Objective This systematic review aims to identify generic prognostic factors for disability and sick leave in subacute pain patients.
Setting General practice and other primary care facilities.
Participants Adults (>18 years) with a subacute (≤3-month) non-malignant pain condition. Eligibility criteria were cohort studies investigating the prediction of disability or long-term sick leave in adults with a subacute pain condition in a primary care setting. 19 studies were included, referring to a total of 6266 patients suffering from pain in the head, neck, back and shoulders.
Primary and secondary outcome measures The primary outcome was long-term disability (>3 months) due to a pain condition. The secondary outcome was sick leave, defined as ‘absence from work’ or ‘return-to-work’.
Results PubMed, EMBASE, CINAHL and PEDro databases were searched from 16 January 2003 to 16 January 2014. The quality of evidence was presented according to the GRADE WG recommendations. Several factors were found to be associated with disability at follow-up for at least two different pain symptoms. However, owing to insufficient studies, no generic risk factors for sick leave were identified.
Conclusions Multiple site pain, high pain severity, older age, baseline disability and longer pain duration were identified as potential prognostic factors for disability across pain sites. There was limited evidence that anxiety and depression were associated with disability in patients with subacute pain, indicating that these factors may not play as large a role as expected in developing disability due to a pain condition. Quality of evidence was moderate, low or very low, implying that confidence in the results is limited. Large prospective prognostic factor studies are needed with sufficient study populations and transparent reporting of all factors examined.
Trial registration number CRD42014008914.
- Prognostic factors
- Early detection
This is an Open Access article distributed in accordance with the Creative Commons Attribution Non Commercial (CC BY-NC 4.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited and the use is non-commercial. See: http://creativecommons.org/licenses/by-nc/4.0/
Statistics from Altmetric.com
Strengths and limitations of this study
This systematic review provides new knowledge on risk factors across pain sites, which may help physicians and researchers when initial referral decisions are made.
The review also provides a solid foundation for planning future high-quality studies on risk factors for poor outcomes in pain patients.
The protocol for the systematic review was registered beforehand in PROSPERO and reported according to the PRISMA statement, with the quality of the evidence judged as recommended by the GRADE Working Group.
Quality of evidence was moderate, low or very low, implying that confidence in the results is limited.
Pain is the most common reason patients consult general practice,1 and long-term disability and sick leave due to a pain condition are associated with huge negative consequences for the individual and for society.2 It would be both costly and unnecessary, however, to offer specialised treatment to all patients presenting in primary care with a pain condition; despite its frequency, pain is in most cases a temporary phenomenon.3 Still, a small group of patients will develop chronic or recurrent pain causing long-term disability and sick leave. It is estimated that approximately 3–10% of patients with acute pain develop a chronic pain condition.3 ,4 Chronic pain conditions are associated with social and family problems, loss of work, and loss of self-esteem and integrity.5–7 Moreover, chronic pain is often associated with other symptoms or comorbidities such as fatigue, concentration and memory problems, sleep disorders, depression and anxiety.5 Once pain has become chronic, treatment is complex and difficult.2 Thus, early identification of pain patients at high risk of developing long-term problems would offer a great opportunity for reducing cost and suffering associated with long-term disability and sick leave because optimal care could be initiated at an early stage.
Most pain research focuses on one specific pain site (eg, low back or shoulder pain8 ,9). As a result, prognostic factor research is normally conducted on each site separately.8 For example, substantial prognostic factor studies on back pain have been carried out, with several systematic reviews reporting prognostic factors for back pain.10–14 However, this single-site approach limits clinical applicability for the general practitioner (GP) because most pain patients have pain at more than one anatomical location.9 ,15 Factors that have predictive value across different pain sites (ie, generic factors) may exist,8 but few attempts have been made to explore prognostic factors across pain sites.8 ,16
This systematic review was conducted as part of a national Danish ‘Health Technology Assessment’ (HTA) aimed at identifying possibilities for early identification and timely treatment of pain patients across pain sites with relevance to a broad range of stakeholders in Denmark.17 The specific aim of the evidence synthesis was to identify potential factors for the development of long-term disability or sick leave in patients with subacute, non-malignant pain in primary care.
The review was conducted in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statement18 on the basis of a predefined protocol available from the International Prospective Register of Systematic Reviews (PROSPERO: CRD42014008914).
