REVIEW ARTICLE
Pain prevalence and its determinants after spinal cord injury: A systematic review S. van Gorp1,2, A.G. Kessels3, E.A. Joosten1, M. van Kleef1, J. Patijn2 1 Department of Anesthesiology/Pain Management, Maastricht University Medical Center, The Netherlands 2 University Pain Centre Maastricht, Maastricht University Medical Center, The Netherlands 3 Clinical Epidemiology and Medical Technology Assessment, University Hospital Maastricht, The Netherlands
Correspondence Sebastiaan van Gorp E-mail: [email protected] Funding sources None. Conflicts of interest None declared. Accepted for publication 6 April 2014 doi:10.1002/ejp.522
Abstract Pain prevalence studies are important as they illustrate the magnitude of pain problems in a certain patient population, such as patients living with a spinal cord injury (SCI). Strikingly, reported pain prevalence rates in SCI patients are found to vary greatly, while determinants for the differences between pain prevalence reports remain unclear. We here aim to identify determinants for the differences (heterogeneity) in pain prevalence reports through a systematic review of all SCI pain prevalence reporting studies. Literature search was done using Medline, Cumulative Index to Nursing and Allied Health Literature, ISI Web of Knowledge and Embase. Data abstraction was performed while blinded and was followed by meta-(regression)-analyses. We identified 82 studies. Study design-related determinants of SCI pain prevalence reports were pain definition strictness (mild, moderate or high), primary study goal (pain study or not), data source (retrospective or not), and in a limited number of cases response/attrition rates. While correcting for these items, population characteristics correlating with pain prevalence rates were both proportion of patients with a depression and average time after injury (positive correlations). Between-study heterogeneity may remain even after the identification/correction of above-mentioned causes of heterogeneity.Pain after SCI does seem to relate to the duration of the injury and depression, yet major causes of bias in reported pain prevalence are found to be related to the primary study goal (pain study or not), choice of pain definition and the use of retrospective data.
1. Introduction Treatment of spinal cord injury (SCI)-related pain is still a major problem (Haisma et al., 2007; Teasell et al., 2010; Finnerup and Baastrup, 2012). The reported prevalence of pain after SCI, however, varies considerably, with rates reported from 19% to as much as 96% (Finnerup et al., 2001; New et al., 2002; Siddall et al., 2003; Raissi et al., 2007; Dijkers et al., 2009). In a recent, and the only published, systematic review of SCI pain prevalence studies, it was concluded that a definite statement about a pain prevalence could not be made due to this heterogeneity of © 2014 European Pain Federation - EFIC®
the study population (Dijkers et al., 2009). Identifying which factors cause this heterogeneity could reveal important information, as their identification supports their role as relevant pain determinants and/or study confounders. Unfortunately, the authors of the review reported to be unable to identify such factors, yet this was not the primary aim of their study. By using a more extensive and somewhat different approach, it might have been possible to identify factors causing the heterogeneity in reported SCI pain prevalence. For example, as pain prevalence has not yet been consistently reported to differ due to SCI aetiology, one can better not exclude studies based on aetiology and Eur J Pain •• (2014) ••–••
1
Spinal cord injury pain determinants
Databases • For this research, a new database was created from data abstracted from manuscripts identified using Medline, CINAHL, ISI Web of Knowledge, Embase, reference searches and contact with authors. • Factors we were able to identify as influential for the size of pain prevalence are depression, time after injury, pain definition, primary study goal (whether it was designed as a pain study or not), data source (retrospective or not) and response/ attrition rates. • Our manuscript provides important insights on (1) pain determinants, (2) design of pain studies, (3) interpretation of pain studies and (4) metaanalyses of prevalence studies. What does this study add? • This systematic review shows that the size of reported spinal cord injury pain prevalence is determined by pain definition, whether pain was the primary study outcome, depression prevalence, time after injury and population response rates.
