Independent validation of PHS601 for prostate cancer age-at-onset stratification in two Norwegian cohorts with comparison to rare pathogenic variant testing

This study provides the first external validation of the PHS601 polygenic hazard score in two independent Norwegian cohorts, demonstrating its superior ability to stratify prostate cancer risk and predict age-at-onset compared to rare pathogenic variant screening.

The Gist

Imagine your risk of getting prostate cancer is like the weather forecast for a specific day in your life. For a long time, doctors have used a single tool (like a PSA blood test) to guess if a storm is coming. But this tool often predicts rain when it's actually just a cloudy day, leading to unnecessary panic and "over-treating" harmless conditions.

This study introduces a much more sophisticated weather map called PHS601. Instead of looking at just one cloud, PHS601 looks at thousands of tiny genetic "clouds" (common variations in your DNA) to predict exactly when you might get sick.

Here is what the researchers did and found, explained simply:

1. The Experiment: Testing the New Map

The researchers took this new "genetic weather map" (PHS601) and tested it on two groups of Norwegian men:

• Group A (HUSK): A large group of regular men from the general population.
• Group B (NFCS): A group of men who already had a family history of cancer (a "high-risk" group).

They wanted to see if the map could correctly predict who would get prostate cancer and, more importantly, who would get the dangerous, aggressive kind.

2. The Results: The Map Works

The results were like finding a compass that actually points North.

• Sorting the Crowd: The map successfully sorted men into "low risk" and "high risk" groups. Men in the top 20% of risk were about 5 to 8 times more likely to develop prostate cancer at a younger age compared to men in the bottom 20%.
• Predicting the Storm: It didn't just predict any cancer; it was also good at predicting the "severe storms" (aggressive cancer). Men with high scores were significantly more likely to develop aggressive disease.
• The Age Factor: The map is most accurate for younger men. Think of it like a weather forecast: it's very good at predicting a storm next week (younger age), but as you get older, other factors (like lifestyle or random cellular changes) start to matter more than your genetic blueprint, making the genetic forecast slightly less dominant.

3. The Big Showdown: The "Genetic Map" vs. The "Rare Defect Detector"

This is the most exciting part of the study. Usually, when we look for genetic risks, we hunt for "rare, broken parts" (like a missing gear in a machine). These are rare pathogenic variants (RPVs), such as a broken HOXB13 gene.

• The Rare Defect: The researchers found that men with this specific broken gene had about a 3.8 times higher risk of cancer.
• The Genetic Map: However, when they looked at the top 1.8% of men with the highest polygenic scores (the ones with the most "storm clouds"), their risk was 8.8 times higher.

The Analogy: Imagine you are looking for a reason a car might crash.

• RPV Screening is like checking if the car has a specific, rare, broken brake pedal. If it does, it's dangerous. But very few cars have this broken pedal.
• PHS601 is like checking the overall wear and tear on the tires, the engine, and the suspension. Even if no single part is "broken," the combination of many small issues can make the car much more likely to crash.

The study found that looking at the overall wear and tear (PHS601) was actually better at finding the most dangerous drivers than just looking for the one broken brake pedal (RPV).

4. Why This Matters (According to the Paper)

The paper suggests that using this genetic map could help doctors decide who needs to be screened and when.

• High-Risk Men: Could start screening earlier and more often to catch the "storm" before it hits.
• Low-Risk Men: Could wait longer or screen less often, avoiding the "false alarms" and unnecessary treatments that happen with current methods.

The Bottom Line

This study is the first time this specific genetic map has been tested on independent Norwegian groups, and it worked perfectly. It proves that looking at the "big picture" of your genetics (thousands of small factors) is a powerful way to predict prostate cancer risk, potentially even better than hunting for rare, single-gene errors. It offers a new way to tailor cancer screening so it's more accurate and less stressful for everyone.

Technical Summary

1. Problem Statement

Prostate cancer (PCa) is a leading cause of cancer mortality, but current screening methods relying on Prostate-Specific Antigen (PSA) are limited by false positives, overdiagnosis, and overtreatment of indolent disease. There is a critical need for risk stratification tools that can identify men at high risk for early-onset and clinically significant (aggressive) PCa to optimize screening intervals.
While Polygenic Hazard Scores (PHS) like PHS601 have shown promise in US cohorts (Million Veterans Program) and the PRACTICAL consortium, they lack independent external validation in European population-based cohorts with distinct healthcare systems and genetic backgrounds. Furthermore, the relative utility of PHS-based stratification compared to Whole-Genome Sequencing (WGS) for detecting rare pathogenic variants (RPVs) in high-penetrance genes (e.g., BRCA1/2, HOXB13) remains unclear in a direct, within-cohort comparison.

