Original Research

CrossFit and the Epidemiology of

Musculoskeletal Injuries
A Prospective 12-Week Cohort Study
Paulo Roberto de Queiroz Szeles,*† MD, MSc, Taline Santos da Costa,† MD,
Ronaldo Alves da Cunha,† PT, MSc, Luiz Hespanhol,‡§k PT, PhD,
Alberto de Castro Pochini,† MD, PhD, Leonardo Addeo Ramos,† MD, PhD,
and Moises Cohen,† MD, PhD
Investigation performed at Sports Medicine Division, Department of Orthopedics and
Traumatology, Universidade Federal de São Paulo (UNIFESP), Escola Paulista de Medicina, São
Paulo, Brazil

Background: Despite the increasing international popularity of CrossFit, there is a paucity of scientific evidence on the risk of
CrossFit-related musculoskeletal injuries (CRMIs).
Purpose: To investigate the incidence (cumulative incidence proportion [IP] and incidence density [ID]) of CRMIs and the asso-
ciation of CRMIs with personal and training characteristics.
Study Design: Descriptive epidemiology study.
Methods: A prospective, 12-week descriptive epidemiology cohort study was conducted in a convenience sample of CrossFit
facilities in a single Brazilian city. Printed baseline questionnaires were distributed to 13 CrossFit boxes. All participants who filled
out the questionnaire and consented to participate in the study were invited to respond to an online follow-up questionnaire every 2
weeks to collect data on CrossFit training characteristics and CRMIs. A CRMI was defined as any self-reported musculoskeletal
injury or pain that prevented an athlete from exercising for at least 1 day. The IP was defined as the number of new cases divided by
the entire population at risk, while the ID was defined as new events divided by the total person-time exposure in hours. Logistic
mixed models were developed to investigate the association of CRMIs with personal and training characteristics.
Results: A total of 515 CrossFit participants filled out the baseline questionnaire and provided informed consent, and 406 (78.8%)
completed at least 1 follow-up measure. There were 133 participants who reported at least 1 CRMI during the study, and a total of
247 unique and new CRMIs were reported over a total estimated person-time exposure to CrossFit of 13,041 hours. The IP was
32.8% (95% CI, 28.4%-37.5%). The ID was 18.9 (95% CI, 16.6-21.3) per 1000 hours of CrossFit exposure. The shoulders (19.0%; n
¼ 47) and lumbar spine (15.0%; n ¼ 37) were most affected. Muscle injuries (45.3%; n ¼ 112) and joint pain (24.7%; n ¼ 61) were
the most common CRMI types reported. Switching between prescribed and scaled down training loads (odds ratio [OR], 3.5 [95%
CI, 1.7-7.3]) and previous injuries (OR, 3.2 [95% CI, 1.4-7.7]) were risk factors for a CRMI, while CrossFit experience was identified
as a protective factor (OR, 0.7 [95% CI, 0.5-1.0]).
Conclusion: In this 12-week prospective study, the ID was 18.9 CRMIs per 1000 hours of exposure; switching between training
loads and previous injuries was associated with 3.5- and 3.2-fold higher odds, respectively, of sustaining CRMIs.
Keywords: sports injury; high-intensity interval training; epidemiology; risk factors; longitudinal studies

Physical benefits related to high-intensity interval train- high-intensity intermittent exercise sessions designed to
ing8,24 and a sense of community have led CrossFit to an improve fitness and health.16,31 One of the most distinctive
unprecedented level of global popularity.39 CrossFit is a elements of CrossFit is that its exercises are multifaceted;
strength and conditioning exercise program with con- that is, activities can include weight lifting, gymnastics,
stantly varied multiarticular functional movements and running, biking, plyometric training, and rowing.16,31
Part of the popularity of CrossFit is its claim that the
The Orthopaedic Journal of Sports Medicine, 8(3), 2325967120908884
exercise program is suitable for everyone: professional ath-
DOI: 10.1177/2325967120908884 letes, pregnant women, the elderly, teenagers, and people
ª The Author(s) 2020 with disabilities.15,16 All participants, called “athletes”

This open-access article is published and distributed under the Creative Commons Attribution - NonCommercial - No Derivatives License (https://creativecommons.org/
licenses/by-nc-nd/4.0/), which permits the noncommercial use, distribution, and reproduction of the article in any medium, provided the original author and source are
credited. You may not alter, transform, or build upon this article without the permission of the Author(s). For article reuse guidelines, please visit SAGE’s website at
http://www.sagepub.com/journals-permissions.

