The Protocol The Biometrics Weekly

Sign in Subscribe

Biostatistics Desk

Statistical trial design and analysis — estimands, adaptive and group-sequential designs, Bayesian and causal methods, missing data, multiplicity, RWE methodology.

Response-Adaptive Randomization Tries to Answer Its Critics

A previously unpublished Zhu–Rosenberger response to the 2011 Korn–Freidlin critique surfaces alongside seven new RAR papers — the field is plugging the operational holes regulators have been pointing at for a decade.

AstraZeneca tests the "subgroups cannot save a failed trial" rule

A prespecified 62% survival benefit in a kappa-isotype cohort of a failed Phase 3 is now AstraZeneca's filing strategy — and a live audit of how much subgroup evidence the agencies will actually accept.

Five Years On, the Estimand Framework Meets Its Legacy Conventions

The 2019 addendum's anniversary stocktaking arrives alongside two Project SignifiCanT forums and a censoring commentary that move the hard problem from defining intercurrent events to dismantling the conventions that contradict them.

The Safety-Monitoring Toolkit Catches Up to the Tape

A 40-point placebo-adjusted remission rate did not save Abivax's share price the week cancer cases surfaced. Five new methodology papers argue ad-hoc safety review is the actual liability.

The MAR Default Is Quietly Becoming Indefensible

A pincer of new PRO papers and a Bayesian causal framing of reference-based imputation are making the MAR-plus-tipping-point SAP template look less like a default and more like a liability.

Master Protocols' Borrowing Question Hits the SAP Before It Hits E20

A single SBR issue, a Statistics in Medicine result, and three live platforms all converge on one unresolved question — and ICH E20 is being written against it.

The Measurement Layer Gets Audited

Five PRO/COA methodology papers and an EMA workshop on Geographic Atrophy endpoints land in the same week — all pointing at thresholds, crosswalks, and surrogates that biometrics teams have been treating as settled.