Sex-Specific Differences in Running Injuries: A Systematic Review with Meta-Analysis and Meta-Regression

Two independent investigators (K.H. and C.E.) conducted a systematic literature search including articles from inception till April 2020. Prospective cohort studies and randomized controlled trials investigating healthy runners from different age groups were included. The search was restricted to articles from peer-reviewed journals published in English, German, or Spanish languages. Furthermore, studies had to report rates of running-related injuries for both sexes. Overall injury rates and injury rates for specific locations, diagnoses or injury mechanisms were considered. Included running disciplines were middle distance and long-distance track as well as cross-country, trail and road running. There was no restriction to a specific injury definition. Reviews, systematic reviews, commentaries, case studies, case series, cross-sectional studies, retrospective studies and interventional arms of randomised controlled trials (RCT) were excluded. For RCTs, only untreated control groups were considered.

The search strategy using specific keywords (running AND injur*) was applied to four different databases (PubMed, Web of Science, PEDro, SPORTDiscus). All databases were searched to identify relevant studies based on keywords, title and abstract. Two independent investigators (C.E. and K.H.) extracted relevant studies based on the inclusion criteria first by reading the title, the abstract and the full text, if available. A third reviewer (A.Z.) was available for consensus decisions. The bibliographical information of included articles was examined for further relevant references (backward search). A forward search was done via citation tracking using Web of Science® (Thomson Reuters).

Study characteristics (design, running discipline, population, age and number of participants) as well as prevalence and incidence rates for both sexes were extracted. For prevalence rates, number of injuries or number of injured runners were related to the number of runners investigated. For incidence rates, number of injuries and specific exposures (in hours, kilometres or athlete exposure) were used. An athlete exposure (AE) is defined as one athlete participating in one practice or competition [24]. When it was not possible to extract the data from an article for specific running distances (e.g., pooling of overall injuries for track disciplines), corresponding authors were contacted by email to obtain the data. If specific data were not able to be obtained, the respective study was included in the systematic review but not in subsequent analyses.

Due to insufficient study quality assessment tools in sports injury epidemiology, a new tool was developed by consensus of K.H., A.J., A.L.R., A.Z. and S.S. on the basis of previously used tools [20, 22, 25, 26]. The modification ensured that all relevant points regarding the quality of the study design and important content-related information would be considered—e.g., differences in methodological approaches such as competition or season, or the type of data collection.

This tool consisted of 15 items on recruitment, reporting, injury and exposure collection, injury definition and drop-out (Table 1).

The identified quality score was used to determine a high (above the median) or low (below the median) study quality of the studies investigated (median score was 18). Two independent reviewers (K.H., A.J.) with a third reviewer (A.L.R.) for consensus assessed the study quality of the included studies.

Publication bias was checked by visual inspection of funnel plots (log risk ratio against standard errors) and regression test for funnel plot asymmetry.

To compare injury risk between male and female runners, risk ratios (RR) with 95% confidence intervals (CI) were computed for each study. Meta-analytic pooling was done using a random-effects model (DerSimonian and Laird method [27]). Between-study heterogeneity was estimated using Cochran’s Q and I² statistics. To reveal potential publication biases, funnel plots were constructed if more than ten studies were available [28]. Besides visually checking their symmetry, Egger’s regression test was applied.

Following the calculation of pooled RRs, we used a mixed-effects meta-regression model to identify variables potentially affecting the outcome of the meta-analysis [27]. The choice of tested moderators (effect modifiers) was based on three criteria: (1) a plausible impact on the tested variables, (2) reporting in the included studies, (3) sufficient variation of the moderators’ values [29]. The following moderators were submitted into the meta-regression model: performance/expertise level (recreational: no competitions, competitive: participating in local competitions, elite: qualifying for national or international competitions); age (youth: < 18, adult: ≥ 18) competition distance (≤ 10 km, > 10 km); study quality (low: study quality score < 18, high: study quality score ≥ 18), training duration (low: < 7.5 h or high: ≥ 7.5 h/week), training mileage (low: < 64 km/week, high: ≥ 64/week). Moderator analyses were performed if ten or more studies were available [28]. If a significant moderator was detected, a subgroup analysis comparing the respective values of the moderator was performed using the meta-analytic procedures described above.