1
2 Szeles et al The Orthopaedic Journal of Sports Medicine

according to CrossFit philosophy, share the same exercise measures. The study was conducted in 2016 and 2017
space, called a “box,” independent of their skill level. Each under the supervision of sports medicine researchers from
box’s head coach designs the workout sequence and inten- a local university. We selected participants from CrossFit
sity using his or her methodology and personal experience, boxes in a single metropolitan area of Brazil after obtain-
and then, individual class coaches help athletes execute ing ethical approval from the relevant local ethics
these workouts, offering additional assistance as needed.16 committee.
However, adaptations and assistance are not always car-
ried out in a standardized way, meaning that while each
box follows the same principles, athletes’ particular work- Population Sample, Setting, and Eligibility Criteria
outs vary according to their perceived skill level, the box
that they attend, and the head coach or class coach with This study population was made up of a convenience sam-
whom they exercise at a given moment. ple of general CrossFit athletes in a single metropolitan
As CrossFit has been gaining in popularity, so has the area of Brazil. Adults aged 18 years and who participated
scientific evidence on this exercise program.2 While several in physical exercise at CrossFit boxes were eligible to par-
authors have demonstrated the benefits of practicing ticipate in this study. Participants were not eligible for this
CrossFit,10,29,31 others have found a disproportionate mus- study if they provided incomplete data on the baseline ques-
culoskeletal injury risk. 3 The prevalence of CrossFit- tionnaire or if they reported an existing injury at baseline.
related musculoskeletal injuries (CRMIs) has been found The informed consent form and research team contact
to range from 19.4% to 73.5% in periods of time between 6 information were attached to the printed baseline question-
and 18 months, while the incidence of CRMIs has been naire. All athletes who voluntarily signed the informed con-
estimated to vary from 2.1 to 3.1 injuries per 1000 hours sent form agreeing to participate in this study and who met
of exposure.12,17,26-28,34,39 However, most of these previous all eligibility criteria were included in this study. Partici-
studies used a retrospective design, which hampers conclu- pants could withdraw from the study at any time either by
sions on incidence estimates.17,26,27,34,38 Prospective stud- communicating their desire to research staff or by neglect-
ies are considered the most appropriate design for studying ing to fill out the follow-up questionnaires.
the incidence of health-related conditions,32,40 including
sports injuries.18,37 In addition, a systematic review has
found that existing evidence on CRMIs is of low quality Data Collection
because of the studies’ high risk of bias.7
All included participants received a paper baseline ques-
We designed and conducted this study after becoming
tionnaire in the CrossFit box where they were training. The
aware of the low-quality evidence on CRMIs and the pau-
baseline questionnaire collected personal variables (ie, age,
city of prospective studies on the epidemiology of CRMIs,
sex, weight, and height), exercise variables (ie, CrossFit
observing a number of CRMIs in our clinical practice, and
experience, coaching variation, use of protective equip-
learning about CrossFit workout variations in greater
ment, preventive exercises, stretching exercises, competi-
detail. The objectives of this study were (1) to investigate
the incidence rates (cumulative incidence proportion [IP] tions, demonstration of the proper form to perform CrossFit
and incidence density [ID]) of CRMIs and (2) to examine exercises, and practice of other sports), and data on previ-
the association between CRMIs and personal and training ous injuries.
characteristics of CrossFit athletes in a metropolitan area An online questionnaire was then administered to collect
of Brazil. follow-up data on time-dependent variables (ie, hours of
CrossFit exposure and type of workload) and the emergence
of new CRMIs and their characteristics (ie, type, body loca-
tion, symptoms, and missing CrossFit training days
METHODS
because of a CRMI). The follow-up questionnaires were
Study Design developed and administered with an online tool (Survey-
Monkey [www.surveymonkey.com]). A total of 6 follow-up
This article presents the results of a prospective, 12-week questionnaires were sent to all included participants, once
observational cohort study with 6 biweekly repeated every 2 weeks, during the 12-week follow-up period.

*Address correspondence to Paulo Roberto de Queiroz Szeles, MD, MSc, Department of Orthopedics and Traumatology and Postgraduate Program in
Sports Science, Federal University of São Paulo, Rua Estado de Israel 656, São Paulo, SP, Brazil (email: drpauloroberto@clinicaszeles.com.br).
†
Department of Orthopedics and Traumatology and Postgraduate Program in Sports Science, Federal University of São Paulo, São Paulo, Brazil.
‡
Master’s and Doctoral Programs in Physical Therapy, City University of São Paulo, São Paulo, Brazil.
§
Department of Public and Occupational Health, Amsterdam Public Health Research Institute, VU University Medical Center Amsterdam, Amsterdam, the
Netherlands.
k
Amsterdam Collaboration on Health and Safety in Sports, Amsterdam Movement Sciences, VU University Medical Center Amsterdam, Amsterdam, the
Netherlands.
Final revision submitted November 11, 2019; accepted December 3, 2019.
The authors declared that there are no conflicts of interest in the authorship and publication of this contribution. AOSSM checks author disclosures
against the Open Payments Database (OPD). AOSSM has not conducted an independent investigation on the OPD and disclaims any liability or responsibility
relating thereto.
Ethical approval for this study was obtained from the Federal University of São Paulo (reference No. CAAE 1.753.942).
The Orthopaedic Journal of Sports Medicine Incidence of CrossFit Musculoskeletal Injuries 3

