Mechanism-hypothesis-driven EHR case-finding: candidate diseases

Date: 2026-07-12 Related: #6245 · Hypothesis-Based Phenotype Algorithms proposal

Why this survey

The Timothy syndrome curation (#6245) turned one zebrafish finding — fever elicits arrhythmia and seizures even in overtly normal cacna1c heterozygotes — into a computable idea: scan EHRs for rhythm disturbance or seizures following a fever, to surface latent/mild CACNA1C carriers. That query is valid only if the fever→CaV1.2 mechanism holds in humans, so running it is simultaneously case-finding and a test of the mechanism.

Timothy is not special in this respect. It is one instance of a recurring archetype. This report catalogs other diseases where a mechanistic hypothesis underpins an EHR case-finding scan, so that (a) the schema proposal is grounded in a general pattern rather than a single example, and (b) we have a concrete worklist of future hypothesis-based phenotype algorithms — most of them attaching to disease entries that already exist in the KB.

The archetype: trigger-provoked latent disease

A physiological or pharmacological trigger transiently unmasks a latent channel/enzyme/pathway defect. Most carriers are asymptomatic at baseline and never meet the classical case definition. An EHR query for trigger-associated events (event shortly after exposure, in patients without prior such events) should be enriched for latent carriers. The enrichment, tested against genotype or a reference standard, is itself evidence for the mechanism.

The classical case definition finds the severe, syndromic tail. The mechanism-hypothesis scan reaches down the severity distribution toward the latent/mild cases — which is precisely where the mechanism, not the syndrome, is what defines membership. This is what DisMech's mechanism-first stance is for.

The epistemic spectrum (why `derivation_basis` matters)

These candidates do not sit at one point on the evidence axis — they span it, which is exactly why the proposed derivation_basis slot is orthogonal to definition_type:

Established trigger biology (derivation_basis: ESTABLISHED_CRITERIA, or near it) — the trigger→unmasking link is textbook and often already used clinically. Brugada fever-unmasking, malignant hyperthermia on volatile anesthetics, G6PD oxidant-triggered hemolysis, drug-induced long-QT unmasking latent congenital LQTS. Here the scan is a well-motivated case-finding tool; the "hypothesis" is mostly about EHR operationalization/yield.
Emerging / model-system-derived (MECHANISTIC_HYPOTHESIS / MODEL_SYSTEM_EXTRAPOLATION) — the trigger→unmasking link is shown chiefly in a model system and unproven in humans. Timothy fever-exacerbation is the worked example: zebrafish evidence, human validity open.

A hypothesis-based algorithm should be born with its point on this spectrum attached, so a downstream consumer never mistakes a speculative scan (Timothy) for a well-grounded one (Brugada), even though both are PHENOTYPE_ALGORITHM.

Candidate register

All "dismech entry" links are existing KB files (real attach points). "Maturity" is the mechanism's human-evidence status, mapping to derivation_basis.

Disease (dismech entry)	Latent trigger	Mechanism (channel/enzyme)	Proposed EHR scan	Maturity
Timothy syndrome (`Timothy_Syndrome`)	Fever / hyperthermia	CaV1.2 (CACNA1C) temperature activation → arrhythmia + seizure threshold	New arrhythmia/QT event or seizure shortly after documented fever	Emerging (zebrafish) — curated, #6245
Brugada syndrome (`Brugada_Syndrome`)	Fever	Nav1.5 (SCN5A) temperature-dependent gating → type-1 ECG unmasking	Fever-associated type-1 Brugada ECG / fever-triggered VF, structurally normal heart	Established — curated as the ESTABLISHED-basis worked example (`derivation_basis: ESTABLISHED_CRITERIA`)
Congenital long-QT (`Long_QT_Syndrome`)	QT-prolonging drug	KCNH2/KCNQ1 reduced repolarization reserve	Marked QT prolongation or torsade on a QT-prolonging drug → latent LQTS	Established (pharmacogenomic) — curated as the pharmacological-trigger example (`derivation_basis: ESTABLISHED_CRITERIA`)
CPVT (`RYR2_CPVT`)	Exercise / catecholamines	RYR2/CASQ2 Ca²⁺-release instability → DADs	Exertion/stress-provoked syncope or bidirectional/polymorphic VT, structurally normal heart	Established
Malignant hyperthermia (`Malignant_Hyperthermia_of_Anesthesia`)	Volatile anesthetic / succinylcholine	RYR1/CACNA1S Ca²⁺-release → hypermetabolic crisis	Peri-anesthetic hyperthermia / masseter spasm / unexplained rhabdomyolysis	Established — curated as the anesthetic-trigger example (`ESTABLISHED_CRITERIA`); strongest genotype yield (50–>70%)
G6PD deficiency (`Glucose-6-Phosphate_Dehydrogenase_G6PD_Deficiency`)	Oxidant drug / fava / infection	G6PD NADPH shortfall → oxidative RBC injury	Acute hemolytic anemia within days of an oxidant drug/infection	Established
Acute intermittent porphyria (`Acute_Intermittent_Porphyria`)	CYP-inducing drug / fasting / hormones	HMBS haploinsufficiency → ALA/PBG accumulation	Recurrent unexplained abdominal pain + neuro/autonomic signs after inducing drugs	Established, under-recognized
MCAD deficiency (`MCAD_Deficiency`)	Fasting / intercurrent illness	ACADM fatty-acid-oxidation block → energy failure	Hypoketotic hypoglycemia / Reye-like decompensation during fasting/illness	Established (also newborn-screened)
Hereditary angioedema (`Hereditary_Angioedema`)	ACE inhibitor / estrogen / trauma	SERPING1 (C1-INH) → bradykinin surge	Recurrent angioedema without urticaria, esp. ACE-inhibitor-associated	Established, under-recognized
Hypokalemic periodic paralysis (`Hypokalemic_Periodic_Paralysis`)	Carbohydrate load / rest-after-exercise / stress	CACNA1S/SCN4A → transient hypokalemic weakness	Recurrent transient weakness with documented hypokalemia after trigger	Established