All calculations were performed using algorithms of the metaphor package embedded in R (R Foundations for Statistical Computing, Vienna, Austria) as well as the software JAMOVI [30] and OpenMeta [Analyst] software (OS X version 10.12 obtained from http://​www.​cebm.​brown.​edu/​openmeta/​).

The search returned 15,914 studies and 29 additional studies were identified through other sources. After removing 3699 duplicates and applying inclusion criteria, a total of 38 studies were considered eligible [7, 19, 24, 31‐65]. Thirty-one of them could be included in the quantitative analysis. Seven studies reported on the same data sets as other included studies and were excluded from the quantitative analysis [19, 31‐33, 39, 40, 65]. The full literature search process is displayed in Fig. 1.

Of the included studies, 36 reported injury data from 35,689 participants (40.8% female). Two studies reporting on injuries from the National Collegiate Athletic Association (NCAA) database did not state the number of athletes but did report the athlete exposure (242,244 athlete exposures with 46.7% females [24] and 276,207 athlete exposures with 50.7% females [59]). Most studies were prospective cohort studies (n = 37), while the control group (not receiving any intervention) from one randomised controlled study met the inclusion criteria [42]. Twenty-three studies investigated road runners, 11 track runners (middle and long distance), 10 cross-country runners and 3 studies reported on trail running/orienteering (Table 2). Studies from major competitions (European or World Championships) reported concurrently on track and road running (half or full marathon) [19, 31‐33, 41]. Regarding competition level, 18 studies reported on novice and recreational runners, 11 on competitive and 9 on elite runners. Study characteristics of all included studies are summarized in Table 2.

The two independent reviewers evaluating study quality agreed on 441 of 570 evaluated items (agreement = 77.4%). The scores for study quality ranged between 9 and 23 out of 24 points with a median of 18 and a mean ± SD of 16.8 ± 4.1. Most studies (> 90%) reported recruitment procedures, injury assessment and documented injury characteristics. Fewer studies achieved maximal points due to not recording individual exposure data (50.0% of maximal points) or exposure data at all (57.9% of maximal points) as well as not using a medical attention definition (56.1% of maximal points). The results of the study quality assessment are presented in Electronic Supplementary Material Table S1.

Except for one outlier [47], visual inspection of the funnel plot (Fig. 2) showed a symmetrical distribution of the log risk ratios and the regression test for funnel plot asymmetry (− 0.150; p= 0.881) suggested no indication of publication bias.

The overall injury rate was 20.4 (95% CI 19.7–21.1) injuries per 100 male runners and 20.8 (95% CI 19.9–21.7) injuries per 100 female runners. Meta-analytic pooling did not reveal differences between female and male runners’ injury rates per runner (n = 21; RR 0.99, 95% CI 0.90–1.10; p = 0.84; I² = 72.31) or per specific exposures (n = 6; RR 0.94, 95% CI 0.69–1.27;  p = 0.669; I² = 85.93) (Figs. 3 and 4). Due to the small number (n = 6) of studies reporting injuries per exposure (athlete exposures (n = 2) or hours (n = 4)), no aggregation of overall injury rates per specific exposures was performed.

Moderator analyses of injury RR rates per runner revealed an association of a higher injury risk in men and competition distances exceeding distances of 10 km (p = 0.002) (Table 3). Specifically, the subgroup meta-analysis of competition distance showed a significantly higher RR of 1.08 (95% CI: 1.04–1.39) for female runners with competition distances ≤ 10 km. For competition distances > 10 km, the comparison approached but failed to reach significance although the RR of 0.77 (95% CI: 0.58–1.02) was suggestive of a lower probability of injury in male runners (Fig. 5). No meta-regression was performed for specific injuries and moderators training duration or training mileage due to absence of more than ten studies reporting these variables [28].

Data for two specific running-related injuries were available for synthesis.

