A preliminary path analysis of expectancy and patient-provider encounter in an open-label randomized controlled trial of spinal manipulation for cervicogenic headache


A preliminary path analysis of expectancy and patient-provider encounter in an open-label randomized controlled trial of spinal manipulation for cervicogenic headache

Mitchell Haas, DC, Mikel Aickin, PhD, Darcy Vavrek, ND

Dean of Research, Center for Outcomes Studies,
Western States Chiropractic College,
Portland, OR 97230, USA.

OBJECTIVE:   The purpose of this article was to present a preliminary model to identify the effects of expectancy of treatment success and the patient-provider encounter (PPE) on outcomes in an open-label randomized trial.

METHODS:   Eighty participants with chronic cervicogenic headache (CGH) were randomized to 4 groups: 2 levels of treatment dose (8 or 16) and 2 levels of therapy from a chiropractor (spinal manipulation or light massage). Providers were instructed to have equal enthusiasm for all care. Structural equation modeling with standardized path coefficients (beta) was used in a path analysis to identify the effects of patient expectancy and the PPE on CGH pain. The model included monthly pain from baseline to 12 weeks. Expectancy and PPE were evaluated on Likert scales. The patient-provider encounter was measured as patient perception of chiropractor enthusiasm, confidence, and comfort with care.

RESULTS:   Baseline patient expectancy was balanced across groups. The PPE measures were balanced across groups and consistent over the 8-week treatment period. Treatment and baseline pain had the strongest effects on pain outcomes (|beta| = .46-.59). Expectations had little effect on pain (abs value(beta) < .15). The patient-provider encounter had a weak effect on pain (abs value(beta)= .03-.27) and on subsequent confidence in treatment success (abs value(beta)= .09 and .12).

CONCLUSIONS:   Encouraging equipoise in the PPE and balancing expectancy across treatment groups may protect against some confounding related to the absence of blinding in a randomized controlled trial of pain. In this trial, their effects were found to be small relative to the effects of treatment and baseline values.

From the FULL TEXT Article:


Patient-Provider Encounter

Overall, the treating chiropractors were able to interact with their patients with equipoise regarding 2 manual therapies and 2 doses of intervention in an open-label randomized controlled trial. Balance was maintained longitudinally between and within groups. We were specifically interested in balancing chiropractor enthusiasm, a characteristic that we felt the chiropractors could be mindful of and deliberately control. This was particularly important because the same chiropractors provided both manual therapies in the study. Our study suggests that studies can be designed where treating chiropractors do not interact with the participants differentially across treatment groups, and systematic bias favoring SMT from provider-created expectancy can be controlled to the extent that it would not be an important contributor to observed treatment effects.

Although we sought balance in the PPE across groups, we did not try to control the level of individual provider enthusiasm and other patient interaction parameters. We assumed variability between providers because of differing chiropractor personalities and variability within providers because of differing patient-provider dynamics. The small SDs of the variables measured indicate that actual provider variability did not greatly influence the patient perception of the PPE. Furthermore, the path coefficients showed that the PPE was at most weakly associated with outcomes. Either there was in fact little variation between doctors or personality differences simply did not affect patient perception of the PPE.

It must be pointed out that this study was not designed to assess the magnitude of the contribution of provider enthusiasm to outcomes (ie, we did not randomize level of chiropractor enthusiasm). Future research would be required to evaluate the possibility that the PPE contributes a uniform ceiling effect that limits patient improvement from study interventions, despite PPE balance across groups and low association with outcomes.


The second major finding of our analysis was that expectation did not appreciably influence subsequent pain improvement outcomes. There are several possible explanations. Expectation may truly have no causal effect on subsequent pain improvement in the population under study conditions of a randomized trial. Finally, different relationships across groups between expectation and outcomes (interaction effect) may have obscured the influence of expectation on pain improvement. In any event, expectation has been shown to be a determinant of disability outcomes in both acute5 and chronic6 low back pain in other studies.

In our study, balance in expectation for success of the assigned treatment across study groups (Table 2) occurred because participants rated confidence in success equally for both treatments (SMT and LM). This was unanticipated. One would guess that SMT would be more appealing than the LM described in the consent form. The expectation for success of the assigned treatment was balanced across groups and unrelated to outcomes was fortuitous. However, it does show that prior expectation does not necessarily affect outcomes in unblinded trials of manual therapies.