CRMI Definition and Classification Statistical Analysis

A CRMI was defined as any musculoskeletal injury or pain Descriptive analyses were conducted to summarize the
(in joints, bones, ligaments, tendons, or muscles) that pre- characteristics of the sample. Results were presented as
vented an athlete from exercising for at least 1 day. This the mean and 95% confidence interval (CI) for numeric
time-loss injury definition is similar to that in a previously variables presenting a normal distribution. The median
reported study on CrossFit.17 Injuries were self-reported and interquartile range (IQR) were used to summarize
and classified by (1) body location (head, back, shoulders, numeric variables not presenting a normal distribution.
elbows, wrists, hands, hips, thighs, knees, legs, ankles, feet, The normal distribution was investigated through an
or other location); (2) type (concussions, cramps, cuts, frac- inspection of histograms and probability distribution func-
tures, muscle soreness, sprains, tendinitis, or other); (3) tions. Frequencies and percentages were used to summa-
severity, defined as the number of consecutive missed rize dichotomous and categorical data. Linear and linear
training days (mild: 1-3 days; moderate: 4-7 days; severe: probability mixed models were conducted to summarize the
>7 days); and (4) pain severity, measured using the 11- follow-up measures data for numeric and dichotomous vari-
point Numeric Rating Scale (NRS), ranging from 0 (no pain) ables, respectively, to account for the dependency within
to 10 (severe pain). participants’ observations.36 Results were expressed as
percentages and presented as weighted means for linear
models and as mean probabilities for linear probability
CrossFit Exercise Sessions models.
The incidence of CRMIs was analyzed using 2 measures:
All exercise sessions were planned by the corresponding IP and ID. The IP was defined as the number of new cases
box’s head coach and were carried out in CrossFit boxes. (participants reporting new CRMIs) divided by the entire
Each CrossFit session was estimated to last, on average, population at risk and reported as a percentage.14 The ID
60 minutes, and 2 levels of training loads were offered: as was defined as new events (number of CRMIs) divided by
prescribed (“Rx”), meaning that the workout weights and the total person-time exposure to CrossFit exercises in
movements should be performed as prescribed, or “scaled,” hours and reported as the number of CRMIs per 1000 hours
referring to weights and movements scaled down by the of CrossFit exposure.18 Furthermore, 95% CIs were esti-
head coach for those not able to perform the Rx workout. mated for all measures to provide greater information
This study used the concept of “training load” as a compos- about the uncertainty inherent in each calculation.
ite measure of exercise frequency, duration, and intensity. Logistic mixed models were developed to investigate the
Therefore, the categories of Rx and scaled workouts were association of CRMIs with personal and training charac-
used to measure external training loads in our partici- teristics. The dichotomous CRMI variable for each time
pants. In CrossFit, workout loads (ie, Rx or scaled) are point’s measure was used as the dependent variable, and
sex-specific, meaning that the actual absolute load is dif- a time-lag technique was applied to ensure that the time-
ferent between sexes. dependent predictors (ie, independent follow-up training
variables in the model) happened in the biweekly period
before the CRMI report (ie, onset).14,20,35 A study identifi-
Primary and Secondary Outcomes cation code for each participant, the CrossFit training box,
and year of data collection (ie, 2016 or 2017) were included
The primary outcome of this study was the incidence of
as random effects to account for the repeated nature of the
CRMIs (measured as both the IP and ID), while the second-
follow-up data and possible CrossFit box and time-period
ary outcomes were characteristics hypothesized to be asso-
cluster effects. There were 2 models constructed: (1) a full
ciated with CRMIs. Characteristics statistically associated
model including all a priori established predictors deemed
with higher odds of developing CRMIs were considered risk
by us to be relevant for the CRMI risk analysis and (2) a
factors, while those statistically associated with lower odds
final model using backward variable selection and the
of developing CRMIs were considered protective factors.
Bayesian information criterion to remove or maintain the
Before running the association analysis, we discussed the
variables in the model. Results were presented as odds
rationale and relevance of including each collected variable
ratios (ORs) and their 95% CIs. All analyses were con-
in the model. The consensual variables were age, sex, ducted in Excel (Excel for Mac 2011, Version 14.7.3;
weight, previous injuries, CrossFit experience, training Microsoft) and R 3.5.0 (R Foundation for Statistical
workload (Rx, scaled, or alternating Rx/scaled), stretching Computing).
exercises, preventive exercises, use of protective equip-
ment, coaching variation, demonstration of the proper form
to perform CrossFit exercises, competitions, and practice of
RESULTS
other sports. Exercises related to injury prevention
included any warm-up exercises, core strength training Participants and Response Rate
exercises, or rotator cuff strengthening exercises performed
in a CrossFit box. Protective equipment included any exter- A total of 800 baseline questionnaires were physically dis-
nal equipment used by participants, such as gloves, knee tributed to 13 CrossFit boxes in the selected city, where
braces, back support, or wrist wraps. coaches had been familiarized with the study design and
4 Szeles et al The Orthopaedic Journal of Sports Medicine

