A Scoping Review of Chronic Low Back Pain Classification Schemes Based on Patient-Reported Outcomes

Abstract

Background:

In 2014 the National Institutes of Health Pain Consortium Research Task Force recommended that patients with chronic low-back pain (CLBP) be stratified by its impact on their lives. They proposed the Impact Stratification Score (ISS) to help guide therapy and facilitate study comparability. The ISS has been evaluated as a continuous measure, but not for use as a stratification or classification scheme.

Objectives:

Identify the characteristics of successful schemes to inform the use of the ISS for stratification or classification.

Study design:

Scoping review of the peer-reviewed literature.

Methods:

Search of PubMed, CINAHL, and PsycInfo to identify patient self-report-based classification schemes applicable to CLBP. Data were captured on the methods used for each scheme’s development, the domains covered, their scoring criteria and what the classification has successfully measured. The study was reviewed and approved by the ____ Human Subjects Protection Committee (2019–0651-AM02).

Results:

The search identified 87 published articles about the development and testing of five classification schemes: 1) STarT Back Screening Tool, 2) Multiaxial Assessment of Pain, 3) Graded Chronic Pain Scale, 4) Back Pain Classification Scale, and 5) Chronic Pain Risk Score. All have been shown to be predictive of future outcomes and the STarT Back has been found useful in identifying effective classification-specific treatment. Each scheme had a different classification scoring structure, was developed using different methods, and three included domains not found in the ISS.

Limitations:

Expanding the search to other databases may have identified more classification schemes. Our minimum number of publications inclusion criterion eliminated dozens of cluster analyses, some of which may have eventually been replicated.

Conclusions:

The methods used to develop these successful classification schemes, especially those that use straightforward scoring schemes, should be considered for use in the development of a scheme based on the ISS.

Introduction

In 2014 the National Institutes of Health Pain Consortium Research Task Force (RTF) on research standards for chronic low back pain (CLBP) recommended that patients with CLBP be stratified by its impact on their lives.(1) The RTF felt that stratification could have “descriptive and prognostic value and could supplement any pathophysiologic description,”(1)^,p2031 and improved “prognostic stratification of patients with CLBP is important clinically to help guide the nature and intensity of therapy, and important for researchers to adjust for confounding and to improve comparability among studies.”(1)^,p2040

The Institute of Medicine 2011 report, Relieving Pain in America noted that “No simple clinical test can assess a person’s subjective experience of pain. Seriousness depends on self-report… [of] pain’s impact on a person’s activities of daily living, ability to work, and quality of life.”(2)^,p86 The National Pain Strategy (NPS) went on to define high-impact chronic pain in 2015 as that “associated with substantial restriction of participation in work, social, and self-care activities for six months or more.”(3)^,p11 The NPS further stated that in order to lower the burden of pain and better target effective interventions: “It is important to differentiate people with high-impact chronic pain from those who maintain normal activities although experiencing chronic pain.”(3)^,p17 (emphasis added)

The RTF proposed the Impact Stratification Score (ISS) as a measure of CLBP impact. The ISS is calculated as the sum of the raw scores from nine PROMIS-29 items covering physical function, pain interference, and pain intensity with a possible range from a low of 8 (least impact) to 50 (greatest impact). The ISS has been evaluated as a continuous measure,(1,4,5) but it has not yet been evaluated for stratification or classification. The RTF offered cutoff scores for classifying patients as having CLBP of mild (ISS 8–27), moderate (ISS 28–34), and severe impact (ISS ≥35), but noted that these cutoffs were “relatively arbitrary.”(1)^,p2037

Useful classification schemes have been identified for many diseases--e.g., breast cancer,(6) hip or knee osteoarthritis,(7) heart failure,(8) and chronic and musculoskeletal pain.(9–12) These schemes use information from a variety of sources including patient history, physical exam, lab tests, imaging, and patient-reported outcome measures (PROMs). Our focus in this paper is to review classification schemes that have been used for CLBP that, like the ISS, depend only on PROMs.

