Article Text
PDF
Abstract
Objective The purpose of this study was to identify patterns of clustering of the 10 health consequences identified in the Relative Energy Deficiency in Sport (RED-S) framework among collegiate female Cross-Country runners. We also assessed risk characteristics associated with each cluster.
Methods This randomly sampled population included 211 current National Collegiate Athletics Association (NCAA) Division I (DI) female cross country runners who completed a quantitative survey. We used latent class analysis (LCA) to group athletes into mutually exclusive classes based on shared response patterns of RED-S consequences. We computed descriptive statistics to identify demographics, personal characteristics, disordered eating and emotional health characteristics associated with each class.
Results The average age of the sample was 21 years with mean body mass index 20.4 kg/m2. The LCA identified three unique classes of potential RED-S presentations: (1) low probability of RED-S consequences; (2) complex physical and psychological concerns with a higher burden of cardiovascular concern and (3) very high probability of anxiety with high burden of menstrual disturbance, bone injury and gastrointestinal concern. All classes were characterised by high levels of menstrual disturbance and distinguished by the number and burden of other potential RED-S consequences and in reported abuse history, emotional regulation and perfectionism.
Conclusion This study identified a high burden of menstrual disturbance in NCAA D1 cross country runners, and three unique presentations of RED-S consequences. Future research is warranted to better understand how early prevention and intervention strategies may mitigate RED-S consequences in distance runners.
- female athlete triad syndrome
- female
- running
- physiology
Data availability statement
Data are available upon reasonable request. Data are available upon reasonable request to the first author (tlcars@umich.edu).
Statistics from Altmetric.com
WHAT IS ALREADY KNOWN ON THIS TOPIC
Collegiate female runners represent a high-risk population for Relative Energy Deficiency in Sport (RED-S), given the dual pressures of excelling at sport and academic work, as well as the high rates of disordered eating among college students. The RED-S framework provides an expansion to incorporate potential effects of low energy availability on additional aspects of physiological function including metabolic, haematological, immune, cardiovascular and psychological health, as well as growth and development. This study adds to the literature on different presentations of RED-S in National Collegiate Athletics Association (NCAA) Division I (DI) female cross country runners.
WHAT THIS STUDY ADDS
This study identified three unique presentations of potential RED-S consequences in NCAA DI female cross country runners. One class characterised by the lowest probabilities of RED-S consequences; the second class characterised by a complex set of physical and psychological concerns, most notably a higher burden of cardiovascular concern; and the third class characterised by a very high probability of anxiety with additional high burden of menstrual disturbance, bone injury and gastrointestinal concern. The three latent classes differed significantly with respect to abuse history, emotional regulation and perfectionism.
HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY
Understanding how RED-S may present differently is vital to understanding different approaches to treatment of RED-S; some athletes may be best suited for immediate medical care and testing, while others may benefit from psychological treatment in combination with indicated healthcare.
Introduction
In 1993, the American College of Sports Medicine published the first position paper on the state of female athlete health, identifying a constellation of conditions termed the Female Athlete Triad (Triad).1–3 The Triad consists of three conditions: low energy availability (EA), functional hypothalamic amenorrhoea and low bone mineral density (BMD). The Relative Energy Deficiency in Sport (RED-S) framework provides an expansion to incorporate potential effects of low EA on additional aspects of physiological function including metabolic, haematological, immune, cardiovascular and psychological health, as well as growth and development.4 Importantly, several components of the RED-S framework require further research to evaluate direct causative links with low EA, while the components of menstrual function and bone health are better understood from research on the Triad.5–14
Collegiate female athletes represent a high-risk population for RED-S, given the dual pressures of excelling at sport and academic work, as well as the high rates of disordered eating among college students.15 The epidemiology of the Triad and RED-S in collegiate female athletes is not well understood, given organisational limitations on accessing this population for research purposes. However, one longitudinal study assessed Triad outcomes among athletes at a large university (n=323) and found that 27% of these athletes reported menstrual disturbance, 6% met criteria for low BMD and 16% reported at least one stress fracture or bone stress reaction.16 That same study reported that the burden of stress fractures was as high as 34% among collegiate female cross country athletes.16 Further, in non-collegiate populations of female distance runners, there are reports of severe menstrual disturbance as high as 36%–69%.17 18 Elite collegiate runners are reported to experience bone stress injuries at a rate exceeding 20% per year.16 No other research to our knowledge has evaluated the burden of RED-S in National Collegiate Athletics Association (NCAA) collegiate female athletes, despite the large size of this at-risk population.