investigate whether they actually influence on reported pain prevalence (Mariano, 1992; McKinley et al., 2002; Ibrahim et al., 2012). Similarly, as pain after SCI is typically chronic and the definition of ‘chronic’ remains ambiguous, one should not have to exclude studies that are not specifically reporting chronic pain (Purves et al., 1998; Cruz-Almeida et al., 2005). The resulting limitation of study exclusions is likely to result in a more valuable analysis of possible determinants on reported pain prevalence (e.g. depression prevalence or individual items of study quality) (Thompson, 1994; Stroup et al., 2000; Huedo-Medina et al., 2006). Therefore, in this systematic review, we wanted to investigate the impact of a wide variety of study design and study population characteristics on the heterogeneity of reported pain prevalence after SCI by using limited exclusion criteria and detailed meta-regression analyses.
2. Methods A systematic literature review was conducted on studies reporting on SCI pain prevalence (i.e. descriptive studies and observational studies). As no protocols are available for performing systematic reviews on prevalence studies specifi-
2 Eur J Pain •• (2014) ••–••
S. van Gorp et al.
cally, we used protocols for systematic reviews of studies on treatment or exposure effects, when applicable (Stroup et al., 2000; Dickersin, 2002; Higgins and Green, 2008). The following databases were used: Medline, Cumulative Index to Nursing and Allied Health Literature, ISI Web of Knowledge, and Embase. The search was wide and the queries comprised the terms ‘pain’, ‘dysaesthesia’ or ‘dysesthesia’, and ‘spinal cord injury/-ies’, ‘spinal cord disease/-s’, or ‘tetra-/quadri- or paraplegia’. The abstract and/or publication title were used to identify articles that might fit the topic. Next, the full document was requested, translated to English and screened whether these met our inclusion criteria or any exclusion criteria. Also, the reference lists from all screened articles were hand-searched for any old and/or missed articles. The major clinical research groups on SCI pain were contacted for any missing publications. Last, the major journals were hand-searched (i.e. Pain, Spinal Cord/Paraplegia, Journal of Spinal Cord Medicine, American Journal of Physical Medicine & Rehabilitation, and Journal of Rehabilitation Research and Development/United States Veterans Affairs).
2.1 Inclusion criteria Studies were included if they reported on the prevalence of pain after SCI, irrespective of the type of prevalence used (i.e. at a single time point or over a certain amount of time). In one case, only an abstract was available, and the data from the abstract were used (Rintala et al., 1991).
2.2 Exclusion criteria If more than 15% of the included patients had missing data on the occurrence of pain, the study was excluded [e.g. as in a study by Kakulas and co-workers (Kakulas et al., 1990)]. In addition, a study was excluded in any of the following conditions: (1) Only a subtype of SCI pain was investigated (such as neuropathic pain). (2) No demographic data were available on the patients without pain. (3) The study population was less than 15 subjects.
2.3 Data collection method For data abstraction, articles were blinded for journal, authors, involved institutions, dates, countries and original language. Any references to these items were removed or replaced with dummy items to preserve story integrity. This was done to prevent subjective interpretations and predispositions becoming a source of bias. Blinding was performed in a digital document, if necessary aided by optical character recognition software (FineReader 10, ABBYY, Moscow, Russia). This resulted in similarly formatted documents, but with original tables and graphics (if they contained study results relevant to this review).
