Will a hard evening session cost me sleep?
In 14,689 active adults, a maximal session ending 2 h before bedtime went with sleep 36 min later and 22 min shorter than a light one; easier ones ending ≥4 h before did not. Observational.
Leota et al. asked whether the timing and the heart-rate strain of a day's exercise go together with that night's sleep and nocturnal autonomic activity.
They took one year (1 September 2021 to 31 August 2022) of data from 14,689 WHOOP subscribers who wore the wrist device on at least 350 days and logged at least 50 exercises ending between 10 h before and 2 h after their habitual sleep onset — 4,084,354 person-nights.
Each exercise was graded by a summated heart-rate-zone score (minutes in each zone × the zone number) into light (<116), moderate (116 to 214), high (214 to 461) or maximal (>461); timing was the end of exercise relative to the person's habitual sleep onset for that calendar month and that type of day (weekday or weekend).
The later a session ended and the higher its strain, the later sleep began, the shorter and less efficient it was, and the higher nocturnal resting heart rate and the lower HRV ran — a graded, dose-response pattern.
Ending 2 h before habitual sleep onset, a maximal session went with sleep beginning 36.0 min later, 22.2 min (5.4%) shorter, a nocturnal heart rate 3.9 beats/min (6.8%) higher and HRV 8.3 units (14.1%) lower than a light session. Light and moderate sessions ending 4 h or more before habitual sleep onset went with sleep onset and duration within a few minutes of rest days.
The abstract says that "regardless of strain" exercise ending 4 h or more before sleep onset was not associated with changes in sleep; the supplementary contrasts show a maximal session ending 4 h before still went with sleep 12.14 min later and 10.43 min shorter than a rest day (both significant), and at 6 h with sleep onset 2.84 min earlier and no difference in duration.
Our reading: a strong, consistent association in a very large sample of active adults, which fits the meta-analysis that found a single evening session harmless (REF-974; REF-1095 found somewhat more waking after evening exercise, at very low certainty) and adds that long or very hard sessions late in the evening are the exception; it cannot show that the late session caused the later night, and nights with late hard sessions (competitions, late gym visits) differ in other ways.
Protocol
Design: retrospective cohort of consumer wearable data, reported to STROBE; approved by the Monash University Human Research Ethics Committee (#32928). Participants consented at registration to research use of de-identified data; location data were not available.
Data window: 1 September 2021 to 31 August 2022.
Selection (CONSORT, Fig. 3): initial sample 20,000 (7,106,256 nights); 1,261 excluded for fewer than 50 exercise days or missing data; nights with exercise outside the window from 10 h before to 2 h after habitual sleep onset (1,418,619 nights) or an invalid exercise type (226,495 nights) removed; 4,050 participants with fewer than 50 exercises inside the window excluded.Final: 14,689 participants, 4,084,354 nights. Eligibility also required age ≥ 18 and at least 350 days of wear (95.6% of possible wear time).
Who (Table S1): 3,844 women (26.17%) and 10,845 men (73.83%); age 37.89 ± 10.81 y (18-44: 73.78%; 44-64: 24.20%; 64-87: 2.02%); 80.39 ± 15.10 kg; 176.39 ± 9.41 cm; resting heart rate 56.41 ± 7.16 beats/min; HRV 59.95 ± 27.31; 122.98 ± 53.74 exercises logged in the analysis window; exercise duration 68.29 ± 55.27 min. Those excluded had a higher resting heart rate (60.52 ± 7.97) and lower HRV (50.49 ± 23.62) (Table S12).
Device: WHOOP 3.0 or 4.0 wrist band (PPG heart rate, accelerometer, three-axis gyroscope). Validation cited by the authors: 99% agreement with ECG for heart rate and 86-89% agreement with polysomnography for two-stage (sleep/wake) categorisation.
Exercise: logged in the app or auto-detected and confirmed by the participant; activities within 60 min of each other merged; activities under 1 min and relaxation activities (yoga, stretching, meditation, massage therapy) excluded.
Strain (summated heart-rate-zone score, SHRZS): minutes in each zone × the zone number (zone 1 = 1, zone 2 = 2 and so on), summed. Zones as % of HRmax: zone 0 < 50%, zone 1 50-60%, zone 2 60-70%, zone 3 70-80%, zone 4 80-90%, zone 5 90-100%. HRmax from 192 − 0.007 × age², set by the participant, or the highest recorded heart rate. The authors note this score differs from the WHOOP Strain shown in the app.
