Peripheral scope (exercise-programming), but the outcome is patient-important: a lower-extremity stress fracture is a bone overuse injury that ends the training attempt for weeks to months (route-(c) harm). Stress fractures are «common overuse injuries with up to 95% occurring in the lower extremities» and account for «15-20% of all musculoskeletal injuries» among runners. (Wright et al., 2015) The decision this page answers: for a runner, what raises stress-fracture risk, by how much — and does that risk-factor set transport to someone who is not a runner (the obese novice; the military recruit)? The headline is a strong negative with two exceptions.
Headline conclusion: «there was a striking lack of conclusive evidence for any risk factor, but with two exceptions: previous history of stress fracture and female sex». (Wright et al., 2015) Only these two are «strongly supported by the data». (Wright et al., 2015)
Evidence-tier note (confidence: low). Single gold SR-MA (Wright 2016; tier gold,
meta-analysis), but the base is thin: «8 articles met the inclusion criteria; 7 were considered low
risk. 4 articles qualified for meta-analysis.» (Wright et al., 2015)
and only four independent cohorts underlie the eight studies. (Wright et al., 2015) Prospective cohorts only; runners defined as cross-country, track-and-field,
triathlon, duathlon, and recreational. The two pooled estimates below are the firm core (both I2=0%); the
rest is conflicting single-study evidence, which is why the page is held low overall.
The two supported risk factors — pooled estimates
Both are intrinsic / non-modifiable (the decision they drive is screening and vigilance, not behaviour change), and both were pooled by random-effects inverse-variance meta-analysis with zero measured heterogeneity:
| Risk factor | Pooled OR (95% CI) | Studies / N | I2 | Read |
|---|---|---|---|---|
| Previous stress fracture | 4.99 (2.91 to 8.56), p<0.001 | 3 / N=939 | 0% | ~5x risk; the strongest factor |
| Female sex | 2.31 (1.24 to 4.29), p=0.008 | 3 / N=480 | 0% | ~2.3x risk (distance runners) |
- Previous stress fracture (~5x). The individual studies ranged OR 2.90 to 6.36; the pool confirmed them. (Wright et al., 2015) A caveat the authors raise and cannot resolve: «it is unknown whether this represents new injuries or a previous stress fracture which was not fully healed» (Wright et al., 2015) — i.e. the predictor may be partly incomplete healing rather than a stable trait.
- Female sex (~2.3x). Significant on pooling, but the underlying studies conflicted (Yagi OR 3.10 significant; Bennell OR 1.08 and Tenforde OR 1.35 both null), and the mechanism is contested — sex per se is likely a proxy for the energy-availability / menstrual / bone-density cluster below rather than an independent cause (inferred from Wright et al., 2015). The pooled OR is reported bare of a located mechanism; treat it as a prognostic marker, not a causal lever.
The rest — conflicting or single-study, none pooled
Every other candidate was either conflicting across studies or rested on a single cohort. The effect sizes below are individual-study estimates, not pooled, and most carry wide CIs:
| Candidate | Direction / estimate (single studies) | Verdict |
|---|---|---|
| Training volume | >32 km/wk OR 2.40 (1.20-4.82); females >32 km/wk OR 3.08 (1.31-7.20); other study null | one study, threshold suspect (below) |
| Low bone density | spine BMD RR 1.89 (1.04-3.45); whole-body BMC RR 1.79 (1.02-3.13) | 1-2 studies, no OR poolable |
| Low BMI / lean mass | BMI<19 OR 2.76 (1.18-6.46); lean mass per-kg-decrease OR 1.14 (1.01-1.28) | single study each |
| Menstrual dysfunction | late menarche OR 5.30 (2.11-13.34), amenorrhoea OR 2.73 (1.16-6.4) in one study; null in two others | conflicting |
| Nutrition / calcium | protective (dairy, milk, calcium HRs ~0.3-0.6) in one cohort; harmful in another | conflicting |
| Oral contraceptives | ORs 0.55-2.22 across 4 studies | null — «no evidence» |
| Biomechanics | — | gap: no included study examined it |
- Training volume — the threshold is suspect. Only one study found a volume association, and the threshold is low: «the reported threshold for increased risk was a relatively low volume (32 km/week) in relation to the training regimens many runners undertake.» (Wright et al., 2015) The authors go further — the received wisdom is unsupported: «clinicians have little evidence to support the widespread anecdote that higher mileages are associated with increased risk of stress fracture, or the corresponding recommendation that reducing running volume can prevent stress fractures.» (Wright et al., 2015)
- Calcium supplement reverse-causation. Tenforde found calcium-supplement users at OR 3.36 (1.4-7.9) — but flagged that «supplementation was also strongly correlated to previous history of stress fracture», so injured athletes may have been prescribed calcium as treatment, not primary prevention. (Wright et al., 2015) A protective exposure read backwards.
