hh-health-ai/healthcare-equity/conference-abstract-handicap

Conference abstract handicap

Extract signal from published abstract material at the moment the embargo lifts, and convert it into a probability revision on the full readout.

Embargo rule, before anything else

Work only with material that has been publicly released. Never seek, accept or act on pre-embargo content. Record the release timestamp in the brief. If a source's provenance is unclear, stop and route to your view layer → comms-compliance.

Workflow

1. Work the title-release stage first. Most large conferences release titles before

full abstracts. Titles alone carry real information:

• Presentation slot: oral > rapid oral > poster discussion > poster. Slot is

assigned by the scientific committee on perceived importance and is one of the cleanest early signals available.

• Late-breaker status: data arrived after the deadline, which usually means an

event-driven analysis read out. Neutral in direction, high in variance.

• Title wording: a title stating a result ("improves overall survival") versus

one stating a design ("a study of X in Y") is informative. Companies generally do not put a negative result in the title.

1. At abstract release, read the design before the numbers. Apply the trial-reading

checklist in references/abstract-checklist.md: population, comparator, primary endpoint, statistical plan, and — critically — whether the reported analysis is the pre-specified primary or something else.

1. Check the data cut-off date against the trial's expected event accrual. An early

cut-off with immature survival data is the most common way an abstract looks better than the eventual full dataset.

1. Anchor on base rates before adjusting. Start from published phase-transition

probabilities by phase and therapeutic area (BIO/Informa/QLS; Wong, Siah and Lo), then adjust for what the abstract shows. Analysts systematically over-update on abstract-level detail; the base rate should still be doing most of the work.

1. Handicap the presentation. The abstract is a subset. Name specifically what will

be shown live that is not in the abstract — subgroup analyses, updated cut-off, safety detail, quality-of-life data — and what each would have to show to change the view. That list is the actual deliverable.

1. Position the trade shape, not the direction. Conference readouts are variance

events. State the historical move distribution for this name around this conference, and note whether the setup argues for an options expression.

1. Hand the probability revision to a catalyst engine → readout-handicap and emit

the brief.

Not-automatic

An abstract does not license a conclusion about the full dataset. Abstracts are prepared months ahead, from an earlier data cut, and are edited by sponsors.

Reference: references/abstract-checklist.md. Contract: ../../references/evidence-brief.md.