A systematic review of early prognostic factors for persistent pain following acute orthopedic trauma

1Monash University, Monash Injury Research Institute, Clayton Campus, Victoria; 2New South Wales Injury Risk Management Research Centre, University of New South Wales, Sydney, New South Wales, Australia Correspondence: Professor Roderick McClure, Monash University, Monash Injury Research Institute, Building 70, Clayton Campus, Victoria 3800, Australia. Telephone 61-3-9905-4372, fax 61-3-9905 4363, e-mail Rod.McClure@monash.edu In Australia and Canada, and in some European countries, chronic or persistent pain is prevalent (1-3), with chronic pain prevalence estimates ranging from 10.1% to 55.2% (2). Chronic pain is associated with work disability and increased health service utilization, and is a major source of costs to the economy, health systems, injury compensation schemes and society (2-5). Acute orthopedic trauma is a common reason for hospitalization (6) and is often associated with persistent pain and pain-related disability (7). Variation in pain outcomes following acute trauma is reported, and there is uncertainty about which patients are at risk of an adverse pain outcome. Because the biopsychosocial determinants associated with variation in long-term pain outcomes following acute orthopedic trauma are poorly understood, the treating clinician is faced with uncertainty when providing prognostic advice regarding long-term recovery. A synthesis of the research evidence on the risk factors associated with adverse pain outcomes should facilitate the early identification of individuals at risk of pain chronicity and, therefore, in need of further management. While biological factors associated with chronic pain are likely to be related to specific injury diagnoses, psychological and socioenvironmental factors should be common to a wide spectrum of acute traumatic injuries. Previous reviews addressing determinants of chronic pain outcomes secondary to acute trauma have primarily focused on specific injury samples related to structures of the spine, including spinal cord injury and whiplash-associated disorders, or have addressed outcomes following limb amputations (8-11). These reviews report a lack of consistency with respect to the important prognostic factors. One systematic review was more generic, focusing on polytrauma including traumatic brain injury (12). However, this review focused on painrelated rather than pain-specific outcomes; it also included many studies that did not report multivariate analyses and did not provide information on the magnitude of effect. This limits the meaningful evidence on prognostic factors that can be drawn from the review. Limiting a review to a specific biological injury potentially decreases the power and generalizability of the findings. This is because acute orthopedic injuries do not reflect single disease entities but include a complex spectrum of pathologies, causes, severities and baseline prognostic risk of adverse long-term outcomes (13). review


A systematic review of early prognostic factors for persistent pain following acute orthopedic trauma
Fiona J Clay PhD MSc BSc Grad Dip Epidemiology and Biostatistics 1 , Wendy L Watson PhD MA Dip Ed BSc(Hons) BA 2 , Stuart V Newstead PhD MSc BSc(Hons) AStat I n Australia and Canada, and in some European countries, chronic or persistent pain is prevalent (1-3), with chronic pain prevalence estimates ranging from 10.1% to 55.2% (2).Chronic pain is associated with work disability and increased health service utilization, and is a major source of costs to the economy, health systems, injury compensation schemes and society (2)(3)(4)(5).Acute orthopedic trauma is a common reason for hospitalization (6) and is often associated with persistent pain and pain-related disability (7).Variation in pain outcomes following acute trauma is reported, and there is uncertainty about which patients are at risk of an adverse pain outcome.
Because the biopsychosocial determinants associated with variation in long-term pain outcomes following acute orthopedic trauma are poorly understood, the treating clinician is faced with uncertainty when providing prognostic advice regarding long-term recovery.A synthesis of the research evidence on the risk factors associated with adverse pain outcomes should facilitate the early identification of individuals at risk of pain chronicity and, therefore, in need of further management.While biological factors associated with chronic pain are likely to be related to specific injury diagnoses, psychological and socioenvironmental factors should be common to a wide spectrum of acute traumatic injuries.
Previous reviews addressing determinants of chronic pain outcomes secondary to acute trauma have primarily focused on specific injury samples related to structures of the spine, including spinal cord injury and whiplash-associated disorders, or have addressed outcomes following limb amputations (8)(9)(10)(11).These reviews report a lack of consistency with respect to the important prognostic factors.One systematic review was more generic, focusing on polytrauma including traumatic brain injury (12).However, this review focused on painrelated rather than pain-specific outcomes; it also included many studies that did not report multivariate analyses and did not provide information on the magnitude of effect.This limits the meaningful evidence on prognostic factors that can be drawn from the review.
Limiting a review to a specific biological injury potentially decreases the power and generalizability of the findings.This is because acute orthopedic injuries do not reflect single disease entities but include a complex spectrum of pathologies, causes, severities and baseline prognostic risk of adverse long-term outcomes (13).
Furthermore, there may be concurrent injury to multiple body parts or systems that result in a range of impairments.The International Mission for Prognosis and Analysis of Clinical Trials in Traumatic Brain Injury (TBI-IMPACT) investigated methodologies for dealing with injury heterogeneity (14).Substantial gains in statistical power followed covariate adjustment, and this was more efficient than analyses targeted to clinically defined prognostic groups (13).
Given the problems associated with persistent pain, the number of people requiring hospitalization following traumatic injury and the limited generalizability of findings from previous reviews, a formal synthesis of the current knowledge was warranted.The specific aim of the present review was to identify the early prognostic factors associated with variation in persistent pain outcomes following acute orthopedic trauma presenting with a spectrum of pathologies.