There are over 15 000 female collegiate cross country runners in all divisions of the NCAA, the non-profit organisation in the USA which manages and regulates collegiate-level sports. Schools within the NCAA are organised into divisions (D); the most prestigious programmes are in Division I (DI). There are approximately 6000 female cross country runners competing in NCAA DI each year.19 While RED-S is not unique to collegiate runners, these athletes represent a vulnerable population for RED-S, given the significant demands to succeed at sport and academics, in addition to factors that affect all females, such as pressures to meet the societal ‘thin’ ideal. Further, as this population of collegiate female runners is well-defined and contained within the regulated body of the NCAA, there is significant potential for early prevention and intervention. Given the lack of scientific understanding of RED-S in the college athlete population, the purpose of this study was to identify how the ten health consequences of the RED-S framework cluster together among NCAA DI female distance runners.
Methods
Study design and protocol
Participants, source population and probability sampling
This sample included 211 participants from a nationally representative, random sample of current NCAA DI female cross country runners from the Female Athlete Study of Health Trajectories (FASHT). To create this sample, a simple random sampling approach was implemented among current NCAA DI female cross country runners. The sampling frame included approximately 6000 female cross country runners from 352 teams and was constructed by collecting 2018–2019 roster lists, consisting of all athletes on each team, from all NCAA DI female cross country teams in the USA. Two teams did not have publicly available rosters and were not included in the sampling frame. Of those that had rosters available, five women from each team were randomly selected using a random number generator, resulting in a sample of 1750 women to be contacted via email and invited to participate in the study.
Of the 1750 sampled women, 881 (50.3%) had unavailable or invalid email addresses, collected from publicly available sources on the respective university’s websites. Of the remaining 869 sampled women who could be contacted via email, a total of 180 responses were received, resulting in a response rate of 21%. The sampling process was repeated, selecting five more women from each team when available, as some teams did not have five additional women on the roster. That second sample included an additional 1117 women, of whom 758 (68%) had unavailable/invalid email addresses. Of the 359 women contacted in the second sample, 42 responses were received (response rate 6%). A total of four follow-up emails were sent to each selected participant, in 2-week increments, to athletes who were non-responsive to prior contact attempts. Of the 222 total responses, data from 211 women who completed at least 40% of the survey were retained for analysis.
Eligibility criteria
To be eligible for the study, athletes had to be currently competing and/or have competed in at least one full season of cross country at an NCAA DI institution and be on the roster for the 2018–2019 season. A total of 372 women participated in the survey (21.2% response rate). Women were excluded from this analysis if they self-reported currently undergoing inpatient treatment for an eating disorder and/or other psychological disorder. Women who completed the demographic data section of the survey only were not excluded from this analysis (n=161). Women who started but did not complete the survey did not differ significantly in their demographic characteristics from women who were included in the final study sample (data not shown).
Survey overview
Self-reported survey data were collected using the Qualtrics online platform from 2018 to 2019 and captured biological, psychological and sociological factors related to health, sport, injury history, academics and EA; these factors were informed by a preceding qualitative study.20 This survey was piloted by members of the public for feedback.
Measures
RED-S indicator variables
Data from the FASHT self-reported survey were used to operationalise the 10 health consequences of the RED-S model, and our approach was modelled after that of Ackerman et al.21 The Ackerman survey constructs were designed to capture potential RED-S consequences and these were reviewed by experts for content validity.21 Cardiovascular health was measured using questions from the six-heart health section of the Pre-participation Examination—Fourth Edition.22 Menstrual dysfunction questions were adapted from the Study of Women’s Health Across the Nation, where menstrual dysfunction was characterised by a ‘yes’ to ‘Did you have primary amenorrhoea (menarche>age 15.0)’ or ‘Have you ever been unable to predict when your menstrual cycle will come?’), or a ‘no’ to ‘Are you currently getting regular periods?’ or ‘Have you had a period bleed about once a month since your first menstrual period?’. Poor fracture history was defined as self-reported ≥1 sport-related bone injury (ie, bone break, bruise, fracture, stress fracture, shin splints or other) in their sport history. Impaired gastrointestinal (GI) health was defined as a score of greater than 10 to the relevant questions from the Low Energy Availability in Females Questionnaire.23 Impaired haematological health was defined as a self-reported history of anaemia, low haemoglobin, iron or ferritin, and/or abnormal bruising. Endocrine dysfunction was defined as self-report of ever having an abnormal thyroid function test result.11 Metabolic dysfunction was defined as self-report of ever experiencing a low resting heart rate. Impaired growth and development were defined as self-report of ‘falling below normal growth curves during childhood, as indicated by doctor’.21 Poor immune health was defined as agreement or strong agreement to the statement: ‘I seem to get sick more often than others’.24 Psychological health was considered as anxiety and depression separately. Anxiety was measured via the Generalised Anxiety Disorder Scale 7 (GAD-7),25 and depression was measured by the Patient Health Questionnaire 8 (PHQ-8).26 Note that additional description of measures are available in online supplemental appendix.
