Pathway-Dominant Polygenic Risk Profiling

Summary¶

A single cardiovascular PRS returns one number, so it can say that someone is at high risk but not why, and therefore not what to do. This proposal builds calibrated, mutually comparable pathway PRS axes and defines a prespecified dominance statistic that flags individuals whose polygenic risk is concentrated in one treatable pathway rather than spread diffusely. The identified subgroup is people at high lifetime risk who are currently invisible to both conventional risk scores and to global CAD PRS, and for whom axis membership names a specific existing intervention.

Status¶

Provenance note: this page's sources are internal grant-folder documents (grant/), not raw/ papers, so sources: is empty per the §3.3 provenance contract; the documents are cited by path in Citations instead.

Proposal in development for MRFF Genomics Health Futures 2026, Stream 3 (Incubator), Topic A — Cardiovascular disease. Minimum data lodged; applications close 19 August 2026. Nothing here is a result. This page exists to hold the method, its scope boundary against ongoing lab work, and its known weaknesses.

Why this shape, and not a decomposition study¶

The call's objective is narrow: develop novel methods for using PRS to identify subgroups of the population at high risk. The lab's requirements analysis distils four indispensable elements — PRS central; method novel; identifies a meaningful population subgroup; identification leads to early intervention — and warns that satisfying only three reads as misaligned.[1]

It also names the failure mode directly. A weak fit is "pathway PRS used to cluster people with established coronary disease, with no prospective risk or intervention consequence", and "a proposal centred on describing the biology of pathway PRS without demonstrating improved identification of actionable high-risk groups would be less directly responsive."[1] A good fit is "pathway-specific polygenic profiles identify a group at particularly high premature myocardial infarction risk who are missed by standard risk scores."[1]

This design is built to the good-fit sentence.

Scope boundary against mCVDRisk¶

The "no extension of ongoing research" and "no duplication" rules bite hardest here, because the lab already holds a funded CVD risk-score programme.[2] The boundary must be explicit in the application.

	mCVDRisk (ongoing, funded)[2]	This proposal
Question	How much risk?	Which pathway, and therefore which treatment?
Output	one 5-year absolute-risk scalar	a profile across calibrated axes + a dominance call
Inputs	250 plasma lipid species + a PRS	genotype only
Target stratum	intermediate AusCVDRisk (5–10%), reclassified upward	low conventional risk, currently untreated
Action triggered	generic therapeutic intensification	a named, axis-specific intervention
Stage	implementation, TGA/NATA, MBS	proof-of-concept / feasibility

The two are sequential, not competing: mCVDRisk sharpens whether to treat; this sharpens what to treat with, in a stratum mCVDRisk does not target. Genotype-only input keeps the instrument independent of the lipidomics platform and its commercialisation pathway.

Declare precisely: all current funding in the CVD risk-score area, the Heart Foundation Catalyst Partnership, and the Baker commercialisation subsidiary.[2] Assessors can exclude for duplication, and the overlap is close enough that silence would be read badly.

Why the low-traditional-risk stratum is load-bearing¶

Targeting people at low conventional short-term risk is not one design choice among several. It does five jobs at once, and the proposal is materially weaker without it.

1. It is what makes the mCVDRisk boundary structural rather than argued. mCVDRisk targets the intermediate stratum (5–10% 5-year AusCVDRisk) by design.[2] The low-risk stratum is missed by design by that pathway. Non-duplication then follows from the study population itself, not from a paragraph asserting difference. The lab's own requirements analysis reaches the same conclusion: the natural differentiated target is "younger people with high inherited (PGS) risk but low conventional short-term risk, a group the AusCVDRisk/mCVDRisk intermediate-risk pathway misses by design."[1]

2. It is where the residual unmet need sits. Up to 65% of events occur in people classified low or intermediate risk.[2] mCVDRisk addresses intermediate. Low is untouched — and it is the stratum that generates SMuRF-less presentations.

3. It defuses the strongest objection to the method. The obvious challenge to a genetic Lp(a) axis is that measured Lp(a) is cheap and guideline-endorsed. But in a low-risk population nobody is being tested for anything — that is the definition of the stratum. The question is therefore not "genetic score versus measured biomarker" but "how do you decide who among millions of low-risk people warrants any further testing at all?" Genotype is a one-off, lifetime-stable triage layer, which is precisely the "sequential or adaptive testing strategy" the call names as a novelty type.[1] The low-risk focus converts the method's weakest point into its rationale.

