In the New York City Marathon, runners as a group paced positively (fastest 5-km split at 5-10 km, slowest at 35-40 km, often a final spurt over the last 2.2 km), the fastest runners paced most evenly (5-km split CV 7.8% in top men and 6.6% in top women vs 8.3-14.4% further down), men slowed more than women (21.1% vs 16.7% from fastest to slowest split), and heat slowed the field: a systematic review of 76 mostly observational race-results studies of one hilly marathon, synthesised narratively; these are descriptions of what runners did, not tests of a pacing plan
Main caveat: Narrative synthesis of mostly retrospective race-results studies: associations only; no pooled effect, no heterogeneity estimate.
Knechtle and colleagues searched five databases for every study of the New York City Marathon and found 76 publications, which they grouped by theme and summarised without pooling.
Participation grew strongly over decades, more among women and age-group runners; elite and top age-group times improved while the mean finish time of the whole field slowed by about 40 min since the 1970s as the field broadened. In one analysis women's times peaked at 29.7 years of age and men's at 34.8.
About 9.5% of runners in one cohort had a major injury in training or the race that stopped them starting or finishing, and higher training volume went with more injury. Temperature was the weather factor most consistently tied to slower times, especially in slower runners and in men aged 30-64 and women aged 40-64.
On pacing, the race-results studies show a positive profile: the fastest 5-km split at 5-10 km, the slowest at 35-40 km where the course climbs into Central Park, often an end spurt over 40-42.2 km; older runners and women paced more evenly, and the fastest runners slowed least.
The authors conclude that the race is a large natural laboratory and call for cardiovascular monitoring and study of advanced shoes. Several of the review's authors wrote many of the included pacing, weather and participation studies.
Participants
Pooled: finishers of the New York City Marathon across editions from 1970 to recent years (for example 1,174,331 finishers from 1970 to 2017 in one study), all ages and levels, both sexes; some included studies combined New York with other marathons.
Also case reports (cystic fibrosis, type 1 diabetes, a nine-time champion's training), clinical cohorts (lipids, kidney stones, menstrual function, asthma, gastrointestinal bleeding) and injury cohorts of runners in training.
No total number of participants across studies is given.
Protocol
Search: Scopus, PubMed, Web of Science, Embase, Cochrane, to February 2026, no language limits (non-English screened by abstract only); New York Marathon terms combined with performance, pacing, injury, physiology, age or heart. 195 records (67 Scopus, 63 PubMed, 37 Web of Science, 25 Embase, 3 Cochrane); 76 included.
Inclusion: original research on runners in the race, any design, age, sex or level; grey literature, book chapters, editorials and conference proceedings excluded.
Extraction: two authors (authors, year, design, sample, variables, main results); selection agreement κ > 0.80.
Quality: Newcastle-Ottawa Scale by two investigators, a third settling disagreements, case studies excepted; no scores, summary or use of the scores in the synthesis appear in the text.
Synthesis: narrative, by domain (participation, performance and age, nationality, injury, environment, pacing, plus medical and miscellaneous topics).
Funding and interests: no funding; no competing interests declared; ChatGPT used to help produce graphical elements of the figures.
