REF-1387 · van Poppel et al., 2021 · Journal of Sport and Health Science · Paper · PubMed · Free full text
Why do I keep getting injured?
The best-graded warning sign is a previous injury: a past running injury if you run over 20 km a week, a non-running one if you run less. No risk score predicts who gets hurt; all were low quality.
Main caveat: No pooling, few studies per factor. Clinical heterogeneity (populations, injury definitions, how factors were reported) ruled out meta-analysis.
van Poppel et al. reviewed prospective cohort studies that used multivariable analysis to look for risk factors or risk models for lower-extremity running injuries, separately for short-distance runners (mean ≤20 km per week and ≤10 km per session) and long-distance runners (more than both), and graded each factor with GRADE.
Searches to February 2019 found 29 studies with 18,853 participants: 17 single-factor studies of short-distance runners, 8 single-factor and 3 risk-model studies of long-distance runners, and 1 risk-model study of both; 38 potential factors were analysed for short-distance and 36 for long-distance runners. No meta-analysis was possible.
In long-distance runners a previous running-related injury was the strongest risk factor (moderate quality; 3 studies, 922 runners, upgraded for effect size).
In short-distance runners a previous injury not attributed to running was the strongest (high quality; 2 studies, 2,705 runners), and higher BMI, higher age, male sex, no previous running experience and a lower weekly running volume in minutes were risk factors at moderate quality, while — in the body text — a previous running injury was moderate-quality evidence for NOT being a risk factor in this mostly novice group.
Running frequency was not a risk factor (moderate), running-shoe age and previous sports participation were not associated (high), and weekly distance was low-quality evidence.
All 11 risk models (4 studies; areas under the curve about 0.65-0.72 for the seven that report one) were in the derivation stage and graded low quality; the authors found "no evidence that these models are predictive".
Our reading: the review supports asking about previous injury first and treating it as the strongest marker; it gives no training-load threshold and no basis for a risk score, and it predates the single-run-spike cohort the base's ramp line rests on (REF-1102).
Protocol
Design: systematic review registered with PROSPERO (CRD42019133799), reported to PRISMA. Searches of PubMed, CINAHL, Cochrane Library, SPORTDiscus and PsychINFO from inception to February 2019, by a librarian; reference lists checked.
Eligibility: prospective cohort studies with multivariable analysis of individual risk factors or risk models for lower-extremity running injuries. Outcome: a running-related injury, "self-reported musculoskeletal complaints, in the lower extremity, caused by running activities".
Classification: short-distance = mean ≤20 km per week and ≤10 km per session; long-distance = more than 20 km per week and more than 10 km per session; by km per week if the two conflicted. Results reported for the general group, men and women separately where available. Two studies sat near the cut-off.
Risk of bias and grading: QUIPS for single-factor studies (kappa 0.80, 95% CI 0.75-0.83), PROBAST for risk models.GRADE for prognosis: cohort evidence starts high and is downgraded for phase, study limitations, inconsistency, indirectness, imprecision and publication bias, upgraded for a large effect (OR > 2 or < 0.5) or a dose-response gradient; single studies (n < 4000) start at low quality.A factor counted as a risk (or protective) factor when most studies, holding more than 50% of participants, found a significant association.
Included: n = 29 studies, n = 18,853 participants; 25 single-factor and 4 risk-model studies. Follow-up 6 weeks to 1 year (short-distance) and 4 weeks to 2 years (long-distance); 67-100% of participants analysed.
Overlap with the base: the short-distance set includes the cohorts behind REF-1313 (Buist 2010, both papers), REF-1312 (Kluitenberg 2015 and 2016), REF-926 (Nielsen 2013, 930 novices) and REF-619 (Nielsen 2014, weekly-distance progression). It postdates REF-564 (2007) and predates REF-1367 (2022) and REF-1102 (2025).
Funding and interests: no funding statement is printed; "The authors declare that they have no competing interests." No device maker, federation or data platform is named.
