Advancing Physical Therapist Interventions by Investigating Causal Mechanisms

Since the inception of patient-centered care,1 the physical therapy profession has emphasized the importance of patient-reported outcomes such as function and quality of life.2 Accordingly, contemporary trials of physical therapist interventions often designate patient-reported outcomes as primary endpoints, while impairment-level outcomes are considered secondary.3,4 In such trials, it is often assumed that intervening on a set of impairments should yield improvement in outcomes that are relevant and perceptible to everyday functioning. This reflects the fundamental basis of physical therapist practice, as encapsulated by most clinical reasoning frameworks5 and by more broadly accepted frameworks from the National Academy of Medicine and World Health Organization.6 Treatments will often have an impact on patient-reported outcomes via the modulation of causal mechanisms at the impairment-level. That is, an effective treatment should alleviate impairments that are causally associated with patient-reported outcomes. Conversely, an ineffective treatment might fail to alleviate key impairments, or the treatment might alleviate impairments that are not causally associated with patient-reported outcomes. These concepts are depicted in a series of causal diagrams in the Figure. This mechanistic approach has informed clinical reasoning, but there is limited evidence to demonstrate that intervening on selected impairments will necessarily yield a clinical effect on patient-reported outcomes.7,8 To date, only a few studies have empirically tested the underlying mechanisms of physical therapist interventions.8 The lack of evidence for treatment mechanisms leaves researchers with little information to refine existing treatments, or to devise new treatments based on new targets. This knowledge gap also leaves clinicians to rely on implicit theories to devise and deliver treatments. In this article, we discuss ways of improving physical therapist interventions by studying causal mechanisms and conclude with a clinical scenario illustrating how a mechanistic study might inform practice. Causal diagrams illustrating potential scenarios of when an intervention would be effective or ineffective. Black solid arrows represent positive causal effects, the red solid arrow represents a negative causal effect, and broken lines represent null effects. “Complex intervention” refers to interventions that are made up of multiple components. The first model on the left panel assumes that the intervention does not have a causal effect on harmful mechanisms; all models on the right panel assume that the intervention does not work through unknown/unmeasured mechanisms. Well-conducted randomized controlled trials can provide robust estimates of the average causal effect of a physical therapist intervention.9 However, conventional analysis of randomized trials alone cannot explain how an intervention works, or why it does not work.10 To answer these questions, mechanistic analyses such as “causal mediation analyses” can be embedded into randomized trials.11 The evidence generated from these analyses can play an important role within a comprehensive program of building, testing, refining, and implementing interventions.12,13 In this issue, we see an example of a mechanism evaluation of a randomized trial where Mansell and colleagues test multiple mechanisms of a pain education intervention that aimed to reduce disability (a patient-reported outcome) in patients with low back pain.14 Because pain education did not offer added benefit over usual care, the aim of their study was to understand why the intervention did not produce the intended effect on disability. In other words, the investigators were interested in identifying where the hypothesized causal mechanisms broke down. Contrary to their hypotheses that pain education would reduce disability via changes in illness perceptions, catastrophizing, and back pain beliefs, the authors found that none of these impairments mediated the treatment effect. For the most part, this was because the treatment did not cause a sufficient change in the hypothesized mediators. This might suggest that a stronger causal effect on the mediators might have increased the overall treatment effect on disability. We also learn from this study that changes in illness perception and catastrophizing were associated with disability, but changes in back pain beliefs were not. This implies that illness perceptions and catastrophizing are important factors that should be targeted in future iterations of pain education interventions. The challenge is to devise interventions that can modulate these mediators to a greater magnitude (and for more individuals) so it can have a greater impact on disability. Furthermore, that changes in back pain beliefs were not associated with improvements in disability suggests that reframing erroneous beliefs is unlikely to reduce disability. However, this finding should be interpreted with caution because the estimate could have been influenced by unknown and unmeasured confounders,15 and suboptimal implementation of the intervention might have led to insufficient variation in the measure of back pain beliefs. To interpret this data with more confidence, these findings should be replicated in future mechanism studies. Notwithstanding some limitations, Mansell and colleagues provide useful information that should guide the development of revised forms of pain education for patients with low back pain. Here we present a clinical scenario illustrating how the findings of a mediation analysis might inform clinical practice. Clinical problem: You are an outpatient physical therapist scheduled to see a patient who has rheumatoid arthritis of her hand. The patient reports marked pain and loss of hand function. You are aware of a randomized trial that has demonstrated efficacy of an exercise program for improving hand function in this population.16 The trial intervention is composed of 7 mobility exercises and 4 strengthening exercises. However, from this you are not sure which of these exercises are essential components to the intervention. To help understand which exercises would most likely lead to functional improvements, you search for evidence of the underlying mechanisms of this multicomponent exercise program. Solution and caveats: A mechanism study suggests that grip strength partially mediated the overall treatment effect.17 You look for the key effects in this study—the total, indirect, and direct effects. The total effect (4.35 [95% CI = 1.51-7.19]) represents the entire effect of the treatment on hand function; the indirect effect (0.85 [0.10-1.60]) represents the effect of the treatment on hand function that is channeled through grip strength; and the direct effect (3.50 [0.71–6.29]) encompasses all other mechanisms that are not channeled through grip strength. Therefore, the proportion mediated through grip strength is the fraction of the total effect that is explained by the indirect effect (0.85/4.35 = 20%). These findings suggest that strengthening exercises should be retained in the implementation of this exercise program. If you had limited time and resources, one option would be to concentrate on strengthening exercises, as there is proof of this causal mechanism. Furthermore, the study suggests that other mechanisms account for a far greater proportion (80%) of the overall treatment effect. One plausible mechanism is joint mobility, but the intervention did not demonstrate substantial effects on 5 different measures of mobility, and these mobility measures were not associated with hand function. You speculate whether this finding was due to a lack of precision in the mobility measures, or whether it was due to a lack of causal effect of the intervention on mobility. Based on these data alone, you decide not to rule out the possibility of mobility as an important mechanism. In summary, these results suggest that unless there are reasons to prioritize treatment components (eg, due to practice restrictions on treatment duration and/or frequency), you should implement the full exercise program as originally designed by the investigators. With limited available evidence, you have a partial understanding for the underlying mechanisms of the exercise program. The unknown mechanisms may be explained by more complex constructs such as dexterity, or alternative psychological or behavioral factors such as self-efficacy or motivation. However, you can only speculate until more data emerge to support or refute these potential causal mechanisms. In this Point of View, we highlighted the merits of using mechanistic evidence to guide the development of physical therapist interventions, and presented a case scenario that illustrates how mechanistic studies might inform clinical practice. The recent shift toward patient-centered care may have downplayed the importance of impairment-level measures in physical therapy trials. A mechanistic approach that links impairment-level measures to patient-reported outcomes provides deeper causal explanations for how physical therapies may (or may not) work. Generating and implementing this evidence will be important for the scientific progression of physical therapist interventions. Concept/idea/research design: H. Lee, S.E. Lamb Writing: H. Lee, S.E. Lamb Fund procurement: H. Lee, S.E. Lamb Consultation (including review of manuscript before submitting): S.E. Lamb H. Lee is funded by the National Health and Medical Research Council (grant no. APP1126767) and receives project funding from the Berkeley Initiative for Transparency in the Social Sciences, a program of the Center for Effective Global Action (CEGA), with support from the Laura and John Arnold Foundation. The authors completed the ICJME Form for Disclosure of Potential Conflicts of Interest and reported no conflicts of interest.