Abstract
Disablement is a multifactorial and complex process that creates a challenge for both the rehabilitation researcher and the clinical practitioner; however, each seeks to improve quality health outcomes for the adult or child with disability. Knowledge translation (KT) is an approach to evidence-based medicine where various evidence sources are aggregated so that clinical decisions regarding intervention selection and dosing result in beneficial care for an individual. Structural equation modeling was used to test a theoretical model of disablement based on the impairment, activity, and participation categories of the International Classification of Functioning, Disability and Health (ICF) level of functioning. Using available cohort data from a randomized controlled trial of people with poststroke walking disability, exploratory and confirmatory factor analysis revealed that the latent variables—impairment and activity—are separate disablement constructs that limit participation for a person with disability. Path analysis revealed that the direct effect of impairment on participation was not statistically significant; however, the indirect path from impairment to participation through activity was significant (indirect effect). The direct effect of activity on participation was significant. Model assumptions were tested with postintervention data from the same cohort. For people with disability after stroke, the probability that functional tasks could be performed with less effort was greater for those individuals who met or exceeded a physiologic walking threshold after a structured, progressive intervention provided by a physical therapist 6 months earlier. This article discusses how structural equation modeling can be used as a statistical method to explore the causal paths from disability to ability. The knowledge inquiry and synthesis phases of the knowledge-to-action KT framework parallel the essential elements of structural equation modeling; knowledge is created that is theoretically driven, supported by prior research, and analyzed, refined, validated, and tailored to address real-world problems. Using a theoretical framework of disablement with clinical judgment and quantitative research methods, a clinically intuitive model of disablement was validated. The positive dimension is a model of recovery where causal paths lead from disablement to ablement. Innovative approaches in rehabilitation research design along with pragmatic application of research to practice are needed to improve today's health outcomes for people with disability.
Evidence-based practice, the International Classification of Functioning, Disability and Health (ICF),1 randomized controlled trials (RCTs), and, most recently, knowledge translation (KT) are all elements that clinicians are expected to understand and apply in the care of patients and their families. Knowledge translation is described in various ways; however, the fundamental purpose of KT research is to close the gap from knowledge to practice.2 For health-related research, the gap between knowledge and practice is a chasm.3 This is especially true in the area of rehabilitation or disablement research, which, to date, has had limited impact on the health and wellness of people with disabilities.4 In fact, trends in the US health care system show an ever-increasing disparity between people who need and people who actually receive or have access to rehabilitation services.5,6
Many factors contribute to the gap between knowledge and practice in rehabilitation. First, the field of rehabilitation research is young, with no concerted rehabilitation research effort at the National Institutes of Health until 1990, when the National Center for Medical Rehabilitation Research of the National Institute of Child Health and Human Development was established.7 The time from concept development to inception of phase II and phase III rehabilitation clinical trials is long, and the primary outcomes of an intervention RCT typically do not address the biopsychosocial realities that a clinician and a patient face in the clinical situation.8,9 Furthermore, nonpharmacologic intervention clinical trials are methodologically complex, compounded by the real-life dynamics of therapist, patient, and family, who all interact in the study of disablement.10 Rehabilitation trials that investigate populations with the highest degree of disability (eg, age-related conditions such as stroke and Parkinson disease, traumatic brain injury, and Alzheimer disease) typically include participants who have both physical and cognitive impairments and complicated personal and environmental contextual factors that can interact to influence intervention outcomes.
Improvement in health outcomes for a person with disability is the ultimate goal of rehabilitation research. Several recent knowledge translation models described processes and strategies specifically designed to improve health outcomes through cumulative knowledge so that research findings are appropriately disseminated and implemented for the benefit of patients and to address public health needs.2 The knowledge-to-action framework described by Graham et al is a knowledge translation model that can be applied to rehabilitation research and practice.11 This model consists of 3 phases of knowledge creation that lead to an action cycle with application to practice (Fig. 1). The initial process is knowledge inquiry, which includes accumulated knowledge from theoretical and mechanistic perspectives, such as basic and clinical health-related research evidence. The second phase is knowledge synthesis, in which knowledge is analyzed, refined, validated, and tailored to address the problems and barriers to the translation of knowledge to the clinical environment. Finally, knowledge is translated to action through the development of knowledge tools, which transform cumulative research evidence into relevant and ecologically valid (eg, economical, efficient, and effective) key messages that are more likely to be adopted and sustained in clinical practice.
Knowledge-to-action process. Reprinted with permission from Graham ID, Logan J, Harrison MB, et al. Lost in knowledge translation: time for a map? J Contin Educ Health Prof. 2006;26:13–24. Copyright 2006, John Wiley & Sons Inc.