Supplemental material
Demographics and personal characteristics
Individual level demographic characteristics included self-reported age, family socioeconomic status (SES) (ie, very poor, had enough but not extras, comfortable or well to do), sexual identity (ie, heterosexual, bisexual, lesbian, questioning, asexual or other), gender identity (ie, woman or other), height and weight (used to calculate body mass index as 
race (ie, white, black, Asian, American Indian, Native American/Pacific Islander, other or mixed), ethnicity (ie, Hispanic/Latina or non-Hispanic/Latina) and year in school. Current and past contraceptive use was defined by binary self-report.
Disordered eating and emotional health
Disordered eating was assessed using the Eating Disorder Inventory Version 3 (EDI-3) Drive for Thinness (range 0–28),27 28 and the Three Factor Eating Questionnaire (range 0–28) Cognitive Restraint Scale.29 Both provided a continuous measure of disordered eating, and higher scores indicated a greater level of disordered eating. Perfectionism was measured as a proxy for disordered eating, using the EDI-3 Perfectionism Scale (range 0–28), where higher scores indicated a greater level of perfectionism.27 28 Finally, Difficulties in Emotional Regulation Scale Emotion Regulation Impulse was reported (scale 0–15), where lower scores indicated lower emotional regulation.30
Analysis
Statistical analyses were completed using R Studio. Descriptive characteristics (means and SD for continuous variables and frequencies and percentages for categorical variables) were calculated for the full sample. Latent class analysis (LCA) was used to group athletes into mutually exclusive classes based on shared response patterns for the RED-S physical health variables. This data-driven and person-centred approach used maximum likelihood estimation to derive classes of individuals from the observed data set based on the response patterns, and estimates the prevalence of each class.16 Based on the estimated latent class (LC) model, each participant was assigned a predicted probability of membership in each class, and then assigned to the one LC for which she had the greatest probability of membership.19 The Bayesian Information Criterion (BIC), Akaike Information Criterion (AIC), maximum log-likelihood and subject matter considerations were used to guide the final choice of the number of classes that best fit the observed data (table 1), while maintaining stable minimum class sizes (eg, N>25).20 21 The conditional probabilities of each indicator variable for each class were categorised as low (0.00–0.19), medium-low (0.20–0.39), medium (0.40–0.59), medium-high (0.60–0.79) and high (0.80–1.00).31
Latent class analysis model fit statistics
Additionally, descriptive statistics were calculated for the individuals assigned to each class, including means and percentages for demographics, personal characteristics, disordered eating, and emotional health. The derived classes were formally compared in terms of these descriptive quantities using one-way analysis of variance and χ2 tests and assessed the assumptions of these statistical tests.
Results
The average age of the sample was 21 years, and all participants self-identified as a woman. The majority of respondents were white (87%), non-Hispanic (90%), heterosexual (93%) and of comfortable (58%) or well to do (23%) SES, consistent with a prior study.32 Most women in the sample were upper-level undergraduate students; only 13% were graduate or professional students (table 2).
Demographic, disordered eating and emotional health characteristics, and one-way analysis of variance and χ2 test for characteristics by latent class
Potential RED-S consequences were highly prevalent in the sample; 75% of women reported a bone injury, 58% reported haematological issues, 54% reported GI concerns, 53% reported menstrual dysfunction and 34% reported cardiovascular consequences. High anxiety, defined as the highest quartile of the GAD-7 scoring scale, was present in 53% of women, and the mean GAD-7 score was 13.68 (moderate anxiety). The average depression score on the PHQ-8 was 13.33 (moderately severe), and 24% of women fell into the highest scoring category for depression. As shown in table 3, other potential RED-S consequences were less common, ranging from 10% of women with reported endocrine consequences to 16% of women with reported impaired growth and development.