Key Results
GRADE summary of the factors found in two or more studies (Table 5) and the body text (quality: high, moderate, low, very low):
| Group | Factor | Studies (participants) | Direction | Quality |
|---|---|---|---|---|
| Long-distance | Previous running-related injury | 3 (922) | risk factor (strongest) | moderate (downgraded for limitations and indirectness, upgraded for effect size) |
| Long-distance | Training volume | 1 with data | not a risk factor (women) | very low |
| Short-distance | Previous injury not attributed to running | 2 (2,705) | risk factor (strongest) | high |
| Short-distance | Higher BMI | 3 (3,138) | risk factor | moderate |
| Short-distance | Higher age | 3 (3,138) | risk factor (one study, hazard ratio 1.02) | moderate |
| Short-distance | Sex (male) | 3 (3,580) | risk factor (abstract and highlights; no body-text section) | moderate |
| Short-distance | No previous running experience | 3 (3,138) | risk factor | moderate |
| Short-distance | Lower weekly running volume (minutes) | 2 (2,526) | risk factor (more than 60 min a week protective in one; less than 2 h a week a risk in the other) | moderate |
| Short-distance | Previous running-related injury | 3 general-group studies (3,138) | not a risk factor in the body text (Table 5 lists it among risk factors) | moderate |
| Short-distance | Running frequency | 2 | not a risk factor | moderate |
| Short-distance | Weekly running distance | 2 (1,366) | one found more than 30 km a week a risk, one no association | low |
| Short-distance | Running-shoe age; previous sports participation | 2-3 | not associated | high |
| Short-distance | Foot morphology (moderate pronation) | 2 (1,384) | not a risk factor | low |
| Short-distance | Hip abduction strength; gait-related factors | single studies | risk factors | very low |
Risk models (Table 6; all derivation stage, all graded low quality):
| Model (cohort) | Variables kept | Performance |
|---|---|---|
| 5-km runners (van Poppel 2016) | previous injury OR 4.1 (2.2-7.6); weekly distance OR 0.95 (0.90-0.99); age OR 0.97 (0.95-0.99) | AUC 0.71 (0.64-0.79) |
| 10-15-km runners (van Poppel 2016) | previous injury OR 3.8 (2.7-5.3); weekly distance OR 0.97 (0.95-0.99); BMI OR 1.1 (1.0-1.2); weekly training frequency OR 1.3 (0.99-1.70); age OR 0.98 (0.97-0.99) | AUC 0.70 (0.66-0.73) |
| Half-marathon runners (van Poppel 2018) | previous injury OR 3.3 (2.3-4.8); weekly distance OR 0.98 (0.97-1.0) | AUC 0.67 (0.62-0.71) |
| Marathon runners (van Poppel 2018) | previous injury OR 4.3 (2.9-6.1); weekly distance OR 0.98 (0.97-0.99) | AUC 0.68 (0.64-0.72) |
| Male marathon runners (van Middelkoop 2008) | injury in the previous 12 months OR 2.62 (1.82-3.78); more than 7 races a year OR 1.66 (1.08-2.56) | AUC 0.65 |
| Male marathon runners, knee injuries (van Middelkoop 2008) | injury in the previous 12 months OR 3.67 (1.79-7.49); interval training (always) OR 0.49 (0.26-0.93) | AUC 0.69 |
Previous injury runs through every model: in each of the van Poppel and van Middelkoop models that kept it, a previous injury carried an odds ratio of 2.62 to 4.3; in Wen 1998 (marathon training) 2.018. More weekly distance went with slightly fewer injuries in all four van Poppel models (odds ratios 0.95-0.98 per unit of weekly distance; the unit is not printed in the table).
No model is ready: "Because all models were in the derivation stage, they were graded low quality"; "There is no evidence that these models are predictive for RRIs." Three studies (8 models) were applicable; one (3 models, men only) had applicability concerns.
Risk of bias: lowest for outcome and prognostic-factor measurement, highest for study attrition and confounding, mainly from poor reporting.
Mechanisms
Measured: previous injury (running-related in long-distance runners, any non-running musculoskeletal complaint in short-distance runners) was the factor most consistently associated with a new running injury across prospective cohorts with multivariable adjustment.
Authors' interpretation: it is unclear whether the risk after a previous injury comes from incomplete healing, biomechanics changed by the injury or something else. In novices a previous running injury cannot exist yet, so a non-running complaint takes its place, perhaps because people with existing complaints have a lower loading capacity.Novices without experience may build up too quickly for their tissues to adapt; a lower BMI may go with stress fractures and a higher BMI with medial tibial stress syndrome; injuries are multifactorial.
Inference: the finding that a lower weekly running volume is a risk factor in mostly novice cohorts is as likely to reflect who keeps running (and logs more minutes) as a protective effect of running more; the authors themselves warn that more than 60 min a week being protective "does not necessarily mean" that less is a risk factor. It is not a reason to add volume.
Practical Application
Why do I keep getting injured? The best-graded warning sign is a previous injury: a past running injury if you run over 20 km a week, a non-running one if you run less. No risk score predicts who gets hurt; all were low quality.
Ask about previous injury first: in runners doing more than about 20 km a week a previous running injury is the strongest marker (moderate quality); in lower-volume and novice runners a previous non-running musculoskeletal complaint is (high quality). This backs the SYNTH-73 line that a previously injured site gets the slowest ramp; the ramp itself was not tested here (inference).
No training-load rule comes out of this review: running frequency was not a risk factor (moderate), weekly distance was low-quality evidence pointing both ways, and in the risk models more weekly distance went with slightly fewer injuries. None of the 29 cohorts tested a single-run spike; the search ended in February 2019, before REF-1102.The base's line "Runners: no run over ~10% longer than your longest of the past 30 days" (SYNTH-29; also SYNTH-47, SYNTH-64, SYNTH-75) rests on that one later cohort: say it is a risk marker from one cohort, not a safe line, next to REF-1367 (36 prospective studies, no consistent link between training parameters and injury).
