A validity-driven approach to the understanding of the personal and societal burden of low back pain: development of a conceptual and measurement model

A grounded approach to conceptualization and the identification of draft items maximizes the likelihood that the resultant tool will fully cover the construct; in this case, the burden of low back pain. Our process for grounded conceptualization included three concept mapping groups that utilized processes modified from the methods developed by Trochim [25]. Concept mapping is a formal group process tool for identifying and organizing ideas on a topic of interest. The steps include development of a seeding statement, generation of statements (brainstorming), sorting of the statements, generation of a concept map and revision of the concept map.

The Cabrini Human Research Ethics Committee approved the study (No. 13-02-03-09) and all patients who participated in the study provided written informed consent.

There are many options for naming groups of items, including clusters, domains, factors, scales and dimensions. We chose not to use the term 'dimensions' because it has a specific meaning when using multidimensional scaling (MDS), which relates to the number of spatial dimensions in which the MDS software seeks to fit the distances between items. We also chose not to use the term 'factors' because it relates to a specific type of statistical technique - factor analysis.

We use the term 'clusters' when we refer to the outcomes of concept mapping and the term 'domains' when we refer to a refined, hypothesized structure for a proposed instrument. These domains are referred to technically as latent variables during psychometric analysis using structural equation modeling. We use the term 'scales' after the psychometric properties of the instrument have been established.

We consider that the matching between clusters, domains (latent variables) and scales is one of the critical elements in demonstrating construct validity of the final tool. We also use the term 'statements' to refer to the ideas generated by participants in the concept mapping groups, and use the term 'items' when we have begun to redraft these statements into a form that is suitable for a questionnaire.

To maximize the richness and depth of the data obtained, we used a nominal group process that is a method for obtaining the most comprehensive possible range of ideas from individuals on a topic of interest [26]. Usual practice in qualitative data collection is to sample to saturation, which is the point at which no new ideas are emerging. The concept mapping process goes to great lengths to be as exhaustive as possible within each group, and therefore saturation is often reached after a small number of groups.

A carefully crafted seeding statement was presented to individuals in each group, who were then asked to work alone for 5 minutes to generate ideas in response to the statement. The seeding statement for patients was: 'Thinking as broadly as you can, generate statements about how low back pain affects your life (considering both yourself and those around you)'. For the health professional group, the seeding statement was slightly different: 'Thinking as broadly as you can, generate statements about how low back pain affects the life of people with the condition and the community'. Participants were asked to write down their responses according to the following rules: one idea per statement, use bullet points, make the statements brief, and work alone. The nominal group technique uses a facilitator who then asks that the ideas be presented to the group in an egalitarian manner, whereby each participant in turn presents one item on their list, starting with the first, until all items have been presented. Participants were discouraged from passing judgments about the statements but were encouraged to seek clarification of the nature or content of the statement if necessary. The critical advantage of this approach is that the perspective of individuals is collected in a manner that is not influenced or biased by the researcher nor influenced by other, and at times dominant, group members.

Once all statements had been presented, participants were asked to sort the statements into conceptually similar groups according to any system that made sense to them. For this step, they were asked to work alone. MDS and cluster analysis were then used to process participants' input and generate two-dimensional maps of key concepts related to low back pain impact and the interrelationships among these clusters.

Participants were asked to independently consider and label each group of statements and to check that each of the statements fit within that group. If a statement or statements were not considered to fit within the group, participants were asked to nominate the appropriate grouping. They were also asked to consider whether any of the groups should be joined. After this had been completed on an individual basis, we again used a nominal group approach to organize the final groupings, their labels and the included statements. We also checked for any missing domains/concepts.

A similar concept mapping exercise was conducted via email and through a face-to-face workshop at the 10th International Forum for Primary Care Research on Low Back Pain held in Boston in 2009. The expertise of the expert international group was broad and included primary care, rheumatology, occupational health, physiotherapy, chiropractics, epidemiology, public health and health policy.

Prior to the Forum, an email was sent to all participants who had been allocated to the workshop (n = 31) asking them to generate statements in response to a similar seeding statement: 'Thinking as broadly as you can, generate statements about ... how low back pain affects the life of people with the condition and those around them'. Forty-five percent (14/31) of participants responded to this task.

