[Paper Review] Bayesian Semiparametric Estimation of Cancer-Specific Age-at-Onset Penetrance With Application to Li-Fraumeni Syndrome

Seung Jun Shin, Ying Yuan+3 more|arXiv (Cornell University)|Jan 6, 2017

Epigenetics and DNA Methylation29 references3 citations

TL;DR

This paper proposes a Bayesian semiparametric competing risk model to estimate cancer-specific age-at-onset penetrance in Li-Fraumeni syndrome, accounting for family structure, missing genotype data, and ascertainment bias. The method uses Bernstein polynomials for flexible baseline hazard estimation and family-wise likelihood with gamma frailty, showing superior fit and predictive performance over parametric alternatives in a large, long-followed cohort of 186 families.

ABSTRACT

Penetrance, which plays a key role in genetic research, is defined as the proportion of individuals with the genetic variants (i.e., genotype) that cause a particular trait and who have clinical symptoms of the trait (i.e., phenotype). We propose a Bayesian semiparametric approach to estimate the cancer-specific age-at-onset penetrance in the presence of the competing risk of multiple cancers. We employ a Bayesian semiparametric competing risk model to model the duration until individuals in a high-risk group develop different cancers, and accommodate family data using family-wise likelihoods. We tackle the ascertainment bias arising when family data are collected through probands in a high-risk population in which disease cases are more likely to be observed. We apply the proposed method to a cohort of 186 families with Li-Fraumeni syndrome identified through probands with sarcoma treated at MD Anderson Cancer Center from 1944 to 1982. Supplementary materials for this article are available online.

Motivation & Objective

• To estimate cancer-specific age-at-onset penetrance in Li-Fraumeni syndrome (LFS) under competing risks of multiple cancers.
• To address ascertainment bias arising from sampling through early-onset sarcoma probands in a high-risk population.
• To model family data with missing genotype information using family-wise likelihood and frailty models.
• To develop a flexible, nonparametric approach to baseline hazard that improves model fit and predictive accuracy.
• To provide reliable penetrance estimates for personalized risk assessment in TP53 mutation carriers.

Proposed method

• Uses a Bayesian semiparametric competing risk model with Bernstein polynomials to nonparametrically estimate the baseline hazard function.
• Incorporates gamma frailty to account for within-family correlation in age-at-onset of cancer.
• Employs family-wise likelihood to integrate over unobserved genotypes and leverage full pedigree information.
• Applies MCMC sampling with data augmentation to handle missing genotype data and estimate posterior distributions.
• Corrects for ascertainment bias using an ascertainment-corrected likelihood that conditions on the proband's sarcoma diagnosis.
• Uses pseudo-marginal likelihood (PsML) and deviance information criterion (DIC) for model comparison and selection.

Experimental results

Research questions

• RQ1 How can age-at-onset penetrance for specific cancers be estimated in the presence of competing risks in a high-risk genetic cohort?
• RQ2 What is the impact of ascertainment bias on penetrance estimation when data are collected through probands with early-onset sarcoma?
• RQ3 How does incorporating family structure and missing genotype data affect the accuracy and precision of penetrance estimates?
• RQ4 Which baseline hazard model—exponential, Weibull, piecewise-constant, or Bernstein polynomial—provides the best fit and predictive performance?
• RQ5 To what extent does the inclusion of frailty terms improve model fit, indicating within-family correlation?

Key findings

• The proposed Bernstein polynomial-based model achieved the lowest DIC (2983.7) and highest PsML (-1499.689), indicating superior model fit and predictive ability compared to exponential, Weibull, and piecewise-constant models.
• The inclusion of gamma frailty improved model fit only slightly, suggesting weak within-family correlation, consistent with large estimated variance components (ν) in the frailty distribution.
• Penetrance estimates were robust to prior sensitivity, with results stable across flat, weakly informative, and moderately informative priors for γm,k and νk.
• The model effectively corrected for ascertainment bias, enabling unbiased inference from a cohort enriched for sarcoma probands.
• The method successfully handled missing genotype data by integrating over all possible genotypes using family-wise likelihood, maximizing information use from extended pedigrees.
• The approach outperformed alternative models in both fit and prediction, as confirmed by cross-validated ROC curves and model comparison metrics.