How much VO2max do I lose if I stop training?
21 small athlete studies: about 4% after up to a month off and 9% after longer (authors' averages), no clear further fall past 90 days. Uncontrolled studies, reporting errors.
Zheng et al. searched Web of Science, SPORTDiscus, PubMed and Scopus (English, to 31 March 2021) for studies that reported VO2max in athletes before and after a defined period without training. Of 3315 records, 21 studies were included; breaks ran from 10 to 730 days. Each study's before-after change was expressed as a standardised mean difference (Hedges' g).
After 30 days or less, VO2max fell by SMD −0.62 (95% CI −0.94 to −0.31; I² 35.3%), and after longer breaks by SMD −1.42 (−1.99 to −0.84; I² about 77%); the long-term fall was larger (Q = 6.5, p = 0.01). Within long breaks, there was no evidence of a difference between 30-90 days and over 90 days (Q = 0.54, p = 0.46).
In long breaks the loss was larger in athletes training at least 5 times a week before the break and with complete rather than partial cessation, and borderline larger in under-20s (three studies, p = 0.05, CI crossing zero); in short breaks there was no evidence of a difference for any of these moderators.
The authors translate the pooled effects into average falls of 3.93% (short) and 9.43% (long) without saying how, and conclude that athletes should keep doing some exercise during a break.
Our reading: the direction and the levelling-off agree with REF-1133, REF-377 and REF-374; the magnitudes are rough, and the paper's errors (see Key Results) mean its subgroup findings should not carry weight on their own.
Protocol
Included: 21 studies; n = 350 by our sum of the sample sizes in Table 1 (the review gives no total; the two Koundourakis 2014 papers come from one group with the same design and may share players — our inference). Sample sizes 5 to 67; two studies of women only.
Search and selection: four databases, English, before 31 March 2021; 3315 records, 1865 duplicates removed, 1450 screened, 179 full texts assessed, 158 excluded (among them 32 whose full text could not be obtained and 51 in non-athletes). Eligible: a specific detraining duration and VO2max values before and after, in athletes of any age, sex, event or level. Two independent reviewers with a third for disagreements.
Coding: training status (higher: at least 5 sessions a week; lower: fewer), duration, sex, age (under 20 or 20 and over), and format — complete cessation (CDT) or partial cessation (PDT).
Quality: STROBE checklist for cohort studies, each of 22 items scored 0-2; included studies scored 28 to 39 of a possible 44 (our arithmetic for the maximum).
Analysis: Hedges' g per study; short-term 30 days or less vs long-term over 30 days, and long-term split into 30-90 and over 90 days; a fixed model, with random effects "if statistical heterogeneity was shown (I2 < 40%)" as printed; extended Egger test; R meta package.
Funding and interests: no funding statement; no conflicts of interest declared.
Key Results
| Comparison | k | SMD (95% CI) | Heterogeneity |
|---|---|---|---|
| Short-term, 30 days or less | 14 (Table 4) | −0.62 (−0.94 to −0.31) | I² 35.3%; Egger p = 0.335 |
| Long-term, over 30 days | not stated | −1.42 (−1.99 to −0.84) | I² 76.3% (abstract) or 77% (Results); Egger p < 0.01 |
| Short vs long | — | Q = 6.5, p = 0.01 | — |
| Long: 30-90 days / over 90 days | 12 / 7 | −1.6 (−2.47 to −0.74) / −1.20 (−2.13 to −0.28) | Q = 0.54, p = 0.46 |
| Long: higher / lower training status | 10 / 4 | −1.91 (−2.57 to −1.25) / −0.85 (−1.83 to 0.12) | Q = 4.24, p = 0.03 |
| Long: under 20 / 20 and over | 3 / 16 | −2.81 (−6.32 to 0.69) / −1.20 (−1.76 to −0.64) | Q = 5.9, p = 0.05 |
| Long: complete / partial cessation | 16 / 4 | −1.69 (−2.41 to −0.96) / −0.65 (−1.42 to 0.11) | Q = 6.23, p = 0.01 |
| Short: higher / lower training status | 7 / 7 | −0.76 (−1.10 to −0.41) / −0.46 (−0.75 to −0.18) | Q = 1.45, p = 0.22 |
| Short: complete / partial cessation | 11 / 3 | −0.54 (−0.82 to −0.26) / −0.65 (−1.00 to −0.30) | Q = 0.36, p = 0.55 |
In percent (authors' figures, method not described): average falls of 3.93% after short-term and 9.43% after long-term cessation; with partial cessation, falls of 0.93% to 4.38% (the same range is given for long and for short breaks) against falls of up to 11.12% (apparently complete cessation in long breaks; the sentence is garbled); with complete cessation in short breaks, from a 21.28% fall to a 0.84% rise.
