REF-1004 · Doherty et al., 2020 · Journal of Science and Medicine in Sport, Vol. · Paper · PubMed

An evaluation of the training determinants of marathon performance: A meta-analysis with meta-regression

A-fairly strong evidenceMeasured on recreational to world-class athletesWritten before October 2026, being re-checked

A systematic review, meta-analysis and meta-regression asking whether a dose-response relationship exists between training behaviours and marathon performance. Observational and interventional studies of elite and recreational marathon runners were included: eighty-five studies with 137 cohorts of runners (25% female).

Seven training behaviours were used as covariates in separately conducted univariate random effects meta-regression models; each showed a negative statistical association with marathon performance (R2 0.38-0.81, p<0.001), whereby increases in a training parameter coincided with faster finish times. The rate of non-finishers was 7.27% (95% CI 6.09%-8.65%).

Participants

Eighty-five studies including 137 cohorts of runners (25% female); elite and recreational marathon (42.2km) runners.

Protocol

Systematic search of multiple literature sources for observational and interventional studies.
Covariates: hours of running per week; average weekly running distance; weekly run count; peak distance run in a single week; average training pace; longest run distance; and the number of runs of ≥32km in the pre-marathon training block.
Separately conducted univariate random effects meta-regression models; meta-analysis of the rate of non-finishers.

Key Results

Every training behaviour listed was negatively associated, statistically, with marathon performance (R2 0.38-0.81, p<0.001): a higher value of any one parameter went with a faster marathon finish time.
The abstract does not say which behaviour had which R2.
Rate of non-finishers in the marathon: 7.27% (95% CI 6.09%-8.65%).
Authors' conclusion: the data can inform training regimes specific to a given target finish time.

Practical Application

"What training goes with a faster marathon?" Across 85 studies and 137 cohorts of elite and recreational marathoners (meta-regression, between-cohort associations, each behaviour modelled separately), more weekly distance, more runs per week, a bigger peak week, more runs ≥32km, a longer longest run, more hours and training pace were each associated with faster finish times (R2 0.38-0.81). Faster cohorts are the ones that train more, on every volume measure the review looked at.
This is not within-person evidence: it does not show what happens when one runner adds distance, and faster runners may train more because they can.
Because the models are univariate, the behaviours overlap (a cohort with high weekly distance also has long runs and many runs); the abstract gives no way to say which matters most.
What was not examined in the abstract: intensity distribution, injury, strength work, physiology.
Inference, not tested here: for a marathoner without a consistent base, volume and frequency are the first things to build.

Limitations

Cohort-level (ecological) associations from observational and interventional studies; not causal and not within-athlete.
Univariate models: covariates are not adjusted for each other or, in the abstract, for sex, age or ability.
The abstract does not report R2 per behaviour or the slope of any relationship.
25% female; heterogeneity between studies is not reported in the abstract.
Written from the abstract and record metadata only (full text not read).
How this entry was checked and changed
Why this level

systematic review of 85 studies, but the associations are univariate and at the level of study cohorts, drawn from observational as well as interventional studies