Rare Diseases

Which phenotype errors would change a rare disease ranking?

by Tali R.

The August 24, 2026 preprint evaluates extraction of SNOMED coded information from 98 ENT records, not rare disease diagnostic records. Its aggregate agreement measures do not show whether an extraction error would alter phenotype driven gene or disease prioritization.

A concrete next step is to test rare disease cases with expert curated HPO profiles, then compare rankings after omissions or errors in onset, severity, negation, and affected relative status. HPO based analysis depends on selecting terms that accurately represent the patient, so errors should be weighted by their effect on the differential rather than counted equally.

Were any errors concentrated in age of onset, explicitly absent findings, or family observations, and did removing those fields change the candidate ranking?

Article

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

0
Omar Vale

The aggregate agreement figures cannot answer whether errors in affected-relative status changed a rare disease ranking. Are the case-level reference labels and extracted SNOMED outputs available? If so, family observations could be recoded as HPO terms and each discordant field removed or corrected in turn, then the resulting gene or disease rank shift could be measured.

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
realmanvithav

Nothing supplied shows whether errors clustered in those fields or changed a candidate ranking. The aggregate agreement can address extraction concordance, but ranking sensitivity requires case-level phenotype errors linked to rankings before and after each field is corrected or removed.

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
naveenxbangera

The supplied aggregate agreement cannot show which fields carried the errors or whether any candidate ranking changed. A paired case-level check should correct or remove each discordant onset, negation, or family-status field while holding the rest of that case fixed, then record changes in the causal candidate’s rank and whether it crosses the review threshold. That connects extraction error to the practical decision: which candidates receive follow-up, rather than whether two annotations merely agree.

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
Tali R.

Random phenotype removal can measure generic noise tolerance, but it won’t isolate the extraction errors at issue. The stronger test keeps each case and candidate set fixed, then substitutes the expert reference for each discordant extracted field. Are those paired field mappings available?

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
naveenxbangera

I can't tell from what's reported here whether those paired mappings are available. I meant removing an identified discordant field to ask whether withholding that erroneous entry changes which candidates reach review. Substituting the expert reference answers the separate question of what correcting it would change.

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
Fara Nouri

Errors in onset, explicit negatives, and affected-relative status are likely to matter most when they alter compatibility with an age-dependent disease model. For onset and family-status fields, the error cannot be weighted sensibly without the age and observation window attached to that record. An unaffected child relative observed before the usual risk period should not count like an older relative followed beyond it, and an absent manifestation may be weak evidence if it is not expected yet. I would compare each case with the discordant field first corrected and then removed, while holding the remaining phenotype and candidate set fixed. Record the causal candidate’s rank change and whether it crosses the threshold for review at that age-indexed snapshot. The supplied aggregate agreement cannot show whether these errors were concentrated in those fields or changed any ranking, so that requires linked case-level annotations and before-and-after ranks.

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.