Data sources and searches
Studies were identified via a systematic literature search in the following databases: PubMed, EMBASE, CINAHL and PEDro. Additional studies were identified through experts and through a review of the included studies’ reference lists. The following search terms were used: ‘Pain’, ‘Prognosis’, ‘Predictor’, ‘Prognostic factor’, ‘Primary Health Care’, ‘General Practice’ and ‘Family Practice’. The search string tailored for the PubMed database is presented below: Search(((‘Pain’[Mesh]) OR ‘Chronic Pain’ [Mesh] OR ‘persistent pain’)) AND (((‘Prognosis’ [Mesh]) OR ‘Outcome Assessment (Health Care)’ [Mesh] OR predict* OR prognost* AND ((English[lang] OR Danish[lang] OR Norwegian[lang] OR Swedish[lang] AND adult[Mesh]))) AND (((‘Primary Health Care’ [Mesh]) OR ‘General Practice’ [Mesh]) OR ‘Family Practice’ [Mesh] OR GP OR ‘primary care’). Filters: Published in the past 10 years; English; Danish; Norwegian; Swedish; Adult: 19+years (full search is available on request). As part of the search and selection strategy, according to the HTA protocol, the major outcome was long-term disability (>3 months) due to a pain condition. A secondary outcome was sick leave, defined as ‘absence from work’ or ‘return-to-work’. The search was restricted to identify studies published in English, Danish, Norwegian or Swedish between 16 January 2003 and 16 January 2014.
Studies were eligible for inclusion if they met the following criteria: prospective cohort study (including randomised controlled trials), with at least 3 months of follow-up investigating the prediction of long-term disability and/or sick leave in adults (>18 years) with a subacute (≤3-month) non-malignant pain condition, visiting GPs or other primary care facilities. ‘Non-malignant pain condition’ was defined as pain conditions of non-cancer origin. If two or more published studies originated from the same patient population, the study with the longest follow-up period was included. Two reviewers (GHV and MSP) independently assessed abstracts and full-text articles for eligibility, and disagreement was solved by a third reviewer (LØ).
Data extraction and quality assessment
Two review authors (GHV and MSP) independently performed data extraction using a customised data extraction form. To summarise the evidence following the systematic review in the HTA, we applied the ‘Grading of Recommendations Assessment, Development, and Evaluation’ (GRADE) approach for rating quality of evidence (ie, our confidence in the estimates).19 Because we anticipated that the evidence base would come from cohort studies, the GRADE approach for prognostic factor research20 was applied. The risk of bias in the individual studies was assessed by two reviewers (GHV and MSP) using the Quality in Prognosis Studies tool (QUIPS).21 The overall risk of bias for each of the studies was judged as: (1) low if there were a low risk of bias in all key domains, (2) unclear risk of bias if there were an unclear risk of bias for one or more key domains and (3) high risk of bias if there were a high risk of bias for one or more key domains.22
Disagreement was resolved by consensus. Publication bias was explored by funnel plots.
Data synthesis and analysis
If a baseline factor was associated with outcomes at follow-up in one or more studies of different pain sites, it was considered a ‘possible prognostic’ factor and the results were presented as part of the evidence profile. When data were available in different formats, data from the ‘fully adjusted’ analyses were given preference and included in the analysis. For each outcome, we prepared an evidence profile using GRADEpro software.23 According to the protocol, we also aimed to combine individual study results with a meta-analysis section. However, given the substantial clinical (as well as statistical) heterogeneity in the individual studies, we decided to downplay the importance of the results from the meta-analysis and focus on the narrative interpretation of the results. A description of the statistical methods and the corresponding results are presented in online supplementary material S1. The narrative synthesis of the results is presented as proposed by Huguet et al.20
Results of the literature search
The search in the selected databases returned a total of 3533 references. A total of 32 references were identified through the additional search. After removing duplicates, 1841 references remained. The 1841 references were screened for eligibility, and 1641 records were excluded. The remaining 200 articles were read in full text; of these, 181 articles were excluded because they did not satisfy the inclusion criteria. A full list of excluded studies and the reason for exclusion are available from the authors on request. A total of 19 studies satisfied the inclusion criteria and were included in the systematic review. However, only 11 were eligible for inclusion in the evidence profile; the other 8 studies were excluded from the evidence profile due to: (1) inadequate statistical analyses,3 ,24 (2) the factors studied were assessed in only one study25–27 or (3) the factors studied were assessed for only one pain site (eg, only studies on back pain).28–30 See figure 1 for a flow diagram of the included studies.
