Despite the long history of usage in medicine, psychology, sociology and other sciences, the terms “moderator”, “mediator”, “predictor” and “prognostic factor” still seem to elicit discussions among researchers. Several authors have described how these terms are used interchangeably, neglecting important careful handling (Baron & Kenny 1986; Clark et al., 2006 and Probyn et al., 2017).
“A moderator is a third variable that modifies a causal effect.” (Wu & Zumbo 2008 and Baron & Kenny 1986) or “A moderator is a factor, measured at baseline, that interacts with the treatment to change outcome for a subgroup of participants” (Probyn et al., 2017).
“A mediator is a third variable that links the cause and the effect.” (Wu & Zumbo 2008, Baron & Kenny 1986).
“A predictor is a factor, measured at baseline, that affects outcome but does not interact with the intervention” (Probyn et al., 2017).
“A prognostic factor may be seen as measurement of natural history.” (Clark et al., 2006 and Clark et al., 2008)
A moderator is like the dimmer of a light, it affects the strength of the lighting / of the causal relationship (positive / negative). It describes “when” or “for whom” an independent variable causes a dependant variable. A moderator variable is more like a characteristic, a background variable, which is relatively unchangeable, e.g. personality type extravert, environment, gender, ethnicity. Moderator analysis is required when questions arise “for whom (person characteristic) a treatment works” or “when a treatment works (environmental characteristic)” (Wu & Zumbo 2008). A moderator has a single relationship as an independent variable and does not correlate with the outcome (Kraemer et al., 2002). Moderators may identify subgroups of patients with potential deviations of their course of rehabilitation / illness (Kraemer et al., 2002) and consequently support clinical decision – making processes for therapy choice.
Interactive effect = the effect of treatment of each individual patient depends of the value of Mo = Moderator.
A mediator is described to act more like domino stones. It answers questions about “why” or “how” a cause elicited an effect. A mediator may be a current health status / a temporary condition and describes indirect effects, intermediate effects, surrogate effects or intervening effects. The mediator correlates with the independent variable and is an observed non-manipulated variable. Its responsiveness may lead to changes in outcome (Wu & Zumbo 2008). Mediators may or may not interact with the intervention (Probyn et al., 2017) and have a dual role (Wu & Zumbo 2008).
A factor can be called a predictor, if a baseline measure shows an effect on the outcome, but does not have an interactive effect (not a moderator). The predictor forecasts the outcome after an intervention, but is not able to provide insight in the effect size of the treatment regardless of its value (Kraemer et al., 2002).
A predictor describes what kind of treatments should be used to achieve most beneficial outcomes for an individual patient (Simms et al., 2013). The predictor may describe the response or lack of response to a certain intervention (Clark et al., 2006). It may identify a subgroup of treated patients with different outcomes. For example: Compliance to treatment is a predictor and not a moderator, since it predicts better outcome combined with either intervention (Kraemer et al., 2002).
“A prognostic factor is a measurement that is associated with clinical outcome in the absence of therapy or with the application of a standard therapy that patients are likely to receive” (Probyn et al., 2017). Prognostic factors may support clinicians to make decisions regarding the treatment effect by answering the questions; when should one initiate, stop or modify the treatment for a patient (Simms et al., 2013).
PhD Candidate University of Antwerp Belgium and clinical specialist for upper extremity Kantonsspital Winterthur Switzerland
2020 Pain in Motion
Baron, R. M., & Kenny, D. A. (1986). The moderator–mediator variable distinction in social psychological research: Conceptual, strategic, and statistical considerations. Journal of personality and social psychology, 51(6), 1173.
Clark, G. M., Zborowski, D. M., Culbertson, J. L., Whitehead, M., Savoie, M., Seymour, L., & Shepherd, F. A. (2006). Clinical utility of epidermal growth factor receptor expression for selecting patients with advanced non-small cell lung cancer for treatment with erlotinib. Journal of Thoracic Oncology, 1(8), 837-846.
Clark, G. M. (2008). Prognostic factors versus predictive factors: examples from a clinical trial of erlotinib. Molecular oncology, 1(4), 406-412.
Kraemer, H. C., Wilson, G. T., Fairburn, C. G., & Agras, W. S. (2002). Mediators and moderators of treatment effects in randomized clinical trials. Archives of general psychiatry, 59(10), 877-883.
Ogundimu, E. O., Altman, D. G., & Collins, G. S. (2016). Adequate sample size for developing prediction models is not simply related to events per variable. Journal of clinical epidemiology, 76, 175-182.
Simms, L., Barraclough, H., & Govindan, R. (2013). Biostatistics primer: what a clinician ought to know—prognostic and predictive factors. Journal of Thoracic Oncology, 8(6), 808-813.
Wu, A. D., & Zumbo, B. D. (2008). Understanding and using mediators and moderators. Social Indicators Research, 87(3), 367.