Strain groups (Table 6): light < 116 (834,363 exercises, 46.5%; e.g. 30 min at 65% HRmax; brisk walk, slow jog); moderate 116 to 214 (643,045, 35.8%; e.g. 45 min at 85%; gym class, 5-mile run); high 214 to 461 (274,459, 15.3%; e.g. 90 min at 85%; hockey training, 10-mile run); maximal > 461 (43,437, 2.4%; e.g. 120 min at 85%; hockey match, half marathon). The table notes that moderate strain could also come from 90 min at 65% HRmax.
Timing: primary — end of exercise relative to habitual sleep onset, defined as the person's mean sleep onset for that calendar month, separately for weekdays and weekends (to account for season and social jet lag); secondary — end of exercise relative to that night's actual sleep onset.
Outcomes: sleep onset; sleep duration (total sleep time); sleep quality = sleep percentage, (sleep period − wake after sleep onset) / sleep period; nocturnal resting heart rate and HRV (RMSSD), each a weighted average from sleep onset to offset with more weight on slow-wave sleep, all from the device's algorithms.
Analysis: generalized additive mixed models in R 4.2.0 with a smooth of exercise timing for each strain level; strain as an ordered factor with five levels (no exercise, light, moderate, high, maximal); covariates weekday vs weekend, the prior night's value of the outcome, general fitness (mean daily exercise strain), gender and a smooth of age; random intercept per participant.Estimated marginal means at 2-h steps from −10 h to +2 h compared by consecutive contrasts (none–light, light–moderate, moderate–high, high–maximal) and by each strain level vs no exercise; multivariate-t adjustment; significance p < 0.005. Exploratory stratification by gender, age and BMI; sensitivity analysis excluding the week after each daylight-saving change.Analysis code public (Zenodo).
Funding and interests: no specific funding for this work.Two authors (E.R.C. and D.M.P.) are affiliated with WHOOP, Inc., which paid their salaries; the competing-interests statement says the company did not otherwise play a role in study design, data collection or analysis, decision to publish, or preparation of the manuscript — while the author contributions say "D.M.P. and E.R.C. acquired the data".Other authors list research support or consultancy from sporting bodies and companies, stated as unrelated.
Key Results
Strain level vs rest days (no exercise), by when the session ended relative to habitual sleep onset (supplementary Tables S7, S8, S10, S11; minutes for sleep, beats/min for heart rate, RMSSD units for HRV):
| Ended | Light | Moderate | High | Maximal |
|---|---|---|---|---|
| 6 h before: sleep onset / duration | −2.24 / +0.92 min | −0.81 / +0.21 min | −0.66 / +0.69 min | −2.84 / +0.28 min |
| 4 h before: sleep onset / duration | −5.68 / −1.00 min | −2.03 / −3.35 min | +3.51 / −4.75 min | +12.14 / −10.43 min |
| 2 h before: sleep onset / duration | +3.99 / −8.35 min | +15.14 / −15.03 min | +25.36 / −20.69 min | +39.99 / −30.62 min |
| At habitual sleep onset: sleep onset / duration | +59.00 / −38.42 min | +91.83 / −59.58 min | +98.50 / −67.18 min | +122.96 / −79.20 min |
| 4 h before: nocturnal heart rate / HRV | +0.33 / −0.84 | +1.06 / −2.48 | +1.86 / −4.14 | +3.20 / −6.91 |
| 2 h before: nocturnal heart rate / HRV | +0.45 / −1.18 | +1.76 / −3.81 | +2.90 / −6.41 | +4.33 / −9.48 |
On rest days the estimated sleep onset was 23:06, sleep duration 6.93 h, sleep quality 88.64%, nocturnal heart rate 56.33 beats/min and HRV 60.20 (Tables S2-S6).
Standard errors of these contrasts were about 0.2-1.5 min for sleep before habitual onset; in the supplement PDF every cell of this table is bold (p < 0.005) except four at 6 h before: moderate duration (+0.21), high onset and duration (−0.66, +0.69) and maximal duration (+0.28). The bold rows above are our emphasis, not significance marks.
Main comparison, maximal vs light strain (Results and Tables 1-5): ending 2 h before habitual sleep onset — sleep onset 36.0 min later, sleep duration 5.4% (22.2 min) shorter, sleep quality 0.87 percentage points lower, nocturnal heart rate 6.8% (3.9 beats/min) higher, HRV 14.1% (8.3 units) lower; ending 2 h after habitual sleep onset — sleep duration 13.9% (42.6 min) shorter, sleep quality 5.6 points lower, nocturnal heart rate 15.0% (9.4 beats/min) higher, HRV 32.6% (14.6 units) lower.
A number that does not add up: the Results give an 80.0-min later sleep onset for maximal vs light exercise ending 2 h after habitual sleep onset; the estimated means in Table S2 (03:49 vs 02:39) and the consecutive contrasts in Table 1 (8.58 + 15.43 + 46.51) both give about 70 min (our arithmetic). The 2-h-before figures do add up.