© 2014 European Pain Federation - EFIC®
S. van Gorp et al.
Spinal cord injury pain determinants
Table 1 Data collected from included studies. Population characteristics
Study characteristics
Proportion of patients with: Pain interference Pain catastrophizing Complete SCI Tetraplegia Traumatic SCI Married/living together Professional work Depression Gender ratio Age at study Age at injury Time since injury Region studied: Europe, Canada and Australia versus the USA
Pain assessment tool Pain definition Pain recall periods Primary study goal (pain study: yes/no) Sample size Attrition/response rate Sample selection method Population source (SCI centre: yes/no) Study method Study method uniformity Retrospective data In- and exclusion criteria Publication year Study duration Description non-responders Pain classification/taxonomy used: IASP [114], Bryce/Ragnarsson [9] or Cardenas [13] Reporting of: (each item: yes/no) Neuropathic or nociceptive pain prevalence Anatomical pain locations Pain descriptors
last column of Supporting Information Table S2 for abstracted definitions) and presented as such to six pain researchers independently. The items that were used to define the pain definition strictness were, among others, duration, interval, frequency, use of descriptors, wish for treatment, evident relation to SCI (in time or aetiology), and the terms used to describe pain: ‘pain’, ‘pain problem’, ‘painful sensations’, ‘unpleasant sensations’, ‘ache’ or ‘unpleasantness’. The researchers were blinded to all other data from the studies (including the pain prevalence), and they individually allocated pain definitions in three groups of increasing restrictedness. Specifically, they were asked to assign 0, 1 or 2 points for the strictness of the pain definition, indicating: 0: Mild strictness (hardly excluding any pain cases) 1: Moderate strictness (excluding some pain cases) 2: High strictness (excluding many pain cases) Final scores were based on consensus when unanimity was not met.
2.6 Study quality
In addition to the reported pain prevalence, the following population and study characteristics were abstracted, when available (Table 1). Studies that separately report on pain prevalence rates for population subgroups of patients who all either have or do not have one of the population characteristics we abstracted (e.g. a complete injury) are also split into ‘substudies’, based on these population subgroups (e.g. one substudy only reflects the pain prevalence of the part of the study population with a complete injury). These substudies are used for separate meta-regression analyses of the study population characteristics. These substudies are replaced for their main study when their specific characteristic (e.g. prevalence of complete injury) is added to the metaregression analysis. This way, the results are more likely to display the actual relation of this characteristic with the pain prevalence rather than a random relation.
From the abstracted data, a quality score for each included study was calculated (maximal score: 18). The quality criteria used were based on those recommended by LeboefYde and Lauritsen (Leboeuf-Yde and Lauritsen, 1995), and Walker (Walker, 2000). As the criteria were originally designed for studies on low back pain, the descriptions were modified to fit the current study on SCI pain (Table 2; more detailed online in Supporting Information Table S1). Besides analysing the effects of between-study differences in total quality scores and individual quality item scores (Table 2) on heterogeneity, non-stratified continuous data were also analysed when applicable (e.g. for response rates, population size; see Table 1). Moreover, in a similar way, we analysed the influence of some additional dichotomous and continuous variables possibly reflecting the study quality. These variables were the following: a difference found in population characteristics between participants and nonparticipants, study had a cohort design (yes/no), year of publication, and items indicating an extensive pain investigation: (1) the use of the International Association for the Study of Pain (IASP)/Siddal, Bryce/Ragnarsson or Cardenas SCI pain classification; (2) the report of neuropathic pain prevalence; (3) the report of musculoskeletal pain prevalence; (4) anatomical pain locations; and (5) pain descriptors.
2.5 Subgroup formation by pain definitions
2.7 Statistical procedures
Because of the potential effect of between-study differences in used pain definitions on the heterogeneity of reported pain prevalence rates, we classified studies in subgroups based on the restrictedness of their used pain definitions. Hence, pain definitions subgroups were made on the basis of the estimated probability that SCI patients with (mostly minor) pain problems fell below the pain detection threshold. To this end, the pain definitions were abstracted (see
For each study, the standard error of the prevalence rates was calculated using the standard formula for the standard error of a proportion and the mean of the reported prevalence rates (Chang et al., 2001). Meta-regression analyses were then used to explore sources of bias and heterogeneity. More specifically, possible reasons for (remaining) heterogeneity (I2) were investigated using a random effects model and assessing linear correlations with reported pain prevalence of
SCI, spinal cord injury.
