Curriculum
Module 06 · 55 min

Polygenic Risk & Early Warning

PRS has crossed the clinical-utility threshold for some diseases — and not for others.

ClinicalResearch
Core topics

What's covered

  • T01Construction: GWAS summary stats → LD reference → score (LDpred2, PRS-CS, MegaPRS)
  • T02Cross-ancestry portability and the ancestry-bias problem
  • T03Integrated risk: PRS + monogenic + clinical (QRISK3 + PRS for CAD, Tyrer-Cuzick + PRS for breast)
  • T04Use cases with clinical utility: CAD, breast cancer, prostate cancer, T2D, AF, hereditary haemochromatosis
  • T05Use cases without it: most psychiatric, most behavioral
  • T06Regulatory landscape: FDA, MHRA, the EU IVDR position
Learning objectives

By the end of this module you will be able to

  • L01Explain why a PRS trained in European-ancestry GWAS underperforms in African-ancestry populations and quantify the magnitude.
  • L02Describe an integrated risk model where PRS adds clinically meaningful discrimination on top of standard scores.
  • L03List two PRS-driven trials currently informing screening guidelines.
Key takeaways

What you should walk away believing

  • For coronary artery disease, the top 5% of polygenic risk carries ~3–4× the lifetime risk — comparable to a strong monogenic FH variant in absolute terms.
  • PRS portability loss across ancestries is real and large (often ~50% R² loss); deploying European-trained scores as 'universal' is malpractice.
  • Adding PRS to QRISK3 reclassifies ~15% of intermediate-risk patients up or down a treatment threshold for statin initiation (UK Biobank).
  • MyOme, Genomics PLC, Allelica, Nightingale are the main clinical-grade PRS implementers in 2025–2026.
Lesson · Foundations emphasis

What this means at your level

Foundations

Polygenic risk scores aggregate many small-effect variants from genome-wide association studies into a single number predicting disease risk. They've become genuinely useful for some common diseases (heart disease, breast cancer) but remain limited for others, and they perform much worse in populations not represented in the underlying GWAS.

Clinician deep-dive

Use PRS where: (1) outcome is common, (2) effective intervention exists, (3) score is calibrated to the patient's ancestry, and (4) it would change management. Best current candidates: CAD primary prevention (statin decision), breast cancer screening start age and frequency, prostate cancer biopsy threshold. Avoid PRS for major depressive disorder, schizophrenia, intelligence — discrimination is real but interventions are not, and ethics are unresolved.

Research note

Multi-ancestry methods (PRS-CSx, MUSSEL, BridgePRS) are closing but not eliminating the ancestry gap. The All of Us, H3Africa, GenomeAsia 100K, and TOPMed cohorts are the substrate for the next generation. Methylation-based polygenic risk (mPRS) and transcriptome-imputation methods (TWAS, PrediXcan) add orthogonal information for some endpoints.

Myth-buster

If your PRS for X is in the top decile, you're going to get X.

Reality

PRS gives a relative risk over the population baseline, not a deterministic prediction. Top-decile breast-cancer PRS lifts lifetime risk from ~12% to ~30% — meaningful, but ~70% still don't get the disease. Counsel as risk modification, not prophecy.

Evidence-graded claims

What the data say

Quick check

Test yourself

Q1The biggest known limitation of most clinically deployed PRS is:
Q2PRS would most appropriately be used today for:
Glossary

Key terms & abbreviations

PRSPolygenic Risk Score
Weighted sum of effect alleles across many loci, derived from GWAS summary statistics.
GWASGenome-Wide Association Study
Population study testing millions of variants for disease association; the substrate for PRS.
LDLinkage Disequilibrium
Non-random correlation between alleles at nearby loci; LD reference panels (1000G, UKB) shape PRS construction.
Net Reclassification Improvement
Metric for whether a new biomarker moves people across clinically meaningful risk thresholds — more useful than ΔAUC for clinical decisions.
Further reading

Anchor references