TABLE 1
800 baseline quesonnaires/
Baseline Characteristics of CrossFit Athletes (N ¼ 406)a
informed consent forms
printed out and le at 13 Value
CrossFit® boxes
285 baseline quesonnaires/ Age, mean (95% CI), y 32.1 (31.4-32.8)
informed consent forms Sex, n (%)
Male 198 (48.8)
never filled out
Female 208 (51.2)
515 baseline quesonnaires/ Weight, mean (95% CI), kg 74.3 (72.9-75.7)
informed consent forms Height, mean (95% CI), m 1.7 (1.7-1.7)
filled out with at least some Body mass index, n (%)
informaon Normal (18.5-24.9 kg/m2) 206 (50.7)
46 baseline quesonnaires/ Overweight (25.0-29.9 kg/m2) 158 (38.9)
informed consent forms Obese (30.0 kg/m2) 42 (10.3)
filled out incompletely and CrossFit experience, median (IQR), y 1.0 (0.5-1.8)
discarded Coaching variation, n (%)
469 baseline quesonnaires/ Always the same coach 174 (42.9)
Coach with assistant 94 (23.2)
informed consent forms
Alternating coaches 138 (34.0)
filled out completely
Protective equipment, n (%)
Yes 336 (82.8)
17 injured at baseline and No 70 (17.2)
excluded Preventive exercises, n (%)
Yes 294 (72.4)
452 entered prospecve No 112 (27.6)
study to receive biweekly Stretching exercises, n (%)
email quesonnaires on Yes 229 (56.4)
exposure and injury 46 excluded from analysis No 177 (43.6)
Competitions, n (%)
39 did not respond to a
Yes 181 (44.6)
single quesonnaire on
No 225 (55.4)
exposure and injury Received a demonstration of the proper form to
406 CrossFit athletes entered 7 withdrew consent perform CrossFit exercises, n (%)
final sample for analysis Yes 229 (56.4)
No 177 (43.6)
Practice of other sports, n (%)
Figure 1. Flow diagram of recruitment, inclusion/exclusion, Yes 200 (49.3)
and final sample composition. No 206 (50.7)
Previous injury, n (%)
Yes 57 (14.0)
No 349 (86.0)
had agreed to participate. Of those 800 paper baseline ques-
tionnaires, 515 were filled out and returned by the same a
IQR, interquartile range.
number of participants (64.4%). Of the 515 who returned
the baseline questionnaire, 63 (12.2%) participants did not
meet the eligibility criteria (46 [8.9%] returned question-
The sex distribution was balanced in our sample: 198
naires with incomplete data, and 17 [3.3%] already had
men (48.8%) and 208 women (51.2%). Participants reported
injuries at baseline). The remaining 452 (87.8%) of the
a mean age of 32.1 years (95% CI, 31.4-32.8), a mean weight
baseline responders who signed the informed consent form
of 74.3 kg (95% CI, 72.9-75.7), and a mean height of 1.7 m
and met all eligibility criteria were included in this study.
(95% CI, 1.7-1.7). The median experience exercising at a
From the 452 participants, 46 were excluded from final
CrossFit box was 1.0 year (IQR, 0.5-1.8). A breakdown of
analyses (dropout rate of 10.2%): 39 (8.6%) because they
all baseline results can be found in Table 1.
did not return a single follow-up questionnaire and 7
(1.5%) because they withdrew consent during the study.
Therefore, the final sample included in the analyses was CrossFit Exposure
composed of 406 CrossFit athletes (78.8% of those who filled
out the informed consent form). The response rate of the The mean exposure to CrossFit was 3.9 days per week (95%
406 participants was, on average, 68.1% (IQR, 54.9%- CI, 3.8-4.0) (Table 2). Most participants (51.5% [95% CI,
83.9%), meaning that each participant returned about 4 46.7%-56.4%]) had a scaled training load, followed by alter-
of 6 biweekly follow-up questionnaires, on average. The nating Rx and scaled (27.1% [95% CI, 22.8%-31.5%]), while
recruitment process, application of inclusion/exclusion cri- the lowest proportion exercised consistently with an Rx
teria, and makeup of the final sample size are detailed in training load (21.3% [95% CI, 17.3%-25.3%]). Each CrossFit
Figure 1. training session is supposed to last approximately 1 hour.
The Orthopaedic Journal of Sports Medicine Incidence of CrossFit Musculoskeletal Injuries 5

TABLE 2 Shoulder 19.0

CrossFit Exposure and CrossFit-Related Musculoskeletal Lumbar spine 15.0
Injuries During Follow-upa
Knee 11.7
Weighted
Leg 7.7
Mean (95% CI)
Hip/thigh 7.7
CrossFit exposure, d/wk 3.9 (3.8-4.0)
Training load, % Wrist/hand 7.7
Rx 21.3 (17.3-25.3) Elbow 6.5
Scaled 51.5 (46.7-56.4)
Alternating Rx/scaled 27.1 (22.8-31.5) Arm 6.1
Cumulative incidence proportion (n ¼ 133/406), % 32.8 (28.4-37.5) Other 5.7
Sought medical attention, % 15.5 (12.3-19.4)
Incidence density,b injuries/1000 h 18.9 (16.6-21.3) Ankle/foot 3.6
Sought medical attention, injuries/1000 h 8.3 (6.7-9.8)
Thoracic spine 3.6
Injury severity
Pain (0-10 Numeric Rating Scale) 5.0 (4.7-5.3) Forearm 2.8
Missed CrossFit sessions, d/injury/participant 2.7 (2.3-3.1)
Cervical spine 2.8
Sought medical attention, % 40.2 (32.8-47.6)
Physician, % 24.3 (17.9-30.6) 0 5 10 15 20
Physical therapist, % 16.4 (10.9-21.8)
Percentage of Injuries Present at Each Anatomic Site
a
Results were obtained from mixed models, except for “missed
CrossFit sessions.” Rx, prescribed. Figure 2. Anatomic sites most affected by CrossFit-related
b
The total number of CrossFit-related musculoskeletal injuries musculoskeletal injuries.
was 247, and the estimated total person-time exposure to CrossFit
was 13,041 hours.
Muscular injuries 45.34

The total person-time exposure to CrossFit was 13,041 ses- Joint pain 24.7
sions, equivalent to approximately 13,041 hours.
Tendinopathies 12.96

CRMI Incidence and Severity Bruises 5.26

A total of 133 of the 406 participants analyzed reported at Shin splints 4.05
least 1 CRMI during this 12-week prospective study. There-
Dislocaons 3.64
fore, the IP of CRMIs was 32.8% (95% CI, 28.4%-37.5%)
(Table 2). There were 247 unique and new CRMIs reported Clinical symptoms 1.62
during this study over a total estimated person-time expo-
sure to CrossFit of 13,041 hours. Therefore, the estimated Abrasion/skin injury 1.21
ID was 18.9 (95% CI, 16.6-21.3) CRMIs per 1000 hours.
These CRMIs presented a mean pain level of 5.0 (95% CI, Stress fracture 0.81
4.7-5.3) on a 0-to-10 NRS, and 40.2% (95% CI, 32.8%-47.6%)
of the CRMIs required medical attention (Table 2). Plantar fasciis 0.4
Shoulders (19.0%; n ¼ 47), the lumbar spine (15.0%; 0 10 20 30 40 50
n ¼ 37), and knees (11.7%; n ¼ 29) were the most affected
Percentage of Each Type of Injury Reported
body locations (Figure 2). Muscle injuries (45.3%; n ¼ 112),
joint pain (24.7%; n ¼ 61), and tendinopathies (13.0%; n ¼
32) were the most frequent types of CRMIs (Figure 3). Figure 3. Most common types of CrossFit-related musculo-
skeletal injuries.