- RED-S reframes the menstrual / nutrition cluster. The conflicting menstrual findings may reflect a stale paradigm: «Menstrual dysfunction may be considered simply as one group of signs which occurs secondary to the issue of ‘relative energy deficit in sports (RED-S)’» — with subclinical energy deficiency unmeasured in the risk-factor studies. (Wright et al., 2015) So low BMD, low BMI, late menarche, amenorrhoea and a protective-nutrition signal are plausibly one underlying construct (low energy availability) seen through different instruments, not independent levers (inferred from Wright et al., 2015).
The interdependence caveat — why single-variable ORs mislead here
The risk factors are confounded with one another, and adjustment collapses them. In the best-measured cohort, «some apparently significant risk factors (eg, training volume, body mass, racing performance) became non-significant when multivariate analysis was conducted». (Wright et al., 2015) This is the standard warning that a univariate OR (training volume, BMI) is likely capturing a shared upstream cause (energy availability, body composition), so the single-factor estimates in the table above should be read as associations entangled with each other, not as separable causal levers (inferred from Wright et al., 2015).
Transportability I — military data do not transfer to runners (within-paper)
Wright’s framing motive is a transportability failure the field commits routinely: most stress-fracture knowledge comes from military cohorts, and does not carry to runners. «Although military data are commonly extrapolated to the general running population, this is likely inappropriate due to major differences between these populations.» (Wright et al., 2015) The support-factor mismatches are concrete: «Studies involving military personnel include smokers and individuals with poor overall physical fitness, both of whom would not be expected in a population of serious runners», recruits run far less (<10 miles/wk vs runners’ 40-100+), carry heavy loads, wear different footwear, and fracture during a 16-week acclimation window unlike a veteran runner’s first fracture after years. (Wright et al., 2015) This is a worked instance of the fabric’s transportability rule — similar enough means similar in the support factors the mechanism needs, not in the label does running.
Transportability II — the obese / sedentary novice: the host-factor set INVERTS
(inferred from Wright et al., 2015) — the fabric’s own exercise-programming taper places the obese / deconditioned novice in the fabric zone for injury (where a wrong prescription causes the injury that ends the attempt). Wright’s risk-factor set must not be imported wholesale into that stratum, because its host factors run the opposite direction:
- Wright’s bone-stress host profile is low BMI (<19), low lean mass, low BMD, low energy availability, female-athlete-triad / RED-S — markers of a lean, often female, energy-deficient athlete.
- The obese novice is the mirror: high BMI, high mechanical loading. Wright states the relevant mechanism — «Mechanical loading is associated with increased bone mineral density … healthy bone mineral density requirements of active running individuals may be greater than that required by the general public». (Wright et al., 2015) Higher habitual body mass loads the skeleton, which raises BMD, so the obese novice is plausibly at lower bone-stress-fracture risk than the lean runner, not higher (inferred from Wright et al., 2015).
- Mechanism, not outcome — and net effect, not the intended one. This is a directional inference from Wright’s stated loading->BMD mechanism, not a measured outcome in the obese stratum (Wright studied runners; no obese-novice stress-fracture cohort is held). What would confirm or refute it: a prospective cohort of obese novices tracking bone stress vs soft-tissue injury by baseline BMI/BMD. And the net injury picture does not favour the obese novice: higher mass that protects bone simultaneously overloads tendon, plantar fascia and joints, so the binding lower-extremity injury constraint for this stratum is likely soft-tissue / joint, not bone stress -> Training Load Progression and Running Injury, Tendon Adaptation to Mechanical Loading, Knee Osteoarthritis and Modifiable Levers. So the decision-change is which injury to screen and program around: bone-stress screening (prior fracture, low BMI/BMD, menstrual/energy status) is for the lean runner; the obese novice’s prescription guards the soft tissue and the joint instead.
