Will a prevention routine stop running injuries?
Not shown: 9 RCTs in 1904 runners found no clear drop in injury risk or rate; the routines were mostly light and unsupervised. Supervised ones may help (post hoc, 3 small trials).
Wu and colleagues asked whether exercise-based injury prevention programmes reduce running-related injuries in endurance runners, defined as people for whom running was more than half of their training.
Searching PubMed, Web of Science and SPORTDiscus on 15 January 2023, they kept randomised trials of a non-running exercise programme against running only or a placebo exercise, with injuries recorded prospectively; eight military studies were excluded (four of their authors confirmed that running was under half of training, and all eight were dropped), and two studies by one first author were excluded because they shared baseline data that could not be reconciled.
Nine trials remained (1904 participants), from 4-week pre-conditioning to 12-month programmes. Across seven trials the pooled log risk ratio was −0.21 (95% CI −0.46 to 0.047; p = 0.110), a risk ratio of about 0.81 (0.63 to 1.05, our conversion); across six trials the log incidence-rate ratio was −0.15 (−0.45 to 0.15; p = 0.329), about 0.86 (0.64 to 1.16).
Seven trials had high overall risk of bias, one some concerns, one low. The three trials with supervision at least weekly had the highest adherence (88% or more by the authors' summary, against 10-72% in five unsupervised trials, Table 3) and, pooled post hoc, a log risk ratio of −0.77 (−1.18 to −0.37; p < 0.001), about 0.46 (0.31 to 0.69).
The authors conclude that there is currently insufficient evidence to recommend exercise programmes to reduce running injuries, that supervision (possibly through adherence) may be essential, and that heavy strength training and foot-and-ankle programmes need proper trials.
Our reading: no evidence of a difference for the light, mostly unsupervised routines that runners are usually given; the confidence intervals still allow a useful benefit, and the supervised signal comes from the three smallest trials in the injury-risk pool.
Protocol
Design: systematic review and meta-analysis; PRISMA and PERSiST; PROSPERO CRD42021211274; two independent reviewers screened (Covidence), extracted data and assessed risk of bias (RoB 2); GRADE added at reviewers' request.
Inclusion: endurance running for health, occupation or performance (5 km to ultra); running more than 50% of training time; RCT; non-running exercise intervention; running-only or placebo-exercise control; injury rate or incidence reported; injuries recorded prospectively alongside training. Excluded: injured participants (rehabilitation), interventions combined with orthotics, massage or nutrition.
Search and selection: 5896 records (1169 duplicates), 4733 screened, 55 full texts, 11 eligible, 2 excluded for irreconcilable duplicate data, 9 included.
Trials (Tables 1-2): Baltich (Canada, 6 months, 129 novice runners, resistance or functional strength vs stretching, home-based, fortnightly coaching); Bredeweg (Netherlands, 4-week walk-and-hop pre-conditioning before a running programme, 362 novices); Edouard (France, 40 weeks, 136 endurance runners from athletics clubs, core, balance, strength, stretching and Nordics, unsupervised, 6 of 59 compliant); Halvarsson and von Rosen (Sweden, 14 weeks, 62 elite orienteers, balance and hopping 4 times a week, unsupervised); Lundstrom (USA, 12 weeks, 34 recreational adults, core or plyometrics once a week, supervised); Mendez-Rebolledo (Chile, 6 weeks, 16 adolescent female runners, neuromuscular training 3 times a week, supervised, 100% adherence); Taddei (Brazil, 12 months, 118 recreational long-distance runners, foot-and-ankle strengthening 4 times a week with one supervised session, vs static stretching); Toresdahl (USA, 12 weeks, 720 first-time marathoners, self-directed 10-minute bodyweight strength routine 3 times a week); van Mechelen (Netherlands, 16 weeks, 327 male recreational runners, warm-up, cool-down and stretching).
Supervision: "supervised" meant online or in-person supervision at least once a week (Lundstrom, Mendez-Rebolledo, Taddei); the other six provided videos, booklets or one demonstration.
Analysis: random-effects meta-analysis (metafor) of the log risk ratio (injured / all participants) and of the log incidence-rate ratio (injuries per 1000 h); intention-to-treat data preferred; multiple intervention arms pooled; Q, τ² and I²; 95% prediction intervals when τ² > 0; outliers by studentized residuals and Cook's distance; funnel plot with rank-correlation and regression tests. Post hoc: the injury-risk analysis restricted to supervised programmes.