CRMI Association Analysis

DISCUSSION
The analysis associating CRMIs with personal and training
characteristics is presented in Table 3. The final model CRMI Incidence
yielded alternating Rx and scaled training loads (OR, 3.5
[95% CI, 1.7-7.3]) and previous injuries (OR, 3.2 [95% CI, The IP reported in our study (ie, 32.8%) lies within the
1.4-7.7]) as risk factors for a CRMI, while a 1-year increase range of those in previous reports: similar to the IP
in CrossFit experience (OR, 0.7 [95% CI, 0.5-1.0]) was found reported by Feito et al9 (30.5%), higher than that in Wei-
to be a protective factor against CRMIs. The associations of senthal et al38 (19.4%) and Montalvo et al27 (26.0%), but
CRMIs with the other characteristics investigated were not lower than that in Mehrab et al26 (56.1%) and Hak et al17
statistically significant. (73.5%). Different study designs, methods, source
6 Szeles et al The Orthopaedic Journal of Sports Medicine

TABLE 3 sports injuries as traditional methods. In addition, we

Association Between CrossFit-Related Musculoskeletal suggested a more inclusive CRMI definition (ie, any mus-
Injuries and Personal and Training Characteristicsa culoskeletal injury or pain that prevented an athlete
from exercising for at least 1 day) compared with the
Full Model Final Model ones used by Weisenthal et al38 and Mehrab et al26 (ie,
Intercept 0.01 (0.00-0.03) 0.04 (0.02-0.08) >7 missed days, training adaptation for >2 weeks, or a
Training load need for medical attention). We believe that identifying a
Scaled 1 1 higher number of minor or overuse injuries may be
Rx 2.2 (0.9-5.1) 2.4 (1.0-5.7) advantageous for secondary sports injury prevention, as
Alternating Rx/scaled 3.6 (1.8-7.4)b 3.5 (1.7-7.3)b it may create a possibility for early identification,
Exposure to CrossFit 0.9 (0.8-1.0) — thereby reducing the risk of an injury progressing.
CrossFit experience 0.8 (0.6-1.1) 0.7 (0.5-1.0)b
Weight 1.0 (1.0-1.0)b —
Previous injury 3.0 (1.3-6.9)b 3.2 (1.4-7.7)b
Risk and Protective Factors for a CRMI
Age 0.998 (0.996-0.999)b —
In this study, alternating between Rx and scaled training
Sex
Male 1 — loads and previous injuries were found to be risk factors for
Female 1.5 (0.8-2.9) — a CRMI. We suggest that alternating training load
Coaching variation intensities—in this case, switching between Rx and scaled
Always the same coach 1 — workouts—could be explained by a lack of skills in
Coach with assistant 0.6 (0.3-1.3) — maintaining Rx intensities or a lack of awareness of the
Alternating coaches 0.7 (0.4-1.4) — athletes’ limitations. CrossFit experience may mitigate the
Protective equipment 1.7 (0.7-4.3) — first argument, and indeed, we saw that more CrossFit expe-
Stretching exercises 2.0 (0.5-7.3) — rience reduced the odds of sustaining CRMIs. Regarding the
Demonstration of 0.30 (0.07-0.97)b —
second argument, recognizing limitations and adjusting
proper form
exercises accordingly are not simple endeavors, as both ath-
Preventive exercises 0.7 (0.3-1.5) —
Practice of other sports 0.9 (0.5-1.7) — letes and coaches must not only adapt weight loads, but also
dynamically adapt, vary, and increase the intensity, fre-
a
Data are shown as odds ratio (95% CI). Results were obtained quency, and duration of workouts to achieve athletes’ goals.
from logistic mixed models. Dashes signify that variables were not Interestingly, while alternating training loads was iden-
included in the final model. Rx, prescribed. tified as a risk factor for CRMIs, athletes training consis-
b
Statistically significant. tently with Rx or scaled workouts were not significantly
different in terms of the CRMI risk in this study. It is likely
that alternating training loads, especially increasing from
populations, and/or contexts may explain these discrepan-
scaled to Rx, increases the number of spikes (ie, acute
cies. For instance, the data in the previously mentioned
increases) in the training load, while maintaining the train-
studies were collected retrospectively, which, according to
ing load category may result in more gradual progress
the literature, may bias interpretations and conclusions
within that specific category. Evidence has shown that
surrounding incidence.18,37 Another possibility is differ-
spikes in the training load may indeed increase the risk
ences between boxes and styles of training. While CrossFit of sports injuries,23,25 which may partly explain why alter-
boxes are not franchises per se, boxes are affiliates in a nating training loads can result in a higher CRMI risk.
confederation, providing the same branded fitness regimen. Most CrossFit novices initiate their training with scaled
However, all boxes receive the same training course, and (or even lower) training loads. We are not suggesting that
there is no evidence that there are marked differences they should not aim to progress to Rx training loads but
between Brazilian and American boxes in terms of how rather that they should pay extra attention during the
they are run or the implications for these findings. transition period in terms of injury prevention.
The ID found in our study (18.9 CRMIs per 1000 hours There is strong evidence that previous injuries increase
[95% CI, 16.6-21.3]) was almost 10 times higher than the risk of future injuries in several sports.1,13,14,33 Our
most previous reports (all per 1000 hours of exposure): results suggested that CrossFit is no exception: in our
2.4, 2.3, and 2.1.12,27,28 We hypothesized that this dis- study, reporting a previous injury was associated with
crepancy might be explained by differences in the meth- about 3-fold higher odds of sustaining a CRMI. This esti-
ods and by the definition of injury used. We used a mate is consistent with those reported by Chachula et al5
longitudinal surveillance method with repeated mea- and Moran et al.28 There are 3 possible hypotheses for this
sures that should have reduced the probability of under- finding: (1) scar tissue, (2) inappropriate acute/chronic load
reporting minor or overuse CRMIs by minimizing recall balance recovery, or (3) diagnostic/treatment factors. Sev-
bias. Given that prospective studies are typically more eral authors have argued that scar tissue can contribute to
sensitive, studies using similar approaches have also future muscular imbalance, a reduction in flexibility, and
found higher sports injury rates. 6,21,22 For example, mechanical or functional instability.11 Others have hypoth-
Clarsen et al6 reported that a surveillance system based esized that athletes may inadequately balance acute and
on a longitudinal and repeated prospective measures chronic loads because of time spent away from exercise pro-
design was able to capture more than 10 times as many grams.4,30 We also hypothesize that several of the previous
The Orthopaedic Journal of Sports Medicine Incidence of CrossFit Musculoskeletal Injuries 7