Numbers that do not add up: (1) three subgroup confidence intervals that cross zero (lower training status, under-20s and partial cessation in long breaks) are printed with p < 0.001; (2) the long-term subgroups add up to different numbers of comparisons (19 by duration, 14 by training status, 19 by age, 20 by format); (3) long-term I² is 76.3% in the abstract and 77% in the Results; (4) the age tests (Q = 5.9 with p = 0.05; Q = 0.27 with p = 0.87) match two degrees of freedom although there are two age groups — with one degree of freedom they give p of about 0.015 and 0.60 (our arithmetic); (5) the discussion states there was "no significant difference" between complete and partial cessation in the paragraph about long breaks, where Table 3 gives Q = 6.23, p = 0.01; (6) "higher trained-state V̇O2max" in the abstract is defined in Table 1 by training frequency, not VO2max; (7) short-term is "≤30 days" in the Methods and "<30 days" in the Results; (8) the partial-cessation range of 0.93% to 4.38% is given word for word for both long and short breaks, and the largest single fall quoted for short breaks (21.28%) exceeds the largest for long breaks (11.12%); (9) Table 1's coding does not reproduce the subgroup k values (short breaks: 8 higher- and 3 lower-status studies in Table 1 against 7 and 7 in Table 4; long breaks: 5 lower-status studies against 4 in Table 3); (10) the abstract's conclusion says there was "no significant change in V̇O2max when the duration of training cessation was more than 30 days", contradicting its own significant long-term fall.
Citation errors: the quality table labels studies inconsistently (by first name, as with Brian, Eileen and Klavs, or by second author, as with Androulakis and Travlos), and one label, Yi-hung, matches no reference; Table 1's "Melchiorri et al. (1999)" row is cited to reference 47, which is the same Mochizuki 1999 paper as reference 35; the García-Pallarés kayak study (REF-950, 2009) is listed as 2000.
Mechanisms
Measured: larger standardised falls after longer breaks, with no evidence of a further difference beyond 90 days; smaller falls with partial than complete cessation in long breaks.
Authors' interpretation: after a certain time off, daily physical activity holds a floor under VO2max, so the loss stops growing; partial activity in short breaks was probably too easy to protect VO2max, and intensity is the key to maintenance (citing REF-587); athletes who trained more often lose more because their gains depended on high-intensity work.
Inference: a before-after SMD is the change divided by the spread between athletes, so an SMD of −1.42 is large in statistical terms but corresponds to the authors' average of about 9%; studies of homogeneous elite groups give bigger SMDs for the same percentage.The analysis appears to treat before and after values as independent groups (the forest plot labels them experimental and control), which ignores the within-athlete correlation — our reading of the figure legend.
Practical Application
How much VO2max do I lose if I stop training? 21 small athlete studies: about 4% after up to a month off and 9% after longer (authors' averages), no clear further fall past 90 days. Uncontrolled studies, reporting errors.
Agrees with the base's time course: single small studies in REF-1133 put the fall at about 4-7% at 12-21 days, about 10% at 5 weeks and 13-20% at about 8 weeks, levelling after 12 weeks; this pooled estimate points the same way (B+, overlapping primary studies, so not independent confirmation).
A few days off is outside this evidence: the shortest break here was 10 days (where Cullinane 1986 found no change); nothing in this review measures a break of under a week.