Search strategy
The Ovid MEDLINE and EMBASE databases were searched for relevant studies published between January 1996 and October 2010.Search terms were mapped to MeSH terms or subject headings and synonyms were grouped using Boolean operators.A prognostic sensitive filter was applied and the search strategy was developed with the assistance of a specialist subject librarian.A detailed description of the search strategy applied to the Ovid MEDLINE database is provided in Appendix 1.The reference lists of all relevant articles were screened for additional publications.
Results of the database searches were downloaded into Endnote X2 (Endnote, USA) and duplicate articles were excluded.One reviewer (FJC) initially screened all titles and abstracts for potential relevance.Full-text articles considered to be relevant, together with articles where there was uncertainty as to their relevance, were screened by two authors (FJC, RM) and a decision made as to which articles should be retained.Disagreements were resolved by consensus.
outcomes Because of the nature of pain, the reported outcomes are many and varied.Because the focus of the present review was on determinants of persistent pain, only patient-focused outcomes that specifically referred to pain were considered.For example, if the outcome was the presence of persistent pain or pain severity, the study was included.Examples of these may include a visual analogue scale or a pain scale that forms part of a questionnaire (eg, the pain or discomfort item from the EQ-5D [EuroQol Group, The Netherlands]).If it was a summary measure of disability postinjury, such as the Short Form-36 Physical Component Summary score that did not identify and analyze an explicit measure of pain, it was not included in the review.

definition of injury
The following definition of injury was applied to studies included in the present review.
Injury was defined as damage to the body "resulting from the transfer of environmental energy to a person at levels that exceed the threshold for tissue damage" (15).Injury was further defined according to the International Classification of Disease ICD-10-AM S and T injury diagnosis codes (16).Orthopedic injury is defined as an injury affecting the body's musculoskeletal system, and includes injuries to bones, joints, ligaments, tendons, muscles and nerves (17).Orthopedic injuries may results from acute or chronic exposures; however, the focus of the present review is on orthopedic injuries resulting from acute exposures.