2.4 Data items
© 2014 European Pain Federation - EFIC®
Eur J Pain •• (2014) ••–••
3
Spinal cord injury pain determinants
S. van Gorp et al.
Table 2 Meta-regression results of pain prevalence rates and individual study quality items. Correlation coefficient (%) w/ pain prevalence Covariable: Pain definition strictness (mild, moderate or high) Coefficient ≤ −8.7%; p ≤ 0.003
Study quality items: None 1. Representative study population 2. Non-responders described 3. Attrition/response rate (continuous values)
Pain prevalence and its determinants after spinal cord injury: A systematic review S. van Gorp1,2, A.G. Kessels3, E.A. Joosten1, M. van Kleef1, J. Patijn2 1 Department of Anesthesiology/Pain Management, Maastricht University Medical Center, The Netherlands 2 University Pain Centre Maastricht, Maastricht University Medical Center, The Netherlands 3 Clinical Epidemiology and Medical Technology Assessment, University Hospital Maastricht, The Netherlands
Correspondence Sebastiaan van Gorp E-mail: [email protected] Funding sources None. Conflicts of interest None declared. Accepted for publication 6 April 2014 doi:10.1002/ejp.522
Abstract Pain prevalence studies are important as they illustrate the magnitude of pain problems in a certain patient population, such as patients living with a spinal cord injury (SCI). Strikingly, reported pain prevalence rates in SCI patients are found to vary greatly, while determinants for the differences between pain prevalence reports remain unclear. We here aim to identify determinants for the differences (heterogeneity) in pain prevalence reports through a systematic review of all SCI pain prevalence reporting studies. Literature search was done using Medline, Cumulative Index to Nursing and Allied Health Literature, ISI Web of Knowledge and Embase. Data abstraction was performed while blinded and was followed by meta-(regression)-analyses. We identified 82 studies. Study design-related determinants of SCI pain prevalence reports were pain definition strictness (mild, moderate or high), primary study goal (pain study or not), data source (retrospective or not), and in a limited number of cases response/attrition rates. While correcting for these items, population characteristics correlating with pain prevalence rates were both proportion of patients with a depression and average time after injury (positive correlations). Between-study heterogeneity may remain even after the identification/correction of above-mentioned causes of heterogeneity.Pain after SCI does seem to relate to the duration of the injury and depression, yet major causes of bias in reported pain prevalence are found to be related to the primary study goal (pain study or not), choice of pain definition and the use of retrospective data.
1. Introduction Treatment of spinal cord injury (SCI)-related pain is still a major problem (Haisma et al., 2007; Teasell et al., 2010; Finnerup and Baastrup, 2012). The reported prevalence of pain after SCI, however, varies considerably, with rates reported from 19% to as much as 96% (Finnerup et al., 2001; New et al., 2002; Siddall et al., 2003; Raissi et al., 2007; Dijkers et al., 2009). In a recent, and the only published, systematic review of SCI pain prevalence studies, it was concluded that a definite statement about a pain prevalence could not be made due to this heterogeneity of © 2014 European Pain Federation - EFIC®
the study population (Dijkers et al., 2009). Identifying which factors cause this heterogeneity could reveal important information, as their identification supports their role as relevant pain determinants and/or study confounders. Unfortunately, the authors of the review reported to be unable to identify such factors, yet this was not the primary aim of their study. By using a more extensive and somewhat different approach, it might have been possible to identify factors causing the heterogeneity in reported SCI pain prevalence. For example, as pain prevalence has not yet been consistently reported to differ due to SCI aetiology, one can better not exclude studies based on aetiology and Eur J Pain •• (2014) ••–••
1
Spinal cord injury pain determinants
Databases • For this research, a new database was created from data abstracted from manuscripts identified using Medline, CINAHL, ISI Web of Knowledge, Embase, reference searches and contact with authors. • Factors we were able to identify as influential for the size of pain prevalence are depression, time after injury, pain definition, primary study goal (whether it was designed as a pain study or not), data source (retrospective or not) and response/ attrition rates. • Our manuscript provides important insights on (1) pain determinants, (2) design of pain studies, (3) interpretation of pain studies and (4) metaanalyses of prevalence studies. What does this study add? • This systematic review shows that the size of reported spinal cord injury pain prevalence is determined by pain definition, whether pain was the primary study outcome, depression prevalence, time after injury and population response rates.
investigate whether they actually influence on reported pain prevalence (Mariano, 1992; McKinley et al., 2002; Ibrahim et al., 2012). Similarly, as pain after SCI is typically chronic and the definition of ‘chronic’ remains ambiguous, one should not have to exclude studies that are not specifically reporting chronic pain (Purves et al., 1998; Cruz-Almeida et al., 2005). The resulting limitation of study exclusions is likely to result in a more valuable analysis of possible determinants on reported pain prevalence (e.g. depression prevalence or individual items of study quality) (Thompson, 1994; Stroup et al., 2000; Huedo-Medina et al., 2006). Therefore, in this systematic review, we wanted to investigate the impact of a wide variety of study design and study population characteristics on the heterogeneity of reported pain prevalence after SCI by using limited exclusion criteria and detailed meta-regression analyses.