Parameter table — a DISTINCTION from Damsted, not a tension
Wright’s training volume -> stress fracture and Damsted’s load-change -> running-related injury look like the same question and are not. The matched parameters fail the same-quantity test on every row, so this is a distinction to record, never a tension to file (op-weave: no cell, no claim):
| Parameter | Wright (this page) | Damsted (Training Load Progression) | Same quantity? |
|---|---|---|---|
| Exposure | absolute weekly running volume (km/wk) + host factors | rate of change of weekly load (% ramp) | NO — level vs rate-of-change |
| Outcome | bone stress fracture (site-specific overuse) | any running-related injury (whole-body RRI) | NO — bone-specific vs general |
| Population | runners (cross-country/T&F/tri/duathlon/rec; often lean/female) | novice & amateur runners | partial overlap, not matched |
| Design | SR-MA, prospective cohorts, 2 factors pooled | SR, 4 heterogeneous studies, no pool | different synthesis depth |
The two are complementary, not opposed, and even agree where they touch: Wright explicitly leaves room for load change — «Rapid changes in training volume and intensity are often considered in the development of injuries, though this is difficult to quantify in epidemiological studies» (Wright et al., 2015) — which is exactly the exposure Damsted isolates. No clash to reconcile; they answer adjacent questions.
Synthesis — the training-load injury heuristics are evidentially unsupported (type-A)
(inferred from Wright et al., 2015) — read together, Wright and Damsted (two independent groups, different exposures, different outcomes) converge on a non-obvious meta-conclusion that neither states alone: the popular training-load rules for preventing running injury lack evidential support.
- Wright: the «widespread anecdote that higher mileages are associated with increased risk of stress fracture» and the «corresponding recommendation that reducing running volume can prevent stress fractures» have «little evidence». (Wright et al., 2015)
- Damsted: no evidence exists for the 10% rule (the ramp-rate heuristic), and the one direct test found 10% and 24% weekly increases equivalent (the verbatim finding and quote live on its page) -> Training Load Progression and Running Injury.
Two different load heuristics — one on absolute volume, one on rate of change — both turn out to be received wisdom without an evidence base. This is independent backing for a structural claim about the field (the heuristics are oversold), not for a single number. It does not license the opposite (train however you like): both SRs are insufficient-evidence verdicts (measurement error and small heterogeneous bases, not demonstrated safety), and the conservative gradual-progression default survives on mechanism and low-regret grounds, as argued on the Damsted page. The convergence sharpens what is unknown, not a permission.
Decision relevance
- Screen the lean runner on two things only: prior stress fracture (~5x) and the energy-availability cluster. A female runner with a prior stress fracture and clinical signs warrants early imaging; the low-BMI / low-BMD / menstrual-dysfunction markers are best read as one RED-S construct, not separate checkboxes.
- Do not transport this risk-factor set to the obese / sedentary novice. Their bone-stress risk is plausibly lower (higher mass -> higher BMD), and their binding injury constraint is soft-tissue / joint, not bone — so the prescription guards tendons and joints (gradual load, strength tolerance), not bone-stress screening.
- Do not treat cut your mileage as evidence-based stress-fracture prevention. The volume-threshold evidence is one study at a suspiciously low cutpoint, and the authors say the reduce-volume recommendation lacks support. (Volume is also a poor proxy for skeletal load — intensity and surface vary within a fixed mileage.)
- Do not transport military stress-fracture findings to runners (or vice versa) — different smoking, fitness, load-carriage, footwear and acclimation-window support factors.
- This is insufficient evidence on most factors, not reassurance. Biomechanics was unstudied; most factors conflict; only four cohorts underlie the field. The firm core is the two pooled ORs; everything else is a named gap.