REF-1368 · Menges et al., 2026 · Frontiers in Sports and Active Living · Paper · PubMed · Free full text

If I swap some runs for rides, will I lose running fitness?

In 3 small trials of runners over 4-5 weeks, cycling in place of some or most runs left 1-mile to 5-km times not clearly different from running alone. Imprecise (A-).

Main caveat: Few, small, old trials. Seven trials, most with fewer than 15 participants per group, five of them from 1974-1996.
fairly strong evidenceMeasured on non-athletes to highly trained athletes

Menges, Dindorf, Dully and Fröhlich asked whether training adaptations transfer both ways between running and cycling.

They included randomised trials of at least four weeks that compared running-only training with cycling-only or combined running-and-cycling training in trained or recreationally active people, and pooled three outcomes: VO2max on a treadmill, VO2max on a cycle ergometer, and track running performance over 1 mile, 3,000 m or 5,000 m.

Seven trials qualified (three from databases, four from citation searching; three further reports could not be retrieved). The same four trials fed both VO2max analyses (Hoffmann 1993, Pechar 1974, Pierce 1990, Ruby 1996), and three fed the running analysis (Mutton 1993, two distances; Paquette 2018; White 2003).

On the treadmill, cycling-only training gained a little less VO2max than running-only (g −0.32, 95% CI −0.76 to 0.13, p = 0.16); on the cycle ergometer, running-only gained a little less than cycling-only (g −0.34, −0.79 to 0.11, p = 0.14); and running performance did not differ between cross-training and running alone (g 0.02, −0.62 to 0.66, p = 0.88).

Statistical heterogeneity was zero in all three, but the authors stress the conceptual heterogeneity (untrained women to competitive runners, 4 to 10 weeks, 70% heart-rate reserve to above lactate threshold) and the small samples.

They conclude that current evidence shows no clear improvement or decrement from cross-training between running and cycling over short to moderate periods, and that the data do not support claims that the modes are interchangeable. Our reading: this is the base's named dissent on the direction of transfer in PHYS-40.

The two older reviews in the base say transfer runs more from running to cycling than back (REF-537, REF-1358); here both VO2max estimates point to a small, non-significant advantage of the mode that matched the test, of almost the same size in each direction (−0.32 and −0.34), which fits specificity and shows no asymmetry.

The review did not test asymmetry directly, and its trials are too few and too small to rule one out.

Protocol

Design: systematic review and meta-analysis, PRISMA 2020; registered on PROSPERO (CRD420261283267).
Search: PubMed, Cochrane Library, ScienceDirect and Springer Nature, searched between 15 August 2025 and 9 February 2026 (RCTs in English or German, 1970 to 9 February 2026), plus forward and backward citation searching; two reviewers screened independently. 438 database records → 15 full texts → 3 included; citation searching → 4 more included (3 reports could not be retrieved).
Inclusion: RCTs in endurance-trained or recreationally active adults with running or cycling experience; structured cross-training of at least 4 weeks; a physiological or performance outcome; no-training controls excluded.
Sample: n = 7 trials (published 1974-2018 per the reference list; sizes 11 to 40 per Table 2), with 4 comparisons in each meta-analysis.
Comparisons: treadmill VO2max — cycling-only (intervention) vs running-only (control); cycle-ergometer VO2max — running-only (intervention) vs cycling-only (control); running performance — cycling-only or cycling plus running (intervention, pooled into one category) vs running-only (control).
Analysis: change-score standardised mean differences (Hedges' g), with an assumed pre-post correlation of r = 0.50 (sensitivity 0.20-0.80: stable); random-effects models with REML; heterogeneity by Cochran's Q and I²; publication bias not assessed (fewer than 10 studies).
Risk of bias: RoB 2 by two raters (weighted κ 0.75; 88.1% agreement). Randomisation "some concerns" in all trials; selective reporting "some concerns" in all (no pre-registration); Pierce 1990 high risk (more than half of enrolled participants dropped out); in the running analysis Paquette 2018 and White 2003 high risk (missing outcome data).The text does not say whether outcome assessors were blinded in any trial (training itself cannot be hidden from participants); measurement of VO2max was rated low risk.
Funding and interests (as printed): no financial support received. One author (T.M.) was employed by Endurance Coach GmbH; the others declare no commercial or financial relationships.

