Many seemingly overlapping terms: MCID, MID, minimally relevant different (MiReDif), smallest effect size of interest (SESOI), smallest worthwhile difference (SWD) ...

Different traditions? Would be nice with a good overview.

Thanks, I'll check it out!

Any specific critical literacy resources from philosophy you would recommend?

In Danish, it is similar with the word for male cousin (perhaps more in humorous context). E.g., "I saw this weird cousin in the train today ..."

@tweis.bsky.social

Just to clarify our definition: active placebo = similarly looking control + imitating the perceptible side effects of the exp drug + not therapeutic. Agree that standard of care needs to be considered for both/control arm(s), e.g. therapeutically active control.

Ah, OK! Stopping rules are not my field so forgive me if mistaken. I think active placebos would likely not mask true disease progression itself, but potentially affect "subjective" assessments, e.g. chronic pain, depression, PTSD. (But ethics important re. imitated side effects' acceptability)

NB. Heterogeneity/meta-regression analyses likely imprecise like the main analysis, so absence of evidence != evidence of absence.

It might. Underlying mechanisms probably relate to, e.g., outcome types (patient vs observer), active placebo matching quality, trial design (crossover vs parallel). Heterogeneity not explained by these factors (nor by condition or drug category).

Can you elaborate on phase and trt strategies?

... For mechanisms for bias for to lack of blinding see, e.g.:
pubmed.ncbi.nlm.nih.gov/11879884/
pubmed.ncbi.nlm.nih.gov/21993424/

If unblinding does matter, probably same mechanisms as complete lack of blinding. E.g. for patient-reported outcomes probably response bias (or placebo effect since these are difficult to disentangle), But also changes in co-interventions, compliance, attrition. Don't know for sure. ...

#EpiSky
Maybe #MetaEpiSky?

References:
Within-trial analysis (first study, 2023): www.cochranelibrary.com/cdsr/doi/10....
Between-trial analysis (second study, in press): www.jclinepi.com/article/S089...
Twitter/X thread for first study: x.com/LaursenDavid...

10/10

Perhaps we are at least now more certain about our uncertainty. Additional future trials with both types of placebo controls would be useful ...

Thanks to all my co-authors and especially my main supervisor Asbjørn Hróbjartsson!

9/10

Tricky to interpret these results. Maybe unblinding doesn't matter, maybe the active placebos are not good enough, or maybe they do matter and we just don't have enough data to detect, or maybe our there is some flaw somewhere in the meta-research methods.

8/10

In our papers, we discuss limitations, e.g. for both studies imprecision and heterogeneity, for within-trial analysis external validity, and for between-trial analysis residual confounding.

7/10

Results: Confidence intervals from both studies were wide and compatible with no average impact and important average impact. Within-trial analysis: ROR 0.86, 95% CI 0.70 to 1.08. Between-trial analysis: ROR 0.98, 95% CI 0.77 to 1.24.

6/10

We identified and meta-analysed randomised within-trial comparisons (first study) and observational between-trial comparisons (second study). We analysed ratio of odds ratios: ROR < 1 meaning standard placebo-controlled trials exaggerated the experimental drug effects, e.g. pain improvement.

5/10

Active placebo controls contain ingredients that attempt to imitate the drug side effects (with varying degree of success). Few studies have actually assessed the impact on estimated drug effects when using active placebo vs using standard placebo. We wanted to investigate exactly this.

4/10