The 10% Mileage Rule Has One Trial Behind It. It Found Nothing.

Short version: I found one randomized trial that tested the 10% rule. In 532 novice runners it saw no difference in injuries. Cohort studies since then also do not back a weekly 10% line. The rule is a habit, and the data do not even show it is wrong.

The question

Many coaches and running magazines say: raise weekly distance by no more than 10%. I want to know what study sits under that sentence. A rule that tells people how to train should have at least one controlled test. Does it?

Data and where it came from

I read four sources. I read the abstracts and records, not every full text, so some details (group sizes, running logs) are not in what I checked.

1. Trial. Buist and colleagues, the GRONORUN trial, published 2008. It randomized 532 novice runners who prepared for a 4-mile (6.7 km) event. The control group followed a standard 8-week programme. The intervention group followed a graded 13-week programme built on the 10% rule. Both groups logged runs online. An injury was any lower-limb or back complaint that restricted running for at least 1 week. [1]
2. Cohort, 2014. Nielsen and colleagues followed 874 novice runners for 1 year with GPS watches. Runners were sorted after each session by weekly distance change: under 10% or a decrease, 10% to 30%, or over 30%. The authors call it exploratory. [2]
3. Cohort, 2025. Frandsen and colleagues followed 5,205 runners for 18 months, with 588,071 recorded sessions. Injury was self-reported overuse injury. [3]
4. Review. Damsted and colleagues, 2018, a systematic review of four studies on changes in training load and running injury. [4]

I mark each by design. Source 1 is a trial. Sources 2 and 3 are cohorts. Source 4 is a review, and it is the one that disagrees with the rule, so I read it as the counter-voice.

Method

I take the trial's own numbers first. Then I add a confidence interval that the abstract does not give. I computed that interval by hand, without the Lab, and a reader can repeat it.

The inputs are 20.8% injured in the graded group and 20.3% in the standard group. [1] I assume an equal split of 266 runners per arm, because I did not read the group sizes. The log risk ratio has this standard error:

$SE = \sqrt{\frac{1-p_1}{p_1 n_1} + \frac{1-p_2}{p_2 n_2}}$

Here $p_1$ and $p_2$ are the injury proportions in the two arms, and $n_1$ and $n_2$ are the arm sizes. With the inputs above, $SE \approx 0.17$. The 95% interval is the log risk ratio plus or minus 1.96 times $SE$, then exponentiated.

Result with numbers and uncertainty

The trial. Injury was 20.8% with the graded programme and 20.3% with the standard one (P = .90). [1] That is about 208 against 203 injured runners per 1,000. The risk ratio is 1.02. My hand-computed 95% interval is 0.73 to 1.43. The difference in absolute risk runs from about 64 fewer to 74 more injured runners per 1,000.

What this does not show. It does not show that the 10% rule is useless. A true 27% cut in injuries sits inside my interval. The trial can say "no large effect either way". It cannot say "no effect".

The comparison is also not clean. The graded programme lasted 13 weeks and the standard one 8 weeks. [1] The arms differ in length, in total exposure and in the speed of progression, not only in the 10% step. The Damsted review reports that the average weekly volume rose about 10% in one arm and about 24% in the other, with an injury hazard ratio of 0.8 (95% CI 0.6 to 1.3). [4] That interval also includes no difference. It also includes a 40% lower rate for the graded arm.

The cohort. In the 874 novice runners, 202 were injured. Weekly progression showed no significant difference in overall injury rate. [2] For distance-related injuries such as patellofemoral pain and iliotibial band syndrome, runners who raised distance by over 30% had a hazard ratio of 1.59 (95% CI 0.96 to 2.66). [2] The interval crosses 1.0. The authors suggest keeping increases below 30%, and they say randomized trials are needed. [2] That is a 30% line, from an exploratory study, with an interval that does not exclude no effect. It is not a 10% line.

The larger cohort. In 5,205 runners, a week-to-week ratio showed no relationship with injury. [3] What did predict injury was a single session that went beyond the longest run of the past 30 days. A session more than 10% longer than that run had a rate ratio of 1.64 (95% CI 1.31 to 2.05). A session more than 100% longer had 2.28 (95% CI 1.50 to 3.48). [3] The number 10% appears here, but it is a rule about one long run, not about weekly distance. The dose-response also does not climb with the size of the spike: the 30% to 100% group (1.52) sits below the 10% to 30% group (1.64), though their intervals overlap.

So the 10% rule as people say it, a weekly cap, has one trial that found nothing and one large cohort that found nothing for the weekly ratio. [1][3] A related rule, "do not suddenly run a much longer single run", has cohort support. I find that more interesting, and it fits an old complaint of mine: weekly mileage is a poor stand-in for load. One 30 km week can be four steady runs or one long run plus filler. The sum looks the same. The stress on the legs does not.

The review calls the whole area "very limited evidence". [4] I agree with that wording.

Sensitivity: which assumption moves the result most

I list the assumptions in order of how much they could change my view.

1. Who the runners were. All trial runners were novices preparing for a 6.7 km event over a few weeks. [1] Novices started from low base volume. A 10% rise from 10 km a week is 1 km. The same rule at 80 km a week is 8 km. The trial says nothing about experienced runners. This assumption moves the result most, because the rule is usually told to everyone.
2. Power. My interval spans 0.73 to 1.43, built on an assumed equal split. If the real arms were uneven, the interval would widen a little. It would not change from "wide" to "narrow". A trial this size cannot settle a 20% effect. Over how many weeks? Eight to thirteen. That is short for tendon and bone injuries.
3. The 13 versus 8 week mismatch. If longer programmes carry extra risk, the trial could hide a benefit of the 10% step. If a gentler start carries a benefit that the longer duration cancelled, the same. I cannot separate these from the abstract.
4. Injury definition. The trial counted any complaint that restricted running for 1 week. [1] A looser definition would catch more minor pain. A stricter one would count fewer injuries and widen the relative interval.
5. Cohort confounding. The two cohorts did not randomize who raised distance fast. [2][3] Runners who feel good may run more. Runners who feel pain may run less, which can make high progression look safe. That bias pushes results toward "no harm" and could hide a real effect.

My present view

I hold a view with moderate confidence, about 0.7. There is no controlled evidence that a weekly 10% cap lowers injury, and the best trial I found cannot exclude a modest benefit or a modest harm. I would not tell anyone to drop the rule on this basis, and I give no training advice here. I would only stop calling it a finding.

Two things would change my mind. One is a trial with at least a few thousand runners that randomizes weekly progression and keeps programme length equal across arms. The other is a pooled analysis of the existing cohorts that reports the 10% to 30% band against a clean reference with an interval narrower than 0.8 to 1.25. The next study I would trust is the first one. I do not know of one running.

I treated the 2014 shoe-trial evidence the same way in the earlier post on shoe trials. The pattern repeats: a rule with a confident tone, one or two modest studies, and intervals that include no effect. I extend that post here. I do not retract it.