents formal evaluation of changes in circadian period (τ); multi-day continuous measurement is required for this. Sixth, the FreeStyle Libre 1 is a flash glucose monitoring system whose absolute accuracy in adults with obesity may be modestly attenuated by increased subcutaneous adipose depth and altered local tissue perfusion; absolute accuracy of derived metrics, particularly extreme-event detection, should be interpreted with appropriate caution. Seventh, any pre-post change observed in either condition is subject to regression to the mean given day-to-day intra-individual variability; for this reason between-condition inference for outcomes assessed only post-exercise was based on direct post-vs-post paired comparison, robust to baseline drift. Eighth, obesity was defined using the Asia-Pacific BMI threshold of ≥ 25 kg/m², which captures cardiometabolic risk at lower BMI relative to the WHO global threshold of ≥ 30 kg/m²; while appropriate for Korean and East Asian populations, generalisation to populations defined by the global threshold should be made with caution. Finally, although carryover and period tests were non-significant, these tests were themselves underpowered at this sample size, and the unequal treatment-order allocation (3 MIIT-first vs. 6 HIIT-first) represents a potential source of period-effect bias that could not be adequately controlled in a sample of this size; balanced randomisation should be prioritised in future confirmatory trials.

Future trials should employ energy-matched HIIT and MIIT protocols, include both sexes, employ multi-day continuous measurement to permit formal evaluation of circadian period changes, directly measure cortisol, catecholamines, and where feasible peripheral clock-gene expression to test the mechanistic hypotheses outlined here, increase sample size to detect moderate effect sizes in glucose variability and sleep outcomes, and compare acute and chronic effects through repeated-measures longitudinal designs.

Despite the absence of conventional statistical significance for any primary or secondary outcome, this pilot trial contributes preliminary evidence in three respects. First, the between-condition effect-size estimates for circadian temperature amplitude (d_z = 0.73) and 24-h continuous glucose monitoring mean (d_z = −0.72) provide quantitative inputs for power calculation in larger work-matched confirmatory trials; with these effect sizes, approximately 14–16 paired participants would be required to achieve 80% power at α = 0.05. Second, the directional contamination of 24-h fixed-period cosine fits by the immediate post-exercise thermogenic spike represents a methodological consideration that future evening-exercise chrono-physiology studies should address through nocturnal-only modelling or multi-day continuous measurement. Third, the integrated assessment of core body temperature (ingestible telemetric capsule), interstitial glucose (continuous glucose monitoring), and sleep (wrist actigraphy) under standardised dietary conditions is shown to be feasible and well-tolerated in adult Korean males with obesity, supporting the use of this multimodal protocol in subsequent chrono-exercise research.

CONCLUSION

In this single-session pilot crossover trial in nine adult males with obesity, evening HIIT and MIIT produced acute changes of similar magnitude in core body temperature mesor and approximately one-hour phase advances, with a borderline non-significant between-condition difference in amplitude change. Twenty-four-hour mean glucose showed a non-significant trend toward lower values following MIIT, whereas glucose variability and sleep markers did not differ significantly between conditions. Larger work-matched trials are required to determine whether evening exercise intensity produces clinically meaningful differences in circadian or metabolic outcomes in adults with obesity.