Deviations from protocol: PEDro dropped; GRADE added; Hedges' g not reported; the planned subgroup by training modality dropped (most programmes mixed modes) and the supervision subgroup added.
Funding and interests: grant from the China Scholarship Council in collaboration with Loughborough University; the authors declare no conflicts of interest.
Key Results
| Analysis | Trials | Pooled log ratio (95% CI) | Ratio, our conversion | Heterogeneity | GRADE (authors) |
|---|---|---|---|---|---|
| Injury risk | 7 | −0.21 (−0.46 to 0.047), p = 0.110 | RR 0.81 (0.63 to 1.05) | Q(6) = 13.00, p = 0.043, I² 65.02%; prediction interval −0.75 to 0.34 (RR 0.47 to 1.40) | moderate |
| Injury rate (per 1000 h) | 6 | −0.15 (−0.45 to 0.15), p = 0.329 | IRR 0.86 (0.64 to 1.16) | Q(5) = 9.23, p = 0.100, I² 28.50%; prediction interval −0.65 to 0.34 (IRR 0.52 to 1.40) | high |
| Injury risk, supervised only (post hoc) | 3 | −0.77 (−1.18 to −0.37), p < 0.001 | RR 0.46 (0.31 to 0.69) | Q(2) = 0.13, I² 0% | not graded |
Risk of bias: seven trials high overall, one some concerns (Halvarsson), one low (Baltich); common reasons were unblinded participants, dropout and self-reported injuries.
Publication bias: regression test for funnel asymmetry p = 0.044, rank-correlation test p = 0.381.
Adherence: supervised trials 88% or more of sessions attended and five unsupervised trials 47-72% meeting their compliance criterion, by the authors' summary; one unsupervised trial 86-93%. Table 3 does not fully match: Edouard, one of the five, had 6 of 59 compliant (10%), and Taddei's 88% sits beside period figures falling from 90.4% to 48.9% (mean about 71%, our arithmetic).
Within-trial adherence signals: elite orienteers doing the exercises at least twice a week had fewer injuries than controls (odds ratio 0.26, 0.07 to 0.92) while those doing less did not (0.88); in the marathon trial compliant runners reported fewer minor injuries than non-compliant ones (41.5% vs 56.2%), a comparison that is not randomised.
The largest trial (first-time marathoners, 720) found major injuries in 27 controls vs 25 on the programme (RR 0.97, 0.57 to 1.63) and minor injuries in 186 vs 163 (RR 0.92, 0.79 to 1.07); it is in the base as REF-924.
The foot-and-ankle trial (Taddei, 118 runners over 12 months, against static stretching): 20 runners injured in the control group and 8 in the intervention group; the authors' Discussion cites a 2.42-fold difference.
Numbers that do not add up:
Taddei's injured percentages (16.9% and 6.7%) are 20 and 8 out of all 118 runners, not of each group; per group they are 20 of 61 (32.8%) and 8 of 57 (14.0%) (our arithmetic).
The same trial's hazard ratio is given as 2.42 with p = 0.035 and a 95% CI of 1.98-3.62; an interval that far from 1 implies p far below 0.001, and it is not centred on 2.42 on the log scale (our arithmetic), so one of the three numbers is wrong.
The Discussion says injuries were self-reported and not clinically diagnosed in "four of the nine studies"; Table 3 marks six trials as not clinically diagnosed, and elsewhere the Discussion says "Six studies asked participants to self-report injuries".
The Discussion lists the interventions "that found a positive effect" as [45, 47, 49], but [49] is the null marathon trial, cited again in the same paragraph among studies with no difference; [48] (Taddei) is most plausibly meant.
Table 1 gives van Mechelen's age groups as 129 + 155 + 137 = 421 runners for a trial of 327.
Mechanisms
Measured: supervised programmes had higher adherence, and within two unsupervised trials the more adherent runners had fewer injuries of one kind (overuse injury prevalence in the orienteers; minor but not major injuries in the marathon trial, major 9.1% vs 8.0%) (non-randomised comparisons).
Authors' interpretation: supervision works through adherence, and so through actual exposure to the exercise; programme content may also have been inadequate, since strength work was low in volume and intensity, whereas higher strength doses reduced overuse injuries in other sports.Loading in endurance running is lower in magnitude and more repetitive than in games sports, so injury mechanisms and the benefit of add-on exercise may differ; the general-sport meta-analyses (13,355-32,254 participants) should not be applied to runners.