Comparison of RED-S health consequences by latent class
Latent class analysis
We used multiple criteria, the BIC, AIC, maximum log likelihood and subject matter consideration, to assess the best model fit, and we ultimately selected the model that did best across all of these criteria. The specified model fit criteria did not provide an agreement on best fit yet narrowed down our selection to a 2-class and 3-class model based on AIC and BIC. We used subject matter considerations when choosing between a 2-class and 3-class model; the 2-class model did not provide classes that were as unique and clinically distinguishable, based on subject matter, as compared with the 3-class model (table 3). The three classes identified in the LCA analysis are shown in table 3 and table 4.
Probability of RED-S consequence by latent class
Endocrine concerns were low and menstrual disturbances were high or medium-high across all three classes. Notable differences were present in the patterning of the other potential RED-S consequences across the LCs. Latent class 1 (LC1), the ‘Low RED-S Consequence Class’, represented 39% of participants and was characterised by low psychological concern and low concern on many physical aspects, but medium concern for GI, haematological and bone health. We considered LC1 the least severe RED-S class, despite notable menstrual health concerns.
Latent class 2 (LC2), the ‘High RED-S Consequence Class’, included 16% of women and was characterised by a complex set of physical health concerns across most domains, including high cardiovascular, haematological and menstrual concerns, as well as medium to medium-high psychological concerns and medium-high bone, and GI concerns. Growth and development and metabolic concerns were medium-low in this class, and there was low immune and endocrine concerns.
Latent class 3 (LC3), the ‘Anxious, High RED-S Consequence Class’, included 45% of the women and had a defining feature of high anxiety. Similar to LC1, LC3 had medium-high menstrual, bone and GI concerns, with other factors including medium haematological and depressive concerns, medium-low growth and development, and low immune, endocrine, cardiovascular and metabolic concerns.
In the overall test of class differences, there were statistically significant differences in class distributions for emotion regulation (p=0.008) and perfectionism (p<0.001). There were significant pairwise differences for emotion regulation (LC1>LC2, p≤0.001; LC1<LC3 p≤0.001; LC2<LC3, p≤0.001) and perfectionism (LC1<LC2, p≤0.001; LC2>LC3, p≤0.001). Further, there were significant pairwise differences for contraceptive use (LC1<LC2, p≤0.001; LC2 vs LC3, p≤0.001) and marginal differences for contraceptive use (LC1>LC3, p=0.085). Additionally, there were marginal differences for year in school (LC1<LC2, p=0.054) and racial diversity (LC1<.LC3, p=0.093). There were no significant pairwise differences between LCs for demographic variables, such as ethnicity, age or SES (table 2). However, LC3 was more ethnically diverse, relative to LC1 and LC2, and SES was marginally lower in LC2 compared with LC1.
Discussion
Overall, this study in a nationally representative sample of NCAA D1 female runners demonstrates that potential RED-S consequences are highly prevalent among these NCAA D1 athletes. Over 50% of women in this sample reported experiencing a bone injury, cardiovascular, haematological, or GI concern, menstrual dysfunction, and high symptoms of anxiety. Notably, when examining clustering of these RED-S concerns, all classes were characterised by high levels of menstrual dysfunction. However, classes differed by number and burden of other potential RED-S consequences, with one class additionally differentiated by the extremely high burden of anxiety complaints. While all three classes, reported low levels of endocrine, immune and metabolic concern, we hypothesise that endocrine and metabolic consequences may have been under-reported, as they require clinical assessments, and this study only obtained information on self-reported clinical assessments of endocrine and metabolic markers.
