Identifying Effect Modification of Latent Population Characteristics on Risk Factors with a Sparse Varying...

Area of Science:

• Epidemiology
• Biostatistics
• Machine Learning
• Genomics

Background:

• Observational data analysis is crucial for understanding disease risk factors and outcomes.
• Traditional models often assume fixed associations, potentially missing complex individual variations.
• Latent features influencing these associations may be partially captured by observed risk factors.

Purpose of the Study:

• To develop a novel regression model that incorporates effect modification by latent features.
• To improve the accuracy of disease risk prediction by capturing dynamic associations.
• To identify latent effect modifications in real-world datasets, such as lung cancer.

Main Methods:

• Developed a novel regression model where coefficients vary as functions of latent features.
• Extracted latent features from observed risk factor data.
• Validated the model using simulation studies and applied it to The Cancer Genome Atlas (TCGA) lung cancer dataset.

Main Results:

• The proposed model demonstrated superior performance across various data settings in simulations.
• Application to lung cancer data showed a significant increase in predictive accuracy (6%-118% AUC-0.5 improvement) compared to lasso and elastic net.
• Identified novel latent effect modifications linked to specific gene pathways in lung cancer.

Conclusions:

• The novel regression approach effectively captures underlying data structures by modeling varying associations.
• This method enhances disease risk prediction and provides deeper insights into disease etiology.
• The findings highlight the importance of considering latent effect modifications in epidemiological studies.