injuries reported in our study could have remained undiag- participate having different characteristics from those who
nosed and untreated until the participants reported them decided not to participate. Self-reported data may be prone
in our online surveillance system. In this case, certain to detection bias, especially when these data include infor-
CRMIs measured during this study could have been mation on health conditions, such as sports injuries. There-
relapses or sequelae of previous injuries. fore, another limitation of this study was the self-reported
We found that a 1-year increase in CrossFit experience method employed to collect the data: there was no medical
reduced the odds of sustaining a CRMI by approximately assessment to diagnose the CRMIs officially. Because of the
half. Our results contradicted the findings of Montalvo large geographic area coupled with a short period of
et al,27 who found that more CrossFit experience (in years) repeated measures, in-person visits by health care profes-
was associated with higher odds of sustaining a CRMI. sionals was not possible with our resources. Our strategy to
However, the study of Montalvo et al27 was retrospective, minimize this bias as much as possible was to develop a
which could have introduced bias in their analysis. It seems multipronged strategy to classify the injured athletes. For
reasonable to assume that more experienced athletes man- instance, athletes were questioned about their CRMIs and
age loads and the injury risk better than inexperienced also asked to describe how many days of training were
ones.19 This should, however, be confirmed with data, as missed to corroborate that the injury actually led to missed
it is also entirely possible that CrossFit athletes who have training. Also, the researchers responsible for cleaning the
been injured or who are prone to injuries have already data checked each response, cross-checking the CRMI
dropped out by 1 year. Therefore, the discrepancy found information with the type and body region reported, to
between our study and the literature in this regard may identify possible inconsistencies. Finally, we recognize that
provide an opportunity for future studies. 12 weeks may be suboptimal to measure CRMIs; now that
we have identified the variables that most affect the CRMI
Strengths incidence, we intend to implement a future study with
these variables and much longer follow-up times. A longer
The main strengths and novelties of this study included (1) study duration may also allow us to draw temporal conclu-
a prospective and repeated measures design and analysis sions around our current finding that more CrossFit expe-
and (2) collecting and monitoring CrossFit-specific charac- rience is associated with a lower risk. This longer study
teristics. This study’s prospective nature and repeated period could clarify if injuries decrease in the same athlete
measures provided an opportunity to collect and monitor over time or if those who sustain more CRMIs abandon the
data with a lower risk of bias, such as recall bias, an issue exercise altogether.
reported in a recent systematic review on the topic.7 Inves-
tigating CrossFit-specific characteristics, such as CrossFit Implications for Practice
experience, training load (ie, Rx and scaled), preventive
exercises, use of protective equipment, coaching variation, Maintaining a physically active lifestyle is beneficial for
and participation in competitions, is relevant and impor- health. We believe that the popularity of CrossFit creates
tant to better describe, report, understand, and explain the an opportunity to engage people in physical fitness or keep
CrossFit practice and its association with the CRMI risk. them physically active, helping to achieve health benefits.
However, no intervention comes without risks, and we hope
Limitations that the results of this study may help the community bet-
ter understand the musculoskeletal risk associated with
The limitations of this study were mainly related to the use CrossFit. We do not believe that this risk surpasses the
of a convenience sample and the self-reported nature of the health benefits that can be achieved with CrossFit, but a
study. A random selection from the source population is the better understanding of the risks associated with CrossFit
recommended method for selecting a study sample. How- may help to implement a safer exercise program. In addi-
ever, epidemiological studies, such as ours, may face tion, we believe that specific information on CrossFit char-
recruitment challenges that make the proper execution of acteristics, a novel aspect of this study, may be useful for
these studies all but impossible. We did not have control future endeavors in developing and implementing tailored
over or access to the entire CrossFit population in our city. CRMI prevention strategies.
Therefore, drawing a random sample from this entire popu- Before concluding, we wish to further contextualize this
lation would not be possible. Our strategy to overcome this study and its motivations. This study was conducted by
issue and reduce the probability of having a nonrepresen- orthopaedic surgeons, many of whom are enthusiastic par-
tative sample was to invite all CrossFit boxes we could con- ticipants in CrossFit and began to see higher injury rates in
tact and include and monitor the highest number of both their daily clinical and exercise practice than those
participants possible using an online tool. We do not have rates reported in the literature. This study was, therefore,
reason to believe that our sample was not representative of designed to better understand the reality of CrossFit ath-
the CrossFit population in our city, but we are aware that letes in Brazil and not to evaluate the exercise program or
our study could have been affected by this possible selection make a value judgment on its philosophy. We believe that
bias. Additionally, given that we left the printed baseline CrossFit provides an important and interesting set of func-
questionnaires/informed consent forms out in the boxes for tional exercises, as well as a community setting in which to
athletes to choose at will, we recognize that there may have exercise with like-minded people interested in maintaining
been response bias caused by those who decided to a healthy lifestyle. We also believe that orthopaedic
8 Szeles et al The Orthopaedic Journal of Sports Medicine

