BDAZ posted a link to the European Federation of Statisticians in the Pharmaceutical Industry - WC500202537

It is worth a read.

My hope for the SSRA is that the independent statistical analysts will be looking at unblinded data in order to determine if the prespecified guidelines are being met and it will be clear sailing based on the planned trial design.

Therefore, I'm surprised about point 2 on page 17.

Also note point 3 on page 27. I interpret that to mean that if the overall trial is not significant at 95% confidence or higher but if a subgroup findings are highly significant there will be some concern by the reviewing committee yet point 1 suggests that if it is prespecified this may be less of an issue. Any thoughts on this?

I've included other pages that may be of interest. Best to read the full pdf.

Toinv

P17 - Sample size re-estimation

1. Uncertainty about sample size assumptions. e.g. size of placebo effect
2. Whenever possible, use blinded sample size reassessment e.g. total number of events
3. Need to pre-specify size of treatment effect to be detected
4. If based on unblinded analysis, need to show control of type I error

P27 - Confirmatory subgroup analysis

1. Generally requires pre-specification that a subgroup is expected to have larger effect
2. Usually expected in the context of an overall positive trial
3. Not usually possible to rescue a trial with overall non-positive result

P28 -Subgroup analysis

Overall concern that the response of the “average” patient may not be the response of the all patients in the study (i.e. sub-groups could vary e.g. rosuvastatin vs. atorvastatin)*

Routine requirement for analysis by subgroup

Aim

1. Identify patient groups with differential treatment effects
2. Assessment of internal consistency
3. License can be restricted if not sufficient evidence of a positive risk-benefit in the subgroup

P40 - Missing data analysis

Increased regulatory focus on missing data

All statistical analyses where data is missing rely on un-testable assumptions about unobserved data

·       Best strategy is avoidance (I think this means make sure all data is available to committees and investigators even, for example, drop outs, missing data, missing treatments, etc)*

Missing data more problematic if imbalance in withdrawal rates across treatment arms or characteristics of withdrawals different to completers

P41 - ITT analysis (De Facto estimands)

Two separate aspects:

1. Including all randomized patients and all available on-treatment data (ITT Population)
2. Assessing outcome regardless of whether the patient remained on the assigned treatment

First principle almost universally agreed

Second principle less well understood, either requires follow-up off treatment or an assumption regarding missing data

* Italics are my comments - Toinv