Although the ‘Low RED-S Consequence Class’ LC1 reported the lowest probabilities of RED-S consequences, they, like all classes, reported moderately high menstrual disturbance. Whether this group of women represents those inherently less affected by the potential RED-S consequences or whether women in this class are earlier in their progression to a fuller and more severe set of potential RED-S consequences observed among other classes could not be determined in this cross-sectional study. Future work is needed to understand whether women move from one LC to another over their career, and if so, what the implications of these changes are for female athletes’ overall health and well-being. The observed menstrual disturbance among this otherwise seemingly low burden class should not be ignored; identifying the root cause of menstrual disturbance is vital to ensure proper treatment.2 33
The ‘High RED-S Consequence Class’ LC2 is characterised by a complex set of physical and psychological concerns, most notably higher burden of cardiovascular concern than the other two LCs. Some research has reported impaired cardiovascular health in female athletes with low EA and amenorrhoea, including lower heart rates and systolic blood pressure compared with eumenorrheic athletes, early atherosclerosis, endothelial dysfunction and unfavourable lipid profiles.34–36 Low EA, in a non-athlete populations with anorexia nervosa, has been associated with severe cardiovascular concerns, including valve abnormalities, pericardial effusion, severe bradycardia, hypotension and arrhythmias.37 Additionally, haematological and menstrual concerns have been associated with low EA and disordered eating/eating disorders.21 38 Additional, yet less prominent, concerns in this class included anxiety, bone and GI complaints that may have been associated with low EA. However, the temporality of these consequences is unknown.
The ‘Anxious, High Red-S Consequence Class’ LC3 is characterised by very high probability of anxiety with additional high burden of the physiological symptoms of menstrual disturbance, bone injury and GI concern. Anxiety and psychological stress are associated with menstrual disturbance and hypothalamic amenorrhoea, in both athlete and non-athlete populations.4 39–42 Further, GI upset is associated with anxiety and life stress among runners43 and general adult populations44 45; however, reverse causation between anxiety and GI problems is possible in the current study. Clark and Mach46 highlighted the need for continued research to understand the specific effects of physical and psychological stress on GI distress during exercise.
Bone injury and menstrual disturbance were of significant concern in all three LCs. The harmful relation between menstrual disturbance and subsequent bone injury and loss of BMD has been well described.3 47 48 Bone injury may be a consequence of the nature of running mechanics and sport training load. The annual incidence rate of bone stress injury (BSI) is approximately 20% among elite collegiate runners, and a study at a large university found that female cross country runners had the highest incidence of bone injury of any university sport.49 50 However, that relationship may have been at least partially explained by low EA, as these studies found an increasing risk of bone injury as EA decreased.49 50
RED-S concerns should be evaluated and monitored, to safeguard athletes’ well-being. Further, it is important to ensure appropriate psychological health evaluations to consider disordered eating and eating disorders, and/or other psychological health concerns, that may be the underlying causes of RED-S consequences. Understanding how RED-S may present differently is vital to understanding different approaches to acute treatment of RED-S; some athletes may be best suited for immediate medical care and testing, while others may benefit from psychological treatment in combination with other medical care.
Strengths
This epidemiological study employs a national, random sample of NCAA DI cross country runners. This sampling approach makes the study findings more representative and generalisable to the population of NCAA DI cross country runners.
Limitations
There are important limitations to address. Small sample size, low power and precision, as well as sparse-data bias are notable limitations of this study, which has the potential to inflate the effect estimates.1 2 We adapted survey measures to capture RED-S constructs from Ackerman et al, and reliability and validity assessments were not performed on these measures; however, Ackerman et al notes that ‘the survey constructs used to capture RED-S consequents were reviewed by experts for content validity’.21 Notably, some RED-S variables were particularly challenging to capture in a self-reported survey, and the consequences of endocrine and metabolic outcomes generated particular concern. Those RED-S variables were likely under-reported in the current sample. Those consequences were not showing up as concerns in the LCs, but that may have been due more to issues with self-reported proxy measures for endocrine and metabolic health markers that are commonly collected in clinical blood samples, rather than lack of importance. For bone fractures, we did not have information about BSI versus non-BSI and the type of imaging used to define bone injury; collecting information about BSI versus non-BSI and type of imaging used is an important step for future research. Additionally, we did not have information about family history of osteoporosis. We also note that our definition of menstrual function status was non-time-specific and thus could represent menstrual irregularity anytime from puberty through college.
We recognise that additional psychometric work, to assess reliability and validity of these RED-S consequences, is an important next step for future research. Additionally, this study did not directly assess energy availability or collect constructs that may affect energy availability. We had a response rate of 21%, and it is quite possible the respondents were biased towards those with more potential RED-S consequences. The self-reported nature of the survey data collection, particularly for medical and health related information, was subject to desirability bias. The cross-sectional nature of the survey also limits our understanding of temporal relations between RED-S consequences.
