Disentangling contextual effects from musculoskeletal treatments

Abstract

When a patient receives a treatment for a musculoskeletal pain condition, the outcome they experience (total effect) is made up of the specific treatment effect and non-specific effects that may include natural history, regression to the mean, and contextual effects1Zou K. Wong J. Abdullah N. Chen X. Smith T. Doherty M. et al.Examination of overall treatment effect and the proportion attributable to contextual effect in osteoarthritis: meta-analysis of randomised controlled trials.Ann Rheum Dis. 2016; 75: 1964-1970https://doi.org/10.1136/annrheumdis-2015-208387Crossref PubMed Scopus (89) Google Scholar. Understanding the specific effect of the treatment is important for patients, clinicians and policy makers, as some treatments have harms or costs that could outweigh positive specific effects. In this article we 1) discuss challenges in identifying and removing contextual effects in pragmatic trials, 2) introduce a way to isolate contextual effects of treatments using causal mediation analysis, and 3) highlight the importance of explicitly defining the contextual effect we want to estimate or control. Randomised controlled trials (RCTs) are used to disentangle non-specific effects associated with natural history and regression to the mean from the total effect of a treatment on a given outcome; and placebo controls are used to disentangle non-specific effects and contextual effects from total effects. By comparing the average outcome under the treatment and placebo groups, the specific effect of the treatment (without non-specific and contextual effects) can be estimated (Fig. 1). Complex interventions that are made up of several interacting components are commonly used to manage musculoskeletal pain conditions. RCTs of complex interventions for musculoskeletal pain rarely use placebo controls. In a recent systematic review of physiotherapy interventions for low back pain only 79 RCTs out of 2,215 (4%) used a placebo control2Cashin A.G. Lee H. Bagg M.K. O'Hagan E. Traeger A.C. Kamper S.J. et al.A systematic review highlights the need to improve the quality and applicability of trials of physical therapy interventions for low back pain.J Clin Epidemiol. 2020; 126: 106-115https://doi.org/10.1016/j.jclinepi.2020.06.025Abstract Full Text Full Text PDF PubMed Scopus (4) Google Scholar. Placebos of complex interventions are not straightforward to design3Boutron I. Tubach F. Giraudeau B. Ravaud P. Blinding was judged more difficult to achieve and maintain in nonpharmacologic than pharmacologic trials.J Clin Epidemiol. 2004; 57: 543-550https://doi.org/10.1016/j.jclinepi.2003.12.010Abstract Full Text Full Text PDF PubMed Scopus (188) Google Scholar and there may not be consensus, even among experts, on what constitutes an appropriate placebo for a given complex intervention4Hancock M.J. Maher C.G. Latimer J. McAuley J.H. Selecting an appropriate placebo for a trial of spinal manipulative therapy.Aust J Physiother. 2006; 52: 135-138https://doi.org/10.1016/S0004-9514(06)70049-6Abstract Full Text PDF PubMed Scopus (45) Google Scholar. For RCTs of complex interventions it is more common to use pragmatic designs to estimate the difference between treatment and usual care/no treatment group. Pragmatic trials test the effectiveness of treatments in real world settings and in theory increase the applicability of the findings to clinical settings5Ford I. Norrie J. Pragmatic trials.in: N Engl J Med. vol. 375. 2016: 454-463https://doi.org/10.1056/NEJMra1510059Crossref Scopus (586) Google Scholar. In pragmatic trials, effects associated with regression to the mean and natural history are removed, but contextual effects remain. Consumers of evidence are therefore left wondering how much of the effect of the complex intervention is attributable to the specific effect of the treatment. A novel type of treatment effect, the “natural direct effect”, can be used to represent the specific effect of a treatment in a pragmatic trial6Vanderweele T.J. Vansteelandt S. Conceptual issues concerning mediation, interventions and composition.Stat Interface. 2009; 2: 457-468https://doi.org/10.4310/sii.2009.v2.n4.a7Crossref Google Scholar. A natural direct effect is just like a standard treatment effect from an RCT, but it partitions out the treatment effect that goes through a defined variable on the causal pathway. So to isolate the specific effects of a treatment, one would first define and measure a variable that represents a contextual effect (such as the patient and therapist beliefs and behaviours, the patient-therapist relationship, or the non-specific clinical procedures and clinical environment in which the treatment is provided7Rossettini G. Carlino E. Testa M. Clinical relevance of contextual factors as triggers of placebo and nocebo effects in musculoskeletal pain.BMC Muscoskel Disord. 