Early sessions: ending 6 h or more before habitual sleep onset, maximal exercise went with earlier sleep onset than lower strain and rest days (−9.12 min at 10 h before) and strain went with longer sleep (maximal +8.50 min at 10 h before).
"4 hours" (abstract; the conclusion recommends ending exercise 4 h or more before sleep onset): "regardless of strain, exercise bouts ending ≥4 hours before sleep onset are not associated with changes in sleep".In the supplementary contrasts every strain level ending 4 h before habitual onset still differed significantly from rest days (all eight sleep-onset and duration cells bold, p < 0.005).The differences were small for light, moderate and high sessions (onset −5.68 to +3.51 min, duration −1.00 to −4.75 min), while a maximal session still went with sleep 12.14 min later and 10.43 min shorter. At 6 h before, a maximal session went with sleep onset 2.84 min earlier (bold) and no difference in duration (+0.28 min, not bold).
Sleep quality (sleep percentage): differences before habitual onset were under 1 percentage point (maximal vs rest day at 2 h before: −0.70); the Results say increased strain went with lower sleep quality when exercise ended 8 h or less before habitual sleep onset.
Nocturnal heart rate and HRV at any time of day: every strain level went with a higher nocturnal heart rate and lower HRV than rest days, also when the session ended 6-10 h before sleep (HRV 1.56-4.52 units lower in the authors' summary; maximal vs rest day at 10 h before: +1.92 beats/min, −4.52 HRV).
Actual sleep onset (secondary analysis, Fig. 2): the same graded pattern for duration, quality, heart rate and HRV when timing was taken from that night's sleep onset.
Subgroups and sensitivity: the associations did not differ by gender, age group (18-43 vs 44-87) or BMI (< 30 vs ≥ 30) on visual comparison of the stratified models (Figs. S1-S6); excluding the week after each daylight-saving change did not change the results.
Mechanisms
Measured: nocturnal resting heart rate rose and HRV fell with both strain and lateness of the session, in step with the sleep outcomes; all strain levels went with a higher heart rate and lower HRV than rest days even when exercise ended 6-10 h before sleep.
Authors' interpretation: strenuous exercise slows the return of parasympathetic (vagal) activity, and because sleep onset and maintenance depend on rising parasympathetic and falling sympathetic activity, a late, high-strain session delays and disturbs sleep; recovery time should be proportionate to the strain.They add behavioural routes: food, caffeine and alcohol around exercise, and bright light in gyms or arenas delaying the circadian clock and suppressing melatonin.
Inference: at 0 h and +2 h the session ends at or after the person's usual bedtime, so part of the later sleep onset there is simply that they were still exercising; the 2-h-before figures are largely free of that.
Inference: the strain score counts minutes in zones, so the "high" group mixes long moderate sessions with shorter hard ones.Under the score's formula 120 min in zone 2 (60-70% HRmax) gives 240, inside the high range (our arithmetic); a long easy evening ride is therefore graded alongside much harder sessions, and the study cannot say whether it weighs on the night as much. REF-161 points the other way for trained athletes: 60 or 120 min below VT1 left HRV back at baseline within 5-10 min in highly trained runners, while threshold or interval work took about 30 min.The study did not separate intensity from duration.
Inference: nocturnal HRV in this study is the night's own value, weighted to slow-wave sleep. A wearable's "morning readiness" built from that night can read low after a late hard session for this reason alone.
Practical Application
Will a hard evening session cost me sleep? In 14,689 active adults, a maximal session ending 2 h before bedtime went with sleep 36 min later and 22 min shorter than a light one; easier ones ending ≥4 h before did not. Observational.
Light evening training went with almost no change: a light session (heart-rate score under 116, e.g. 30 min at 65% HRmax; less than about an hour in zone 2) ending 2 h before habitual sleep onset went with sleep about 4 min later and 8 min shorter than a rest day; ending 4 h before, within about 6 min (B+, observational, active adults).A moderate one, which in this score includes an easy hour or more in zone 2, ending 2 h before went with sleep about 15 min later and 15 min shorter. This agrees with REF-974 (A-, single sessions); REF-1095 (A-) found somewhat more waking in the night after evening exercise, at very low certainty.
Long or very hard sessions are the exception: for high or maximal strain (in this study's terms a heart-rate score above 214: e.g. about 55 min or more at 80-90% HRmax, or close to 2 h at 60-70%; the authors' examples are a 10-mile run, hockey training, a hockey match and a half marathon), finishing 4 h or more before the usual bedtime went with little change, 6 h with none.Ending 2 h before, they went with sleep 25-40 min later and 20-30 min shorter than on a rest day (Tables S7-S8), and more if it ends later still.