The general goal of all classification schemes is to segment large diverse patient populations (e.g., patients with CLBP) into relatively homogeneous subgroups. Homogeneity can be defined in at least three ways. The subgroups could be similar in their level of current severity and concomitant effects—e.g., subgroups with similar levels of chronic pain impact according to several measures have been shown to have similar healthcare costs, unemployment and absenteeism.(13–16) The subgroups could also be defined by having similar future outcomes or recovery (i.e., prognosis), regardless of treatment.(10,17,18) Additionally, the subgroups could have similar response to specific treatments—i.e., vary by factors that are treatment modifiers.(10,17,18) An implicit goal (the “Holy Grail”(17)) of defining more homogeneous groups is to guide treatment. However, only the last definition of homogeneity (treatment modification) identifies the best treatments for each subgroup. The second (prognostic stratification) could also provide a more limited guide to treatment—e.g., by avoiding unnecessary treatment for those who were going to improve on their own. Prognostic stratification would also be useful for designing studies that minimize the heterogeneity of treatment effects. The first (severity) guides treatment only in the sense of identifying those most in need. There is no guarantee that the same classification scheme can generate subgroups with all three types of homogeneity.(10,17)

As a first step in evaluating the ISS as a classification scheme for CLBP we review other PROM-based schemes to determine how they were developed, the domains they measure, the way their classification categories are determined, and whether they have been shown useful in creating categories with similar severity, prognosis, and/or benefit of particular treatment. This information will be used to inform future research on the use of the ISS as a classification system.

Methods

This study followed the Preferred Reporting Items for Systematic Reviews and Meta-Analyses extension for Scoping Reviews (PRISMA-ScR) protocol(19) and the checklist is included as Appendix A. To identify existing PROM-based classification schemes for CLBP we searched the abstracts of articles in Medline (Ovid), CINAHL and APA PsycInfo from their inception through September 7, 2021. The full search for Medline is shown in Appendix B, but in general we looked for articles whose abstracts included either back pain or chronic pain and variations on stratification, classification, categorization, grading, subgrouping, or clustering. We restricted the search to human studies published in English.

We chose for consideration studies that described, used and/or evaluated classification schemes (i.e., methods by which patients are classified into mutually exclusive homogenous groups) that were used for adults with CLBP; required only information from PROMs for classification; and were the topic of at least three publications—i.e., the classification scheme was of enough interest to warrant more than one other article. Schemes that required information from physical exam, lab tests or imaging studies were excluded as were those with no more than two publications. Inclusion/exclusion criteria and whether individual studies met these criteria were collectively agreed upon by the first three authors. For each scheme the first author extracted the number of items used, the domains included, the method(s) used to develop the scheme, the formula used to classify, the variables the scheme’s results could discriminate, and the situations in which the scheme was shown to be useful for prognosis or as a guide for treatment.

Results

The database search resulted in 7,550 articles to consider (Figure 1(20)). After removal of duplicates and articles excluded based on reading the title and abstract, we reviewed 161 full-text articles and identified 87 articles describing five classification schemes that met our inclusion criteria. These are included in our narrative review.

Figure 1.

Flow of articles into the review

BPCS = Back Pain Classification Scale; CLBP = Chronic low back pain; CPRS = Chronic Pain Risk Score; GCPS = Graded Chronic Pain Scale; MAP = Multiaxial Assessment of Pain; PROM = Patient reported outcome measure; SBST = STarT Back Screening Tool

The domains and scoring rules used in the ISS and the five classification schemes are shown in Table 1. Details on each scheme’s background, development approach, scoring, ability to differentiate baseline characteristics, and use for predicting outcomes and guiding treatment are below.

Table 1.

Items, domains, scoring, and demonstrated usefulness of the Impact Stratification Score and five existing classification schemes