Authors' interpretation: adherent runners may also differ in other health behaviours (sleep, nutrition), which no trial measured.
Inference: the three supervised trials were also the three smallest in the injury-risk pool (16, 34 and 118 runners; Halvarsson's 62 was not in it), had different populations (adolescent girls, college-age adults, experienced long-distance runners) and, in the largest of them, a stretching control; with funnel asymmetry in the main analysis, a small-study effect is as plausible an explanation as supervision.
Practical Application
Will a prevention routine stop running injuries? Not shown: 9 RCTs in 1904 runners found no clear drop in injury risk or rate; the routines were mostly light and unsupervised. Supervised ones may help (post hoc, 3 small trials).
Do not sell a home routine as injury insurance (A-, runners): a 10-minute self-directed bodyweight programme, a walk-and-hop pre-conditioning month or a warm-up and stretching routine each showed no evidence of fewer injuries. If the athlete likes such a routine, keep it for its own sake; do not promise fewer injuries.
If prevention is the reason, make it something the athlete will actually do, ideally coached: the three supervised programmes (one to four sessions a week, 6 weeks to 12 months, adherence 88% or more) went with about half the risk of injury, but post hoc and in 168 runners (about B weight). Not from a study; the base's own practice: a programme the athlete completes twice a week beats a longer one they abandon.
Load is still the lever the base holds: single-run spikes are associated with injury (REF-1102), weekly jumps only early and imprecisely in one smaller cohort (REF-623; the larger cohort found no weekly link); no exercise programme in this review was shown to offset them (inference; the review did not test load).
Limitations
Quality of the trials. Seven of nine at high risk of bias (unblinded, dropout, self-reported injuries); the authors still grade certainty moderate for risk and high for rate. A "high" grade with four of the six rate trials at high risk of bias, and an interval reaching an IRR of 0.64, looks generous to us.
What was tested. Mostly light, multimodal, bodyweight routines with poor adherence; four trials used strength exercises, none heavy. The null is about these programmes, not about strength training in general.
Different injury definitions (more than a week of restricted running, a single missed run, or any impairment of training), many self-reported; interventions and follow-up from 6 weeks to 12 months (authors), with equal weight regardless of duration.
Supervised subgroup: added after the protocol, three trials and 168 runners, confounded with population, control type and trial size; a hypothesis, not a finding.
Search date 15 January 2023; trials published later are not included.
Reporting: the Taddei percentages and hazard-ratio interval, the self-report count and one misplaced citation do not match the tables (see Key Results).
Interests. China Scholarship Council funding; no conflicts declared.
How this entry was checked and changed
Why this level
systematic review and meta-analysis of RCTs read from the full text (open access, CC BY 4.0); held at A- rather than A because seven of nine trials were at high overall risk of bias, heterogeneity for injury risk was substantial (I² 65.02%) and the programmes were mostly light and poorly adhered to, so the null answers "do typical light programmes work" rather than "does strength training work". The null itself is not a reason for the level. The supervised-subgroup result is post hoc, from three small trials, and carries about B weight within this entry
Full text (checked 2026-10-27)
A total of nine articles containing 1904 participants were included in analysis (full text)
The average log risk ratio was −0.21 (95% CI −0.46 to 0.047), a statistically non-significant effect (z = −1.60; p = 0.110); Q(6) = 13.00, p = 0.043, I2 = 65.02% (full text)
The pooled log incidence rate ratio was −0.15 (95% CI −0.45 to 0.15), not statistically significant (z = −0.98; p = 0.329), I2 = 28.50% (full text)
Post hoc, three supervised studies: pooled log risk ratio −0.77 (95% CI −1.18 to −0.37), z = −3.75, p < 0.001 (full text)
Seven had high overall risk of bias, one was with some concerns, and one had low overall risk of bias (full text)
Supervised studies reported attendance of ≥ 88%; five unsupervised studies reported compliance of 47–72%, and one 86–93% (full text)
Taddei: Con 20 participants (16.9%), Int 8 participants (6.7%); hazard ratio 2.42, p = 0.035, 95% CI 1.98–3.62 (full text)
Search conducted on 15 January 2023; funding from the China Scholarship Council (full text)