2. Methodology

The study utilized two independent Norwegian cohorts to validate PHS601 and compare it against RPV screening:

• Study Populations:
  • HUSK (Hordaland Health Studies): A population-based cohort ($N=14,688$ males) recruited in western Norway (1992–1999).
  • NFCS (Norwegian Familial Cancer Study): A clinical cohort enriched for family history of cancer ($N=2,850$ males), recruited nationwide (1980–1995).
• Phenotyping:
  • PCa cases were identified via ICD-10 code C61 from the Norwegian Cancer Registry (98.8% complete).
  • Aggressive PCa was defined using specific metastasis codes indicating lymph node involvement, distant metastasis, or invasion of neighboring structures.
  • Time-to-event analysis used age as the time scale, with follow-up from birth to diagnosis or censoring.
• Genotyping & PHS Calculation:
  • HUSK: Genotyped on Illumina GSA v3 arrays; imputed using HGDP + 1KG reference panels.
  • NFCS: Genotyped on Illumina OmniExpress 24 v1.1; imputed using HRC v1.1.
  • PHS601: Calculated using 519 (HUSK) and 557 (NFCS) SNPs from the original 601-SNP model (21 X-chromosome SNPs were excluded due to lack of imputation data). Scores were centered and scaled.
• Whole-Genome Sequencing (WGS) & RPV Screening:
  • A subset of the NFCS cohort ($N=503$ with both WGS and genotyping) underwent WGS (average coverage 37.2X).
  • RPV Identification: Used the Cancer Predisposition Sequencing Reporter (CPSR) to screen for pathogenic variants in 12 high-risk genes (BRCA1, BRCA2, ATM, PALB2, CHEK2, HOXB13, MLH1, MSH2, MSH6, PMS2, TP53, EPCAM) based on ClinVar classifications.
• Statistical Analysis:
  • Cox Proportional Hazards Models: Used to estimate Hazard Ratios (HR) per standard deviation (SD) increase in PHS ($HR_{SD}$), between top/bottom quintiles ($HR_{80/20}$), and top 5% vs. median ($HR_{95/50}$).
  • Comparison: Direct comparison of PHS stratification vs. RPV carrier status (specifically HOXB13, the only gene showing enrichment) in the WGS subset.
  • Age Interaction: Tested for time-dependent effects of PHS using Schoenfeld residuals and time-varying covariates.

3. Key Contributions

• First External Validation: Provides the first independent validation of PHS601 in Norwegian, European population-based cohorts, confirming its transferability across different genetic backgrounds and healthcare systems.
• Direct PHS vs. WGS Comparison: Offers a rare, direct within-cohort comparison of genotyping-based polygenic risk stratification versus WGS-based rare variant screening for PCa.
• Aggressive Disease Stratification: Demonstrates that PHS601, trained on any-stage disease, effectively stratifies risk for aggressive/metastatic PCa, a crucial metric for clinical utility.
• Age-Dependent Risk Modeling: Quantifies the attenuation of genetic risk effects with age, suggesting PHS is most actionable for younger men.

4. Key Results

A. PHS601 Performance in General Populations

• HUSK Cohort:
  • $HR_{SD} = 1.87$ (95% CI [1.79, 1.95]) for any PCa.
  • $HR_{80/20} = 5.74$ (95% CI [5.06, 6.45]).
  • For aggressive PCa: $HR_{80/20} = 4.60$ (95% CI [3.19, 6.45]).
• NFCS Cohort (High Risk/Family History):
  • $HR_{SD} = 2.09$ (95% CI [1.87, 2.33]) for any PCa.
  • $HR_{80/20} = 7.79$ (95% CI [5.70, 10.59]).
  • For aggressive PCa: $HR_{80/20} = 3.14$ (95% CI [1.49, 6.63]).
• Clinical Impact: By age 75, men in the top PHS quintile had an estimated cumulative incidence of ~30% vs. ~8% for the bottom quintile.

B. PHS vs. Rare Pathogenic Variants (WGS Subset)

• RPV Enrichment: Among 12 screened genes, only HOXB13 showed significant enrichment in cases (OR = 9.65). The carrier frequency was 1.8%.
• Risk Comparison:
  • HOXB13 Carriers: $HR = 3.77$ (95% CI [1.75, 8.11]) compared to non-carriers.
  • Top 1.8% of PHS: Individuals in the top 1.8% of the PHS distribution (matching the HOXB13 carrier frequency) had an 8.78-fold higher risk ($HR = 8.78$, 95% CI [5.00, 14.40]) compared to the median.
• Discrimination: The PHS model had a C-index of 0.702, significantly outperforming the HOXB13 model (C-index = 0.519).

C. Age-Dependent Effects

• A negative trend was observed where the hazard ratio of PHS declined with age (approx. 1.1–1.6% per year).
• Genetic risk stratification is strongest at younger ages, suggesting the highest clinical utility for early screening decisions.

5. Significance and Implications

• Population-Level Utility: Genotyping-based PHS601 provides stronger population-level risk stratification than WGS-based screening for known rare variants. High polygenic risk affects a much larger proportion of the population (50x more than BRCA2 mutations) than rare pathogenic variants, making it more suitable for broad screening programs.
• Precision Screening: The study supports a shift from uniform PSA screening to risk-stratified screening. High-PHS men could benefit from earlier and more intensive surveillance (e.g., MRI), while low-PHS men could undergo reduced or deferred screening to mitigate overdiagnosis.
• Cost-Effectiveness: Since genotyping arrays can simultaneously calculate PHS and capture common variants like HOXB13 (via imputation), a single genotyping test may suffice for initial population screening, avoiding the high costs of WGS for the general population.
• Future Directions: The findings support the implementation of PHS in prospective clinical trials (e.g., the P-CARE trial) and highlight the need for cost-effectiveness analyses to determine optimal screening thresholds and age ranges, particularly focusing on younger men where genetic risk is most predictive.

Limitations: The study lacked family history data for all participants (though NFCS was enriched for it), and WGS was limited to a subset of one cohort, potentially underpowering the detection of extremely rare variants. Additionally, the NFCS cohort is not fully representative of the general population, necessitating reliance on HUSK for absolute risk estimates.