2. Methods A systematic literature review was conducted on studies reporting on SCI pain prevalence (i.e. descriptive studies and observational studies). As no protocols are available for performing systematic reviews on prevalence studies specifi-
2 Eur J Pain •• (2014) ••–••
S. van Gorp et al.
cally, we used protocols for systematic reviews of studies on treatment or exposure effects, when applicable (Stroup et al., 2000; Dickersin, 2002; Higgins and Green, 2008). The following databases were used: Medline, Cumulative Index to Nursing and Allied Health Literature, ISI Web of Knowledge, and Embase. The search was wide and the queries comprised the terms ‘pain’, ‘dysaesthesia’ or ‘dysesthesia’, and ‘spinal cord injury/-ies’, ‘spinal cord disease/-s’, or ‘tetra-/quadri- or paraplegia’. The abstract and/or publication title were used to identify articles that might fit the topic. Next, the full document was requested, translated to English and screened whether these met our inclusion criteria or any exclusion criteria. Also, the reference lists from all screened articles were hand-searched for any old and/or missed articles. The major clinical research groups on SCI pain were contacted for any missing publications. Last, the major journals were hand-searched (i.e. Pain, Spinal Cord/Paraplegia, Journal of Spinal Cord Medicine, American Journal of Physical Medicine & Rehabilitation, and Journal of Rehabilitation Research and Development/United States Veterans Affairs).
2.1 Inclusion criteria Studies were included if they reported on the prevalence of pain after SCI, irrespective of the type of prevalence used (i.e. at a single time point or over a certain amount of time). In one case, only an abstract was available, and the data from the abstract were used (Rintala et al., 1991).
2.2 Exclusion criteria If more than 15% of the included patients had missing data on the occurrence of pain, the study was excluded [e.g. as in a study by Kakulas and co-workers (Kakulas et al., 1990)]. In addition, a study was excluded in any of the following conditions: (1) Only a subtype of SCI pain was investigated (such as neuropathic pain). (2) No demographic data were available on the patients without pain. (3) The study population was less than 15 subjects.
2.3 Data collection method For data abstraction, articles were blinded for journal, authors, involved institutions, dates, countries and original language. Any references to these items were removed or replaced with dummy items to preserve story integrity. This was done to prevent subjective interpretations and predispositions becoming a source of bias. Blinding was performed in a digital document, if necessary aided by optical character recognition software (FineReader 10, ABBYY, Moscow, Russia). This resulted in similarly formatted documents, but with original tables and graphics (if they contained study results relevant to this review).
