1. Is High Placebo Response Really a Problem in Clinical Trials?
Mark Whitlock
Head of Statistics for Internal Medicine RU
Early Clinical Development
Pfizer

2. The Placebo Response “Problem”
Typical response to a failed drug trial “The study failed due to an unexpectedly high placebo response……”

3. Contents Placebo Response – Terminology / Additivity
Inherent correlation between treatment effect and placebo response
Analysis of MDD data
Is there evidence for placebo non-additivity?
Conclusions

4. Placebo Response - Terminology
Treatment effect is difference between what happened to patients as result of treatment and what would have happened if no treatment.
Counter-factual (cannot be observed)
Estimated by difference between estimated treatment response and estimated placebo response
True placebo response = “natural response” + placebo effect.
Natural response is the average measurement that would have been obtained from the subjects if they had not been recruited into a study
The placebo effect is caused by all the factors associated with being in the clinical trial apart from, unknowingly, not receiving the active treatment, e.g. change in behaviour such as diet and exercise, psychological factors such as expectation of benefit.
True treatment response = “natural response” + placebo effect + treatment effect
Assumption that the placebo and treatment effects are independent  Additivity
i.e. the placebo effect is the same in placebo and treatment arms

5. When is high placebo response a problem?
If placebo response is not additive
If there are floor/ceiling effects, i.e. small window to see drug effect over placebo response.
Response
Baseline
Additive
Placebo
Active
Drug
Effect
Observed Drug
Effect
Non-Additive
Placebo
Active
True Drug
Floor/Ceiling
Placebo
Active
True Drug
Effect
Observed Drug Effect

6. Are studies failing because of high placebo response?
Even if Additivity holds, a high estimated placebo response is likely to lead to a smaller estimated treatment effect
In any given study we estimate the treatment effect (∆) from the observed data.
The estimated ∆ may be more or less than the true ∆.
We are more likely to see a smaller estimated ∆ if the estimated placebo response is high
Results in correlation between estimated ∆ and placebo response
Whether the estimated placebo response is lower/higher than the true response cannot be predicted in advance (if random sampling variability)
Some published work on the problem of high placebo response can be explained largely by this inherent correlation.
This correlation cannot be “designed-out”.
Measures to reduce the placebo mean will not remove random variability about the mean.
Measures to reduce the placebo mean will only increase the ∆ if they don’t also reduce the treatment mean by an equivalent amount.
True Pbo Mean
True Drug Mean
True Drug Effect (∆)
True Pbo Mean
True Drug Mean
True Drug Effect
True Pbo Mean
True Drug Mean
True Drug Effect
True pbo response
True ∆

7. Placebo Response
Many measures have been proposed to reduce the placebo effect:
focus on patient population
improved screening procedures
blinded placebo run-in
number of visits
Design approaches intended to remove or down-weight the impact of subjects with high placebo effect
Placebo lead-in
Sequential Parallel Comparison Design (SPCD).
If the placebo effect is additive then these efforts are likely to be ineffectual at increasing the treatment effect since they would be expected to reduce the placebo effect in both the placebo and treatment arms by an equivalent amount.

8. Are studies failing because of increasing placebo response?
Is there evidence that the placebo response has increased?
And, if so, without an additive increase in active response?
Widely reported as a fact, particularly in psychiatric indications, e.g. Major Depression Disorder (MDD).
Undurrage & Baldessarini 2012 “a secular increase in sites and participants per trial was associated, selectively, with rising placebo associated response rates, resulting in declining drug–placebo contrasts or effect-size”
FAVA 2003 SPCD paper “…analysis has shown that the response to placebo has increased significantly in recent years, with a significant positive relationship between year of publication and response rate”
SPCD Presentation at Pfizer 2015 (Ivanova) “rising placebo response and failed trials”
Not just in psychiatric disorders:
Tuttle et al, Pain 2015 “…we extracted data from published randomized controlled trials (RCTs) of drugs for the treatment of chronic neuropathic pain over the years 1990 to We find that placebo responses have increased considerably over this period, but drug responses have remained stable, leading to diminished treatment advantage”.

9. Are studies failing because of increasing placebo response? Depression
Walsh 2003 paper: Review of 75 MDD trials between 1981 and 2000.
“Response to placebo has increased significantly in recent years, as has the response to medication”
“There was no statistically significant correlation between effect size and year of publication”
Undurraga 2012 paper (used in recent SPCD presentations): Review of 122 trials up to 2010.
Concluded increase in placebo response not matched by increase in active response.
Linear regression fits cannot be replicated!

10. Are studies failing because of increasing placebo response? Depression
Undurraga 2012 paper (used in recent SPCD presentations): Review of 122 trials up to
Re-analysis of data fitting locally-weighted loess
Placebo response and active response increased up to ~1998 (consistent with Walsh paper).
No significant increase in placebo response rates between 1998 and 2010 although trend for smaller treatment response (are recent compounds less effective?)
Smaller treatment effects on average in recent years. Could be due to the higher placebo response or other covariates (e.g. larger recent studies, different compounds (SNRI v TCA), publication bias).

11. Are studies failing because of increasing placebo response? Depression
Smaller treatment effects on average in recent years.
Within MDD treatment class, there is not strong evidence of smaller treatment effects due to higher placebo responses (effects appear additive)
Overall reduction in average treatment effects could equally be explained by different compounds e.g. SNRI v TCA.
Further breakdown of SRI by compound given in back-ups

12. Are studies failing because of increasing placebo response? Depression
Treatment response and placebo response are highly correlated
Analysis of data like these often naively fit OLS regression and conclude non-additivity (e.g. slope of 0.6)
Ignores random variability in both x and y
decrease in estimated treatment effects at higher estimated placebo rates can be explained by the inherent negative correlation between treatment effect and placebo response
Orthogonal regression (assuming equal variability in x and y) gives slope of 1.0
Supports additive assumption
More recent studies tend to have smaller treatment response for a given placebo response
Even in studies with low placebo response

13. Are studies failing because of increasing placebo response? Epilepsy
Review of 63 studies evaluating 20 AEDs between 1989 And 2009.
“Our results indicate that in adjunctive therapy trials in partial epilepsy the increase in response to placebo during the last two decades was clearly associated with a parallel increase in response to active medication. As a result, the relative risk for a positive response was not significantly affected…”

14. Are studies failing because of increasing placebo response? Pain
Tuttle 2015 paper
84 clinical trials for chronic neuropathic pain.
Placebo response increased but drug responses remained stable.
Smaller treatment effects on average in more recent studies.
Internal experience
NeP: No evidence of placebo response increase or effect decrease (at least between 1998 and 2007)
OA: Evidence of placebo response increase but with additive active (naproxen) response increase.

15. Conclusions
Placebo response is perceived as a problem throughout Pharma – particularly in psychiatric indications.
There will be negative correlation between estimated treatment effect and estimated placebo response caused by random variability. This cannot be designed out!
Design strategies to reduce placebo response will only increase the treatment effect if the placebo response is non-additive or if there are ceiling/floor effects.
Re-analysis of historical MDD data, appropriately accounting for variability in active and placebo response, does not support non-additivity.
There are other measures that should be considered that may be more effective in reducing unwanted placebo effects or increasing trial sensitivity/efficiency, e.g.
Patient selection
Better pre-study characterisation of patients (e.g. readily available cohorts)
Patient phenotyping
Incorporating historical knowledge of placebo/SOC response through Bayesian informative priors
Can be used to reduce sample size
Can moderate effects of unexpectedly high (or low) observed placebo response

16. QUESTIONS?