Modeling intra-tumor protein expression heterogeneity in tissue microarray experiments.

Area of Science:

• Biostatistics
• Cancer Research
• Proteomics

Background:

• Tissue microarrays (TMAs) are crucial for measuring tumor protein expression and validating cancer biomarkers.
• Variability in TMA data, arising from repeated measurements within tumors, is often overlooked in survival analyses.
• Ignoring TMA variability leads to biased hazard ratio estimates in proportional hazards models.

Purpose of the Study:

• To develop and evaluate methods for accurately associating TMA protein expression data with patient survival outcomes.
• To address the bias introduced by inherent variability in TMA measurements.
• To compare a novel joint modeling approach with traditional two-stage methods.

Main Methods:

• Proposed a Latent Expression Index (LEI) as a surrogate for protein expression in a two-stage analysis.
• Compared empirical Bayes, full Bayes, and varying replicate number estimators for LEI.
• Developed and implemented a joint model with shared random effects for survival and TMA expression data.
• Utilized Markov chain Monte Carlo (MCMC) for Bayesian estimation.
• Conducted simulation studies to compare the performance of two-stage and joint models.

Main Results:

• Two-stage methods reduced bias compared to naive approaches but still underestimated hazard ratios.
• The joint model consistently outperformed two-stage methods in terms of bias and coverage probability across simulations.
• Case studies on prostate cancer TMA data showed variable performance of two-stage methods, highlighting the joint model's robustness.

Conclusions:

• The joint model provides a more accurate and reliable method for analyzing TMA data in cancer survival studies.
• It effectively accounts for the biological and experimental variability inherent in TMA measurements.
• Recommends using joint model inference, especially when results from two-stage methods differ, for robust biomarker association.