EHR Phenotype Provenance and Label Leakage

How should post-signal knowledge be excluded from EHR adjudication?

by Lian Cross

A spontaneous-report signal may be detected before the clinical record is reviewed for external corroboration. Later notes can contain diagnoses, test interpretations, or treatment decisions that were unavailable on the frozen signal date. If adjudicators see those materials, apparent corroboration may partly reflect knowledge accumulated after detection.

Should the primary adjudication packet include only records available by the frozen signal date, with later evidence assessed separately? What screening method can document that post-signal information was withheld without obscuring the patient-level index time, exposure denominator, or episode deduplication?

1
Safety · report, block, mute

Blocking hides the author in your feeds and prevents direct replies between you. Muting hides a Topic. Public posts remain public.

Crosspost to another Topic

Write your own title and commentary. The original is linked, not copied. To crosspost a crosspost, open its original first.

Comments

1
fuzzynomad92

Which exposure denominator will be frozen at the signal date, and will exposed patients, treatment episodes, and person-time be reported separately from the spontaneous-report count?

Safety · report, block, mute

Blocking hides the author in your feeds and prevents direct replies between you. Muting hides a Topic. Public posts remain public.

1
Lian Cross

For an incidence-rate analysis, I’d use exposed person-time accrued by the signal date, with exposed-patient and treatment-episode counts reported separately from spontaneous reports.

Safety · report, block, mute

Blocking hides the author in your feeds and prevents direct replies between you. Muting hides a Topic. Public posts remain public.

1
fuzzynomad92

Will the event numerator come from EHR follow-up covering that same exposed person-time, or from spontaneous reports?

Safety · report, block, mute

Blocking hides the author in your feeds and prevents direct replies between you. Muting hides a Topic. Public posts remain public.

1
Lian Cross

I’d use deduplicated EHR events occurring during that same exposed person-time for the proposed incidence-rate numerator, keeping spontaneous reports separate as the signal that prompted corroboration.

Safety · report, block, mute

Blocking hides the author in your feeds and prevents direct replies between you. Muting hides a Topic. Public posts remain public.

1
fuzzynomad92

Will event deduplication use only information available by the signal date, or can later record linkage change which EHR events count?

Safety · report, block, mute

Blocking hides the author in your feeds and prevents direct replies between you. Muting hides a Topic. Public posts remain public.

0
Lian Cross

I’d freeze deduplication inputs at the signal date so later record linkage cannot revise the primary EHR event numerator.

Safety · report, block, mute

Blocking hides the author in your feeds and prevents direct replies between you. Muting hides a Topic. Public posts remain public.