Clinical practice is informed from various sources of research evidence (ie, basic science, social science, ethics) synthesized into clinical practice guidelines, clinical expertise, and the patient-family perspective. The knowledge-to-action framework is an approach to knowledge translation in which all sources of health-related research evidence are aggregated and synthesized in the knowledge inquiry phase to converge on high-impact treatment strategies.9 The current prevailing approach to evidence-based practice is to rely on RCTs and systematic reviews as the highest level of evidence to guide practice. Clinicians want to be informed of the most beneficial treatment for a person, but rehabilitation clinical trials report differences in treatment effects between groups. Kaplan et al12 recently noted that overreliance on RCTs as the predominant source of evidence for clinical decision making can be as unstable as a “one-legged stool.” One of the advantages of using the KT approach to evidence-based practice is that multiple, cumulative methods can be used to understand disease, recovery, and the clinical strategies that lead from disease to improved health outcomes.
Jette and Latham, the editors of this special issue on disability intervention, have challenged both researchers and clinicians to address the “persistent lack of evidence about strategies to reduce activity limitations and improve participation for people with disability.”13(p324) We propose that there is adequate evidence if multiple sources of rehabilitation research evidence are aggregated into a plausible, coherent model of disablement to guide clinical practice. The primary purpose of this article is to propose a theoretical model of disablement and recovery with causal paths that lead from disease or injury to disability (disablement) and causal paths that lead to improved participation for people with disability (recovery). We introduce structural equation modeling as a statistical method for quantifying and validating the model.
Structural equation modeling is a multivariate statistical method that often is used by psychologists and social scientists to test complex theoretical models.14 Structural equation modeling is useful for rehabilitation research because it permits the measurement and assessment of simultaneous, multiple dependent relationships between variables. An essential element of structural equation modeling is that it requires sound theory based on empirical and practical evidence to establish a model. The theoretical model provides a consistent, comprehensive, and cohesive explanation of actual phenomena. In this article we demonstrate (1) how structural equation modeling can quantify and validate an evidence-based conceptual model of disability and recovery with hypothesized causal paths between clinical measures of impairment, activity, and participation; (2) that available data instead of new data can be used to apply structural equation modeling to various populations with disability; and (3) how structural equation modeling can be an effective way to translate the complex, multifactorial phenomenon of disablement into a model of recovery for RCT hypothesis generation and to guide clinical practice.
Theoretical Model of Physical Disablement
Constructs of Physical Disability
In the United States, the National Health Interview Survey is used to collect from randomly sampled households health-related information regarding the health status of children and adults.15 On the basis of the disability constructs of the National Health Interview Survey, a person is considered to have a physical disability if some form of limitation in basic actions (eg, ability to walk) interferes with the ability to perform functional tasks (eg, independently perform self-care) or to participate in complex activities (eg, live at home independently). If a child or adult needs the assistance of another person to complete age-expected basic mobility tasks, perform basic or complex activities of daily living (ADL), or engage in social or work-related activities as a result of a health condition that lasts 3 months or more, then that person is considered to have a severe physical disability.16
The role of physical therapists within the health care service delivery system is to diagnose and provide treatment for acute, episodic, and chronic health conditions that result in movement dysfunction.17 An essential element of the physical therapist evaluation, at the level of body structure and function impairment, is to determine the effect of impaired musculoskeletal, cardiovascular, integumentary, or neuromuscular physiologic capacity on the ability to complete body movements. Thus, a primary rehabilitation goal of physical therapy is to provide interventions that increase physiologic capacity in impaired body systems; however, the ultimate goal of rehabilitation is to restore or maintain physiologic capacity or slow its rate of decline so as to decrease disability. Thus, we propose that the effectiveness of physical therapy interventions is optimal when the interventions efficiently and effectively increase physiologic capacity in body systems to the threshold needed to support the biomechanical, neuromuscular, and cardiovascular demands of functional task performance. For a child or adult with physical disability, increased physiologic capacity in body functions and structures and greater capability to perform functional tasks increases the likelihood for improved ability to participate in home, social, or community life.
Disablement Framework
At present, there is worldwide acceptance of the ICF as a common reference framework for characterizing the multidimensional factors that affect a person's experience of disablement.1 Because the ICF framework captures both physiologic and social determinants of health, it is used for disability classification, health assessment categorization, and rehabilitation research. The advantage of the ICF framework for rehabilitation research is that variables associated with physiologic outcomes (ie, the consequences of a health condition for impairment severity, functional limitations, and participation restrictions) can be separated from other determinants of health (ie, environmental and personal contextual factors). Although it is widely accepted that the body function, activity, and participation levels of functioning are conceptually linked with both positive (ability) and negative (disability) dimensions,18 it has not been empirically confirmed with clinical measurement data assessed by clinicians that the impairment and activity constructs are independent with causal relationships to participation as reported by the person with disability. On the basis of the ICF conceptual framework, we developed a theory-driven, hypothesized structural model of disablement with causal paths between body function impairments and activity limitations that lead to restricted participation and disability (Fig. 2). Available RCT data were used to quantify and validate the hypothesized model.