inclusion and exclusion criteria
The present review was limited to studies published in English and available as full-text articles.Only prospective and retrospective cohort studies were considered.Articles were selected for inclusion according to the following criteria: • The primary research question addressed the determinants of persistent pain, the study design was longitudinal (prospective or retrospective [18]) and the article reported results with multivariate statistical analyses appropriate to prognostic studies.• Samples included persons of working age, defined as 15 to 70 years of age.• Because factors collected in the early stages of recovery following an injury are the most useful to guide rehabilitation, only factors measured proximal to the injury either preinjury or the early postinjury phase were included (defined as within 30 days of the date of injury).• The International Association for the Study of Pain defines chronic pain as pain without biological value that has persisted beyond the normal healing time.This may be less than one month to more than six months, with three months being considered a convenient point of division between acute and chronic pain (19).
In keeping with this definition, pain outcomes had to be measured at least three months following the date of injury.The terms 'chronic' or 'persisting' pain are used interchangeably throughout the present review.The following studies were excluded: • Studies of pain related to the spinal column, because these studies have received consideration in recent reviews.These include studies of back pain, spinal cord injury, neck pain and whiplashassociated disorder.• To rule out findings with low statistical power that may not be broadly representative of the biopsychosocial nature of pain or that may be due to chance variation in the data, studies were excluded if the sample size analyzed with respect to the pain was fewer than 80. • Studies that did not sufficiently recognize the biopsychosocial nature of pain.The biopsychosocial nature of persistent pain recognizes that the experience and impact of pain is determined through the interaction of three main groups of personal factors, namely, biological, psychological and socioenvironmental factors (4).To prevent unnecessary pain-related disability, as many of these factors as possible should be addressed in interventions (4).A study that would be excluded is one that examined the determinants of persistent pain outcomes and included in multivariate models only clinical data from the initial inpatient postinjury episode (21).

Quality assessment
The quality of the included studies was rated according to the 12 criteria outlined in Table 1.The criteria were adapted from a review by Hayden et al (22), and address the six major sources of bias in prognostic reviews (22,23).The quality assessment addressed potential bias in the form of study participation, the measurement of outcome, prognostic factor measurement and statistical analysis.Two authors independently assessed the quality of each included study.The 12 criteria were answered 'yes', 'no' or 'unsure'; only scores of yes contributed to the overall quality score.The quality assessment criteria are further outlined in a systematic review of factors associated with return to work by the first author (24).Studies that scored 10 or more points were rated as high quality, those that achieved seven to nine points were considered moderate quality and studies with less than seven points were rated as low quality.No studies were excluded on the basis of quality.

Level of evidence
The strength of the evidence of prognostic factors was determined according to a rating system similar to that used by Clay et al (24).The levels of evidence are outlined in the Table 2.For the level of evidence to be considered strong, findings had to be consistent in at least two high-quality studies.When the findings were from only one study, the level of evidence was considered to be limited.A positive effect of a factor in one study and a negative effect in another were rated as an inconsistent finding.A significant effect of a factor in one study and a nonsignificant association in another were rated as consistent findings provided that the effect sizes were in a similar direction.

data extraction
The following data were extracted as they appeared in the original publication: study population and setting, sample size included in the analysis, nature of injuries, inclusion and exclusion criteria, study design, outcome definition, follow-up time points, prevalence of pain, loss to follow-up and missing data (if retrospective study), analytic method, and significant and nonsignificant factors identified in multivariate analyses.Only factors measured proximal to the injury are reported.Factors were considered to be significant determinants at P<0.05.Due to the potential for confounding to invalidate the findings associated with factors reported as significant at the univariate level, only factors included in multivariate analyses are reported.

Study selection
The search strategy resulted in an initial yield of 979 references from the two databases.A further 13 potentially relevant references were identified from the reference lists of articles.After removal of duplicate references, 928 references remained.Of these, 883 references were excluded based on either title or abstract.The full text of 45 articles was retrieved for detailed evaluation; 27 did not meet the inclusion criteria.Reasons for exclusion included that the study did not report multivariate analyses, the sample size analyzed was less than 80, the sample excluded acute trauma patients, the outcome was measured before three months postinjury, the study was cross-sectional or the study did not adequately consider the biopsychosocial nature of pain.After independent review by two authors, a further eight articles were excluded, leaving 10 articles in the review.The search strategy is outlined in Figure 1.

Quality assessment
The quality assessment revealed methodological problems with many studies including insufficient justification of measures and constructs, analyses lacking power to achieve statistical significance with the available sample size, lack of reporting on the extent of missing data, and insufficient reporting on the detail of the statistical analysis including the precise form of regression analysis carried out and whether the assumptions associated with the regression technique were met.Three studies were rated as high quality (30,33,34), five as moderate quality (25,26,29,31,32) and two as low quality (27,28).Table 3 summarizes information on the studies including the country, setting, nature of the injuries, inclusion and exclusion criteria, study design, outcome definition and follow-up time points.Table 4 summarizes the quality assessment score.