How each would be represented (the Timothy template, reused)

Every row follows the same shape now committed for Timothy — all achievable with existing slots today, and upgraded to structured epistemic markers once the #6245 schema extension lands:

mechanistic_hypotheses entry with a stable hypothesis_group_id and a status matching the maturity column (EMERGING for Timothy; CANONICAL/ ALTERNATIVE for the established-trigger rows where the biology is settled but the EHR operationalization is the new claim).
A trigger pathophysiology node ("Fever-triggered …", "Anesthetic-triggered …", "Oxidant-triggered …") whose downstream edges opt into hypothesis_groups: [<id>].
A hypothesis-based PHENOTYPE_ALGORITHM definition — the EHR case-finding query — that (post-#6245) sets derivation_basis, a structured validation_status object, and attaches_to the trigger node so the basis is inferred from the pathograph. Until then, the unvalidated status is stated in the name/scope/notes (as done for Timothy).
A discussions / HUMAN_MODEL_MISMATCH (or KNOWLEDGE_GAP) entry whose proposed_experiments is the genotype-enrichment test of the scan.

Near-term opportunities

Brugada + fever — done. Curated as the established-biology mirror image of the Timothy case: Brugada_Syndrome now carries a fever-unmasking pathophysiology node and a PHENOTYPE_ALGORITHM definition with derivation_basis: ESTABLISHED_CRITERIA / validation_status: UNVALIDATED. Timothy (MECHANISTIC_HYPOTHESIS / PROPOSED) and Brugada now sit at opposite ends of the derivation_basis spectrum as paired worked examples.
Drug-induced long-QT → latent congenital LQTS — done. Curated on Long_QT_Syndrome as the pharmacological-trigger example (a QT-prolonging drug unmasking reduced repolarization reserve; derivation_basis: ESTABLISHED_CRITERIA), broadening the archetype from physiological (fever) to drug triggers.
Malignant hyperthermia / anesthesia (Malignant_Hyperthermia_of_Anesthesia) — done. Curated as the anesthetic-trigger example and the first to leave cardiac electrophysiology (skeletal-muscle RYR1/CACNA1S Ca²⁺ release). It is the archetype's strongest established case: a discrete codeable exposure (the anesthesia record) and an unusually high genotype yield (50–>70%). The definition attaches_to the entry's existing trigger node — showing that when the pathograph already models the trigger, the algorithm just attaches to it.
CPVT / exercise (RYR2_CPVT) is the remaining cardiac established-trigger candidate; G6PD / oxidant drug (Glucose-6-Phosphate_Dehydrogenase_G6PD_Deficiency) would move the archetype further out, into a hematologic/metabolic system.

New Approach Methodologies (NAMs) as the other validation arm

The EHR scan is a population-level test of a trigger hypothesis. Its natural complement is a mechanistic test in a human-relevant model — a New Approach Methodology (human iPSC-derived cells, organoids, organ-on-chip, heterologous channel expression, in-silico electrophysiology) rather than an animal model. NAMs contribute to these hypotheses in three ways:

Bridge the HUMAN_MODEL_MISMATCH. When a hypothesis rests on an animal model (Timothy's zebrafish fever effect), the open question is human relevance. A human iPSC / organ-on-chip / in-silico version of the same trigger challenge is the most direct way to close that gap — and it is what moves a definition's derivation_basis from MODEL_SYSTEM_EXTRAPOLATION toward a human-validated mechanism. dismech already represents this via Experiment.model_systems (ExperimentalModel, NAMO-aligned: experimental_model_type ∈ {IPSC_DERIVED_MODEL, ORGANOID, ORGAN_ON_CHIP, …} + optional namo_type).
Generate hypotheses. Unbiased iPSC/organoid phenotyping and in-silico channel modeling can surface trigger-sensitivities before any clinical signal exists — candidate EMERGING mechanistic_hypotheses.
Provide the mechanistic readout the EHR cannot. Persistent I_CaL, APD prolongation, EADs, MEA network hyperexcitability, RyR1 Ca²⁺-release — direct causal readouts, not billing-code proxies.