When the only slot is late: if a long or hard session cannot end 4 h before bed, the authors' advice is a lighter one in that window (their recommendation, not tested as an intervention). The study measured sleep, not fitness or performance, so it says nothing on whether moving a key session costs more than a shorter night; that trade-off is the athlete's (not from a study; the base's own practice).
Read the next morning's readiness with the evening in mind (inference): a maximal session went with a nocturnal heart rate about 4 beats/min higher and HRV about 8 units lower than a light one when it ended 2 h before habitual sleep onset, and about 9 and 15 when it ended 2 h after.A low HRV or a high resting heart rate on a wearable the morning after such a session is expected after such a session and may be its footprint rather than accumulated fatigue; compare it with a similar session's night, not with a rest day (SYNTH-30).The authors themselves read the lower nocturnal HRV as delayed parasympathetic recovery, so it is no proof that recovery is complete.
Scope (the authors' limitations): the findings are from active adults wearing a wrist device; the authors expect fitter people to recover faster, and no athletes were selected. Strength training is underestimated by a heart-rate score, so a late heavy lifting session is not covered.
Limitations
Observational. Nights after a late hard session are not chosen at random: competitions, late gym visits, social evenings, travel, meals and light come with them.The model adjusted for the weekday or weekend, the prior night, general fitness, gender and age, and allowed each person their own baseline through a random intercept (which does not make the comparison strictly within-person), but no unmeasured cause of both a late session and a late night can be ruled out. The authors name bright light as an uncontrolled confounder.
One consumer device for everything. Sleep, nocturnal heart rate, HRV and the strain score all come from WHOOP's algorithms; two-stage sleep agreed 86-89% with polysomnography and sleep stages less well (REF-364: 50-65%), and nocturnal HRV is weighted to automatically detected slow-wave sleep.
No sleep-onset latency. The device does not know when the person tried to sleep, so "later sleep onset" mixes going to bed later with taking longer to fall asleep (the authors' limitation).
The strain score is heart-rate time. It underestimates brief high-effort work such as strength training (the authors' limitation), uses an age-formula HRmax unless the participant set one, and treats a long easy session and a short hard one as alike.
Who. Physically active WHOOP subscribers, about three in four men, mean age about 38, wearing the device nearly every day; not athletes, and those excluded had a higher resting heart rate and lower HRV. The authors expect the relationship to be stronger in less fit people and do not test fitter ones.
Reporting. The abstract's "regardless of strain … ≥4 hours" overstates the supplementary contrasts for maximal sessions (12 min later, 10 min shorter at 4 h); the 80.0-min figure for exercise ending 2 h after habitual sleep onset does not match Table S2 or Table 1 (about 70 min, our arithmetic).The bold (p < 0.005) marks in the supplement PDF show that at 4 h before habitual onset every strain level, not only maximal, differed significantly from rest days in sleep onset and duration, by under 6 min for light to high strain. The abstract's sentence holds in size, not in significance.
Interests. Two authors are paid by WHOOP; the competing-interests statement says the company had no role in data collection, while the author contributions credit the two WHOOP authors with acquiring the data. No specific funding.
How this entry was checked and changed
Why this level
large observational cohort (n = 14,689), read from the full text and the supplementary tables (open access, CC BY-NC-ND 4.0). The rubric places a large cohort at B+; associations only, no randomisation, the outcomes and the strain measure come from one consumer device, and two authors are employed by its maker
Full text (checked 2026-10-26)
The study population included 14,689 participants logging an average of 122.98 exercises across the 1-year study, 10,845 (73.83%) men (full text)
Engaging in maximal exercise instead of light exercise is associated with a 36.0-minute later sleep onset if the exercise ended 2-h before habitual sleep onset (full text)
Maximal instead of light exercise ending 2-h before habitual sleep onset: a 5.4% (22.2 minute) shorter sleep duration, a 6.8% (3.9 beats/min) increase in RHR and a 14.1% (8.3 unit) decrease in HRV (full text)
Maximal instead of light exercise ending 2-h after habitual sleep onset: an 80.0-minute later sleep onset and a 13.9% (42.6 minute) shorter sleep duration (full text)
Maximal vs no exercise ending 4 h before habitual sleep onset: sleep onset 12.14 min later and sleep duration 10.43 min shorter; at 6 h before: −2.84 and 0.28 min (supplement, Tables S7-S8; −2.84 significant, 0.28 not; full text)
Strain groups: light <116 (46.5%), moderate 116 to 214 (35.8%), high 214 to 461 (15.3%), maximal > 461 (2.4%) (full text)
E.R.C. and D.M.P. are affiliated with WHOOP, Inc., which provided support in the form of salaries; D.M.P. and E.R.C. acquired the data (full text)