The 19 included studies consisted of 17 cohort studies and 2 randomised controlled trials. Fourteen of the studies referred to patients with back pain4 ,24–26 ,28 ,29 ,31–38 and one referred to patients with pain in the neck or back.3 Two studies referred to patients with neck pain,27 ,39 one referred to patients with headache40 and one referred to patients with shoulder or back pain.30 From this last-mentioned study, only the cohort with back pain was included in the synthesis because the cohort with shoulder pain comprised both patients with subacute pain and patients with chronic pain.30 Outcome measures were disability in 16 studies,3 ,4 ,24 ,28–40 sick leave in 3 studies3 ,27 ,37 and return-to-work in 2 studies.25 ,26
Characteristics of the 19 included studies are presented in table 1. The total number of patients included in the 19 studies was 6266. The median number of patients per study was 184 (range 56–2662). In 11 of the studies, more women than men participated. Age was reported in 18 studies, with a median average of 42 years (range 34–52 years). Pain duration at baseline was reported in eight studies with a median average of 12.6 days (range 1–27 days). The follow-up period ranged from 3 months to 22 years (with a median of 9 months). Most of the studies recruited patients from general practice.3 ,4 ,24 ,26–28 ,30–33 ,35–38 ,40 The remaining studies recruited patients from physiotherapy or chiropractor clinics29 ,34 ,39 and the workers compensation board.25
Risk of bias within studies
Risk of bias in the included studies was assessed using QUIPS (figure 2). Overall, the agreement between the two assessors (GHV and MSP) on the different aspects of risk-of-bias assessment was 85.5% (weighted κ 0.49), which corresponds to a moderate degree of agreement. In all cases, any disagreement between the assessors was settled by consensus discussion. The domain ‘Study Confounding’ carried the highest risk of bias. In this domain, 3 studies were judged as having a high risk of bias, and 11 studies were assessed as having a moderate risk of bias. The high number of studies judged as having a high or moderate risk of bias in this domain was mainly due to the insufficient description of the factors that were included in the multivariable analysis. On the basis of the judgement of the 6 domains, 11 studies were judged to have a low risk of bias,4 ,26 ,28–30 ,33 ,35–39 3 studies had a moderate risk of bias27 ,31 ,32 and 5 studies had a high risk of bias.3 ,24 ,25 ,34 ,40
Publication bias was assessed by funnel plots for all eight prognostic factors. No obvious asymmetry was found (see online supplementary material S2).
Prognostic factors for disability
Prognostic factors for disability were assessed in 16 studies.3 ,4 ,24 ,28–40 A total of 81 factors were assessed in the unadjusted analysis (see online supplementary material S3). Of the 81 factors assessed, 53 were included in the multivariable analysis of the primary studies. Of these factors, the following eight factors were assessed in two or more studies and for at least two different pain sites: multiple site pain, higher baseline pain severity, previous pain episodes, older age, longer baseline pain duration, baseline disability, anxiety and depression. A total of 11 studies were included in the evidence profile (table 2).4 ,31–40 For these potential prognostic factors, the results are synthesised and summarised below.
Multiple pain sites
The association between multiple pain sites and disability at follow-up was assessed in three studies, including patients with headache,40 low back pain36 and neck pain.39 Multiple pain sites were significantly associated with disability in all three studies when adjusted for: age,36 ,40 sex,36 ,40 baseline disability36 and recruitment.36 In the study by Leaver et al,39 factors adjusted for were not described.
Higher pain severity
Six studies including patients with headache40 and low back pain31 ,32 ,34 ,36 ,37 investigated the association between higher pain severity at baseline and disability. Higher pain severity was consistently associated with disability at follow-up when adjusting for: age,36 ,40 sex,36 ,40 baseline disability36 and recruitment36 (in three studies, factors adjusted for were not specified31 ,34 ,37). However, a substantial heterogeneity between study results was detected. Consequently, the quality of evidence was downgraded due to inconsistency.
