Does catastrophizing mediate the relationship between patients’ knowledge of pain and levels of pain intensity and function? Statistical analysis protocol

Hopin Lee, James H McAuley, Markus Hübscher, Adrian C Traeger, Steven J Kamper, G Lorimer Moseley

1.1      Research questions

  1. Does catastrophization mediate the relationship between patients’ knowledge of pain biology and pain intensity?
  1. Does catastrophization mediate the relationship between patients’ knowledge of pain biology and function?

2        Methods

2.1      Participants and procedures

Data were collected as part of a retrospective clinical audit conducted across 7 physiotherapy clinics in Australia, USA, and the UK between 2001 and 2010.

Inclusion criteria: data from patients with chronic pain (duration >3 months) who received pain education sessions were included in the study. All patients could read and write English, were 18 years of age or older and gave consent for their information to be used for research.

Exclusion criteria: data from patients was not included if they had undergone surgery in the preceding 3 months, had a current diagnosis of cancer, CRPS, a psychiatric disorder (not including depression or anxiety), a neurological disorder (MS, Parkinson’s, dementia, spinal cord injury, stroke, or whose primary complaint was migraine or headache).

The pain education intervention involved individualised or group face to face sessions. The intervention was delivered over 2 to 3 sessions that ranged between 30 to 60 minutes. In line with developments of pain education techniques, it is likely that the intervention changed over the course of the data collection period (9 years) with supposedly better techniques of delivering pain education. In parallel to pain education, the patients received usual care. The intervention was completed before the first follow-up assessment point (T1). The type or amount of care additional to the education sessions was not recorded.

A total of 7 clinics (11 clinicians) were involved in the clinical audit. Data from the 810 patients who were accessible at follow-up were used in this study. Patients not providing follow-up data before 12 months were treated as having withdrawn their consent. All assessments were self-reported by the patients.

Ethical approval to access and analyse the de-identified clinical data used for this study was provided by the University of South Australia Human Research Ethics Committee.

2.2      Measures

Participants were assessed at baseline (T0), 1 month follow-up (T1), 6 month follow-up (T2), and 12 month follow-up (T3). All of the following variables were measured at each time-point.

2.2.1     Patients’ knowledge of pain biology

The Neurophysiology of Pain Questionnaire (NPQ) [1] is a tool used to assess knowledge about the neurophysiology of pain. In its original format, it was used to assess postgraduate medical students’ understanding of pain mechanisms. The language of the original NPQ has been adapted to suit the patient population, and is used in clinical practice and research to assess patients’ understanding of the biological mechanisms that underpin his/her pain. The NPQ presents 9 true and 10 false statements about the neurophysiology of pain. The respondent is required to indicate whether each statement is true or false. The outcome is calculated by summing the correct responses. The total score ranges from 0 to 19, with higher scores representing better knowledge. The NPQ has demonstrated acceptable internal consistency (person separation index = 0.84) and good test-retest reliability (ICC = 0.97) in patients with chronic pain [2].

2.2.2     Catastrophizing

The Catastrophic Thoughts about Pain Scale (CATS) measures the severity of catastrophizing [3]. It has 7-items, with each item scored on an 11-point scale. The total score ranges from 0 to 70, with higher scores indicating greater levels of catastrophizing.

2.2.3      Pain intensity

Pain intensity was measured on a 100mm visual analogue scale [4]. The following question was asked “What was your average pain level over the last two days?”. The scale was anchored by the statements ‘No pain’ (0) and ‘Pain as bad as it can be’ (100).

2.2.4     Function

Function was measured using the Patient Specific Functional Scale [5]. Patients were asked on initial assessment to pick 3 to 5 activities they were unable to do or had difficulty with due to their pain. Then they rated their ability to do them on a 0 – 10 NRS, where 0 = unable to perform activity to 10 = able to perform activity at the same level as before injury or problem. The outcome was taken as the average, expressed as a percentage of possible maximum (i.e. the sum of the activity scores divided by the number of activities). This scale has demonstrated sufficient validity, reliability, and responsiveness in a range of musculoskeletal conditions [6].