surgeons and physical therapists should have access to modalities: a pilot study. J Sports Med Phys Fitness. 2017;57(9):
clear and transparent data on CRMIs to better advise their 1227-1234.
11. Fuller CW, Ekstrand J, Junge A, et al. Consensus statement on injury
patients and provide preventive counseling. This study
definitions and data collection procedures in studies of football (soc-
should, therefore, be interpreted in this light: as an objec- cer) injuries. Br J Sports Med. 2006;40(3):193-201.
tive presentation of the true injury experience of CrossFit 12. Giordano B, Weisenthal B. Prevalence and incidence rates are
athletes in several boxes of the São Paulo greater metropol- not the same: response. Orthop J Sports Med. 2014;2(7):
itan area in Brazil. 2325967114543261.
13. Girometti R, De Candia A, Sbuelz M, et al. Supraspinatus tendon US
morphology in basketball players: correlation with main pathologic
models of secondary impingement syndrome in young overhead
CONCLUSION
athletes. Preliminary report. Radiol Med. 2006;111(1):42-52.
14. Giroto N, Hespanhol Junior LC, Gomes MR, et al. Incidence and risk
In an urban Latin American population, about 1 in 3 Cross-
factors of injuries in Brazilian elite handball players: a prospective
Fit athletes sustained CRMIs in a 12-week training period. cohort study. Scand J Med Sci Sports. 2017;27(2):195-202.
The estimated ID was 18.9 CRMIs per 1000 hours of Cross- 15. Glassman G.The purpose of CrossFit: parts 1-2. CrossFit Journal.
Fit exposure (95% CI, 16.6-21.3). CRMIs presented, on 2011. http://journal.crossfit.com/2011/05/afpurposeofcf.tpl. Accessed
average, low to mild severity in terms of pain and training October 20, 2016.
time loss. CrossFit participants with alternating Rx and 16. Glassman G. Understanding CrossFit. CrossFit Journal. 2007. http://
library.crossfit.com/free/pdf/CFJ_56-07_Understanding.pdf.
scaled training loads had 3.5-fold higher odds of sustaining
Accessed October 20, 2016.
CRMIs than those training with scaled loads. Similarly, 17. Hak PT, Hodzovic E, Hickey B. The nature and prevalence of injury
those with previous injuries had 3.2-fold higher odds of during CrossFit training [published online November 22, 2013]. J
sustaining CRMIs. On the other hand, the odds of sustain- Strength Cond Res. doi:10.1519/JSC.0000000000000318.
ing a CRMI by those who have practiced CrossFit for at 18. Hespanhol Junior LC, Barboza SD, van Mechelen W, et al. Measuring
least 1 year were about half that of novice athletes. sports injuries on the pitch: a guide to use in practice. Braz J Phys
Ther. 2015;19(5):369-380.
19. Hespanhol Junior LC, Costa LO, Carvalho AC, et al. A description of
training characteristics and its association with previous musculo-
ACKNOWLEDGMENT skeletal injuries in recreational runners: a cross-sectional study. Braz
J Phys Ther. 2012;16(1):46-53.
The authors thank Elena Atkinson for her assistance in 20. Hespanhol Junior LC, Pena Costa LO, Lopes AD. Previous injuries
translating, editing, and formatting this work for and some training characteristics predict running-related injuries in
publication. recreational runners: a prospective cohort study. J Physiother. 2013;
59(4):263-269.
21. Hespanhol Junior LC, van Mechelen W, Postuma E, et al. Health and
REFERENCES economic burden of running-related injuries in runners training for an
event: a prospective cohort study. Scand J Med Sci Sports. 2016;
1. Agresta CE, Krieg K, Freehill MT. Risk factors for baseball-related arm 26(9):1091-1099.
injuries: a systematic review. Orthop J Sports Med. 2019;7(2): 22. Hespanhol Junior LC, van Mechelen W, Verhagen E. Health and eco-
2325967119825557. nomic burden of running-related injuries in Dutch trailrunners: a pro-
2. Beers E. Virtuosity goes viral. CrossFit Journal. 2014. http://library. spective cohort study. Sports Med. 2017;47(2):367-377.
crossfit.com/free/pdf/CFJ_2014_06_10000_Beers3.pdf. Accessed 23. Hulin BT, Gabbett TJ, Blanch P, et al. Spikes in acute workload are
October 20, 2016. associated with increased injury risk in elite cricket fast bowlers. Br J
3. Bergeron MF, Nindl BC, Deuster PA, et al. Consortium for Health and Sports Med. 2014;48(8):708-712.
Military Performance and American College of Sports Medicine con- 24. Hussain SR, Macaluso A, Pearson SJ. High-intensity interval training
sensus paper on extreme conditioning programs in military personnel. versus moderate-intensity continuous training in the prevention/
Curr Sports Med Rep. 2011;10(6):383-389. management of cardiovascular disease. Cardiol Rev. 2016;24(6):
4. Blanch P, Gabbett TJ. Has the athlete trained enough to return to play 273-281.
safely? The acute:chronic workload ratio permits clinicians to quantify 25. Malone S, Roe M, Doran DA, et al. Protection against spikes in work-
a player’s risk of subsequent injury. Br J Sports Med. 2016;50(8): load with aerobic fitness and playing experience: the role of the acute:
471-475. chronic workload ratio on injury risk in elite Gaelic football. Int J Sports
5. Chachula LA, Cameron KL, Svoboda SJ. Association of prior injury Physiol Perform. 2017;12(3):393-401.
with the report of new injuries sustained during CrossFit training. Athl 26. Mehrab M, de Vos RJ, Kraan GA, et al. Injury incidence and patterns
Train Sports Health Care. 2016;8(1):28-34. among Dutch CrossFit athletes. Orthop J Sports Med. 2017;5(12):
6. Clarsen B, Myklebust G, Bahr R. Development and validation of a new 2325967117745263.
method for the registration of overuse injuries in sports injury epide- 27. Montalvo AM, Shaefer H, Rodriguez B, et al. Retrospective injury
miology: the Oslo Sports Trauma Research Centre (OSTRC) Overuse epidemiology and risk factors for injury in CrossFit. J Sports Sci Med.
Injury Questionnaire. Br J Sports Med. 2013;47(8):495-502. 2017;16(1):53-59.
7. Claudino JG, Gabbett TJ, Bourgeois F, et al. CrossFit overview: sys- 28. Moran S, Booker H, Staines J, et al. Rates and risk factors of injury in
tematic review and meta-analysis. Sports Med Open. 2018;4(1):11. CrossFit: a prospective cohort study. J Sports Med Phys Fitness.
8. Eddolls WTB, McNarry MA, Stratton G, et al. High-intensity interval 2017;57(9):1147-1153.
training interventions in children and adolescents: a systematic 29. Murawska-Cialowicz E, Wojna J, Zuwala-Jagiello J. CrossFit training
review. Sports Med. 2017;47(11):2363-2374. changes brain-derived neurotrophic factor and irisin levels at rest,
9. Feito Y, Burrows EK, Tabb LP. A 4-year analysis of the incidence of after Wingate and progressive tests, and improves aerobic capacity
injuries among CrossFit-trained participants. Orthop J Sports Med. and body composition of young physically active men and women. J
2018;6(10):2325967118803100. Physiol Pharmacol. 2015;66(6):811-821.
10. Fisher J, Sales A, Carlson L, et al. A comparison of the motivational 30. Murray NB, Gabbett TJ, Townshend AD, et al. Calculating acute:
factors between CrossFit participants and other resistance exercise chronic workload ratios using exponentially weighted moving
The Orthopaedic Journal of Sports Medicine Incidence of CrossFit Musculoskeletal Injuries 9