Key Results

Outcome (k = comparisons)Intervention vs controlHedges' g (95% CI)pHeterogeneity
Treadmill VO2max (k = 4)cycling-only vs running-only−0.32 (−0.76, 0.13)0.16I² 0%, Q(3) = 1.44, p = 0.70
Cycle-ergometer VO2max (k = 4)running-only vs cycling-only−0.34 (−0.79, 0.11)0.14I² 0%, Q(3) = 2.68, p = 0.44
Running performance, 1 mile to 5,000 m (k = 4)cycling or cycling + running vs running-only0.02 (−0.62, 0.66)0.88I² 0%, Q(3) = 0.67, p = 0.88

The trials (Table 2):

TrialParticipantsTrainingResult in the trial
Hoffmann 1993n = 16 healthy menrun vs cycle, 4×/week, 9 weeks, 80-85% HRmaxboth raised treadmill and cycle VO2max; no between-group difference
Pechar 1974n = 40 healthy young menrun vs cycle, 3×/week, 8 weeks, 85% HRmaxtreadmill VO2max +7.1% running, +2.7% cycling; cycle VO2max +6.8% running, +7.7% cycling; no significant group difference
Pierce 1990n = 16 healthy men (high dropout)run vs cycle, 4×/week, 10 weeks, 89% of VO2peakboth raised both VO2peaks; the only estimate favouring cycling on the treadmill
Ruby 1996n = 18 untrained womenrun vs cycle, 4×/week, 10 weeks, 70-80% HR reserveboth raised both VO2max values; larger gains in the matching mode
Mutton 1993n = 11 moderately fit male runnersrun vs run + cycle, 4×/week, 5 weeks, 85-90% HRmax1-mile and 5,000-m times improved alike (g 0.00 for both)
Paquette 2018n = 31 high-school cross-country runners (15-16 y)two easy runs a week replaced by cycling, 4 weeks3,000-m times improved in both; g −0.49; post-test on grass two weeks later, 7 runners excluded
White 2003n = 11 female college distance runnersrun vs run + cycle (2:1 cycling-to-running time), 5 weeks between seasons, 75-80% HRmax3,000-m times slightly slower in both; g 0.36; one overuse-injury dropout

Numbers that do not add up: the Results say the trials were "published between 1974 and 2003" and that sample sizes ranged "from N = 11 to N = 60"; the reference list dates Paquette et al. to 2018, and the largest sample in Table 2 is 40 (Pechar). For Pechar, Table 2 gives treadmill gains of 7.1% (running) and 2.7% (cycling), the Discussion 6.8% and 2.6%.

Mechanisms

Measured (pooled): small, non-significant deficits when the VO2max test did not match the trained mode, of similar size in both directions; no difference in running performance when cycling replaced part of running.
Authors' interpretation: the trend for each mode to score best on its own test is consistent with training specificity, but it is an interpretive pattern, not a tested moderator; variation between trials is more likely due to training status and dose (sufficiently high cardiovascular strain may blunt mode-specific differences) than to mode.Earlier literature's claim of more transfer from running to cycling is cited (higher peak heart rates in running, stretch-shortening efficiency, upright posture) but not tested.
Authors' interpretation: the absence of declines over 4-10 weeks suggests that replacing part of running with cycling may help hold central aerobic capacity, an indirect reading because no trial compared cross-training with stopping training.
Inference: a symmetric pair of estimates (−0.32 and −0.34) is the opposite of what an asymmetric transfer would predict, but with 90 participants and confidence intervals spanning about 0.9 units it cannot exclude a moderate asymmetry either way.