Theory-based, hypothesized structural model of physical disablement. (1) Theoretical rationale for model constructs. Structural model based on International Classification of Functioning, Disability and Health levels of functioning. Latent variables represent unobservable, yet significant, contributors to a complex state such as disablement. In this structural model of disablement, the unobservable, latent variables are represented by the constructs “impairment” and “activity.” These variables are the exogenous (ie, independent) variables that exert an influence on other variables in the quantitative model. In this model, the endogenous (ie, dependent) variable is “participation”; thus, the endogenous variable participation is influenced by the latent variables impairment and activity. Impairment is an unobserved construct that represents disease-specific severity. Physical therapists determine the degree of disease-specific severity through musculoskeletal, neuromuscular, cardiovascular, and integumentary assessments of body structures and functions needed to perform movements. Activity is an unobserved construct that represents basic physical actions of the whole person. Physical therapists determine activity limitations through observation of performance in mobility activities such as sitting, standing, walking, running, and climbing stairs. The endogenous variable participation represents actual observed or self-reported physical actions and functional tasks needed for age-expected participation, such as play, school, work, homemaking, and living at home independently. (2) Theoretical rationale for causal paths. The directional arrow between impairment and activity indicates the correlation between the 2 independent variables. Directional arrows from impairment and activity to participation indicate direct effects of the independent variables on the dependent variable. The dotted line indicates the indirect path from impairment through activity to participation.
Establishing the Disablement Model: Structural Equation Modeling
Participants
Model testing was done with available baseline (preintervention) data from 80 participants with moderate to severe walking impairments 4 months to 5 years after a unilateral, first-time stroke; these data were from the Strength Training Effectiveness Post-Stroke (STEPS) study. The results of the STEPS RCT, including CONSORT compliance, were reported previously.19 The STEPS RCT was 1 of 5 clinical trials of the Physical Therapy Clinical Research Network funded by the Foundation for Physical Therapy.20
The STEPS study was designed on the basis of literature from multiple research sources that attributed walking disability after stroke both to the severity of stroke-specific impairments, such as loss of selective motor control, weakness, and balance deficits, and to limitations in walking ability; thus, both the severity of sensorimotor impairments and limitations in walking ability are factors that may independently contribute to participation restrictions after stroke.21,22 As previously reported, the measurements used in the STEPS study have established psychometric properties recognized as valid and reliable clinical indicators of motor, balance, and walking abilities after stroke-related disability; physical therapists trained in standardized testing procedures completed all clinical measures.19
Theoretical Model of Disablement
Figure 2 depicts a structural model of disablement that identifies the hypothesized relationships between the impairment and activity categories from the ICF model and the dependent variable (participation). The testable hypotheses of this theoretical model are that: (1) impairment, activity, and participation are separate constructs of disablement and (2) there are causal paths between movement-related impairment (ie, strength, joint mobility, and pain) and limitations in functional mobility (ie, inability to sit, stand, walk, or run) that influence the ability to participate in functional life tasks for the person with a disability.
A model of disablement based on the ICF levels of functioning can be used to guide clinical practice. Physical therapy interventions that increase physiologic capacity and functional ability should decrease disablement. Physical therapy interventions that increase physiologic capacity and functional ability should increase ablement and decrease disablement. Ablement interventions provided by physical therapists include the use of therapeutic exercise and activity to increase physiologic capacity, assistive device prescription and training needed when physical capacity is insufficient to perform physical activity, or recommendations for environmental adaptation when structural or architectural barriers limit access needed to participate fully in life activity. However, the theoretical model depicted in Figure 2 is specific to the hypothesized relationships between movement-related impairments and activity limitations that are presumed to be modifiable with physical therapy interventions, resulting in increased participation for a person with disability.
The model hypothesizes that there is a direct relationship between participation and the severity of disease-specific impairments because of reduced physiologic capacity within the musculoskeletal, neuromuscular, integumentary, or cardiovascular system to support age-expected movement (Fig. 2, directional arrow from “impairment” to “participation”). In addition, there is a direct relationship between participation and the level of functional mobility (Fig. 2, directional arrow from “activity” to “participation”). In other words, when the ability to perform activities is limited or another person is needed to complete essential basic or instrumental ADL (IADL), the ability to participate in age-expected life roles is restricted.