4. It forces a two-stage design, which is a strength. SMuRF-less myocardial infarction is a case phenotype — membership is only known after the event — so it cannot be the prospective target. It is, however, the ideal derivation set:

Derive which axes are enriched in SMuRF-less MI versus matched low-risk controls, with collider-bias control (see Index-Event (Collider) Bias in Disease-Subtype Genetics — conditioning on case + absence of risk factors is textbook collider stratification).
Apply prospectively in the low-risk general population, before events.
Validate externally against incident events in that stratum.

This separation is what the call asks for when it requires the subgroup to be predefined or reproducibly discoverable, with derivation distinguished from external reproduction.[1]

5. Sex and gender become substantive rather than compliance. SMuRF-less STEMI carries higher 30-day mortality, particularly among women, and conventional risk scores already underperform in women.[2] Whether pathway dominance differs by sex — and whether the missed low-risk group skews female — is a real question inside this design with a direct equity payoff. The lab's requirements analysis flags this as a natural strength to integrate in both the Sapphire form and Component B.[1]

Duplication exposure to manage. Team members hold closely-related funded work in exactly this space: a consensus clinical pathway for SMuRF-less CAD across three states with a registry (GNT1196629-CRE) and a PRS-in-primary-care pathway piloted in 1,000 patients (GNT2005791-Partnership).[2] These are assets for Criteria 5 and 7 — they make the "credible next step" a proven pathway rather than a promise, and they supply the post-identification clinical route. They are also the sharpest duplication risk in the application. The line to draw: those implement existing scores and pathways; this develops a new method for a stratum they do not target. Declare both precisely.

The method¶

The problem, stated as the call requires¶

Current methods cannot do X. Conventional risk scores use phenotypic markers and miss subclinical disease; the lab's own analysis puts up to 65% of events in people classified low or intermediate risk.[2] A global CAD PRS improves discrimination but is a sum: a person whose entire polygenic burden sits in one extreme pathway can have an unremarkable total score, because that pathway contributes a small fraction of genome-wide variance. Neither tool names a mechanism, so neither indicates a treatment.

Our new method does Y. Construct pathway PRS axes on a common, calibrated scale and compute a prespecified dominance statistic that detects concentration of risk in one axis.

This allows population group Z to be identified. Adults at low conventional short-term risk — currently untreated — whose risk is concentrated in a single treatable axis.

For intervention A. Axis membership maps to a named action, below.

Step 1 — Axis definition (prespecified, not discovered)¶

Five axes, each chosen because a licensed or late-stage therapy exists. This is deliberate: the call warns that the subgroup must be "predefined or reproducibly discoverable" and cautions against exploratory pipelines generating post-hoc clusters.[1]

Axis	Action triggered if dominant
Lipoprotein(a)	Measure Lp(a); cascade screening of first-degree relatives; intensified LDL targets; trial eligibility for Lp(a)-lowering agents
ApoB / LDL	Early lipid-lowering; FH-style family cascade
Blood pressure	Earlier and more intensive antihypertensive management
Inflammation (IL-6 signalling)	hsCRP measurement; anti-inflammatory therapy assessment
Thrombosis / platelet	Antiplatelet consideration

Lipoprotein(a) is the flagship case: it is largely determined by a single locus, absent from conventional risk equations, diluted inside a global CAD PRS, and directly actionable today through measurement and cascade screening.

Step 2 — Calibration (the actual statistical novelty)¶

Pathway PRS are not comparable off the shelf. They differ in SNP count, heritability, and variance, so "dominance" is meaningless without putting them on a common footing. The method's technical core is:

Variance standardisation of each axis within ancestry stratum, so axis percentiles are interpretable against each other.
Mutual orthogonalisation — axes share variants and are genetically correlated; residualise each axis on the others so dominance reflects genuine concentration rather than shared signal.
A dominance statistic with a prespecified threshold: the criterion is that axis k exceeds a stated percentile and the remaining axes do not, formalised as a concentration measure over the calibrated axis vector.