Given our recruitment criteria which excluded any athletes currently receiving inpatient treatment for an eating disorder, our sample may be missing those athletes at highest risk for an eating disorder. Thus, our findings may not be generalisable to all athletes, and more work is required to understand the patterning of RED-S consequences across different at-risk athlete groups. Additionally, selection bias may have impacted the LCs we observed; we did not observe a true healthy class of athletes, and it is possible that women who would have been assigned to a true healthy class did not participate in the survey. Alternatively, we may not have captured women who would have been assigned to a more severe RED-S class, as compared with LC2 and LC3, if those women dropped out of the sport prior to the current study or were selected for the study but uninterested in participating. It is possible that women who had more severe RED-S consequences experienced bone injury and/or other career-ending RED-S consequences, such as psychological health impacts, that removed them from sport, and therefore, they were not captured in the current data. Further, we excluded women from this analysis if they self-reported currently undergoing inpatient treatment for eating disorder and/or other psychological disorders, and this may have resulted in a more conservative estimate of RED-S, if any effect, in the case that women excluded for inpatient psychological had comorbid RED-S consequences.
It is important that future studies improve methods of measuring self-reported RED-S consequences, particularly those that are not as apparent to an individual, unlike a broken bone or cessation of menses. Further, this study did not collect constructs that may affect energy availability, including food intake, exercise training (ie, running mileage, etc), dietary patterns, as well as disordered eating presentations beyond drive for restraint and thinness, and we suggest that these constructs be investigated in future research.
Conclusion
The current study identified three distinct LCs of potential RED-S consequences in this population of NCAA DI female cross country runners. The ‘Anxious, High RED-S Consequence Class’ highlights the need for further attention and treatment of psychological concerns in the population; the ‘High RED-S Consequence Class’ is characterised by significant cardiovascular concern and several additional physiological and psychological complaints; and ‘Low RED-S Consequence Class’ is characterised by moderate menstrual concern. Future research is needed to understand if (1) athletes can be accurately classified into different LCs in clinical practice and (2) if this classification allows for more targeted interventions and/or improved health outcomes.
Data availability statement
Data are available upon reasonable request. Data are available upon reasonable request to the first author (tlcars@umich.edu).
Ethics statements
Patient consent for publication
Ethics approval
This study involves human participants and was approved by University of Michigan IRB (HUM00161740). Participants gave informed consent to participate in the study before taking part. Electronic informed consent was collected prior to participation in the survey. All participants were provided with mental health resources.
Acknowledgments
I would like to thank Dr Kamryn Eddy, Dr Jenny Thomas and Dr Deb Franko for their mentorship and feedback on this project during my summer fellowship with the Eating Disorders and Clinical Research Programme at Massachusetts General Hospital. Second, thank you to Emily Zheutlin, Halimat Olaniyan and Cassie Gaskins for their assistance with creating the sampling frame for this study. Finally, my sincerest thanks to all the participants who shared their stories and made this research possible. This manuscript originated from the following dissertation: The Culture and Consequences of Low Energy Availability in National Collegiate Athletic Association Division One Female Distance Runners: A Mixed Methods Investigation, T. Carson et al. 2020.
References
Supplementary materials
Supplementary Data
This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.
Footnotes
Twitter @traci_Carson
Collaborators Thank Dr Kamryn Eddy, Dr Jenny Thomas and Dr Debra Franko with the Eating Disorders and Clinical Research Programme at Massachusetts General Hospital.
Contributors The first author is responsible for the overall content as the guarantor, conducted all analyses and was the sole writer of this paper. The subsequent authors provided essential guidance in the design of the study, as well as substantive feedback and edits to the final manuscript.
Funding The authors have not declared a specific grant for this research from any funding agency in the public, commercial or not-for-profit sectors.
Competing interests None declared.
Patient and public involvement Patients and/or the public were involved in the design, or conduct, or reporting, or dissemination plans of this research. Refer to the Methods section for further details.
Provenance and peer review Not commissioned; externally peer reviewed.
Supplemental material This content has been supplied by the author(s). It has not been vetted by BMJ Publishing Group Limited (BMJ) and may not have been peer-reviewed. Any opinions or recommendations discussed are solely those of the author(s) and are not endorsed by BMJ. BMJ disclaims all liability and responsibility arising from any reliance placed on the content. Where the content includes any translated material, BMJ does not warrant the accuracy and reliability of the translations (including but not limited to local regulations, clinical guidelines, terminology, drug names and drug dosages), and is not responsible for any error and/or omissions arising from translation and adaptation or otherwise.