Summary features of the included studies
The 10 studies included in the final review evaluated four different pain outcomes: presence of pain, presence of pain or discomfort, pain severity and pain interference with activities.Of the six different measures used to assess pain outcomes; only two were developed with a primary focus on pain (Chronic Pain Grade and Graded Chronic Pain questionnaires) (30,34).Nine of the 10 studies included in the present review were based on inception cohorts.Participant follow-up occurred between six months and seven years postinjury.Using criteria of 80% follow-up, only two of the studies had acceptable loss to follow-up (29,30).Given the extended period of follow-up in the study by Castillo et al (34), 28% loss to follow-up appears reasonable.Most of the injury samples consisted of trauma with an injury severity score (ISS) of 10 or higher (35).No studies were identified that only included individuals with ISS minor

Continued on next page
Clay et al injuries (ISS: 1 to 8).Depending on the nature of the injured sample as well as the time point and method of outcome measurement, the prevalence of pain following acute orthopedic trauma for studies included in the present review ranged from 11% to 77%.Two studies were based on the same time period of recruitment from the same orthopedic trauma registry (25,26).Each study selected subsamples of the registry cohort using different inclusion and exclusion criteria but used the same outcome measures and analytic approach techniques.For the remainder of the present review, they will be described as subset analyses of a single cohort.

Assessment of statistical analysis
An inclusion criterion for studies in the present review was that the publication reported results with statistical analysis techniques appropriate to prognostic studies.A number of studies used stepwise elimination multivariate regression techniques and did not report information on factors that were not statistically significant in the model (22,23,26,28,34) such as the estimated effect size and standard deviation before elimination from the model or change in model fit on removal of the factor.For those studies, it was not possible to assess whether the analysis was sufficiently powered to be able to test the study hypotheses.A number of studies used multivariate linear regression to assess the determinants of pain severity but did not test for the normality of the residuals (30,31,33).At least one moderate-quality study appeared underpowered because no statistically significant associations were detected even though the estimated effect sizes for a number of factors were of clinically interesting magnitude (25).In a number of studies, insufficient detail was provided to be able to fully understand the analytic method, make an appropriate assessment of the statistical quality or to fully understand the results (28,(30)(31)(32)34).Some studies failed to identify potential colinearity between included factors.For example, the inclusion of two closely related measures of posttraumatic symptoms in the same MANOVA model is potentially subject to problems with collinearity (28).Data pooling was not possible due to the heterogeneity of outcome definitions, follow-up time points, covariate classifications and methods.Features of the statistical analysis including the number of observations included in analyses, the regression technique used and the significant and nonsignificant factors as reported in the multivariate analyses are reported in Table 4.

Levels of evidence
The evidence for specific factors identified in studies included in the present review will now be outlined.Only generic factors that would be relevant to multiple contexts are reported.Factors related to specific injury subsets in the study by Castillo et al (34) are not included.The classifications that were used are arbitrary; some factors could potentially be included under more than one classification.While more than 70 individual factors were examined in studies included in the present review, only 17 factors were considered in more than one cohort.An overview of the evidence for factors measured in more than one cohort is presented in Table 5.

Prognostic factors Sociodemographic factors:
Age and sex were investigated in all studies included in the present review.One high-quality study and two moderate-quality studies reported an association between age and pain (30)(31)(32); no association was identified in the remaining studies.Two high-quality studies reported that female sex was associated with an increased risk of pain (32,33), and five lower-quality studies reported no such association.There was strong evidence from three studies that more years of education was associated with better pain outcomes including less severe pain and less interference with activities (26,30,34).One high-quality and two moderate-quality studies found no association between education and pain (25,29,33).The evidence relating to the consumption of alcohol was inconsistent.Current hazardous drinking was associated with less severe pain at 12 months postinjury in one high-quality study (30), and in a second high-quality study it was associated with greater pain interference with activities (34).Past alcohol dependence was associated with an increased risk of pain in one study (31).injury-and treatment-related factors: ISS was examined in six lowerquality studies with only one moderate-quality study reporting a statistically significant association between the new ISS and pain (25)(26)(27)29,31,32).The need for surgery were associated with an increased risk of pain but was only evaluated in a single study (30).
There was strong evidence from four studies that high pain intensity at discharge or early postinjury was associated with adverse pain outcomes (26,30,31,34).There was limited evidence for the effect of drug treatment on pain outcomes.In one study, treatment with narcotics at an early time-point was associated with better outcomes (34) while a second study reported that ongoing opiate treatment was associated with an increased risk of severe pain (31).A lower extremity injury was associated with an increased risk of pain in one lower-quality study (29) but not in other studies (31,33).Psychosocial factors: There was strong evidence from two studies that preinjury anxiety or depression was associated with poorer pain outcomes (30,33).One moderate-quality study found no such association (26).There was moderate evidence from two studies that postinjury anxiety or depression was associated with persisting pain (28,34).One high-quality and one moderate-quality study reported limitations in preinjury instrumental activities of daily living as being predictive of adverse pain outcomes (26,30).Self-efficacy was considered in one high-quality study and was associated with less pain interference with activities (34).Compensation status was associated with pain severity in one moderate-quality study (26), while in a second moderate-quality study no statistical association was found between compensation and poor pain outcomes (31).

