The present review is the first to systematically review the literature on the incidence rate of running-related injuries in different types of runners. The weighted estimate of 17.8 (95 % CI 16.7–19.1) running-related injuries per 1000 h of running in novice runners was significantly greater than the incidence rate of 7.7 (95 % CI 6.9–8.7) in recreational runners. One study reported the incidence of running-related injuries in ultra-marathon runners as 7.2 per 1000 h . In track and field athletes, two studies reported the incidences of running-related injuries from 2.5 to 26.3 per 1000 h [19, 22]. In the latter, track and field athletes were subdivided into sprinters, middle-distance runners, and long-distance runners, which may be relevant as the reported running-related injury incidence per 1000 h was greater in sprinters and middle-distance runners than in long-distance runners .
In Fig. 2, a summary of the results in different types of runners is presented. The healthy participant effect may play a role when grouping novice versus recreational runners . In novice runners, the five studies are heterogeneous since the estimates reported by three of the studies [11, 20, 24] range from 30.1 to 33 and are significantly higher than those reported by the two remaining studies [25, 26]. A possible explanation for the discrepancy is the follow-up time in the respective studies. The non-injured runners accumulate relatively more exposure time in studies with a long follow-up, while the injured runners are censored. This will mathematically explain the overall decrease in running-related injuries per 1000 h of running in studies with longer follow-up amongst novice runners. The two studies with the lowest incidence of running-related injuries per 1000 h of running had 81 and 52 weeks of follow-up, while the three studies with the greatest injury incidence had follow-up periods of 8–13 weeks (Table 1).
The link between a relatively short follow-up time and a high incidence rate of running-related injuries versus long follow-up time and a lower incidence rate of running-related injuries indicates the possibility that runners classified as novice runners at the beginning of a study may reasonably be classified as recreational runners as time passes. If novice runners exceed 8–13 weeks without injury, they may well have adapted to running and face a lower injury risk after this period, even though they may spend more time running. Novice runners exceeding 8–13 weeks’ follow-up may then be considered as recreational runners instead. Based on this, it may be appropriate to identify a cut-off distinguishing a novice runner from a recreational runner.
In contrast, the injury incidences are homogeneous in recreational runners and the weighted estimate is unaffected by bias.
The strengths of the present review are mainly the systematic search of the literature and the use of meta-analyses to compare the injury incidences. The searches were performed thoroughly in five databases, in cooperation with a certified librarian. Moreover, all reference lists of the included full-text articles were checked for additional studies and, to the authors’ knowledge, one article  was also able to be included for analysis, although it was not indexed in any of the five databases searched. Evaluation of the quality of all articles presenting estimates of running-related injuries per 1000 h was accomplished and meta-analyses on these data were conducted. Thus, the present systematic review and meta-analyses represent rigorous evaluations and provide estimates of running-related injury incidences in novice runners, recreational runners, ultra-marathon runners, and track and field athletes.
The present study has a number of limitations, including differences in definitions of injury, definition of type of runner, and outcome measures used. First, definition of injury varies considerably across studies. Eight studies used time-loss definitions, but even within this definition there is a lack of consensus of the amount of time needed to classify time loss from running as a running-related injury. One study did not define the amount of time , some studies used 1 day in their definition [11, 14, 15], while other studies used 1 week [19, 20, 22, 24]. The only study  solely defining injury as the need for medical attention was reporting on ultra-marathon runners, and as these data were collected in real time while the runners participated in the ultra-marathon, this method was reasonable. No studies exclusively defined a running-related injury as physical pain alone, but, in four studies, physical pain was incorporated as part of the injury definition [16, 22, 25, 28]. Second, runners from the included studies were classified into four groups according to the type of runner, enabling relevant intergroup comparison. No exact definition of each category was made, but the baseline characteristics leading to grouping in one of the four types of runners are listed in Table 1. Third, the method of gathering data on exposure time may be questionable. In many studies, runners were asked to self-report their training exposure in web-based running diaries. This approach may lead to training hours or distance being estimated wrongly, possibly because of recall bias and time spent self-reporting . The quality assessment tool accounted for this, and awarded no star when exposure was registered by written self-report (item 5). However, it is questionable whether the risk of bias was, in reality, higher in the study by Bovens et al. , which received no star in item 5 because running exposure was collected in diaries, than in the study by Benell et al. , in which a star was awarded for a retrospective personal interview completed by one of the researchers at the end of the 12 months of follow-up. Lack of agreement in the way exposure time was calculated was another challenge. In some studies [11, 14, 15, 19, 23–25, 28], exposure time was calculated from the time a participant started the running programme until the time they reported a running-related injury (injured runners) or until the end of the programme (non-injured runners). This way of calculating exposure time was ideal due to that fact that the same runner could only contribute exposure time as long as he had not been injured. Thus, the risk of registering the same injury twice, if re-occurring, was avoided. Additionally, an injured person could not add exposure time after the injury occurred, and the number of injuries would be the same as the number of injured runners. Other studies did not mention whether study participants were censored if an injury occurred [12, 20, 26]. Further, some studies specified the premise that the same runner was included and was contributing exposure time, if running was resumed after an injury occurrence [21, 22]. Due to the varying ways of calculating exposure time in the included studies, the most appropriate comparison of the incidence of running-related injuries across all included studies was to use the total number of registered injuries instead of the total number of injured runners. This approach made it possible for one runner to figure twice or more in the pooled count of injuries. However, it would have been preferable if all studies had used the ideal method of calculating exposure time since this would have meant that one single runner could not accumulate exposure time after a first-time injury and have a recurrent injury counted twice.
Of the 13 studies providing estimates on running-related injuries per 1000 h of running, not all provided raw data on exact exposure time or 95 % CIs of the reported estimates. Corresponding authors from the respective articles [12, 14, 15, 26] were contacted, and data were received from Malisoux et al. , Theisen et al.  and Nielsen et al. . Moreover, the estimate of 30.1 running-related injuries per 1000 h used in the meta-analysis relating to novice runners derives from the complete study population of runners in the prospective study of Buist et al. . Overall, 155 of these 629 runners were described as runners already participating in running at baseline, running a mean of 1.2 h per week. Unfortunately, we were unable to obtain data that allowed us to calculate estimates for each of the groups of runners separately. Consequently, we decided to include the estimate of 30.1 running-related injuries per 1000 h in the category of novice runners; therefore, the true incidence of running-related injuries in novice runners might be even higher.
The present study constitutes a thorough and fully updated literature review presenting data regarding the incidence rates of running-related injuries, and outlining relevant key issues, which limit the comparison of studies in running-related injury research. The included meta-analyses form new estimates showing variations in the incidence rates of running-related injuries among different types of runners, and can be used as a starting point in future running-related injury research.