2018; 19: 27https://doi.org/10.1186/s12891-018-1943-8Crossref PubMed Scopus (84) Google Scholar,8Doherty M. Dieppe P. The “placebo” response in osteoarthritis and its implications for clinical practice.Osteoarthritis Cartilage. 2009; 17: 1255-1262https://doi.org/10.1016/j.joca.2009.03.023Abstract Full Text Full Text PDF PubMed Scopus (87) Google Scholar). Then the effect that goes through that contextual variable can be partitioned out using causal mediation analysis9Lee H. Herbert R.D. McAuley J.H. Mediation analysis.J Am Med Assoc. 2019; 321: 697-698https://doi.org/10.1001/jama.2018.21973Crossref PubMed Scopus (53) Google Scholar. Although this approach requires stronger assumptions than the placebo controlled trial10VanderWeele T. Mediation analysis: a practitioner's guide.Annu Rev Publ Health. 2016; 37: 17-32https://doi.org/10.1146/annurev-publhealth-032315-021402Crossref PubMed Scopus (448) Google Scholar, it offers several advantages. First it forces the investigator to be specific about the contextual effect that is being partitioned out. This is not common practice in placebo controlled trials11Kamper S.J. Williams C.M. The placebo effect: powerful, powerless or redundant?.Br J Sports Med. 2013; 47: 6-9https://doi.org/10.1136/bjsports-2012-091472Crossref PubMed Scopus (11) Google Scholar. Second, it allows the investigator to answer the question of “what would the treatment effect be if the contextual effect was set at the natural value it would take under the control (or intervention)?”. This is advantageous because it allows the specific effect of treatment to be estimated in isolation of the contextual effect. Finally, the benefits of a pragmatic trial can be maintained, thereby allowing the estimation of specific treatment effects under more realistic conditions. The trade-off of these benefits to the degree of robustness of the natural direct effect, and the challenges in measuring and estimating contextual effects should be weighed. In the Back Skills Training Trial (n = 701), Lamb et al.12Lamb S.E. Hansen Z. Lall R. Castelnuovo E. Withers E.J. Nichols V. et al.Group cognitive behavioural treatment for low-back pain in primary care: a randomised controlled trial and cost-effectiveness analysis.Lancet. 2010; 375: 916-923https://doi.org/10.1016/S0140-6736(09)62164-4Abstract Full Text Full Text PDF PubMed Scopus (282) Google Scholar. showed that a 6-week group cognitive behavioural intervention for people with low back pain reduced disability more than best practice advice alone (total treatment effect = 1.3 [95% confidence interval (CI), 0.6 to 2.1] points on the 24 point Roland Morris Disability Questionnaire at 12 months). This pragmatic, multicentre RCT, recruited adults with troublesome subacute or chronic low-back pain from 56 general practices in England. We used causal mediation analysis13Imai K. Keele L. Tingley D. Yamamoto T. Causal mediation analysis using R.Adv Soc Sci Res Using R. 2010; : 129-154https://doi.org/10.1007/978-1-4419-1764-5_8Crossref Google Scholar to determine the size of the specific treatment effect when the mediator, patient satisfaction (contextual effect mechanism), was set to be the value it would take (naturally) under the control intervention (Fig. 2). Our complete-case mediation analysis (n = 426), indicated that 5% of the treatment effect was mediated by patient satisfaction (natural indirect effect = 0.1, [95% CI, −0.1 to 0.2]) (Table I). The natural direct effect suggests that the majority of the total effect worked through the specific effects of the intervention and possibly other contextual effects unrelated to patient satisfaction (natural direct effect = 1.1 [95% CI, 0.1 to 2.2] (Table I).Table IEffect decomposition for the total treatment effect of cognitive behavioural treatment on disability (measured 12 months post-randomisation) with patient satisfaction (measured 3 months post-randomisation) as the contextual effect mechanismEffectWhat the effect representsEstimate (mean difference, 95% confidence interval)Total effectThe entire effect of the intervention (including specific and contextual effects)1.2 (0.2 to 2.3)Natural direct effectThe specific effect of the intervention without the contextual effect through patient satisfaction1.1 (0.1 to 2.2)Natural indirect effectThe effect of the intervention that goes through patient satisfaction0.1 (−0.1 to 0.2)Proportion mediatedThe proportion of the total effect that is explained by the natural indirect effect5% (−10% to 33%) Open table in a new tab Researchers should consider embedding mediation analyses into RCTs so that research consumers can better understand the specific and non-specific effects of treatments. Contextual effect mechanisms should be explicitly defined and measured. A greater understanding of specific effects of treatment will help guide the implementation of effective treatments into clinical practice, and ultimately improve outcomes for people with musculoskeletal pain. All authors contributed to the concept, design and writing of the article, and have approved the final version. None declared.