averages provides a more sensitive indicator of injury likelihood than 36. Twisk JWR, Smidt N, de Vente W. Applied analysis of recurrent
rolling averages. Br J Sports Med. 2017;51(9):749-754. events: a practical overview. J Epidemiol Community Health. 2005;
31. Paine J, Uptgraft J, Wylie R. CrossFit study. Command and General 59(8):706-710.
Staff College. 2010. http://library.crossfit.com/free/pdf/CFJ_ 37. Verhagen E, van Mechelen W. Sports Injury Research. Oxford, UK:
USArmy_Study.pdf. Accessed November 17, 2016. Oxford University Press; 2010.
32. Samet JM, Muñoz A. Evolution of the cohort study. Epidemiol Rev. 38. Weisenthal BM, Beck CA, Maloney MD, et al. Injury rate and patterns
1998;20(1):1-14. among CrossFit athletes. Orthop J Sports Med. 2014;2(4):
33. Saragiotto BT, Yamato TP, Hespanhol Junior LC, et al. What are the 2325967114531177.
main risk factors for running-related injuries? Sports Med. 2014;44(8): 39. Whiteman-Sandland J, Hawkins J, Clayton D. The role of social cap-
1153-1163. ital and community belongingness for exercise adherence: an explor-
34. Sprey JW, Ferreira T, de Lima MV, et al. An epidemiological profile of atory study of the CrossFit gym model. J Health Psychol. 2018;23(12):
CrossFit athletes in Brazil. Orthop J Sports Med. 2016;4(8): 1545-1556.
2325967116663706. 40. Yach D, Botha JL. Epidemiological research methods, part V: follow-
35. Twisk JWR. Applied Longitudinal Data Analysis for Epidemiology. 2nd up studies. S Afr Med J. 1987;72(4):266-269.
ed. Cambridge, UK: Cambridge University Press; 2013.