Developing an optimal stratification model for colorectal cancer screening and reducing racial disparities in multi-center population-based studies

Abstract

Background: Early detection of colorectal neoplasms can reduce the colorectal cancer (CRC) burden by timely intervention for high-risk individuals. However, effective risk prediction models are lacking for personalized CRC early screening in East Asian (EAS) population. We aimed to develop, validate, and optimize a comprehensive risk prediction model across all stages of the dynamic adenoma-carcinoma sequence in EAS population.

Methods: To develop precision risk-stratification and intervention strategies, we developed three trans-ancestry PRSs targeting colorectal neoplasms: (1) using 148 previously identified CRC risk loci (PRS₁₄₈); (2) SNPs selection from large-scale meta-analysis data by clumping and thresholding (PRS₁₈₃); (3) PRS-CSx, a Bayesian approach for genome-wide risk prediction (PRS_Genomewide). Then, the performance of each PRS was assessed and validated in two independent cross-sectional screening sets, including 4600 patients with advanced colorectal neoplasm, 4495 patients with non-advanced adenoma, and 21,199 normal individuals from the ZJCRC (Zhejiang colorectal cancer set; EAS) and PLCO (the Prostate, Lung, Colorectal, and Ovarian Cancer Screening Trial; European, EUR) studies. The optimal PRS was further incorporated with lifestyle factors to stratify individual risk and ultimately tested in the PLCO and UK Biobank prospective cohorts, totaling 350,013 participants.

Results: Three trans-ancestry PRSs achieved moderately improved predictive performance in EAS compared to EUR populations. Remarkably, the PRSs effectively facilitated a thorough risk assessment across all stages of the dynamic adenoma-carcinoma sequence. Among these models, PRS₁₈₃ demonstrated the optimal discriminatory ability in both EAS and EUR validation datasets, particularly for individuals at risk of colorectal neoplasms. Using two large-scale and independent prospective cohorts, we further confirmed a significant dose-response effect of PRS₁₈₃ on incident colorectal neoplasms. Incorporating PRS₁₈₃ with lifestyle factors into a comprehensive strategy improves risk stratification and discriminatory accuracy compared to using PRS or lifestyle factors separately. This comprehensive risk-stratified model shows potential in addressing missed diagnoses in screening tests (best NPV = 0.93), while moderately reducing unnecessary screening (best PPV = 0.32).

Conclusions: Our comprehensive risk-stratified model in population-based CRC screening trials represents a promising advancement in personalized risk assessment, facilitating tailored CRC screening in the EAS population. This approach enhances the transferability of PRSs across ancestries and thereby helps address health disparity.

Keywords: CRC early screening; Colorectal neoplasm; Lifestyle factors; Polygenic risk score; Trans-ancestry.

Fig. 1

Overview of the study design. First, three trans-ancestry PRS approaches were implemented. After assessment and validation in the East Asian and European ancestry screening populations, the best-fitting PRS model together with generated ERS for CRC screening was determined. At last, the prediction effect of constructed PRS and ERS was further evaluated in cohorts

Fig. 2

Quality control of GWAS meta-analysis and functional annotation of the newly discovered risk variant rs140356782 in 10q23.31. A Manhattan plot of GWAS meta-analysis results, the P values (-log10) of the SNPs (y-axis) are presented according to their chromosomal positions (x-axis). B QQ plot of GWAS meta-analysis. C Epigenetic tracks obtained from the ATAC-seq peaks of our own three-stage tissues (CRC, adenoma, and normal) and ENCODE database show the enrichment of enhancer marks (DNase modification peaks, H3K4me1, and H3K27ac peaks) in the rs140356782 region. D Hi-C plots reveal the interaction of the region containing rs140356782 with PANK1 promoter in our own three-stage tissues (CRC, adenoma, and normal). E–G eQTL analyses demonstrate the correlation between rs140356782 genotype and the expression of PANK1 in the GTEx normal transverse colon samples (E), TCGA colorectal cancer samples (F), and our own colorectal cancer tissues (G). Data are shown as the median (minimum to maximum). P values were calculated by a two-sided Student’s t test in own colorectal cancer tissues, respectively