	Impact Stratification Score (ISS)	[Keele] STarT Back Screening Tool (SBST)	Multiaxial Assessment of Pain (MAP)	Graded Chronic Pain Scale (GCPS)	Back Pain Classification Scale (BPCS)	Chronic Pain Risk Score (CPRS)
How developed	Selected from PROMIS	Clinical advisory panel review of statistically promising items; ROC curves for cutoffs	Cluster analysis of responses to Multidimensional Pain Inventory	Mokken analysis to develop Guttman scale	Stepwise discriminant analysis	Latent transition regression analysis
Number of items	9	9	52	7	13	8 items + the # used for depression scale
Pain intensity	X	Bothersome	X	X	pain descriptor words	X
Pain interference	X		X	X		X
Physical function	X	X		X		X
Pain diffusion, frequency & duration		X				X
Beliefs about pain		X	X
Emotional well-being/ distress		X	X			X
Social support			X
Scoring	Sum scores 8–27 = mild 28–34 = moderate >=35 = severe	Sum score ≤ 3 = low risk; Sum score 4+ and subscore ≤ 3 = moderate risk; subscore 4+ = high risk	Group means from the original study or a computer program to allocate individuals into groups	Disability points 3–4 = Grade III; Disability points 5–6 = Grade IV; Disability points 0–2 and pain intensity ≥ 50 = Grade II; Rest = Grade I	The sum of the weighted values given to each checked word is compared to mean scores for each classification	Low risk = 0–7; Intermediate risk = 8–15; Possible chronic pain = 16–21; Probable chronic pain = 22+
Severity	X	X	X	X	X
Prognostic		X	X	X	X	X
Specify treatment		X

ROC = Receiver operating characteristic

STarT Back Screening Tool

The Subgroups for Targeted Treatment (STarT) back screening tool (SBST)(21) was developed in the UK for use in primary care with adults experiencing the spectrum of non-specific back pain. The goal was to develop and evaluate a brief and easy to score tool that used treatment-modifiable indicators to allocate primary care patients into one of three a priori initial treatment options based on their risk: 1) low risk group suitable for primary care management (e.g., analgesia, advice, and education); 2) medium risk group with high levels of physical indicators, appropriate for physiotherapy; and 3) high risk group with consistently high levels of psychosocial indicators, appropriate for a combination of physical and cognitive–behavioral management.

The set of prognostic constructs included in the tool were identified through analysis of existing datasets using forward stepwise binary logistic regression to predict the reference standards listed below and a review of the literature. A clinical advisory panel reviewed the list of identified constructs, excluded those considered rare or nonmodifiable in primary care, and helped choose the final constructs based on strength, independence, consistency of association with outcomes, and perceived face validity. Items were selected for each construct based on receiver operating characteristic (ROC) curves to identify optimal items (items that identify patients above the median scores seen using the full questionnaires) from multi-item constructs and input from the expert panel. Items to include in a psychosocial subscale were also identified.

In addition to estimating internal consistency and test-retest reliability of the tool, discriminant validity was assessed using area under the curve (AUC) from ROC curves for the overall tool scores and the score of the psychosocial subscale against the following dichotomized reference standards: back pain disability (Roland-Morris Disability Questionnaire score(22) or RMDQ >=7), whether there was referred leg pain, very or extremely bothersome back pain, catastrophizing (Pain Catastrophizing Scale score(23) >=20), fear avoidance (Tampa Scale of Kinesiophobia(24) >=41), and depression (Patient Health Questionnaire-2(25) score >=2).

To identify cutoff scores for each risk subgroup, ROC curves were examined. First, the optimal overall score (highest average sensitivity and specificity) that most consistently discriminated between reference standard cases and non-cases in terms of patients pre-defined as being suitable for standard primary care management (low disability, no leg pain, low bothersomeness) was determined. Then the psychosocial subscale score that best discriminated between the medium and high-risk groups was identified using predefined psychosocial reference standards (catastrophizing, fear avoidance, depression). Emphasis was given to maximizing specificity for the psychosocial subscale because it was believed that physiotherapy could help lower distress and that there could be negative impacts from cognitive-behavioral approaches in those without distress. The validity of the tool in terms of predicting 6-month disability (RMDQ >=7) was then evaluated using standard contingency table indices. The result was that individuals were classified as “low risk” if their total SBST score (out of 9 possible) was 3 or less, and “high risk” if their psychosocial subscale score (out of 5 possible) was 4 or above. The rest were considered “medium risk.”

Although the SBST was developed to be useful for patients who present to primary care with all types of non-specific low back pain, and despite studies showing that it can be a better predictor of future outcomes in patients with longer pain duration,(26) it has also been considered by some to be a screening tool to predict whether acute or subacute LBP would become chronic.(27) Since our review specifically focused on classification schemes used for adults with CLBP, we included and focus here only on studies where the SBST has been used in CLBP (or majority CLBP) samples.