The association between baseline disability and disability at follow-up was examined in seven studies relating to patients with neck pain39 and low back pain.32 ,34–38 The following factors were adjusted for in the multivariable analysis of the primary studies: age,36 ,38 sex,36 ,38 body mass index (BMI),38 duration (hours) between pain debut and inclusion,35 job status,35 previous spine surgery,35 compensation status35 and self-rated health status.35 In four studies, the factors adjusted for were not specified.32 ,34 ,37 ,39 Baseline disability was associated with a higher risk of disability at follow-up in five studies.35–39 Two studies did not demonstrate a statistically significant association between baseline disability and disability at follow-up.32 ,34 The lack of statistical significance in the study by Heneweer et al may reflect the lower power of the analysis (small study sample).
Higher age was significantly associated with disability at follow-up in four studies.32 ,33 ,37 ,39 Three studies did not demonstrate significant associations between higher age and poor outcome.4 ,31 ,36 However, although statistically insignificant, visual inspection of the remaining three studies’ results indicated a similar trend. A dose–response effect was observed; an increase by 10 years of age was associated with a higher risk of disability than was a 1 year increase.37 The following factors were adjusted for in the multivariable analysis of the primary studies: sex,4 ,33 job,33 BMI,33 baseline pain severity,33 recruitment,36 depression,33 anxiety,33 fear avoidance,33 activity level prior to current pain episode33 and baseline disability.32 ,36 Three studies did not specify the factors adjusted for.31 ,37 ,39
Four studies including patients with headache40 and low back pain4 ,31 ,32 investigated the association between previous episodes and disability. The association between previous episodes and disability was inconsistent. Two studies detected a significant association between previous episodes and disability.31 ,40 One reported a non-significant positive association,32 and one study reported a non-significant negative association (previous episodes were not associated with disability at follow-up).4 The following factors were adjusted for in the multivariable analysis of the primary studies: age,4 ,40 sex4 ,40 and baseline disability.32 One study did not specify the factors adjusted for.31
Longer pain duration
The association between longer pain duration at baseline and disability was assessed in three studies including patients with headache40 and low back pain.32 ,38 Longer pain duration was defined as <4 vs >24 h in one study.40 The two other studies did not provide information on their definition of pain duration at baseline.32 ,38 All three studies found that longer pain duration was associated with disability at follow-up when adjusted for: age,38 ,40 sex,38 ,40 BMI38 and baseline disability.32
The association between anxiety and disability was assessed in two studies concerning patients with headache40 and low back pain.36 Both studies reported a significant association between anxiety when adjusted for age,36 ,40 sex,36 ,40 recruitment36 and baseline disability.36 Although ‘anxiety’ was significantly associated with outcome in the study by Grotle et al,36 the size of the association was fairly low, accounting for <2% of the explained variation, implying that the prognostic value in clinical practice may be limited.
Three studies assessed the association between depression and disability. Depression was found to be associated with disability at follow-up in two studies when adjusted for age,36 ,40 sex,36 ,40 recruitment36 and baseline disability.36 However, although significant, Grotle et al36 stated that the size of the association for the factor ‘depression’ was low, accounting for <2% of the explained variance in the study. No association was found in the remaining study when adjusted for age, sex and BMI.38
Quality of evidence for the risk of developing disability
The quality of evidence for the potential prognostic factors for the risk of developing disability is presented in table 2. All the included studies in the evidence profile were phase 1 studies, which are characterised as predictive modelling studies or explanatory studies conducted to generate a hypothesis.20 Thus, the quality of evidence was moderate as a starting point.20 Reasons for upgrading or downgrading the quality of evidence for the given prognostic factor are described below in table 2. The quality of evidence was graded as moderate, low or very low.
