Table 1 Key recommendations for studies attempting to assess contextual effects
From: Contextual effects: how to, and how not to, quantify them
Possible study design | What does it do? | Potential limitations |
|---|---|---|
Three arm study: (1) intervention (2) placebo/sham (3) untreated control. | • Adjusts for non-specific effects, via untreated control, when measuring contextual effects. | • Untreated control group may increase the risk of bias (e.g. attrition bias, response bias, compensatory rivalry, resentful demoralization) when measuring outcomes. • Assumes that an additive model of treatment effects is true or at least that this assumption does not impact the statistical analysis. |
(Modified) Zelen Design – three arm study, randomization before informed consent: (1) intervention (2) placebo/sham (3) untreated control. | • Same as three arm design and lessens the risk of bias through attrition bias, response bias, compensatory rivalry and resentful demoralization. • Allows monitoring of natural disease course without randomizing participants to a no-treatment control group. | • Fewer research questions are ethically appropriate for this design • There still might be risk of bias (e.g. attrition bias, response bias, compensatory rivalry, resentful demoralization) when measuring outcomes. • Assumes that an additive model of treatment effects is true or at least that this assumption does not impact the statistical analysis. |
Two arm study: (1) placebo/sham: participants informed that they are receiving a “real” treatment (2) placebo-control group: informed they are receiving a placebo treatment that has no effect. | • The placebo-control group is used to control for potential confounders related to the treatment context. • Might lessen risk of bias as both groups receive “treatment”. | • As patients are told that they are treated with an actual placebo treatment that has no effect they might also develop compensatory rivalry and/or resentful demoralization. • Assumes that an additive model of treatment effects is true or at least that this assumption does not impact the statistical analysis. |
Three arm study (1) placebo/sham: see above (2) placebo-control group: see above (3) untreated control. | • In comparison to two arm study (row above), inclusion of an untreated control group to control for non-specific effects. • Might lessen risk of bias as both groups receive “treatment”. | • As patients are told that they are treated with an actual placebo treatment that has no effect they might also develop compensatory rivalry and/or resentful demoralization. • Untreated control group may increase the risk of bias (e.g. attrition bias, response bias, compensatory rivalry, resentful demoralization) when measuring outcome measures. • Assumes that an additive model of treatment effects is true or at least that this assumption does not impact the statistical analysis. |