Across reviewed studies the subgroups at baseline consisted of patients with different levels of pain intensity, activity limitations, disability (RMDQ), trunk motion, medication use, and a number of psychological measures(28–33) The SBST classification has also been shown to predict future Oswestry Disability Index (ODI(34))(35,36) and RMDQ(26,28,30,32,37–40) scores, pain intensity,(28,35,36,41,42) fear of movement,(35) work ability,(43) preference-based health-related quality of life (EQ-5D(44)),(32,36,45), 6-week ODI scores from an exercise program(46) and 2-year Graded Chronic Pain Scale grade and SF-12 physical and mental health composite scores from a comprehensive health program.(47)

Treatment assigned based on the SBST has also been shown to improve RMDQ scores(31,48,49) and preference-based health-related quality of life (EQ-5D)(45,49,50), reduce time off work,(45,48) and is likely cost-effective.(45,48–51) However, one trial of the SBST in a large health system where clinicians were trained on the tool and it was incorporated into the electronic health record system found no significant effect 2 or 6 months later on patients’ back-related physical function, pain severity or healthcare utilization.(52) It also had limited effect on clinician behavior; clinicians used it to assess risk in only about half of their patients and the treatments they recommended did not change.

Multiaxial Assessment of Pain

The goal of the original Multiaxial Assessment of Pain (MAP) study was to see whether psychosocial and behavioral measures could be used to derive a reliable and valid classification system for patients with chronic pain.(53) The author conducted cluster analyses of the nine scale scores of the West Haven-Yale Multidimensional Pain Inventory (MPI)(54) in two samples of patients with chronic pain referred to an outpatient pain clinic. Three profiles were identified from the cluster analysis: dysfunctional, interpersonally distressed, and adaptive copers. Relative to those in the other two groups, those in the dysfunctional profile reported higher pain severity and interference and psychological distress, and lower general activity levels and ability to control their lives. Those in the interpersonally distressed profile were more likely to report that their families and significant others were not very supportive of them and their pain, and those in the adaptive copers profile reported lower levels of pain severity and interference and psychological distress, and higher levels of daily activity and control of their lives relative to the other two groups.

Two methods can be used to classify patients into these profiles. There is a computer program developed by Rudy(55) that assigns those who have completed the MPI to one of the three groups or to an “other” group. A second more ad-hoc approach is to classify individuals based on how their MPI scores compare to the group means from the original study.(53)

The stability of the three-cluster solution was confirmed by using cluster analysis on two applications of the MPI in the same sample(56) and replication in samples with different subgroups of chronic pain patients,(57,58) including those with different chronic pain syndromes (i.e., low back pain, headache, and temporomandibular disorders).(59) The clusters were also evaluated by third-party reports.(60) One study used the Comprehensive Pain Evaluation Questionnaire, a shorter measure modelled after the MPI, and generated clusters very similar to the three patient profiles identified using the MPI.(61) Another study proposed that a fourth cluster (defensive repressors) was needed(62) and increased the applicability of the MAP groups.(63)

In reviewed studies the clusters at baseline were able to discriminate between patients with different levels of pain intensity, disability, affective distress, anxiety, depression, pain behaviors, fear avoidance, endurance coping, catastrophizing, functional self-efficacy, personality types, psychopathology, and medication use.(58,64–73)

MAP chronic pain profile status has been found to predict future sickness absence,(74,75) cost of lost productivity,(74) reductions in pain intensity and interference, and improvement of mental health and coping in response to various pain management programs,(58,66) whether someone completes treatment (a functional restoration program),(64) absence from work, general health status and use of healthcare resources following a vocational rehabilitation program,(76) and outcomes from Interdisciplinary Multimodal Pain Rehabilitation Programs.(77) However, one study found that targeting specific treatments to each profile was not more effective than standard care.(78)

Graded Chronic Pain Scale

The Graded Chronic Pain Scale (GCPS) was developed to offer a classification of chronic pain based on global (across-domain) measures of its severity.(16) The authors used Mokken analysis to test whether a set of pain-related items form a Guttman severity scale. From previous work they hypothesized that the lower range of pain severity would be measured by pain intensity and persistence and that the upper range would be measured by pain-related disability.