In addition, the model hypothesizes an indirect pathway from impairment to participation (Fig. 2, directional arrows and dotted line from “impairment” to “activity” and “activity” to “participation”). In other words, impairment severity has both a direct effect on participation (ie, greater impairment severity is associated with more restricted participation) and an indirect effect through the mediating variable activity (ie, greater impairment severity contributes to limitations in activity performance) that restricts participation. Thus, a person's actual, self-reported participation is a summation of the direct and indirect effects of impairment severity.
Defining the Latent Factors
Figure 3 shows the relationships between the assumptions that clinicians or clinical scientists make regarding clinical decision making and intervention effectiveness. They select measurements, based on the ICF framework, that are presumed to quantify impairment severity and activity limitations. Clinicians assume that people with disability will have an improved ability to participate in age-expected ADL if their interventions reduce impairment severity or if functional activity can be performed with less assistance.
Clinician assumptions for a measurement model of poststroke physical disablement: (1) Clinical assumptions. Measurements are selected that have sound psychometric properties, presumed to be representative of the impairment and activity constructs. (2) Categories of the International Classification of Functioning, Disability and Health (ICF). In the available data set, research assumptions were that the selected measures of lower-extremity Fugl-Meyer motor score (LE-FM), lower-extremity (LE) paretic isometric torque, and Berg Balance Score are objective measures of motor selectivity, strength, and balance severity associated with the construct impairment; walking speed (comfortable, fast) and distance walked are objective measures of walking capability, capacity, and endurance associated with the construct activity. (3) Patient-centered outcome. The ADL/IADL score of the Stroke Impact Scale (SIS) was selected as an objective measure of participation ability from the perspective of the patient.
On the basis of these clinical assumptions, the observable measures presumed to be most associated with the latent variable “impairment” included the Fugl-Meyer lower-extremity (LE-FM) motor score as an indicator of movement selectivity,23 the isometric torque of the paretic limb as an indicator of strength for the flexors (hip, knee, and ankle dorsiflexor torques combined) and extensors (hip, knee, and ankle plantar-flexor torques combined),24 and Berg Balance Scale (BBS) score as an indicator of balance.25
The measures associated with the latent variable “activity” included comfortable walking speed as an indicator of preferred walking ability, fast walking speed as an indicator of maximal walking capacity, and distance walked in 6 minutes as an indicator of walking endurance.19
The dependent variable was the level of difficulty in completing basic ADL or IADL, as reported by a person with disability, using the ADL/IADL score of the Stroke Impact Scale (SIS).26
Data Analysis
The correlation among all measurements was examined with Pearson correlation analysis. From a group of measurements with a strong correlation (r≥.8), only 1 measurement was selected for further analysis. Exploratory factor analysis (EFA) was used to explore and extract the latent factors from the clinical measurements. We used varimax rotation to maximize the independence between latent factors, and we used promax rotation to examine the cross-loading of clinical measurements. Measurements with severe cross-loading (equally loaded on multiple latent factors) were removed from the final model. Exploratory factor analysis was conducted with SAS 9.2 (SAS Institute Inc, Cary, North Carolina). Structural equation modeling was used for confirmatory factor analysis (CFA) and path analysis to test the overall fitting based on the model structure derived from EFA. Structural equation modeling was conducted with the Mplus4 package (Muthén & Muthén, Los Angeles, California). For structural equation modeling estimation, the maximum-likelihood mean-adjusted chi-square test was used because it is robust for non-normality. Model fit indices used included the Comparative Fit Index (CFI), Tucker-Lewis Index (TLI), and Standardized Root Mean Square Residual (SRMR).
Results of the Exploratory Phase
All observable variables were standardized to z scores for the exploratory and confirmatory analyses. The correlation matrix of z scores for all observable variables is shown in Table 1. Fast gait speed and 6-minute walk distance showed extremely strong correlations with comfortable gait speed (r=.94 and r=.97, respectively). Because of the severe colinearity between walking variables, comfortable gait speed was used as the single representative for walking capacity from the 3 activity measures in the hypothesized model (Fig. 3). Comfortable gait speed was selected because it is practical for clinical use in all settings.
After removal of the fast gait speed and 6-minute walk distance variables, the remaining measurements were standardized to z scores (X̅=0, SD=1) and entered into the EFA with varimax rotation. Two factors were identified during the first round of EFA on the basis of a scree plot and the Kaiser-Guttmann rule (eigenvalue of >1). Because the paretic LE flexor and extensor strength variables loaded on one factor and gait speed loaded on another factor, we assigned factor 1 as the impairment factor and factor 2 as the activity factor.