Do not offer a risk score or a screening verdict: of eleven models in four studies, the seven with a reported area under the curve reached about 0.65-0.72 (the rest gave only goodness of fit or R²) and none was validated; the authors found no evidence that they predict injury.
Novices and older or heavier beginners: higher BMI, higher age, no running experience and (by the abstract) male sex were moderate-quality risk factors in short-distance runners. Not from this paper; the base's own practice (SYNTH-75): "weekly totals up gradually (under ~20%, never by a third or more) and symptoms, not a percentage, as the gate".
What it is good for: which runner characteristics are graded risk markers, separately for low- and higher-volume runners, and that prediction models are not usable. What it does not show: that any factor causes injury, that changing a factor (losing weight, running more) prevents injury, anything about elite, trail or ultra runners, or any injury-specific risk (the outcome is any lower-extremity running injury).
Limitations
No pooling, few studies per factor. Clinical heterogeneity (populations, injury definitions, how factors were reported) ruled out meta-analysis; most factors rest on 2-3 studies, and single studies start at low quality under the authors' GRADE rules.
The short-distance results are mostly novices. About two-thirds of the studies are short-distance and many of those are novice runners; the authors say the factors may be specific to short-distance novices. Two studies sat near the 20 km cut-off, and the cut-off itself is "arguable" (the authors' word).
Self-reported, mixed injury definitions; "previous injury" was defined differently across studies (from missed practice with no timeframe to a running injury in the 12 months before an event).
Internal inconsistencies (our reading of the full text): Table 5 lists "Previous RRI (generic)" among the significant risk factors for short-distance runners at moderate quality, while the body text grades it moderate-quality evidence for not being a risk factor.Male sex appears as a risk factor in the abstract, highlights and conclusion and in Table 5 (3 studies, 3,580 participants), but the body text has no section on it, so its direction cannot be checked in the main text.Table 3's row for Messier 2018 repeats the age, sex split and BMI of Nappier 2018 (36.3 ± 8.4 years, 0/74, 22.7 ± 2.5), and the risk-factor labels of the two Kluitenberg papers in Table 2 appear swapped relative to their titles. The methods sentence "the evidence was downgraded when more than 75% of the participants were in low RoB studies" reads inverted as printed.
Effect sizes are small where named: "strong risk factors" in the abstract means the best-graded ones; age carried a hazard ratio of 1.02 in its one significant study.
Funding and interests: no funding statement; no competing interests declared. No device maker, federation or data platform is named.
How this entry was checked and changed
Why this level
a systematic review whose included studies are all prospective cohorts with multivariable analysis and whose evidence is graded per factor; held below A because no pooling was possible (clinical heterogeneity of populations and injury definitions), most factors rest on 2-3 studies, and the strongest factor for long-distance runners rests on 3 studies of 922 runners, two at high risk of bias. Read from the full text (Europe PMC, CC BY-NC-ND 4.0); Supplementary Tables 1-8 (search, criteria, per-factor results, full GRADE table, risk models) not read
Full text (checked 2026-10-28)
29 studies with a total of 18,853 participants were included in this review; 25 studies presented risk factors (single factor studies) and 4 studies presented risk models (full text)
Short-distance (mean running distance ≤20 km/week and ≤10 km/session) or long-distance (mean running distance >20 km/week and >10 km/session) (full text)
A previous running-related injury was the strongest risk factor for an injury for long-distance runners, with moderate-quality evidence; Table 5: Previous RRI (generic), 3 studies, 922 participants (full text)
Two studies (low RoB) found that a previous injury (musculoskeletal complaint) not attributed to running is a risk factor (high quality evidence) for new RRIs in short-distance runners; Table 5: 2 studies, 2705 participants (full text)
We found moderate quality evidence (downgraded for inconsistency) that previous RRIs is not a risk factor (short-distance runners) (full text)
We found moderate quality evidence (downgraded for study limitations) for running frequency not being a risk factor for RRIs (full text)
One study (medium RoB) found a weekly volume of more than 60 min to be a protective factor, and the other study (high RoB) found a weekly volume of less than 2 h to be a risk factor for injuries (full text)
1 study (low RoB) found a higher running distance (>30 km/week) to be a risk factor for RRIs. We found low quality evidence (downgraded for inconsistency and imprecision) for running distance as a risk factor (full text)
Because all models were in the derivation stage, they were graded low quality; There is no evidence that these models are predictive for RRIs (full text)
Previous injury (yes/no): OR = 4.3 (95%CI: 2.9-6.1), weekly distance OR = 0.98 (95%CI: 0.97-0.99), AUC = 0.68 (95%CI: 0.64-0.72) in marathon runners (full text)
The authors declare that they have no competing interests (full text)