Key Results
Pacing (race split data):
| Finding | Value | Source study in the review |
|---|---|---|
| Overall profile | positive pacing with a final spurt at 40-42.2 km | Santos-Lozano 2014; Nikolaidis & Knechtle 2018 |
| Fastest and slowest 5-km split | 5-10 km fastest, 35-40 km slowest | Nikolaidis & Knechtle 2017, 2019 |
| Speed loss, fastest to slowest split | men 21.1%, women 16.7%; at 25-30 km women sped up while men slowed | Nikolaidis & Knechtle 2018 |
| Split variability by level | CV 7.8% (top men), 6.6% (top women) vs 8.3-14.4% in less successful runners | Santos-Lozano 2014 |
| By performance group | slowest group: largest drops at 5-20 km and largest rises at 35 and 40 km; fastest group: smallest drops throughout and smallest rise at 40 km | Nikolaidis & Knechtle 2019 |
| Age | older runners more even, especially in slower groups | Nikolaidis & Knechtle 2017 |
| End spurt by nationality | American women 87.2%, men 78.2%; Ethiopian 54.3%, Kenyan 45.8% | Aschmann 2018 |
| Per-mile analysis | final-mile performance correlated with the 4th mile | Thuany 2025 |
Weather, injury and age:
| Domain | Finding |
|---|---|
| Temperature | most consistent weather factor; high temperature linked to an average loss of about 8 min (Knechtle 2021); stronger in men aged 30-64 and women aged 40-64 (Weiss 2022) |
| Who slows more in heat | slower runners lost disproportionately more (Ely 2007), but one study found the fastest runners' relative loss larger at the same exposure (Gasparetto & Nesseler 2020); in Berlin some top finishers ran faster on warmer days (Knechtle 2021) |
| Other weather | humidity, dew point, pressure mostly not associated; higher humidity and wind went with faster running in some analyses |
| Injury | major injury (prevented start or finish) about 9.5%; minor about 49.2% (Toresdahl 2022); overuse injury in preparation about 36.1-40%; race-day injury about 14.1-16%; more training runs per week and higher volume went with more injury; longer longest run and a prior half-marathon with less |
| Strength training and injury | a hip-abductor-focused programme went with 64% lower overuse-injury risk (Toresdahl 2025); a self-directed programme in first-timers did not reduce injury causing non-completion (Toresdahl 2020) |
| Age of peak performance | 29.7 y (women), 34.8 y (men) by 1-year intervals; 30-34 and 35-39 by 5-year groups (Nikolaidis 2018); another study 29 y (women) and 27 y (men) (Lara 2014) |
| Field time | mean finish time about 40 min slower than in the 1970s; recent means about 4:40 |
| Shoes | world-class runners in advanced shoes about 2.0% (men, about 2.8 min) and 2.6% (women, about 4.3 min) faster (Senefeld 2021) |
Practical Application
In a hilly city marathon the typical runner went out fastest in the first 10 km and slowed to the 35-40-km hills, while the fastest runners held the most even pace; these are observed splits from race results, not a test of what pacing would make any runner faster. The pattern is consistent with the base's marathon rule (first half at or slightly below goal pace, never faster: SYNTH-64, SYNTH-10), but fitter runners may pace evenly because they are fitter, and the late slowdown here coincides with the course's hills.
The slowest runners' pace swung most, the fastest runners' least: in one included analysis the slowest group had the largest percentage speed changes, downward over the first 20 km and upward at 35 and 40 km, and the fastest group the smallest (the review does not say what the changes are measured against).These are not data on a too-cautious start: the slowest runners lost the most speed early. SYNTH-10 reads a much faster finish as a start that was too cautious (REF-820); this review neither supports nor tests that reading.
Heat costs time, and who pays most is not settled here: the included studies disagree on whether slower or faster runners lose more. For pacing on a warm day the base's heat rule in SYNTH-10 applies; this review adds no per-degree correction.
Injury data in the review are prevalences from training cohorts and are not treatment advice; a runner with pain should see a physician or physiotherapist.
Limitations
Narrative synthesis of mostly retrospective race-results studies: associations only; no pooled effect, no heterogeneity estimate.
The Newcastle-Ottawa appraisal is stated in the methods, but no scores or summary appear in the text read, and the synthesis does not weigh studies by quality; the per-study tables and figures were not in the file read.
Several of the review's authors wrote many of the included pacing, weather, nationality and participation studies, so the selection and reading of those studies is not independent.
Some included studies mix New York with other marathons, and the discussion draws on Berlin, Boston and Oslo data, which the authors flag as context rather than New York evidence.
One hilly course: late slowing cannot be separated from the Central Park hills, and the findings may not transfer to flat courses.
Race databases carry registration errors, missing splits and timing changes over decades (the authors).
How this entry was checked and changed
Why this level
a systematic search, but of observational studies (mostly retrospective race-results databases, cross-sectional) of one race, retold without meta-analysis and without a reported risk-of-bias result, so it cannot stand above the cross-sectional design it summarises; not a meta-analysis of trials