© 2014 European Pain Federation - EFIC®
S. van Gorp et al.
Spinal cord injury pain determinants
Table 1 Data collected from included studies. Population characteristics
Study characteristics
Proportion of patients with: Pain interference Pain catastrophizing Complete SCI Tetraplegia Traumatic SCI Married/living together Professional work Depression Gender ratio Age at study Age at injury Time since injury Region studied: Europe, Canada and Australia versus the USA
Pain assessment tool Pain definition Pain recall periods Primary study goal (pain study: yes/no) Sample size Attrition/response rate Sample selection method Population source (SCI centre: yes/no) Study method Study method uniformity Retrospective data In- and exclusion criteria Publication year Study duration Description non-responders Pain classification/taxonomy used: IASP [114], Bryce/Ragnarsson [9] or Cardenas [13] Reporting of: (each item: yes/no) Neuropathic or nociceptive pain prevalence Anatomical pain locations Pain descriptors
last column of Supporting Information Table S2 for abstracted definitions) and presented as such to six pain researchers independently. The items that were used to define the pain definition strictness were, among others, duration, interval, frequency, use of descriptors, wish for treatment, evident relation to SCI (in time or aetiology), and the terms used to describe pain: ‘pain’, ‘pain problem’, ‘painful sensations’, ‘unpleasant sensations’, ‘ache’ or ‘unpleasantness’. The researchers were blinded to all other data from the studies (including the pain prevalence), and they individually allocated pain definitions in three groups of increasing restrictedness. Specifically, they were asked to assign 0, 1 or 2 points for the strictness of the pain definition, indicating: 0: Mild strictness (hardly excluding any pain cases) 1: Moderate strictness (excluding some pain cases) 2: High strictness (excluding many pain cases) Final scores were based on consensus when unanimity was not met.
2.6 Study quality
In addition to the reported pain prevalence, the following population and study characteristics were abstracted, when available (Table 1). Studies that separately report on pain prevalence rates for population subgroups of patients who all either have or do not have one of the population characteristics we abstracted (e.g. a complete injury) are also split into ‘substudies’, based on these population subgroups (e.g. one substudy only reflects the pain prevalence of the part of the study population with a complete injury). These substudies are used for separate meta-regression analyses of the study population characteristics. These substudies are replaced for their main study when their specific characteristic (e.g. prevalence of complete injury) is added to the metaregression analysis. This way, the results are more likely to display the actual relation of this characteristic with the pain prevalence rather than a random relation.
From the abstracted data, a quality score for each included study was calculated (maximal score: 18). The quality criteria used were based on those recommended by LeboefYde and Lauritsen (Leboeuf-Yde and Lauritsen, 1995), and Walker (Walker, 2000). As the criteria were originally designed for studies on low back pain, the descriptions were modified to fit the current study on SCI pain (Table 2; more detailed online in Supporting Information Table S1). Besides analysing the effects of between-study differences in total quality scores and individual quality item scores (Table 2) on heterogeneity, non-stratified continuous data were also analysed when applicable (e.g. for response rates, population size; see Table 1). Moreover, in a similar way, we analysed the influence of some additional dichotomous and continuous variables possibly reflecting the study quality. These variables were the following: a difference found in population characteristics between participants and nonparticipants, study had a cohort design (yes/no), year of publication, and items indicating an extensive pain investigation: (1) the use of the International Association for the Study of Pain (IASP)/Siddal, Bryce/Ragnarsson or Cardenas SCI pain classification; (2) the report of neuropathic pain prevalence; (3) the report of musculoskeletal pain prevalence; (4) anatomical pain locations; and (5) pain descriptors.
2.5 Subgroup formation by pain definitions
2.7 Statistical procedures
Because of the potential effect of between-study differences in used pain definitions on the heterogeneity of reported pain prevalence rates, we classified studies in subgroups based on the restrictedness of their used pain definitions. Hence, pain definitions subgroups were made on the basis of the estimated probability that SCI patients with (mostly minor) pain problems fell below the pain detection threshold. To this end, the pain definitions were abstracted (see
For each study, the standard error of the prevalence rates was calculated using the standard formula for the standard error of a proportion and the mean of the reported prevalence rates (Chang et al., 2001). Meta-regression analyses were then used to explore sources of bias and heterogeneity. More specifically, possible reasons for (remaining) heterogeneity (I2) were investigated using a random effects model and assessing linear correlations with reported pain prevalence of
SCI, spinal cord injury.
2.4 Data items
© 2014 European Pain Federation - EFIC®
Eur J Pain •• (2014) ••–••
3
Spinal cord injury pain determinants
S. van Gorp et al.
Table 2 Meta-regression results of pain prevalence rates and individual study quality items. Correlation coefficient (%) w/ pain prevalence Covariable: Pain definition strictness (mild, moderate or high) Coefficient ≤ −8.7%; p ≤ 0.003
Study quality items: None 1. Representative study population 2. Non-responders described 3. Attrition/response rate (continuous values)