The results of the first round of EFA showed that the LE-FM motor score loaded weakly on both factor 1 and factor 2. When promax rotation was used, the cross-loading of the LE-FM motor score between factors was more pronounced but was still relatively low (factor 1 loading weight=.15, factor 2 loading weight=.07). Because the LE-FM motor score loaded weakly on both factors, this score was eliminated from the subsequent factor analysis. Furthermore, the LE-FM motor score is not generally accepted as being practical for clinical use.
The remaining 4 measures (paretic LE flexor strength, paretic LE extensor strength, BBS, and gait speed) were entered into the final EFA. The results of the final factor loading from the factor analysis with varimax rotation are shown in Table 2. Contrary to our original hypothesized model, the BBS did not load on the impairment factor but did load on the activity factor. We were not overly surprised because the BBS, as a measure, represents a continuum of tasks that increase in difficulty—starting with basic balance abilities, such as static sitting and standing, and advancing to balance abilities during task performance, such as transferring and reaching. The EFA results showed that the BBS is a strong indicator of the activity construct for people with walking disability after chronic stroke.
As determined by the EFA process, the final structural model included the observable variables paretic LE flexor strength and paretic LE extensor strength, assigned to the latent variable impairment, and the observable variables BBS and gait speed, assigned to the latent variable activity.
Results of the Confirmatory Phase
Confirmatory factor analysis was used in the confirmatory phase to validate the goodness of fit of the theoretically derived structural model defined in the exploratory phase. According to Kline,27 a model is considered a good fit if the value of the chi-square test is insignificant, at least one incremental fit index (CFI or TLI) is greater than .90, and one “badness-of-fit” index (SRMR) is less than .08. On the basis of these criteria, all model fitting indexes showed excellent goodness of fit for our hypothesized structural model of disablement (χ2 of model fit=4.7, df=3, P=.2; incremental fit indexes: CFI=.99, TLI=.96; badness-of-fit index: SRMR=.02).
The BBS and gait speed loaded on the same factor. The high factor loading for each observable variable indicated that the measures were independent and were highly associated with the latent variable activity (factor loading: BBS=.88, gait speed=.95). Loading on the latent variable impairment was high for both observable variables: paretic LE flexor strength and paretic LE extensor strength (factor loading: paretic LE flexor strength=.95, paretic LE extensor strength=.88).
The results of CFA and path analysis are shown in Figure 4. The results indicated that the direct effect of impairment on participation was not statistically significant (r=.25, P=.10); however, the indirect path from impairment to participation, mediated through activity, was significant (indirect effect, r=.21, P<.05). The direct effect of activity on participation was significant (r=.33, P<.05). This confirmatory model suggested that participation is influenced by and weighted toward the direct effect of the latent variable activity on participation and the indirect effect of impairment severity, mediated through activity, on participation.
Confirmatory model of disablement after stroke. Standardized estimates for the confirmed structural model are shown for the significant and nonsignificant direct (solid lines) and indirect (dashed lines) effects with causal paths (directional arrows). Directional arrows between the Impairment and Activity constructs and the measures that actually loaded on each factor are shown. All model fitting indexes showed excellent goodness of fit for the hypothesized structural model of disablement (χ2 of model fit=4.7, df=3, P=.2; incremental fit indexes: Comparative Fit Index [CFI]=.99, Tucker-Lewis Index [TLI]=.96; badness-of-fit index: Standardized Root Mean Square Residual [SRMR]=.02). The direct effect of impairment on participation was not significant (not significant [ns], P=.10). The indirect effect of impairment on participation, mediated through activity, was significant (*P<.05). The direct effect of activity on participation was significant (*P<.05). The total effect is the combined direct and indirect influences on the dependent variable participation. The statistical convention is to report the regression coefficient (beta). However, the beta value from the standardized variable (transformed to a z score) in the regression model is identical to the correlation coefficient and is more easily interpreted by clinicians. The model suggests that participation is weighted toward the direct effect of the latent variable activity on participation and the indirect effect of impairment severity, mediated through activity, on participation. LE=lower extremity, e=error term.
Summary of the Model of Disablement
Structural equation modeling including EFA, CFA, and path analysis showed that the model of disablement is statistically sound. The structural model of disablement, validated with clinical data from the physical therapist evaluation, confirms that the ICF levels of functioning (impairment, activity, and participation) are separate disablement constructs. Path analysis confirmed that the impairment and activity constructs are separate but make related contributions to participation. Activity has an independent influence on participation (eg, a direct, causal relationship between slow walking speed and difficulty performing ADL/IADL tasks) but also mediates the influence of impairments on participation (eg, a poor ability to generate LE force limits walking ability, and this impairment contributes to difficulty performing ADL/IADL tasks).