They found that three variables formed a Guttman scale: pain intensity measured as the mean of present, and worst and average pain in past 6 months; number of days in past 6 months kept from usual activities because of pain; and disability measured as pain interference with daily activities, changes in ability to take part in recreational, social, and family activities, and changes in ability to work, including housework all in the past 6 months. Scoring of the GCPS yields four chronic pain grades. Grades I and II are defined as those with fewer than 3 disability points (determined by the number of disability days and the disability score, with a maximum of 6 points), and either pain intensity <50 on a 0–100 scale (Grade I - low disability-low intensity) or pain intensity ≥50 (Grade II - low disability-high intensity). Grades III and IV are defined by disability, regardless of pain intensity: 3 or 4 disability points define Grade III and 5 or 6 disability points define Grade IV.

At baseline the grades were associated with significant and monotonic increases in the proportion of patients with depression, fair-poor self-rated health, frequent opioid use, frequent pain visits, unemployment and high pain impact (defined as 8 or more “yes” answers to a list of 16 pain-related functional limitation items).(16) Baseline chronic pain grades were also significantly associated with pain duration and physical and psychosomatic comorbidity,(79) health-related quality of life (single summary score of SF-8(80)) and somatization,(81) job change,(82) and with back pain advice and misconceptions,(83) days of sick leave, doctor visits, nights in hospital and unemployment.(84)

Chronic pain grade at baseline has also been found to predict 1-year pain grade, depression, fair-poor health status, frequent opioid use, frequent pain visits, high pain impact, and unemployment;(16) and pain grade and high pain impact at 3 years.(16) Baseline chronic pain grade predicted 6-month functional capacity, pain and the SF-36 Physical Component Summary score;(79) 1-year healthcare costs, number of visits and admissions, number of radiologic procedures and pain medication fills;(13) and healthcare costs and future chronic pain grade at 2 years after a back exercise program.(85)

A revised version of the GCPS was published in 2020 that categorizes those with chronic pain into mild (Grade 1), moderate (Grade 2) and high-impact (Grade 3) chronic pain.(86) Based on work for the US National Pain Strategy(87) and by the National Center for Health Statistics cognitive library,(88) those with high-impact chronic pain were identified based on responses of “most days” or “every day” to an item asking how often pain limited life or work activities. A summary score of 12 or greater on the Pain, Enjoyment, and General activity (PEG) scale(89) identified those with moderate chronic pain and those with lower scores had mild pain. The 12 or greater cutoff was chosen to represent a mean of 4 or higher across the scale’s three 0–10 items. A 4 on a 0–10 pain scale has been shown by others(90–92) to be the lower bound in identifying those with moderate pain. At baseline the Revised GCPS grades were associated with coping beliefs, reported health status, depression/anxiety, activity limitations and pain medication, including long-term opioid use.(86)

Back Pain Classification Scale

The Back Pain Classification Scale (BPCS) was developed to provide an easy to administer indicator of whether a patient had functional (psychological) or organic (physiological) CLBP. (93) The measure was developed using CLBP patients referred to neurosurgeons and orthopedic surgeons in the UK with probable intervertebral disc disease. These patients’ clinical and laboratory findings were reviewed by board-certified surgeons and assigned to one of the two groups. Patients were shown a list of 71 pain descriptor words from the Low Back Pain Questionnaire(94) and asked to choose the words that best describe how their pain typically feels. The authors then used stepwise discriminant analysis to identify the best combination of pain words that would distinguish between the functional and organic groups. The resulting set of 13 pain words were able to correctly classify patients as organic or functional with an overall 94% accuracy. Applying the discriminant scores to a second validation sample resulted in an accuracy rate of 83% overall. Another study team using a different sample found 80% accuracy for patients with chronic, intractable back pain.(95)

Patients classified as functional at baseline were found to have a higher incidence of neurotic disorders than those classified as organic,(96) and they were especially higher on the Minnesota Multiphasic Personality Inventory hypochondriasis scale. Classification at baseline using the BPCS was also associated significantly with medication need, patients’ rating of improvement and change in pain over a 12-month period.(97)

Chronic Pain Risk Score

The goal of the Chronic Pain Risk Score (CPRS) was to “discard the notion that ‘chronic’ means unlikely to change” and shift to predicting “the likelihood that clinically significant back pain will continue and, by extension, to [shift the focus to] steps that might be taken to reduce future risks of significant pain and dysfunction.”(98)^p305 The CPRS was developed using a sample of patients in the US with a history of primary care back pain visits.