diSCuSSioN
In the present review, individuals who were older, female, or who reported preinjury anxiety or depression at recruitment or high pain intensity in the acute postinjury period were more likely to report the presence of persisting pain and other pain-related outcomes.In contrast, persons with higher levels of education had a reduced risk of adverse pain outcomes.While many factors were considered across the 10 studies, the small number of factors considered in more than one cohort limits the conclusions that can be drawn on individual factors.While a comprehensive literature search was performed, only 10 studies were identified that fulfilled the inclusion criteria for the present review.
Seven of the 10 studies had a primary analytic question that focused on pain (26,28,(30)(31)(32)(33).Two studies were based on data from the same orthopedic trauma outcomes registry cohort (25,26), and while such registries provide a valuable base by which to chart postinjury recovery, the data they collect are generally limited and do not substantively support hypotheses-driven research analyses.The need for additional high-quality research to evaluate a comprehensive range of determinants of persisting pain following acute orthopedic trauma is a key outcome of the present review.
Many of the studies included in the present review had methodological shortcomings.The most common of these was the lack of detail provided on the statistical analyses.This made it difficult to interpret the results effectively, to assess the adequacy of the sample size or to evaluate the possibility of bias associated with loss to follow-up and missing data.The selective reporting of only statistically significant associations in many of the articles meant that it was not possible to determine whether the factors that were not significant were due to low statistical power or may have represented potentially important clinically significant results worthy of additional study.The need to improve the statistical rigor associated with pain outcome studies is apparent.While a number of acute trauma outcome studies that were not included in the present review reported the prevalence of persisting pain, it was surprising how few studies assessed the role of potential determinants of persisting pain in multivariate analyses.This is particularly so given the large volume of research literature that exists in relation to persisting pain of nonspecific etiology.It is possible that because acute trauma is, for the most part, diagnostically specific, pain is an accepted symptom and, as such, the primary focus has been on the immediate medical management of the injury rather than the consideration of long-term pain outcomes.This approach appears more in keeping with a biomedical model that centres on the medical treatment of the pathological causes related to the injury and ensuing disability (36).Support for this proposition comes from a review of studies examining chronic pain secondary to disability including spinal cord injury and traumatic amputations (10).Ehde et al (10) questioned whether persons with chronic pain secondary to disability were regularly identified as having a chronic pain problem.
While the biopsychosocial model of pain has provided a useful conceptual framework that has advanced our understanding of the determinants associated with the variation in pain outcomes (4,37), the majority of studies included in the present review evaluated a simplistic set of factors not well linked to the biopsychosocial model.Findings from studies of chronic pain-related disability as a primary condition highlight the importance of psychological factors in the development and maintenance of pain chronicity (38).While mixed orthopedic injury samples should facilitate the identification of psychological and social factors in common to a range of orthopedic injuries, the studies outlined in the present review considered only a small subset of psychosocial factors.The review identified strong evidence for preinjury anxiety or depression, and moderate evidence for postinjury depression or anxiety as a risk factor for the development of persisting pain.Given the evidence for these factors, it seems reasonable that identifying injured persons with these symptoms may lead to psychological treatment being initiated.The six studies included in the present review that measured depression or anxiety used five different measures of depression or anxiety including self report, Brief Symptom Inventory, Short Form-36, Hospital Anxiety and Depression Scale and the Symptom Checklist-90.The validity of the findings is open to question given that most of the studies included in the present review did not well specify cut-offs that would indicate or enable comparisons between measures or the degree to which the study measured symptoms of minor or major depression.For the prognostic determinants associated with adverse pain outcomes to be of use in clinical care, clear evidence on prognostic determinants established from the inclusion of the same attributes and measures across multiple studies is needed.While the review concluded that there was evidence for pre-existing depression or anxiety symptoms as being associated with the development of persisting pain, further research is needed to determine whether the presence of such symptoms reflects a causal association or act simply as a marker of persisting pain.
Pain is a complex subjective phenomenon with multiple dimensions; the expression of pain often occurs via behaviour including pain intensity, activity, medication use, litigation, work disability and health service utilization (39)(40)(41).The present review identified a lack of consistency in the timing and assessment of pain outcomes.The inclusion of a single pain measure or dimension measured on only one occasion may provide an incomplete picture with respect to the problems associated with pain chronicity (40).Composite measures of pain intensity and pain interference serve to indicate a problem but may not well identify the specific dimension(s) associated with the pain.The majority of the studies included in the present review measured pain at only one time-point and used a single pain measure.Given that the personal assessment of pain is highly subjective and subject to recall bias (42), repeated assessments with more than one measure of pain are considered to be critical (40).