The statements from the patient group, from the clinician/health policy group and from the Forum workshop participants were then combined and redundancies were removed. This final set of statements were then sent to Forum participants in a second email requesting that they sort the statements into conceptually similar groups according to any system that made sense to them. They were also asked to rank each of the statements in order of importance. Fifty-eight percent (18/31) completed this task.

The same process of multidimensional scaling and cluster analysis was used to process participants' input and generate two-dimensional maps of key clusters of low back pain impact and the interrelationships among these clusters.

At the Forum we presented the results of the patient and clinician/health policy-maker workshops and the final concept map that was generated by the Low Back Pain Forum workshop participants. Participants were asked to independently consider and label each group of statements and to check that each of the statements fit within that group. If a statement or statements were not considered to fit within the group, participants were asked to nominate the appropriate grouping. They were also asked to consider whether any of the groups should be joined. After this had been completed on an individual basis, the group worked together to organize the final groupings, their labels and the included statements. We also checked for any missing domains or concepts.

At this point we had three concept maps: two from the initial groups and one from the international expert group. The process of integrating the three maps included a number of steps. In addition to the two-dimensional MDS that underlies the concept maps, we undertook three-dimensional and four-dimensional MDS using the Clustan software [27] and repeated the cluster analysis on the outputs of these analyses. Sometimes a three-dimensional or four-dimensional MDS can more accurately capture the similarities between statements and leads to cleaner (more self-evidently homogeneous) clusters. The output of the MDS and cluster analysis is viewed as a tree diagram; a diagram that allows all cluster solutions from a single cluster to a number that equals the number of items to be examined. This diagram allows us to examine the division of items each time a cluster is split into two smaller clusters to determine whether this split has substantive meaning. Through this process we looked to determine the smallest number of clusters (most general concepts) that made sense, the largest number of clusters (most refined concepts) that made sense, and the items that are considered most typical of each refined concept.

At the level of the most general concepts, the results from different concept mapping groups tend to be similar. This means that the results can be combined at this level and the results from the different concept mapping groups provide different details under these high-level concepts. These results for each group analysis are displayed as mind maps (Mindjet Mind Manager software, 2010, MindJet Ltdr, Sydney, New South Wales, Australia). The mind maps are then combined so that the common general concepts form the first level of detail and the branches represent each substantively meaningful split identified through examination of the tree diagrams.

Throughout this process the researchers attempted to use the cluster names assigned by the original group participants. The mind map aims to provide a clear hierarchical overview of the burden of low back pain as seen by the participants. This hierarchical representation does not, however, show the richness of the relationships between the clusters as well as the original maps. For this reason, the integrated mind map needs to always be considered in conjunction with the original maps.

The next step in refining the structural model was to check the proposed domains against the original item pool. The researchers classified every statement produced by the three concept mapping groups according to the proposed domains. In performing this classification we were looking for: items that cannot be classified - these may indicate the need for additional domains; items that seem to relate to more than one domain - these may be ambiguous items or may indicate a relationship between the hypothesized domains; domains that still seemed to contain multiple concepts and may need to be split; and match between domain names and the item content - a poor match may require renaming the hypothesized domain.

Our approach to construct definition and instrument development is based on the tenet that construct validity needs to be the primary concern of all instrument development activities and of all proposed applications of instruments. This is consistent with the descriptions provided by Pedhazur and colleagues [28], and the Standards for Educational and Psychological Testing developed jointly by the American Educational Research Association, the American Psychological Association and the (American) National Council on Measurement in Education [29, 30]. The Standards describe validation as an ongoing process that commences with the conceptualization and continues each time someone proposes an additional interpretation or application of the tool [29].