Despite pooling data from all available epidemiological studies, no differences in overall injury rates between female and male runners were found in this systematic review. This was the case for both studies reporting injuries per runner and injuries per specific exposures. The injury rates of 20.4 (male) and 20.8 (female) per 100 runners are in accordance with summaries of injury rates from the last three decades [4]. Nonetheless, these rates are at the lower spectrum of published injury rates that were reported to be up to 79.3% [66].

Injury rates depend on several factors that need to be taken into consideration, such as systematic factors (age, BMI), running-/training-related factors (training frequency, training and racing distance, experience, level of running, footwear, biomechanics), health factors (injury history) and lifestyle factors (drinking, smoking) [66‐70].

Not all of these factors were reported in each study and may vary between investigated populations. Therefore, the moderator analysis was incorporated into this study. Only competition distance was a statistically significant moderator for an increased risk of female runners compared to male runners when running competition distances of 10 km and shorter. Furthermore, the subanalysis revealed a tendency of increased injury risk for male runners for longer distances than 10 km. This is in accordance with the finding that male runners had a higher risk of sustaining injuries compared to female runners when running high mileages (> 64 km/week) [18]. While running higher mileages are associated with longer competition distances, this can only be used as an estimate for this discussion. Unfortunately, there was insufficient reporting of training load (time or mileage) in the included studies. For future studies reporting data on injury epidemiology or risk factors, it is strongly recommended to report the training load [71, 72].

Female runners had twofold higher risk of having a bone stress injury compared to male runners in this review. A bone stress injury is an injury pattern with known sex differences for epidemiology and risk factors [73]. Bone stress injuries are common running-related overuse injuries due to cumulative microtrauma to the bone [74]. Especially in younger ages, females seem to have a higher risk for bone stress injuries compared to male runners. For example, Changstrom et al. [24] reported a twofold risk and Plisky et al. [56] a 2.5-fold risk for female high school runners of sustaining a bone stress injury compared to male high school runners in cross-country. In older collegiate athletes, female cross-country runners were found to have 28.6 injuries per 100,000 athlete exposures (AE) compared to 16.4 injuries per 100,000 AE in males, representing a statistically significant rate ratio of 1.8 [59]. In outdoor track (100 m–1500 m), this difference was even higher (22.3 injuries/100,000AE for females and 7.2 injuries/100,000AE for males, risk ratio of 3.1) [59]. One possible explanation that has been discussed was the association of bone stress injuries with the female athlete triad (low energy availability, menstrual disturbance and low bone mineral density) to explain the higher risk for bone stress injuries in female runners [35, 75, 76]. However, while the term female athlete triad is used only for female athletes, the more current and more detailed concept of relative energy deficiency of sports (RED-S) has also been described for male athletes [77‐79]. Despite using the same initial treatment (activity modification, protected or non-weight bearing) of bone stress injuries for both sexes, the further treatment differs between female and male runners, depending on specific risk factors, such as elevated RED-S risk, biomechanics (load rates, hip adduction, rearfoot eversion), altered hormonal status or calcium and vitamin D intake [14, 15, 73, 80]. In summary, bone stress injuries are more prevalent in female runners and treatment/rehabilitation strategies should incorporate sex as an important variable. Nonetheless, in the prevention of bone stress injuries consideration of the sex would probably benefit from awareness of RED-S, including screening for low energy availability and low mineral bone density.

Data from two studies showed that male runners had almost twice the risk of having an Achilles tendinopathy as female runners [49, 51]. This is in accordance with a systematic review on the pathogenesis of Achilles tendinopathy [81]. The Achilles tendon transmits the generated forces from the gastrocnemius-soleus muscle complex and, thus, is an important tendon for propulsion during running. However, the Achilles tendon has a poor blood supply and, therefore, is prone to overuse injuries, such as a tendinopathy [81]. The lifetime prevalence has been reported as high as 40–50% in runners [13, 82] and a recent 1-year prospective study determined the incidence rate in a cohort of recreational runners to be 5.2% [51]. While the amount of loading is the key factor in the etiology of Achilles tendinopathy, there are several intrinsic (age, stress, genes, biomechanics, body composition) and extrinsic factors (footwear) modulating the risk for this injury [83]. Recent studies found biomechanical (footstrike pattern, ankle dorsiflexion moments) and training-related parameters (changes in training, cold weather, footwear, use of compression socks, mileage) as possible risk factors [10, 51, 84‐86]. This summary of (possible) risk factors does not directly explain the increased probability for male runners to have an Achilles tendinopathy. Therefore, we can only speculate about the possible mechanisms. One recently published study discusses the mechanism of the lifetime cumulated load (together with running years) which might be higher in male runners than in female runners [87]. Chronic loading needs to be taken into account when evaluating the risk for Achilles tendinopathies.