The quantitative measurement model revealed that paretic LE strength, balance, and gait speed are high-impact clinical indicators that equally account for walking disability after stroke. On the basis of path analysis, clinical hypotheses can be developed and tested to determine whether physical therapy interventions can improve participation for people with disability after stroke. Such clinical hypotheses are as follows:
-
If activity is directly and positively associated with participation ability, then task-specific interventions that increase balance capacity (ie, positive change in BBS score) and walking capacity (ie, positive change in gait speed) likely will have a positive impact on the capability to participate in ADL and IADL tasks (ie, positive change in the SIS ADL/IADL score).
-
If paretic-limb capacity is positively associated with participation ability, mediated through activity, then interventions that increase force production needed to support task performance likely will have a positive impact on the capability to participate in ADL and IADL tasks.
Application of the Structural Model to Practice
The disablement model described in this article is a structural model of the negative dimension that leads to disability. For example, the negative consequence of poststroke hemiparesis, impaired balance and limited walking ability, lead to disability by restricting the ability to participate in life activities. However, the positive dimension of the same structural model is a model of recovery. The recovery model assumes that increased capacity to generate force and perform functional activities leads to increased ability to participate in life activity. Both negative and positive dimensions of this structural model can be used to describe the negative or positive dimensions of disease on health status, that is, disablement or recovery.
Approaching rehabilitation from the positive dimension emphasizes recovery; thus, interventions that improve physiologic capacity and functional performance should lead to greater participation for people with disability. On the basis of therapeutic principles from exercise science, we can pose testable clinical questions. Will physical therapy interventions that increase physiologic capacity to a critical threshold result in the ability to perform functional tasks with less effort? This approach is fundamentally different from that of traditional clinical decision making, goal setting, and treatment planning because the therapy goal is determined on the basis of a physiologic performance threshold rather than a finite dose of treatment sessions determined on the basis of practice guidelines, individual therapist preference, or funding restrictions.
Reference values such as the minimal clinically important difference (MCID) are used by clinicians to assess the minimal change in performance needed in a clinical activity measure for a patient to perceive a beneficial effect on health status. In previous work,28 we determined the MCID for gait speed to be 0.16 m/s at 20 to 60 days after stroke by using anchor-based methods to define the threshold at which an improvement in gait speed is most likely to result in a reduction in disability level. On the basis of the premise of the present article, the MCID would represent a capacity marker of clinical performance that is associated with a change in capability.
Typically, RCTs compare the mean difference between primary outcomes (eg, gait speed) rather than person-centered participation outcomes (ie, SIS ADL/IADL score) for intervention groups that receive a fixed number of treatment sessions based on a study protocol. To test our model of recovery, we used available longitudinal data from measurements of comfortable gait speed and SIS ADL/IADL scores at baseline and at a 6-month follow-up in a rehabilitation RCT. In this demonstration, intervention effectiveness is not based on group membership; rather, effectiveness is determined by the capability of a person to perform ADL tasks with less effort 6 months after receiving a physical therapy intervention. We hypothesized that the capability to perform ADL tasks would be proportionately greater for participants who met or exceeded the capacity threshold than for participants whose performance was below the capacity threshold.
Participants
Model testing was done with available baseline and 6-month longitudinal data from the STEPS study. The data were from 61 participants who received 1 of 4 interventions including task-specific training with or without a progressive resistive exercise program for 24 sessions over 6 to 8 weeks and for whom both baseline and 6-month follow-up measurements of comfortable gait speed and SIS ADL/IADL scores were obtained. Walking speed change scores (from baseline to 6-month follow-up) were calculated to determine whether participants met or exceeded the walking capacity threshold (ie, ≥0.16 m/s). Nineteen participants met or exceeded the capacity threshold; the performance of 42 participants was below the capacity threshold. The Wilcoxon rank sum test was used for comparison of the capacity threshold groups.
Results
Baseline walking speed and SIS ADL/IADL scores were not significantly different between the groups (Wilcoxon rank sum test: walking speed, P=.26; SIS ADL/IADL score, P=.21). Individual SIS ADL/IADL change score data for participants who achieved or did not achieve the capacity threshold are shown in Figure 5. The median SIS ADL/IADL change score was significantly higher for participants who met the capacity threshold (achieved threshold: median=2.5, range=−22 to 40; below threshold: median=−2.5, range=−17 to 25; Wilcoxon rank sum test: P=.04).
Box plots of Stroke Impact Scale (SIS) ADL/IADL change scores (from baseline to 6-month follow-up) for participants with gait speed changes of greater than or equal to 0.16 m/s (the functional capacity threshold). Threshold achieved: n=19, median=2.5, first-third quartile range=−5 to 12.5 (shaded box), interquartile range=−22 to 40. Below threshold: n=42, median=−2.5, first-third quartile range=−7 to 5 (open box), interquartile range=−17 to 25. Medians were significantly different between the capacity threshold groups (P=.04). The dashed line indicates the clinically important change in the SIS ADL/IADL score. The probability of a change in the capability to perform ADL tasks was significantly greater when the capacity threshold was met (P=.01).