Latent transition regression analysis(99) was used to empirically identify four pain severity classes (no pain, mild pain, moderate pain and limitation, and severe, limiting pain), and then estimate the probabilities of transitioning between these pain severity classes from one year to the next. Pain severity class was estimated using pain intensity, disability days, pain interference, pain impact score, unable to work for any health reason and kept from full-time work due to back pain. The first three of these were elements of the GCPS. Transition probabilities were based on three prognostic variables (depression, pain duration, and diffuse pain) which were chosen because they “have been consistently found to have prognostic value in predicting pain outcomes in longitudinal outcome studies.”(98)^p305 Clinically significant back pain was defined as having Chronic Pain Grades II, III or IV—i.e., intense back pain accompanied by mild to severe dysfunction.(16)

The resulting CPRS (possible range 0–28) is calculated as the sum of items from the GCPS(16) and prognostic variables. Days of activity limitation due to back pain from the GCPS was coded 0–4 and the pain intensity and pain interference items were each recoded from 0–10 to 0–2. The prognostic variables included the SCL-90-R depression score (recoded to 0–4), number of other pains (0–4), and the number of days with back pain in the prior six months (recoded to 0–4). The item scores were summed and the risk subgroups for the CPRS formed based on these cutoffs: ≥22 = probable chronic pain (≥80% probability of future clinically significant back pain); 16–21 = possible chronic pain (≥50% probability of future clinically significant back pain); 8–15 = intermediate risk of chronic pain (≥20% probability of future clinically significant back pain); and 0–7 = low risk of chronic pain (<20% probability of future clinically significant back pain).

The chronic pain classification based on the CPRS at baseline was designed to determine risk of clinically significant back pain in the future (Chronic Pain Grades II, III or IV)(98), but it has also been found to predict unemployment at 6 months and long-term opioid use.(100)

One study in the UK used the 14-item Hospital Anxiety and Depression Scale(101) (HADS) instead of the SCL-90-R (90 items) as their measure of depression and found slightly different cutoff points to be optimal in patients with low back pain.(102) The probable chronic pain cutoff in the low back pain population was the same, but the cutoff for possible chronic pain was increased and the cutoffs for intermediate or low risk chronic pain were slightly reduced.

Another study used a sample of US patients initiating primary care for back pain to examine whether the original CPRS could be improved by adding additional variables.(103) Their results are not directly comparable because they used a more stringent definition of a negative outcome (Graded Chronic Pain Grades III or IV), but the success of their models indicates that adding other variables (e.g., college graduate, recovery expectations) may enhance prediction.

Discussion

There are compelling reasons to classify chronic low back pain patients into homogeneous subgroups. From a research perspective this subgrouping could reduce patient heterogeneity and enhance trial efficiency, be used to report on the heterogeneity of treatment effect and would allow adjustment for baseline sample differences for standardized outcome comparisons. For providers and patients, classification would contribute directly to both diagnosis and prognosis and could help guide treatment.

This study’s search identified five established PROM-based classification schemes for CLBP. All have been shown to be useful for predicting future outcomes for patients, but only one (SBST) has shown benefit in being used to guide class-specific treatment. It has been noted that it was designed to do this,(104) whereas, one other (CPRS) was specifically designed for prognosis.(98)

Each scheme was developed using different methods. The SBST identified constructs through the analysis of existing datasets and a review of the literature and then used a clinical advisory panel and ROC curves to identify the items to use and the cutoff scores for each risk subgroup.(21) The MAP used cluster analysis on the nine MPI scale scores to identify three profiles.(21) These clusters proved remarkably replicable, especially in contrast to the dozens of other cluster analyses found that were never replicated more than once. Mokken analysis was used to test whether GCPS items covering pain intensity and pain-related disability formed a hierarchical severity scale.(16) They then used inflection points on the relationships seen between pain intensity and disability to identify their cutoff points for grades. The BPCS used stepwise discriminant analysis to identify the best combination of pain descriptor words that would identify those with functional and organic CLBP.(93) The CPRS used latent transition regression analysis to identify four pain severity classes and to estimate the probabilities of transitioning between these pain severity classes from one year to the next.(98) In summary, a variety of analytic approaches have been used to develop and evaluate classification schemes. One or more of these approaches might be useful in evaluating the ISS. The common element is to identify meaningful subgroups that are associated with differences in CLBP impact and/or can predict future outcomes.