Another explanation might be found in the hormonal differences between women and men. For example, estrogen is associated with collagen synthesis and could therefore influence tendon healing capacity [88, 89]. Furthermore, estrogen deficiency has been reported to negatively affect tendon metabolism and healing [90]. Hormonal fluctuations that are typical for the menstrual cycle have not been associated with modifications of tendon function [90, 91]. A review summarizes that high or low levels of sexual hormones (estrogen, progesterone and testosterone) are not directly causing tendinopathies but may play a role in tendon pathologies [92]. Therefore, individual hormonal status should be taken into account for injury risk of female and male runners as well as for their therapies and prevention [92].

This was the first systematic review on sex-specific differences in running injuries incorporating a meta-regression analysis to determine moderating variables and shall be discussed in light of other systematic reviews on this topic.

This systematic review contrasts the findings of the systematic review by van der Worp et al. [18], who found female runners at a lower overall risk of sustaining an injury than male runners. This finding was particularly found in men under 40 years. However, when assessing the evidence level the authors called for caution in the interpretation of their findings since these were based on only five high-quality and one low-quality studies. In contrast, our review included epidemiological studies reporting injury rates separately for both sexes. With this approach, 26 studies were included and meta-analyses showed no sex differences for overall running injuries when calculated per runner or per exposure (hours or AE). Furthermore, we were able to conduct a meta-regression analysis showing a higher injury risk for female runners in competition distances of 10 km and shorter as well as a tendency for a higher injury risk for male runners in competition distances longer than 10 km. This is a new finding and in line with the increased risk for male runners with a high weekly mileage (> 64 km), which is typically needed for longer competition distances [18].

The systematic review by Wright et al. [93] found female sex to be a primary risk factor for lower extremity bone stress injuries despite conflicting evidence using an exploratory meta-analysis incorporating three etiological studies [6, 94, 95]. The meta-analysis found a similar 2.3-fold increased probability for female runners. Our meta-analysis supports these findings and underlines the evidence for female runners to be more prone to bone stress injuries based on five included prospective studies [24, 35, 49, 56, 59]. Female sex as a risk factor for medial tibial stress syndrome has also been described by a meta-analysis in active individuals (not exclusively runners [96]).

This systematic review summarised data from 38 prospective studies representing more than 35,689 participants (from 36 studies) and 518,000 athlete exposures (from 2 studies). While the distribution between female and male runners (40.8–50.7% females) was similar and no overall differences were found, breakdown of injury data regarding sex and according to location or diagnosis was only possible in six studies. Consequently, the available literature included in this systematic review did not allow conclusions on the sex-dependent epidemiology of pathologies other than bone stress injuries and Achilles tendinopathies.

The meta-regression approach of this study included several potential moderators. However, considering the multifactorial aetiology of running-related injuries, other confounding bias such as biomechanical or psychological variables may have influenced the injury risk. Another limitation was the moderate to high heterogeneity of studies included in the overall injury meta-analyses, emphasizing the need for further studies with a clear injury definition and uniform data collection methods [71, 97].

Regarding quality, future studies would benefit from documenting exposure data and using medical attention injury definitions. Furthermore, moderator analysis was only possible for 1 outcome (overall injuries per 100 runners) due to missing descriptive information on study populations (such as mileage, training duration, competition distances). As seen in Table 2, there are several different data collection methods applied and injury definitions used to determine a running injury. In accordance with recent consensus statements in injury epidemiology [71] and a Delphi consensus on running injuries [98], we encourage future running injury research to follow these guidelines to improve the homogeneity of studies. From this, future meta-analyses would benefit from comparing rates of injuries between studies [97, 99].