Randomization can be used to balance expectation for each of the interventions across treatment groups in a trial. There is no guarantee, however, that randomization will balance baseline expectation for the actual assigned treatment. This is because if there is a systematic difference in prior expectation between treatments, then randomization that balances expectation of both treatments across groups will preserve the systematic difference in expectation of “my own” treatment success. This can have several consequences. First, despite randomization, an expectation that influences outcomes in participants aware of their treatment will introduce bias in an unblinded randomized trial, unless mean expectations for both study interventions are fortuitously the same. Second, assigned treatment expectation or assigned minus comparison treatment expectation can be determinants of outcomes, whereas expectations of the individual treatments or difference between treatments might not.

It is advantageous to evaluate expectations in randomized trials of manual therapy and investigators should consider balancing them across study groups. One methodological question that remains to be answered is which form of expectation variables should be used as baseline covariates in the analysis of efficacy studies seeking to isolate the effect of a single component of care, such as SMT. Potential baseline covariates fall into 2 categories: those that cannot be used in treatment allocation and those that can (eg, in stratification or design-adaptive allocation). Possibilities for those that cannot be allocated include expectation of the assigned treatment or the difference between assigned treatment and a comparison intervention. Alternatively, variables that can be used in treatment allocation are difference between 2 treatment expectations or expectations for both treatments. Another important question is whether covariates should be included in the primary or a sensitivity analysis. Finally, a more interesting methodological question is whether the PPE can mitigate the effect of expectation on outcomes after baseline.


There are important limitations to our structural equation models. Although sample size in this study is small for a structural equation analysis, the exercise was worthwhile nonetheless, for indicating potentially large or potentially nonexistent effects. Standardization of variables is justified because the indicative results were to serve as general guides rather than definitive results.

Our model must be considered preliminary because precise and stable estimates of the path coefficients for a model of this complexity requires at least several hundreds of patients. [33] Second, path analysis is only as good as the variables in the model and the presumed pathways included. Other important determinants, confounders, and paths between variables could change the relationship of outcomes with expectancy or the PPE. For example, different measures of expectancy and the PPE might yield different results. In addition, expectations might influence subsequent PPE; these paths were not included. Also, there could be feedback loops between concurrent outcomes, expectations, and the PPE, that is, there can be mutual causal relationships between these variables. A larger sample is required to construct a more sophisticated model to take all these possibilities into account.

In general, latent variables must be used with caution. Our constructs for treatment and the PPE appear reasonable, however, because of the strength of the path coefficients connecting treatment to intervention and dose and connecting the PPE to patient perception of the interaction with the chiropractor. In addition, we wanted to make an assessment of whether expectancy or PPE could provide alternative explanations for treatment effects. This would have been the indication if treatment effects had vanished when these factors were included in the model, and the only way to test this was to include treatment effects.

It must also be emphasized that our findings are only applicable to studies of efficacy where the PPE and associated expectations are considered confounders to be controlled. In other study designs, nonspecific effects are considered beneficial to the patient and considered a component of the therapy under investigation. These include, for example, studies of efficacy of whole systems and effectiveness of real-world practice. The design of these studies may avoid suppression of the influence of the PPE or even seek to enhance it. The path coefficients connected the PPE to outcomes would be expected to be significantly larger for these study designs.


Several expected relationships were found supporting confidence that the analysis contains no serious bias in estimating the effects of baseline or treatment on outcomes. There was evidence for a coherent PPE variable, as judged by the patient. There was indication that successful treatment leads to increased expectation of further benefit. There was no evidence for strong consistent effects of either patient expectation or patient assessment of patient-practitioner encounter on the reported CGH pain improvement and number of headaches. On the basis of an elaborate structural equation model, in a relatively small sample, we did not find that either expectation or patient-practitioner encounter effects on outcomes provide better explanations than are provided by the effects of balanced treatment assignments.

Clearly, blinding is often not possible in efficacy and relative efficacy studies seeking to evaluate the independent effects of a single component of care (such as SMT). It is therefore important to control the effects of the patient-provider interaction on study outcomes to help optimize study internal validity. It appears that equipoise by the same providers across intervention types can be accomplished. It also appears that it is possible to reduce the confounding effect of the PPE to a relatively small proportion of the treatment effect found for the interventions under study. A challenging methodological issue that remains is determining to what extent equipoise in the PPE across treatment arms can serve as a surrogate for double blinding in randomized controlled trials.

Funding Sources and Potential Conflicts of Interest

No conflicts of interest were reported for this study. This study was supported by the National Center for Complementary and Alternative Medicine, National Institutes of Health, Department of Health and Human Services (grant no. R21 AT002324).