Prognostic factors for long-term sick leave or return-to-work
Three of the included studies had long-term sick leave as an outcome.3 ,27 ,37 Two of the studies, referring to patients with pain in the neck27 and back37 had performed multivariable analysis. The follow-up period was 3 months in both studies. Long-term sick leave was defined as sick leave for more than 30 days in one study37 and more than 7 days in the other.27 Baseline disability was the only potential prognostic factor that was assessed in both studies. In the study by Karjalainen et al,37 however, the results were described only as being ‘non-significant’. Thus, it was not possible to synthesise the results. In the study by Karjalainen et al, factors such as blue-collar work and long-term sick leave at baseline were associated with an increased risk of long-term sick leave at follow-up in patients with subacute back pain. Vos et al27 found that factors such as previous pain episodes, a follow-up appointment scheduled with a GP, and the GPs referring the patient to treatment and baseline disability were all associated with an increased risk of long-term sick leave among patients with acute neck pain. Two studies described ‘return-to-work’ as an outcome.25 ,26 Follow-up was 325 and 12 months.26 Both studies related to patients with back pain, so potential generic factors could not be extracted.
Five potential generic prognostic factors for developing disability following a subacute pain condition were identified. Risk factors across different pain sites included multiple site pain, higher pain severity, higher age, baseline disability and pain duration at baseline. There was inconsistent evidence regarding the association between previous pain episodes and disability. Although a few studies found anxiety, and depression to be associated with disability at follow-up the prognostic value of these factors may be low. Owing to the limited number of studies, it was not possible to identify potential generic risk factors for long-term sick leave or return-to-work. Quality of evidence was low or very low, implying that confidence in the estimate is low.
Comparison with other studies or reviews
Despite the sparse literature in this field, there is some evidence to support our findings. In concurrence with our findings, a previous review reported factors such as multiple site pain, higher pain severity, higher age, baseline disability and longer pain duration at baseline as being potential prognostic factors for a poor outcome in patients with musculoskeletal pain.8 A strong association between the number of pain sites and disability was also demonstrated in a previous cross-sectional study.9 Similarly, a prospective cohort study from 2008 found that the number of pain sites were a strong predictor of work disability 14 years later, regardless of the diagnosis.42 Furthermore, a recent systematic review found some evidence suggesting that the number of somatic symptoms and baseline severity of the condition influenced the future course in patients with medically unexplained symptoms.43 Thus, despite the low quality of evidence of the results, we find it reasonable to believe that the factors identified in our systematic review may act as central prognostic factors for the development of disability across pain sites. Therefore, future research should focus on confirming the role of these factors.
Interestingly, our review found limited evidence that psychosocial factors were associated with disability at follow-up. These findings are surprising because psychosocial factors, also known as ‘yellow flags’, are widely accepted as being key factors in the transition from acute to chronic pain conditions.44 ‘Yellow flags’ include depression, anxiety, catastrophic thoughts and pain-related fear of movement/fear avoidance among others.45 ,46 Several national and international guidelines recommend that clinicians screen for the presence of these factors in the early phase.47–49 In addition, several well-established screening tools for the risk of chronicity are based on the presence of these factors (eg, the Orebro Musculoskeletal Pain Questionnaire,50 the Fear-Avoidance Health Beliefs Questionnaire51 and the STarT Back Screening Tool52). Most studies, however, have not included pain duration at baseline when the importance of ‘yellow flags’ was assessed.16 A plausible explanation for the apparent discrepancy between our results and the widely accepted ‘yellow flags’, therefore, could be the inclusion criteria in our review regarding short pain duration at baseline. It is likely that psychosocial factors are of greater importance once pain has become chronic. Another explanation for the limited evidence could be that our review focuses on risk factors for future disability and sick leave and not on risk factors for developing a chronic pain condition. Nonetheless, future research should address and clarify the role ‘yellow-flag’ factors play in the various phases of pain.
Strength and limitations
It is considered a strength of our systematic review that we followed a rigorous protocol (registered in PROSPERO) prespecifying all the outcomes and analyses; our adherence to the protocol most likely strengthens the credibility of the evidence synthesis. We reported our findings as recommended by the PRISMA statement18 and judged the quality of the evidence based on the recommendations from the GRADE Working group. We believe that the GRADE framework applied to prognostic factor research was valuable for assessing and transparently reporting the quality of the evidence of the possible prognostic factors. To the best of our knowledge, this is the first time GRADE has been used in the evaluation of prognostic studies.
