Modeling Individual-Level Uncertainty From Missing Data in Multifactorial Breast Cancer Risk Prediction.

Area of Science:

• Oncology
• Biostatistics
• Genetics

Background:

• Multifactorial breast cancer (BC) risk models are essential for personalized risk assessment.
• Incomplete risk factor data introduces significant uncertainty into BC risk predictions.
• Accurate quantification of this uncertainty is crucial for effective risk communication and clinical decision-making.

Purpose of the Study:

• To quantify the uncertainty in 10-year BC risk estimates for individuals with missing risk factor data.
• To develop and apply a framework for estimating risk uncertainty distributions and reclassification probabilities.
• To identify the impact of missing data on BC risk stratification.

Main Methods:

• Utilized Monte Carlo simulation methods with the BOADICEA model to estimate BC risk distributions.
• Employed multivariate imputation by chained equations using large reference datasets to handle missing covariates.
• Developed a framework to calculate uncertainty intervals (UIs) and probability of reclassification for individuals with incomplete data.

Main Results:

• Incomplete risk factor data led to considerable uncertainty in BC risk estimates, with 95% UIs spanning all risk categories.
• Moderate-risk women, particularly those with family history or pathogenic variants, showed high reclassification probabilities (up to 57.5%).
• Risk certainty improved substantially with additional data, especially genetic information and mammographic density.

Conclusions:

• Missing data can lead to substantial probabilities of risk reclassification, impacting clinical decisions.
• The presented methodology effectively identifies situations where additional data collection is most beneficial.
• Improved risk stratification through data collection supports more informed clinical decision-making in breast cancer risk assessment.