Comparison with other reviews
In a review of chronic pain secondary to disability, Ehde et al (10) noted how few studies in samples of disabled persons had tested the biopsychosocial model.The small number of studies identified in the current review confirms this observation.While pain coping or catastrophizing were associated with adjustment to phantom limb pain following traumatic amputation, many of these studies were cross-sectional and therefore more prone to bias (10).A number of systematic reviews of work-related neck and back pain have identified strong evidence for the role of job demands, job stress, job insecurity and low social support in the development of chronic pain (43,44).The consistent evidence for these factors suggests that they are worth evaluating in an acute trauma cohort.In common with a review of recovery following whiplash-associated disorder (8), the present review identified evidence for the association between early high pain intensity and persisting pain.

Strengths and limitations
The strength of the present review is that the methodology closely followed the Meta-analysis Of Observational Studies in Epidemiology (MOOSE) guidelines for systematic reviews of observational studies (45).Other strengths include the limiting of inclusion criteria to studies reporting multivariate analyses and, hence, were better controlled for confounding effects in what is a multifactorial problem as well as the exclusion of studies that primarily focused on clinical factors.The multifactorial nature of pain disability is well recognized and clinical factors, when considered alone, have limited power to predict longterm outcome.While every effort was made to include all relevant articles, it is possible that articles were missed due to the terms employed in the search strategy, although the likely miss rate is considered small due to the comprehensive coverage of the databases searched.While studies in which the sample size analysed was less than 80 were excluded, it is possible that these studies, while very small, may have contributed data of potential significance.Excluding studies related to injuries of the spinal column and to traumatic amputation subsets means it may not be possible to generalize our findings to persisting pain following acute injuries of this nature.

implications for research
The search strategy identified many studies that had small sample sizes, only reported the prevalence of chronic pain or only included univariate analyses.The 10 studies included in the present review displayed marked differences in the types of injuries, follow-up time periods, demographic characteristics of the study participants, definitions of persistent pain, methods of data collection and the measurement of persistent pain outcomes.In common with other systematic reviews (2), it is likely that the variability in the range of persistent pain outcomes reported for studies included in the current review (11% to 77%) will have been influenced by the heterogeneity apparent in the different study designs and methodology.Furthermore, the heterogeneity highlights the need for more systematic and uniform prospective epidemiological outcome studies to estimate the determinants of persistent pain outcomes.We have provided our recommendations for the methods and reporting of future studies on chronic pain following acute orthopedic trauma (Table 6).The present review identified no discrete studies of the determinants of pain following non-life-threatening, less serious, injuries.While it might be expected that these injuries will be associated with improved outcomes, including a lower prevalence of pain and painrelated disability, because these injuries are common the burden associated with them may be considerable.The ISS was only associated with adverse pain outcomes in one of the included studies; therefore, given this lack of study, it is unknown whether predictors of pain in samples of major trauma or specific serious injuries will also be associated with chronic pain following less-serious injuries.