While it is common practice in health research to refer to a tool as either validated or unvalidated, it is not tools but only their interpretations and applications that are validated. To maximize the likelihood of producing valid data in relation to a range of possible interpretations and applications of a tool, there are development processes that seek to protect the instrument against two categories of error; measuring less than the proposed construct (construct underrepresentation) or measuring more (construct irrelevant variance) [29]. Protection against the first type of error requires rigor in the processes of conceptualization and definition and the identification of a range of indicators. Protection against the second type of error requires rigor in psychometric analysis. We believe that three disciplines help achieve this necessary rigor: the use of grounded approaches for construct definition; the development of a priori structural hypotheses (that define relevant versus irrelevant variance); and the development of a priori, relational hypotheses as a basis for future construct validation.

The Standards contain 24 standards related to validity of a measure, but the first four of these specifically relate to the linkage between validity and possible interpretations (Table 3). It is clear that the authors of the Standards place a significant onus of responsibility on the developers of instruments to clarify the interpretations that are supported by available evidence at any point in time.

An important initial step in scale development, and the final step in development of the hypothesized model, involves writing (hypothesized) descriptors about characteristics of people with a high score and people with a low score on scales related to each hypothesized domain. This exercise helps to clarify whether the domain can be represented as a scale or whether it is simply a checklist of possible characteristics, the desired range of item difficulty, and possible relationships between scale scores and other variables (other scales, demographic and clinical variables, outcomes of interventions). This final point is an important and often neglected step in preparing for construct validation by developing a broad range of a priori hypotheses about the behavior of the scales in relation to other variables (the so-called nomothetic web) [28, 31].

In considering the ongoing process of validation once the instrument has been developed, it is necessary to specify the interpretations and applications that we are seeking to validly and reliably achieve. These interpretations and applications are presented in Table 4, together with some of the evidence - or the processes to obtain the evidence - that is required to support validity of each type of interpretation/application. Table 4 shows that the expansion of valid interpretations and applications occurs in a number of stages that build upon each other. The first two stages are integral to the process of psychometric development through the application of draft tools to a construction and validation sample (see below). Evidence in relation to the second two proposed applications accrues through use of the tool, while the step from interpretation of data at the group level to interpretation at the individual level usually requires additional technical analysis as well as a body of evidence about the meaning and behavior of each scale acquired through widespread use with groups. There are also steps that can be taken during the psychometric development phase to increase the likelihood that the tool will be usable with individuals. These steps relate to ensuring that the scales have certain properties in relation to the range of difficulty that the items cover and the extent to which they can give scores spread evenly across this range. While the meaning of difficulty is clear in academic tests in this situation, a difficult item would be one where few people give the most positive possible response. There are also different reliability requirements related to each level of use, with individual applications having the most stringent requirements.

One of the primary reasons for conducting this research was the observation that existing instruments inadequately capture the range of impacts of low back pain that are commonly reported by people with low back pain and the clinicians that work with them. This project has produced a conceptual framework that includes many concepts not included in the tools most commonly used to assess needs and/or outcomes for people with low back pain.

At one end of the spectrum, because low back pain has until recently been thought mainly a work-related problem, outcome measures have often been limited to occupational aspects of burden: most of all, measures of absence from work, and the consequent financial costs. Such measures only capture part of the burden of low back pain.

At the other end of the spectrum, Deyo and colleagues proposed a core set of six indicators for routine clinical use that included pain symptoms, function, well-being, disability, social role and satisfaction with care [32]. Another core set of measures proposed for evaluating the effectiveness of treatment in clinical trials and routine care was proposed by Bombardier [33]. Recognizing the importance of the patient's perspective, she proposed the following five domains: back-specific function, generic health status, pain, work disability, and patient satisfaction [33]. Similar to these proposals, the Initiative on Methods, Measurement, and Pain Assessment in Clinical Trials group recommended a core set of six outcome domains be considered in chronic pain clinical trials: pain, physical functioning, emotional functioning, participant ratings of global improvement and satisfaction with treatment, symptoms and adverse events, and participant disposition [34].

More recently, Kopec and colleagues proposed a web-based computerized adaptive test (CAT-5D-QOL) to measure five domains of health-related quality of life (Daily Activities, Walking, Handling Objects, Pain or Discomfort, and Feelings) for patients with back pain based upon item banks developed for these domains relevant to arthritis [35]. Many measures have been developed to specifically quantify the limitations that low back pain places upon functional status. For example, in a 2004 systematic review Grotle and colleagues identified a total of 36 back-specific questionnaires [36]. The authors classified the content of the questionnaires based upon the World Health Organization's International Classification of Functioning, Disability and Health (ICF); they found that while most of the questionnaires had a focus on activity limitations, there was a wide variation in their underlying constructs and content. Many questionnaires also included constructs of pain and symptoms, sleep disturbances, psychological dysfunction, physical impairment and social functions.