Reporting must include how the subgroup is generated, whether membership is prespecified or data-derived, how stability is measured, how it reproduces externally, what threshold defines membership, and what action it triggers — the call lists all six.[1]

Step 3 — Derivation and external validation¶

Derive calibration and thresholds in a large genotyped cohort with incident CVD outcomes; validate in independent Australian cohorts. No new recruitment is required for the core method, which is what makes this feasible inside $1M / 2 years as an Incubator proof-of-concept.[1]

Step 4 — Comparators¶

The call is explicit that novelty must be shown against real alternatives.[1] Minimum comparator set:

conventional clinical risk assessment (AusCVDRisk) alone;
global CAD PRS alone;
clinical risk + global CAD PRS;
and, as an honesty check, the same axes without the dominance step — i.e. does concentration add anything over simply reporting five percentiles?

The last comparator is load-bearing. If dominance adds nothing over the raw axis profile, the method reduces to reporting several PRS and the novelty claim fails.

Step 5 — Endpoints¶

Discrimination alone is insufficient; the call warns against reducing performance to AUC.[1] Primary endpoints should be:

Yield — how many people per 1,000 screened are flagged who are not identified by clinical risk or global PRS;
Event enrichment in the flagged group versus matched non-flagged low-risk individuals;
Actionability — proportion of flagged individuals for whom the indicated intervention is currently licensed and available;
Reproducibility of axis assignment across cohorts and ancestries;
Calibration within sex and ancestry strata.

Known weaknesses¶

These are stated because assessors will find them anyway, and Criterion 8 rewards a risk plan that names real risks.

The Lp(a) objection, largely answered by the low-risk framing above but still requiring explicit handling: measured Lp(a) is cheap and guideline-endorsed, so the sequential-testing rationale (genotype as the triage layer deciding who is tested at all) must be stated in Component A, not left implicit.
Dominance is a ratio-like statistic and is unstable when all axes are near the mean. The threshold must be set where the statistic is stable, and that stability must be demonstrated, not assumed.
Pathway PRS have modest effect sizes. The claim is about identifying who is missed and what to do, not about large hazard ratios. Overclaiming discrimination will be scored down.
Ancestry transferability is a known weakness of all PRS and is worse for pathway-restricted scores with fewer variants. This must be addressed substantively given the call's priority-population alignment, not waved at.
Low conventional risk is a hard stratum to validate in. Event rates are low, so power comes from cohort size and follow-up duration, not from effect size. This constrains which cohorts are usable and is the single biggest feasibility question for the 2-year envelope — it must be answered with actual event counts, not assertions.
The SMuRF-less derivation arm conditions on being a case, so collider bias is a live threat, not a theoretical one. The correction strategy must be specified in Component B rather than cited.

Cross-cutting requirements (scored, not boilerplate)¶

Sex and gender. CVD has well-evidenced sex differences in risk and presentation, and conventional scores underperform in women.[2] Whether pathway dominance differs by sex is a genuine scientific question inside this design, not a compliance statement.
Consumer involvement is assessed against all four criteria.[1] For this method the real design question is how a person is told their risk is concentrated in a specific pathway, and what they want done about it — particularly for an axis whose treatment is not yet licensed.
ELSI — genetic risk disclosure, cascade screening implications for relatives, and insurance/discrimination concerns.
Priority populations — pathway PRS transferability is the equity issue; if pursued, it requires demonstrated leadership by that population.[1]

Open decisions¶

Which derivation and validation cohorts, and whether the ~3×1,000 CTCA cohorts have a role (they are not the primary validation set — they contain people with established or subclinical disease, which is the call's stated weak-fit design).
Whether to include a priority-population focus properly or not at all.
Whether to include a sixth, non-druggable axis as a negative control.

Citations¶

[1] Lab requirements analysis for MRFF GHFM 2026 Stream 3 Topic A, compiled from the published Grant Opportunity Guidelines. grant/reference/GHFM-2026_Stream3-TopicA_Outline.md and grant/reference/GHFM-2026_Stream3-TopicA_Outline_chatgpt.md. Supports: the four indispensable elements; the good-fit / weak-fit subgroup examples; the six subgroup-reporting requirements; the comparator set; the warning against AUC-only endpoints; the sequential-testing novelty type; Incubator scope, $1M / 2-year envelope, and cross-cutting requirements.

[2] Meikle, P. et al. (2025). Translation and implementation of an evidenced-based, omics-supported clinical pathway for CVD risk stratification in primary care. MRFF CHM 2025 Cardiovascular Health, Application ID 2053571. grant/reference/meikle-2025-mcvdrisk-implementation-grant.md. Supports: mCVDRisk composition and purpose; the 65%-of-events figure; the intermediate-risk reclassification design and Busselton validation; partner and commercialisation arrangements defining the duplication boundary.