The data shown in the box plots (Fig. 5) revealed the high variability across participants in the capability to perform ADL tasks. However, the interquartile range was higher for participants who achieved the capacity threshold (interquartile range=−5 to 12.5) than for those who did not (interquartile range=−7 to 5). A change score of 5.9 is considered to represent a clinically important difference for the SIS ADL/IADL score29; 47% of the participants who met the capacity threshold had change scores that exceeded the clinically important difference for the SIS ADL/IADL score, whereas the value for participants who did not was 16%. Thus, the probability of achieving a clinically important change in the capability to perform ADL tasks is significantly greater if the capacity threshold is met (χ2=6.37, df=1, P=.01).
Summary of the Model of Recovery
In this demonstration, we provide evidence that physical therapy interventions that increase walking capacity above a critical physiologic threshold can improve the capability of people with disability after stroke to participate in ADL and IADL tasks. The structural model revealed the critical causal paths from impairment to activity to participation and from activity to participation. Physical therapy interventions that increase physiologic capacity to meet the neuromuscular, musculoskeletal, integumentary, and cardiovascular demands of functional task performance in daily life activity should be the essential therapeutic goal for any person when the consequences of disease or injury interfere with the ability to participate in expected life roles.
Learning to ride a bicycle illustrates the interrelatedness of capacity and capability and, coincidentally, may be meaningful to patients, students, researchers, and clinicians in the translation of an important therapeutic principle from a theoretically driven structural model. Lower-extremity power, balance control, and cognitive capacity to scan the environment are all physiologic requirements for riding a bicycle safely in the community. However, building physiologic capacity for strength, balance, and cognition will not result in the ability to ride a bicycle. A fundamental change in the capability to perform a newly acquired functional skill can occur only with task practice, such as riding the bicycle. The model of recovery applied to practice reveals that impairment-based interventions (ie, strength, balance, and mobility exercises) are unlikely to develop fundamental changes in functional performance without task practice. Likewise, the greatest change in functional performance is unlikely to occur if interventions focus only on task-specific training and are not accompanied by interventions to build capacity.
Discussion
The primary purpose of this perspective article on disablement was to demonstrate that structural equation modeling may be a useful method for knowledge translation from rehabilitation research to clinical practice because it permits the measurement and assessment of simultaneous, multiple, dependent relationships between variables. Despite the accepted use of the ICF as a theoretical framework for research and practice, the ICF levels of function have not been proven as independent constructs with causal relationships between impairment and activity to participation on the basis of assessments that are commonly used by clinicians in clinical practice. We used available rehabilitation RCT data to demonstrate that the ICF levels of functioning are separate constructs with causal paths to disablement or recovery. Theory-driven models validated with clinical data can provide meaningful insights to guide clinical practice and hypothesis generation for rehabilitation and disablement research.
Structural equation modeling can be an effective way to translate a complex, multifactorial phenomenon such as disablement into a model that can guide clinical practice. We suggest that this method may be a potential approach to knowledge translation, as suggested by Graham et al,11 because structural equation modeling relies on reliable and valid clinical measurements, a confirmatory process that is based on theoretical and pragmatic knowledge, and the identification of causal paths that can guide clinical decision making. The structural model of disablement presented here is simple, with causal paths that are clinically intuitive. Application of the recovery model as a framework for practice suggested that improved participation ability for a person with disability is likely to be associated with rehabilitation interventions that increase physiologic capacity and functional perfomance along causal paths that lead from disablement to ablement.
One of the limitations of health-related research is the relatively limited translation of research evidence to practice despite systematic reviews, clinical practice guidelines, or other efforts to close this gap.9 The evidence-based practice approach has involved major efforts in training clinicians to access and appraise research literature, understand research methodology, and synthesize a body of knowledge into interpretable results that can be applied to the delivery of care to individual patients.11 However, knowledge translation models have emerged as a potential approach for improving health outcomes in medicine1; perhaps they can be applied to improve rehabilitation outcomes for people with disability.30
How can the disablement and recovery model inform rehabilitation research, including innovations in RCT design? Typically, an RCT is the gold standard for evidence regarding intervention effectiveness. An RCT usually compares differences between intervention groups in a primary outcome that is a measure of activity (ie, walking speed, Oswestry Disability Index score) instead of participation (ie, 36-Item Short-Form Health Survey questionnaire, SIS). In addition, nonpharmacologic RCTs are constrained by CONSORT requirements, with fixed endpoints defined by the group response (such as the between-group mean velocity change score from 2 to 12 months).10,31 Nonpharmacologic clinical trial design is beginning to move away from rigid CONSORT requirements, which were developed primarily for drug trials, toward mixed-method research.32 A mixed-method design, particularly for complex health conditions such as depression33 and Alzheimer disease,34 involves theory-driven algorithms and clinical reasoning as the clinician reacts to the participant's response.