Table 6
Recommendations for future studies of factors associated with chronic pain outcomes following acute orthopedic trauma • The study design should be based on an inception cohort (defined as a cohort of injured persons assembled early in the development of their clinical diagnosis) • The study population should be clearly described to understand the heterogeneity of participant characteristics.There should be adequate adjustment of participants' characteristics for expected outcomes • The sample size should be adequate to build the model.The number of cases with the outcome of interest and the total number of cases in the sample should be reported • Selection bias should be kept to a minimum.Ideally, studies should be population or hospital based • The nature of the injuries should be adequately described.Common injury severity measures should be included to enable comparison of injury severity among studies • Time period of participant follow-up should be a minimum of 3 months and ideally longer.Pain outcomes should be measured at more than one time point.
• The dependent variable selection and analytic method should be clearly described and be appropriate.To determine the strongest evidence on predictors, analyses should be multivariate based • Consistent, well-validated outcome measures specifically developed for pain should be used.The inclusion of closely related measures potentially subject to collinearity should be avoided.Ideally, more than one pain measure should be included • Important confounders must be included in the model • The intended purpose of the model should be clearly explained.Is the model intended to be useful in clinical practice?Is the model intended to explain the relationship between each explanatory factor and chronic pain outcomes?
• Explanatory factors and measures should be clearly described, including the time point at which the variable was measured, how it was measured and in what form it was entered into the model (did the variable require transformation?).There should be an adequate number of participants in the study sample with respect to the number of risk factors being entered into the model.The choice of explanatory factors should be based on a conceptual framework • The methods for building multivariate models should be appropriate.The process and rationale for explanatory factor selection and model building should be clearly described including testing for: correlations between variables; multicollinearity; statistical interactions; and tests of normality of residuals, the use of stepwise model building methods and whether the reported model is a main effects model • The extent of missing data and loss to follow-up should be reported • The reporting of models should include confidence intervals around odds ratios and regression coefficients.The number of predictor variables that were included in the model should be indicated and the effect sizes for all factors included in multivariate models should be reported

Table 2 levels of evidence for individual factors in the present review level of evidence Criteria for evidence level
StrongConsistent findings in at least 2 high-quality cohortsModerateOne high-quality cohort and consistent findings in one or more low-quality cohorts Limited Findings of one cohort or consistent findings in one or more low-quality cohortsInconsistentInconsistent findings irrespective of study quality

Table 3 Characteristics of included studies author (ref), year Country Setting Nature of injuries Inclusion criteria exclusion criteria Study design Outcome measurement Outcome definition Follow-up time-point
AIS Abbreviated injury scale; AUS Australia; ED Emergency department; EQ EuroQol; FRAN France; GCS Glasgow coma score; ICU Intensive care unit; ISS Injury severity score; NISS New injury severity score; NL Netherlands; NOR Norway; Pros Prospective; ref Reference; Retro Retrospective; SF36 Short Form-36; SWITZ Switzerland; TBI Traumatic brain injury; USA United States

Table 4 -continued Results from the included studies
† Numbers in parentheses after ORs are 95% CIs.ADL Activities of daily living, BMI Body mass index; FIM Functional Independence Measure; GCS Glasgow coma score; HADS Hospital Anxiety and Depression Scale; IADL Instrumental activities of daily living; ICU Intensive care unit; IRR Incidence rate ratio; ISS Injury severity score; LTF Lost to follow-up; NISS New injury severity score; PTSD Posttraumatic stress disorder; ref Reference; SAPS Severity Acute Physiology Scale; SIP Sickness Impact Profile

Table 5 level of evidence for individual factors measured in more than one cohort Prognostic factor level of evidence Direction of association with adverse pain outcomes* Reference(s) of included studes Reference(s) in which no statistical association with P<0.05 was found
Increased or decreased likelihood of an adverse pain outcome relative to the criterion selected as the reference category; † Different constructs used to measure prognostic factor *