Limitations regarding the interpretation of the results from this study should be taken into consideration. A total of 14 of the 19 included studies in our review referred to patients with back pain. The high number of studies concerning patients with back pain may affect the external validity of the results to patients with pain at other sites. However, the vast majority of pain patients visiting general practice suffer from back pain, and the large number of studies on back pain included in the present review therefore reflects the distribution of patients seen in general practice.53 Future studies assessing prognostic factors for non-spinal pain are needed. Publication bias is one of the most common biases in systematic reviews. Therefore, we conducted funnel plots to explore whether publication bias was present in our analysis. No obvious asymmetry was found. In accordance with current knowledge, the use and appropriate interpretation of funnel plots are, however, controversial because of questions about statistical validity, disputes over appropriate interpretation and low power of the tests.54 For instance, a funnel plot can be symmetrical even in the presence of publication bias.54 Hence publication bias might be present in our analyses although undetected. Another common limitation in systematic reviews is the risk of selective reporting of primary study results. Our review was based primarily on observational cohort studies on prognostic factors (phase 1 studies). Such studies harbour a high risk that non-significant findings are not reported or only included in the first (unadjusted) part of the analysis. Any non-reporting of non-significant results invites a risk that the findings in the synthesis were overestimated. We attempted to account for such bias due to selective outcome reporting by listing all the studies that examined a specific prognostic factor in the unadjusted analysis. If a factor was investigated in eight studies, for example, but included only in the adjusted analysis of five, the quality of the evidence was downgraded.
Implications for clinical practice
No high-quality evidence was provided for any of the potential prognostic factors; therefore, no definite clinical conclusion can be made about how to identify patients at high risk of long-term disability or sick leave at an early stage in general practice. However, the empirical evidence illustrates what kind of prognostic factor research would be relevant to pursue in order to increase value and reduce waste in prognostic factor research on long-term disability among patients with subacute non-malignant pain.55 It appears that multiple site pain, high-baseline pain severity, older age, baseline disability and pain for a longer duration are associated with future disability across pain sites in subacute pain patients. Therefore, it may be helpful for GPs to have these factors in mind in clinical decision-making.
Implications for future research on prognostic factors
Correctable weaknesses in biomedical and public health research studies can produce misleading results and waste valuable resources.56 During the preparation of this review, it became clear that the current literature in this field falls short on a number of counts. As suggested by Ioannidis et al,56 this area of research also has weaknesses, such as selective reporting of results; lack of prespecified defined prognostic factors to be assessed; inadequate description of methods; inadequate or poor quality of statistical analysis; failure to distinguish between prognostic factors among patients with acute, subacute and chronic pain; and lack of published studies on patients suffering from non-spinal pain. We suspect most of these limitations can be related to the absence of detailed written protocols and poor documentation of research in general.56–59
Although good research ideas often yield unanticipated but valuable results, much research fails to effect worthwhile achievements. As long as the way in which research projects are prioritised for research is transparent and warranted, the disappointments should not be deemed wasteful; they are simply an inevitable feature of the way science works.55 In order to gain further knowledge on which factors are central prognostic factors (subacute phase), future studies should take into account baseline pain duration at the time patients are enrolled. Future studies on prognostic factors in chronic pain should be conducted as large, prospective, registered and protocol-based prognostic factor studies with sufficient study populations and transparent reporting of all factors studied. Once sufficient knowledge on risk factors has been obtained, documentation of effective treatment for high-risk pain patients is needed. Further, the effect of offering stratified care to pain patients based on their risk profiles should be tested in randomised controlled trials.60
The authors wish to thank Danish Regions for their support. They also thank Hanne Caspersen from the State and University library for conducting the literature search and Professor Troels S Jensen from the Danish Research Center for commenting on the manuscript. Finally, RC acknowledged The Oak Foundation and Copenhagen University Hospital at Frederiksberg for their support.
Review history and Supplementary material
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.
- Data supplement 1 - Online supplement
Contributors All the authors were responsible for the design and search strategy. GHV and MSP were responsible for conducting the search. GHV, MSP and RC conducted the data analysis and produced the tables and graphs. RC provided input into the data analysis and interpretation. The initial draft of the manuscript was prepared by GHV, then circulated among all authors for critical revision. All authors helped to evolve analysis plans, interpret data and critically revise successive drafts of the manuscript.
Funding This research received no specific grant from any funding agency in the public, commercial or not-for-profit sectors.
Competing interests None declared.
Provenance and peer review Not commissioned; externally peer reviewed.
Data sharing statement No additional data are available.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.