Mixed-method research for complex health conditions increases the likelihood that research evidence will be driven more by the “logic of clinical practice”35 than by the artificial constraints of the gold standard RCT. The overgeneralization of RCT findings can harm an individual who does not receive the optimal treatment or can result in underutilization or overutilization of health care resources if a standardized dose is applied to all people.12 Our demonstration with STEPS data illustrated this point. The STEPS study provided evidence that walking speed and distance increased and that the increases were sustained at a 6-month follow-up for people who had walking disability after stroke and who received 24 sessions of task-specific walking training with or without an additional exercise program19; however, the levels of long-term changes in individual participation ability were low across the whole cohort. In contrast, we demonstrated that improved participation was proportionately higher for people who achieved a functional capacity threshold after a physical therapy intervention. Thus, perhaps physical therapy intervention effectiveness should be assessed on the basis of an increase in functional capacity to a critical threshold rather than the total number of intervention sessions. An innovative approach to mixed-method rehabilitation clinical trial design would include intervention arms in which the achievement of a physiologic threshold based on activity performance criteria triggers a change in intervention strategy and primary outcomes are based on a person's ability to participate in real-world activities.
Limitations
The work described in this article was not an a priori, hypothesis-driven study; rather, we used structural equation modeling to demonstrate that existing data can be used to identify models of disablement and recovery that can inform medical rehabilitation research and practice. Unlike recent structural equation modeling studies in which questionnaire or survey data were used to investigate disablement and the presumed constructs of the ICF model,36,37 our structural model was derived from measurement data collected through evaluations by clinicians, and the primary participation outcome was reported by people with disability.
Our sample size was smaller than what is typically recommended for structural equation modeling; thus, it is conceivable that our analysis lacked the power to detect a true relationship between impairment and participation (direct path) (ie, a possible type II error). Despite this limitation, we were able to demonstrate, using a simple structural model and a few parameter estimates, that the impairment, activity, and participation constructs were separate. The model was supported by the strong convergence of multiple goodness-of-fit indicators. It is important to note that the clinical measurements of impairments and activities and the weights of the causal paths between these constructs and person-reported disability most likely will differ depending on the specific health condition. Thus, existing clinical data from populations with other conditions (ie, spinal cord injury, low back pain, and patellofemoral pain) should be used to quantify and validate the causal paths of disablement for other disease-specific health conditions that cause disability.
In a recent structural equation modeling analysis with presurgical questionnaire data from more than 400 people who were scheduled to receive a total hip or knee replacement because of chronic osteoarthritis pain, Pollard et al36 demonstrated that the ICF impairment, activity, and participation constructs were separate when disability was associated with musculoskeletal lower-limb joint pain. Despite the differences between that study36 and our analysis in terms of health conditions, methods used for measurement data collection, and sample sizes, both investigations revealed that the direct causal path from impairment to participation was not significant even though CFA validated the notion that the impairment, activity, and participation constructs were separate. One of the strengths of our approach to modeling of the constructs and causal paths of disablement was the apparent face validity between our model of disablement and its applicability to clinical practice. The measurement model included data from psychometrically sound assessment tools commonly used by clinicians to measure impairment severity and activity limitations; however, the structural model with causal paths shifted the primary outcome of therapy to the level of participation, as reported by a person with disability.
Conclusion
Despite the well-recognized value of rehabilitation, the disparities in health care for children and adults with physical disabilities have never been greater in the history of medical rehabilitation.4,5,13,38 The knowledge-to-action framework is a knowledge translation model that may provide a means to bridge the gap between knowledge and practice in rehabilitation. Innovative approaches involving mixed-methods design and statistical methods such as structural equation modeling should be considered for disablement and rehabilitation research. Advancements in health-related research design that prioritize a person-centered response as the primary outcome, coupled with collaboration between rehabilitation researchers and clinical experts to answer clinically relevant questions, increase the likelihood that knowledge creation will translate to improved participation for people with disability.
Footnotes
-
Both authors were involved in the concept/idea/research design, writing, and data analysis of this report.
-
This work was supported, in part, by the Foundation for Physical Therapy and the Physical Therapy Clinical Research Network.
- Received January 10, 2011.
- Accepted September 14, 2011.
- © 2011 American Physical Therapy Association
References
Issue highlights