The brief and comprehensive ICF core sets for low back pain, based upon the ICF framework, are further attempts to develop a standardized set of indicators to encompass the key functional problems of patients with low back pain envisaged to be used for a variety of purposes including clinical studies and multidisciplinary assessment in clinical care [37]. These were formed by consensus among a group of international clinical experts comprising physicians, occupational and physical therapists, who integrated evidence from a Delphi exercise to identify the most relevant ICF categories in patients with chronic health conditions including back pain [38], a systematic review to identify the concepts contained in outcome measures in clinical trials of musculoskeletal disorders and chronic widespread pain [39], and a study in a convenience sample of people undergoing rehabilitation for one of several chronic conditions including low back pain who were administered the ICF checklist [40]. The comprehensive and brief ICF core sets include 78 and 35 categories, respectively, which cover not only aspects related to pain but also a wide spectrum of activities, social and environmental factors that affect functioning. In keeping with our conceptual model, these core sets recognize the importance of support and relationships, attitudes of significant others and health professionals as predictors of disability in people with low back pain.

A Norwegian study in a convenience sample of 118 patients with low back pain, however, has identified gaps in the comprehensive ICF core set with respect to capturing problems of importance to patients [41]. This study compared the relationship between health problems rated by health professionals using the comprehensive ICF core set and patients' self-reported health problems identified by the Oswestry Disability Index and the World Health Organisation Disability Assessment Schedule II. Relevant domains not covered by the ICF included the subjective domain related to the impact of back pain and the feeling of being a burden to their family, while problems with sexual functions and relationship were poorly reflected in the health professionals' assessments.

Our model for the measurement of the burden of low back pain aims to comprehensively capture all of the various impacts of this condition on the individual. The model includes several domains that have not until now been considered important to measure in patients with low back pain, although they may contribute significantly to the individual's burden; for instance, loss of independence, worry about the future, negative or discriminatory actions by others, and secondary health effects, among others.

The new tool will have a wide range of potential applications for researchers, clinicians, policy-makers and insurance agencies; and for a range of purposes, including needs identification, service planning, evaluation, research and, eventually, for individual clinical assessment and monitoring. In suggesting such a range of applications, we are aware of our responsibility to consider the evidence for validity in relation to each interpretation and application [29, 30].

To strengthen potential generalizability, we have used both a local approach and an international approach to scope and define low back pain burden, nominal group approaches and concept mapping. The questionnaire is being developed with input from an international team of experts in the field. To facilitate comparison of the burden of back pain between countries and between studies, steps are being taken to ensure its wide applicability and cross-cultural generalizability.

In assessing health priorities, allocating resources, and evaluating the potential costs and benefits of public health interventions, governments often consider the burden of a disease and its contribution to the overall health of the population. Information obtained from a single comprehensive measure of back pain burden will greatly enhance research efforts to identify major determinants of back pain burden and population groups that are most affected and to ensure efficient allocation of resources. This information may also inform the development and evaluation of novel new interventions that could improve patient-relevant outcomes.

While the measurement model (Figure 3) does test for a single underlying latent variable, which we have called the burden of low back pain, we expect the questionnaire will be used as a multidimensional tool providing a profile of scores across the various scales. We will not be attempting to provide a scoring mechanism to gain a single overall score. In our experience it is more useful to be able to use profiles of scores to describe the needs of different patient groups and to distinguish the benefits of different types of interventions than to generate a global indicator that is at such a high level of abstraction no-one will be clear what it means. A profile of scores will also serve to highlight the critical psychosocial aspects of the burden of low back pain that have not been adequately addressed in existing tools. It is hoped that this profile of scores will support a greater clinical emphasis and increased research focus on these aspects of the burden experienced by people with back pain.