Fig. 3

Risk score distributions and effect comparison of three approaches to PRS construction and in assessment and validation set. A–C Distribution of PRS₁₄₈ (A), PRS₁₈₃ (B), and PRS_Genomewide (C) in ZJCRC case–control set respectively. D–F Distribution of PRS₁₄₈ (D), PRS₁₈₃ (E), and PRS_Genomewide (F) in ZJCRC cross-sectional screening set respectively. G–I Distribution of PRS₁₄₈ (G), PRS₁₈₃ (H), and PRS_Genomewide (I) in PLCO cross-sectional screening set respectively. J ORs of three PRS models for each PRS decile in ZJCRC case–control set. K–N ORs of three PRS models for each PRS decile through different groups and comparisons in ZJCRC cross-sectional screening set. O–U ORs of three PRS models for each PRS decile through different groups and comparisons in PLCO cross-sectional screening set

Fig. 4

The risk of colorectal cancer screening according to PRS and ERS categories across three assessment and validation set. A ORs for colorectal neoplasms in low, intermediate, and high environmental risk groups in the ZJCRC case–control set. B ORs for colorectal neoplasms in low, intermediate, and high environmental risk groups through different groups and comparisons in ZJCRC cross-sectional screening set. C ORs for colorectal neoplasms in low, intermediate, and high environmental risk groups through different groups and comparisons in PLCO cross-sectional screening set. D ORs for colorectal neoplasms according to genetic and environmental categories in the ZJCRC case–control set. E–H ORs for colorectal neoplasms according to genetic and environmental categories through different groups and comparisons in ZJCRC cross-sectional screening set. I–L ORs for colorectal neoplasms according to genetic and environmental categories through different groups and comparisons in PLCO cross-sectional screening set. ORs are adjusted for age, sex, family history, principal components, and genotype platform. 95% confidence intervals are shown for all analyses

Fig. 5

Evaluation of absolute risk predictions of incident colorectal neoplasm according to PRS and ERS in the PLCO and UK Biobank cohort. A Distribution of PRS when non-advanced adenoma vs normal group within PLCO incident adenoma cohort. B, C Distribution of PRS when advanced adenoma vs normal group within PLCO incident adenoma cohort (B) and UK Biobank cohort (C). D Inverted Kaplan–Meier plot of incident colorectal neoplasm by PRS when non-advanced neoplasm vs normal group within PLCO incident adenoma cohort. E, F Inverted Kaplan–Meier plot of incident colorectal neoplasm by PRS when advanced neoplasm vs normal group within PLCO incident adenoma cohort (E) and UK Biobank cohort (F). G Inverted Kaplan–Meier plot of incident colorectal neoplasm by ERS when non-advanced adenoma vs normal group within PLCO incident adenoma cohort. H, I Inverted Kaplan–Meier plot of incident colorectal neoplasm by ERS when advanced neoplasm vs normal group within PLCO incident adenoma cohort (H) and UK Biobank cohort (I). Participants were divided into most, moderately, and least healthy groups. J Inverted Kaplan–Meier plot of incident colorectal neoplasm according to genetic and environmental categories when non-advanced adenoma vs normal group within PLCO incident adenoma cohort. K–L Inverted Kaplan–Meier plot of incident colorectal neoplasm according to genetic and environmental categories when advanced neoplasm vs normal group within PLCO incident adenoma cohort (K) and UK Biobank cohort (L). Participants were divided into 9 risk groups. The cumulative events table under the plot showed the cumulative incident events of incident colorectal neoplasm cases at years of follow-up. M Per 100,000 person-year at risk separately in 9 risk groups in the context of non-advanced adenoma vs normal group within PLCO incident adenoma cohort. N, O Per 100,000 person-year at risk separately in 9 risk groups in the context of advanced neoplasm vs normal group within PLCO incident adenoma cohort (N) and UK Biobank cohort (O)