Keeping some training in a long break went with a smaller loss (B+, four comparisons from three soccer off-season studies, subgroup CI crossing zero): in breaks over 30 days, partial cessation went with SMD −0.65 against −1.69 for complete cessation; in breaks of 30 days or less it made no difference.The maintenance rule itself — keep the intensity, cut volume — is tested more directly in REF-587 (B+, narrative review, general populations) and REF-951 (B, 12 subjects trained by the study), neither in athletes.
Not from this paper: how quickly VO2max returns on retraining; the review covers only the loss.
Limitations
Uncontrolled inputs. All 21 studies measured the same athletes before and after a break, without a training control group; seasonal and test-learning effects cannot be separated from detraining.
Repeated time points. Studies with several time points (Coyle 1984 at 12, 21, 56 and 84 days; Miyamura 1990 at four points; Petibois and Gamelin at three) contribute more than one comparison from the same athletes, and the pooled analysis appears to treat them as separate.
Errors in the report. Confidence intervals that cross zero printed with p < 0.001, subgroup totals that do not match, age-test p-values that fit the wrong degrees of freedom, a random-effects rule printed backwards ("I2 < 40%"), a quality table with inconsistent study labels, and a duplicated reference (see Key Results). We report the pooled main effects, which are internally consistent, and give the subgroups little weight.
Who. Mostly men; mixed sports with soccer prominent; youth subgroups rest on three studies; "training status" is session frequency.
Bias. Egger's test indicated small-study effects in the long-term pool (p < 0.01); English only; 32 full texts could not be obtained and were excluded.
No funding statement; no conflicts of interest declared.
How this entry was checked and changed
Why this level
read from the full text (Europe PMC, open access, CC BY 4.0). The rubric's A- line is an SR/MA with serious heterogeneity or mostly small trials; graded one step lower because the inputs are not trials but small uncontrolled before-and-after studies (sample sizes 5 to 67), quality was scored with a reporting checklist (STROBE) rather than a risk-of-bias tool, repeated time points from the same athletes enter as separate comparisons, the long-term pool shows small-study bias (Egger p < 0.01), and the report has internal errors (subgroup confidence intervals that cross zero are given p < 0.001; the quality table labels studies inconsistently and one label matches no reference; one study appears to be cited to the wrong reference). It sits at the same level as REF-1133, which reviews many of the same primary studies without pooling
Full text (checked 2026-10-27)
"21 of 3315 potential studies met the inclusion criteria" (full text)
Short-term: "ES = −0.62 [95% CI -0.94; -0.31], p < 0.01; within-group I2 = 35.3%, Egger′s test = −1.22, p = 0.335" (full text)
Long-term: "ES = −1.42 [95% CI -1.99; -0.84], p < 0.01; within-group I2 = 76.3%" in the abstract and "I2 = 77%" in the Results; "Egger′s test = −3.369, p < 0.01"; short vs long "Q = 6.5, p = 0.01" (full text)
"no significant difference regarding V̇O2max change between 30-90 days detraining and larger than 90 days detraining (Q = 0.54, p = 0.46)" (full text)
Subgroups: training status "Q = 4.24, p = 0.03"; age "Q = 5.9, p = 0.05"; format "Q = 6.23, p = 0.01" (full text)
"the average V̇O2max decreased by 3.93% in the short-term period and by 9.43% in the long-term period" (full text)
"Detraining periods varied between 10 and 730 days across the studies" (full text)
Table 3: lower training status "-0.85 | -1.83; 0.12 | <0.001"; PDT "-0.65 | -1.42; 0.11 | <0.001" (full text)
Searches "prior to 31 March 2021"; excluded "full text could not be obtained (n = 32)" and "nonathletes (n = 51)" (full text)
"A fixed model was used for analysis; however, if statistical heterogeneity was shown (I2 < 40%), meta-analyses were performed using a random-effects model" (full text)
Training status footnote: "higher: regular training will be conducted more than or equal to 5 times a week; lower: training will be less than 5 times a week" (full text)
"only two studies are female athletes"; "The authors declare no conflicts of interest" (full text)