Practical Application

If I swap some runs for rides, will I lose running fitness? In 3 small trials of runners over 4-5 weeks, cycling in place of some or most runs left 1-mile to 5-km times not clearly different from running alone. Imprecise (A-).
What it is good for as a source: the dissent on the direction of transfer: the latest pooled trials show the same small mode-matching advantage both ways (g −0.32 and −0.34), not more transfer from running to cycling. It is evidence of no detected difference, not of equivalence.
What it does not show: that the modes are interchangeable (the authors say so), anything in highly trained athletes or over a season, or effects on running economy or injury.
How a principle should cite it: "a 2026 meta-analysis of 7 small trials found no clear difference in either direction, with similar small deficits when the test did not match the trained mode (Menges 2026)".
What it means for training: Not from this paper; the base's own practice: when running must shrink (injury, load), a similar sport such as cycling holds aerobic fitness better than stopping, and race-like work in the race mode stays in the last weeks (SYNTH-48, SYNTH-52, SYNTH-76). In the included running trials cycling replaced about 20-50% of running sessions, sometimes at a 2:1 time ratio.

Limitations

Few, small, old trials. Seven trials, most with fewer than 15 participants per group, five of them from 1974-1996; the training protocols do not reflect modern intensity distribution or interval work (the authors' limitation).
Wide confidence intervals crossing zero in every analysis; the authors warn that non-significance is not equivalence.
Mixed populations. VO2max evidence is mostly in untrained or healthy adults; running-performance evidence in three small runner groups, two at high risk of bias (missing data; test conditions changed in Paquette 2018).
Pooled interventions. Cycling-only and cycling-plus-running were combined for running performance.
Eligibility vs sample. The criteria name endurance-trained or recreationally active adults, yet the VO2max trials include untrained women and healthy men, and one running trial is in 15-16-year-olds.
Assumed correlation (r = 0.50) for change-score SDs; the sensitivity analysis found the estimates stable.
Reporting slips (publication years, sample-size range), noted under Key Results; the supplementary appendix (risk-of-bias figure, pre and post values) was not read.
Interest: as printed, "Author TM was employed by Endurance Coach GmbH" (TM is the first author); no financial support was received.
How this entry was checked and changed
Why this level

a registered systematic review and meta-analysis of mostly small trials (the rubric's line): each pooled estimate rests on 4 comparisons (VO2max analyses: 4 trials, 90 participants; running performance: 3 trials, 53 participants), the trials date from 1974 to 2018, their risk of bias ranged from some concerns to high, and every confidence interval crosses zero, so the null result is imprecise rather than a demonstrated equivalence

Full text (checked 2026-10-28)
Randomized controlled trials with intervention durations of at least four weeks were included; studies comparing run-only training with cyc-only or combined running–cycling interventions in trained or recreationally active individuals were eligible (n = 7) (full text)
For treadmill-assessed VO2max: Hedges' g = −0.32, 95% CI (−0.76, 0.13), p = 0.16; for cycle ergometer–assessed VO2max: Hedges' g = −0.34, 95% CI (−0.79, 0.11), p = 0.14 (full text)
Running performance outcomes demonstrated no meaningful differences between interventions [Hedges' g = 0.02, 95% CI (−0.62, 0.66), p = 0.88] (full text)
Between-study heterogeneity was low in all three analyses (I^2 = 0%), and all confidence intervals included the null effect (full text)
Pierce et al. was judged as high risk of bias due to a substantial dropout rate; in the running analysis Paquette et al. and White et al. were rated high risk for missing outcome data (full text)
Seven studies were included, published between 1974 and 2003; sample sizes ranged from N = 11 to N = 60 (full text)
The available evidence does not support conclusions regarding the interchangeability of training modalities, but rather indicates that no clear between-group differences were detected (full text)
Author TM was employed by Endurance Coach GmbH; financial support was not received for this work (full text)