2.2.5     Other measures

  • Perceived attributes of clinician
    • How would you rate your clinician’s expertise in your condition after your first session? (11 point NRS; 0 = ‘no expertise’, 10 = ‘Very high expertise’)
    • How much empathy or care did your clinician show you? (11 point NRS; 0 = ‘nil’, 10 = ‘total’)
    • How attentive to you was your clinician? (11 point NRS; 0 = ‘not at all attentive’, 10 = ‘completely attentive’)
  • Gender of clinician
  • Patient characteristics
    • Age
    • Gender
    • Pain duration (months)
  • Outcome expectancy; “How well do you expect to recover from this in six months? (11 point NRS; 0 = ‘No recovery at all‘, 10 = ‘Complete recovery‘)

2.3      Analysis

We will test whether change (D) in the patients’ knowledge (T0 to T1) leads to a reduction in catastrophizing (T1-T6), which then leads to a subsequent reduction in pain/function (T6 to T12).

To explore the temporal sequence of the hypothesized mechanism, we will conduct a temporal sensitivity analysis to test alternative models with reversed causal paths. These alternative models are theoretically plausible, but the temporal sequence of the independent (ie. knowledge), mediator (ie. catastrophizing), and outcome variables (ie. pain intensity, function) are set-up in reverse order to the mechanism specified in our main hypotheses. The three alternative models will test if the change in the independent variable precedes the change in the mediator, change in the mediator precedes the change in the outcome, and change in the independent variable precedes the change in the outcome.

If significant indirect effects are found for the alternative models, the temporal sequence of our hypothesis will not be supported. This strategy of using alternative causal sequences to explore alternative models have been used in other fields [7–11].

The main analysis and the alternative models will be conducted with pain and function as the outcome (i.e. total of 8 models).

Model 1: Mediation models in hypothesized causal order

This model will test whether a change in the patients’ knowledge (T0 to T1) leads to a reduction in catastrophizing (T1-T6), which then leads to a subsequent reduction in pain (T6 to T12).

Mediation models in hypothesized causal order

Alternative models:

The next 3 models will test pathways with alternative temporal orders of the change in the independent, mediator, and outcome variables.

Model 2a: This model tests whether an initial change in catastrophizing leads to subsequent change in knowledge, followed by change in pain/function.

Mediation models in hypothesized causal order1

Model 2b: This model tests whether an initial change in knowledge leads to subsequent change in pain/function, followed by change in catastrophizing.

Mediation models in hypothesized causal order2

Model 2c: This model tests whether an initial change in pain/function leads to subsequent change in catastrophizing, followed by change in knowledge.

Mediation models in hypothesized causal order3

Analysis Summary

Model no.∆ T0-T1∆ T1 – T2∆ T2 – T3
1KnowledgeCatastrophizingPain/Function
2aCatastrophizingKnowledgePain/Function
2bKnowledgePain/FunctionCatastrophizing
2cPain/FunctionCatastrophizingKnowledge

 T0 = baseline; T1 = 1-month follow-up; T2 = 6-month follow-up; T3 = 12-month follow-up

Confounders

Potential confounders were selected based on a process using directed acyclic graphs (DAGs) that represent the theorised causal structure of the mechanism in this proposed investigation (appendix). We will control for the following potential confounders: pain duration, diagnosis, patients’ expectation of outcome, patients’ perception of the clinician (empathy, attentiveness, and expertise).

The PROCESS macro (http://www.processmacro.org/index.html) based on ordinary least squares regression will be used to calculate indirect, direct, and total effects with 95% confidence intervals using 1000 bootstrapped resampling. All analyses will be conducted on SPSS V22.0.0.0. The outputs will be presented as unstandardized and standardized regression coefficients.

3.    Interpretation

A significant indirect effect for model 1 would support the hypothesis that catastrophizing mediates the relationship between knowledge and pain/disability.

If significant indirect effects are found for the alternative models, the temporal ordering in our hypothesis will not be supported. This would suggest that reversed causal orders might explain the relationship between knowledge, catastrophiziation, and pain/disability.

If model 1, plus any one of the alternative models show significant indirect effects, then we would support the hypothesis that catastrophizing mediates knowledge and pain/disability, while acknowledging the possibility of a reciprocal causal mechanisms between the variables. The alternative models are exploratory analyses and the intention is not to provide confirmatory evidence for the causal directions that are implied in these models.

If neither model 1 nor any one of the alternative models show significant indirect effects, our hypothesis about the mechanism and the temporal sequence will not be supported.

The indirect effect quantifies how much two cases that differ by one unit on the independent variable (knowledge) are estimated to differ on the outcome (pain) as a result of the effect of knowledge on catastrophizing, which in turn affects the outcome (pain).