Per-example NAM (each a proposed_experiment with NAMO-aligned model_systems):

All four examples now carry a NAM proposed_experiment (curated):

Example	NAM test	experiment_id
Timothy (fever)	iPSC-cardiomyocyte + iPSC-neuron + heart-on-chip hyperthermia challenge on CACNA1C-G406R vs. isogenic control + in-silico human ventricular-AP model	`exp_ts_fever_ipsc_nam_challenge`
Brugada (fever)	iPSC-cardiomyocyte temperature challenge + heterologous Nav1.5 temperature-gating assay	`exp_brs_nav15_temperature_nam`
Long QT (drug)	CiPA — the Comprehensive in vitro Proarrhythmia Assay, a flagship regulatory NAM (in-vitro hERG/multi-channel + in-silico O'Hara-Rudy/ToR-ORd) + iPSC-cardiomyocyte drug challenge	`exp_lqt_cipa_nam_confirmation`
MH (anesthetic)	RyR1 Ca²⁺-release assay + patient-iPSC-myotube caffeine/4-CmC/halothane challenge (scalable in-vitro complement to the invasive CHCT)	`exp_mh_ryr1_ipsc_calcium_nam`

Note the role differs by derivation_basis. For Timothy (the one MECHANISTIC_HYPOTHESIS), the NAM bridges the HUMAN_MODEL_MISMATCH — it tests whether the zebrafish effect is human-relevant at all, so its discussion pairs the NAM with the population EHR scan as the two arms of testing one open hypothesis. For the three ESTABLISHED_CRITERIA examples the mechanism is already settled, so the NAM instead functionally confirms and risk-stratifies the individual carriers (resolving variants of uncertain significance) that the EHR query surfaces — an OPEN_QUESTION about operationalizing the algorithm, not about the mechanism.

Validation pathways & datasets

A recurring question: can these algorithms be tested on public EHR data (e.g. MIMIC)? Separate two very different bars:

Prototype — does the phenotype logic run, and does it surface plausible cases? (tune the OMOP query, temporal windows, code sets; detect the ECG morphology.)
Validate — does a hit actually enrich for genotype-positive carriers? (the PPV / gold-standard test the proposed_experiments describe.)

Bar 2 needs linked germline DNA, which is what gates dataset choice.

MIMIC — a prototyping substrate, not a validation substrate. Two hard limits: (a) no linked genotype, so carriership can never be confirmed; and (b) the ICU/ED population is a worst-case confounder — fever there mostly means sepsis/critical illness and new arrhythmias are rampant for unrelated reasons, the opposite of the "overtly normal carrier" population the scan targets. Its genuine value is narrower: MIMIC-IV-ECG (~800k linked 12-lead ECGs on PhysioNet) lets you build and sanity-check a type-1 Brugada morphology detector against charted temperature (the signal-detection half of the Brugada query), plus fever→event base-rate mechanics. Useful for Brugada prototyping; of little use for Timothy (needs DNA, no clean ECG signature).

Genotype-linked EHR biobanks — where the enrichment test actually lives:

Dataset	Genotype	Longitudinal EHR	ECG waveforms	Role here
eMERGE Network (+ PheKB)	Yes	Yes	Some	Purpose-built for EHR-linked computable-phenotype validation — the canonical home
All of Us (NIH)	WGS (100k+s)	Yes (OMOP)	Growing	Directly query SCN5A/CACNA1C carriers vs. their fever/arrhythmia/seizure events
UK Biobank	WGS/WES 500k	Linked HES + primary care	Rest/exercise ECG (imaging subset)	Strong for the genotype-enrichment test
MVP / BioVU / Genomics England 100k	Yes	Yes	Varies	Same shape, more access-gated

Because dismech definitions are already OMOP-shaped, the same query can be run federated across the OHDSI/OMOP network, including genomics-linked sites.

Mapping to the schema. This is exactly what validation_status is for: an algorithm starts PROPOSED/UNVALIDATED; a genotype-anchored study on eMERGE / All of Us / UK Biobank is what flips it to VALIDATED_AGAINST_GOLD_STANDARD, cited as validation_status.evidence (a standard PMID + PPV excerpt). MIMIC work would generally support only the phenotype-detector step, not that transition.

Scope and evidence discipline

This is a design/landscape survey, not KB evidence. The trigger biology cited here is well-established textbook knowledge; the disease/gene/trigger framings were cross-checked against the referenced dismech entries. Per-claim PMIDs with verified exact-quote snippets are added at curation time, following the standard anti-hallucination workflow (fetch abstract → verify snippet substring → validate terms) — no citations are asserted in this survey to avoid unverified references. Candidate selection favored disorders already in the KB so each has a concrete attach point.