At present no published studies have shown the ISS to be capable of prognostic stratification or treatment modification. All five identified classification schemes have been shown to be good for prognosis and most included domains not included in the ISS—e.g., at least three of the five schemes included measures of emotional well-being/distress. Two included measures of pain diffusion, frequency, and duration, and two included measures of pain beliefs. Therefore, it may be worth adding one or more of these domains to the ISS. It should be noted that to enhance prognosis the CPRS added measures of depression, duration of pain and number of pain sites to the GCPS.(98)

The schemes also vary widely in the number of items required, and thus, in patient burden. The GCPS has 7 items, the SBST has 9, the BPCS has 13, and the MAP has 52 items. The number of items in the CPSR depends on the instrument used to measure depression. Different studies used the SCL-90-R,(98,100) HADS(102,105,106) or the PHQ-8(103) resulting in 98, 24 or 16 items, respectively. The ISS has 9 items.

Three of the five classification schemes used empirically derived cutoff scores to identify their subgroups. The MAP was developed using cluster analysis so that subgroup classification requires calculation of the pattern of scores across subscales. Classification based on the BPCS, developed using stepwise discriminant analysis, requires the summation of weights applied to each selected pain descriptor words. These schemes are more difficult to apply clinically than those that apply upper and/or lower-bound cutoffs to simple total scores. Of the three schemes using cutoff scores, one (CPRS) applies cutoffs to a total score to define subgroups, but the two others also apply cutoffs to sub-scores. The SBST applies one cutoff to the total score to identify the lowest risk subgroup, and then applies a different cutoff to the score of a subset of items to identify the moderate and high-risk groups. The GCPS applies two cutoffs to a disability score—one that identifies Grade IV and one that separates Grade III from Grades I and II. Another cutoff is applied to pain intensity to differentiate between Grades I and II. Different scoring systems could be developed for the ISS in the future.

It should also be noted that refinement of the ISS-based classification scheme will require a different study design depending on the type of homogeneity desired.(18) If one only needs subgroups to be homogeneous in terms of the current severity of CLBP, cross sectional data will be sufficient. If, instead, we want the subgroups to be homogeneous in how their members change over time (prognosis; where the members of some subgroups do better no matter the treatment), then longitudinal data and analysis will be needed. However, if it is important to identify the best treatment for each subgroup (i.e., whether subgroup membership modifies the effect of a treatment), prospective randomized controlled trials are required. These designs can either take the shape of having those in each subgroup randomized to treatment or control or randomizing all participants to receive either usual care or treatment matched to their subgroup. The SBST(49,50) and the MAP(78) were both tested using this last design.

This study identified PROM-based classification schemes through a detailed review of the literature. However, it is not without limitations. The search only included PubMed, CINAHL, and PsycInfo. It is possible that the inclusion of other databases may have identified more classification schemes. We did not include classification schemes that had not been described, utilized and/or evaluated in at least three published studies. This exclusion criterion mainly eliminated cluster analyses on various instruments and groups of instruments. Finally, since we were only seeking to identify previously developed classification schemes, we did not perform a critical appraisal of the quality or validity of the identified studies.

Conclusions

The ISS is made up of nine items from the PROMIS-29 and was proposed by the RTF to stratify CLBP patients by the level of impact their condition has on their lives. The goals of this stratification were prognosis, to help guide therapy, and to aid researchers in identifying more homogeneous samples for trials and in comparing across studies. Nevertheless, to date the ISS has only been evaluated as a continuous measure. This study presents a review of the literature that identified five classification schemes that have been developed for CLBP and that have achieved one or more of these goals. The results of this search identified the methods used to develop these classification schemes and differences between the ISS and these schemes which can inform the use of the ISS for classification. Methods that result in classification according to empirically derived cutoff scores are favored for clinical ease of application. Like the Mokken scale analysis used in the development of the GCPS, item response theory may be useful in identifying levels of the ISS that represent clinically important differences in pain impact. Regression analyses including latent transition analysis may be useful in evaluating how well ISS subgroups predict future outcomes. It also may be worthwhile to supplement the ISS items to include domains included in the other schemes. Further work is needed to achieve the goals originally put forth for impact stratification using the ISS.