References

1             Moseley GL. Unraveling the barriers to reconceptualization of the problem in chronic pain: the actual and perceived ability of patients and health professionals to understand the neurophysiology. J Pain 2003;4:184–9. doi:10.1016/S1526-5900(03)00488-7

2             Catley MJ, O’Connell NE, Moseley GL. How good is the neurophysiology of pain questionnaire? A rasch analysis of psychometric properties. J Pain 2013;14:818–27. doi:10.1016/j.jpain.2013.02.008

3             Lefebvre J, Keefe F, Caldwell D, et al. The development of the catastrophizing-state (CATS) scale in a clinical sample. In: 10th World Congress on Pain. San Diego: 2002.

4             Price DD, McGrath PA, Rafii A, et al. The validation of visual analogue scales as ratio scale measures for chronic and experimental pain. Pain 1983;17:45–56.

5             Stratford P, Gill C, Westaway M, et al. Assessing Disability and Change on Individual Patients: A Report of a Patient Specific Measure. Physiother Canada Published Online First: 24 March 1995.http://www.utpjournals.press/doi/abs/10.3138/ptc.47.4.258 (accessed 25 May2015).

6             Kowalchuk-Horn K, Jennings S, Richardson G, et al. The Patient-Specific Functional Scale: Psychometrics, Clinimetrics, and Application as a Clinical Outcome Measure. J Orthop Sports Phys Ther 2011;42:30–40. doi:10.2519/jospt.2012.3727

7             Greitemeyer T, McLatchie N. Denying humanness to others: a newly discovered mechanism by which violent video games increase aggressive behavior. Psychol Sci  a J Am Psychol Soc / APS 2011;22:659–65. doi:10.1177/0956797611403320

8             Morano M, Colella D, Robazza C, et al. Physical self-perception and motor performance in normal-weight, overweight and obese children. Scand J Med Sci Sport 2011;21:465–73. doi:10.1111/j.1600-0838.2009.01068.x

9             Guendelman MD, Cheryan S, Monin B. Fitting in but getting fat: identity threat and dietary choices among U.S. immigrant groups. Psychol Sci  a J Am Psychol Soc / APS 2011;22:959–67. doi:10.1177/0956797611411585

10          Usborne E, Taylor DM. The role of cultural identity clarity for self-concept clarity, self-esteem, and subjective well-being. Personal Soc Psychol Bull 2010;36:883–97. doi:10.1177/0146167210372215

11          Oishi S, Seol KO, Koo M, et al. Was he happy? Cultural difference in conceptions of Jesus. J Res Pers 2011;45:84–91. doi:10.1016/j.jrp.2010.11.018

Appendix

Uncontrolled DAG depicting the relationship between change in knowledge (exposure) and change in pain/function (outcome) with catastrophizing as the proposed mediator

Figure 1a. Uncontrolled DAG depicting the relationship between change in knowledge (exposure) and change in pain/function (outcome) with catastrophizing as the proposed mediator, when potential confounding variables (red ovals) are unadjusted. Directional arrows indicate a causal relationship between variables or groups of variables. Green oval = exposure, blue oval = outcome or ancestor of outcome, red oval = potential confounding variable or groups of variables, grey oval = unmeasured variable. Red path = biasing path, green path = causal path.

Controlled DAG depicting the relationship between change in knowledge (exposure) and change in pain/function (outcome) with catastrophizing as the proposed mediator

Figure 1b. Controlled DAG depicting the relationship between change in knowledge (exposure) and change in pain/function (outcome) with catastrophizing as the proposed mediator, when selected covariates (white ovals) are adjusted for. Entering covariates into a statistical model removes some confounding paths (red paths in Figure 1a, now black paths in Figure 1b). Therefore, if we adjust for outcome expectancy, clinician’s attributes, and diagnosis, this will reduce bias in the estimated effect of knowledge on pain/disability. For the effect of catastrophizing on pain/disability, bias could still be present through unmeasured confounders (prior beliefs). This model assumes that catastrophizing, fear avoidance behaviour, and self-efficacy lie on the causal path between knowledge and pain/disability. Controlling for variables on the causal path (blue oval) will bias the estimate of total effect.

Uploaded 13/06/15; updated 21/08/15 (amended measure of catastrophizing)

Comments

  1. Geert Crombez says:

    Congratulations with the DAG approach, which we currently also apply.

    [Reply]

Speak Your Mind

*