A Bayesian model for estimating multi-state disease progression.

Area of Science:

• Biostatistics
• Medical Imaging Analysis
• Cancer Epidemiology

Background:

• Population cancer screening is increasing for high-risk individuals.
• Harms like radiation exposure and overtreatment necessitate improved screening models.
• Accurate estimation of mean sojourn time (MST) is crucial for optimizing screening frequency and policy.

Purpose of the Study:

• To develop a probabilistic modeling approach for periodic cancer screening data.
• To address limitations of traditional methods, including ignoring observation error and sparse data issues.
• To provide more accurate and stable estimation of MST.

Main Methods:

• Utilized a three-state continuous Markov model (CMM) to represent cancer state transitions.
• Incorporated observation error directly into the model.
• Employed a Bayesian framework for joint estimation of MST and observation error, including covariates for individualized progression rates.

Main Results:

• The proposed Bayesian approach provides more accurate and sensible MST estimates compared to Maximum Likelihood Estimation (MLE).
• The model effectively handles observation error, a common issue in medical imaging.
• Demonstrated improved performance on data from the National Lung Screening Trial (NLST).

Conclusions:

• The developed Bayesian probabilistic model offers a superior alternative for estimating MST in cancer screening.
• Accounting for observation error and using a Bayesian framework enhances the reliability of screening policy recommendations.
• This approach holds promise for refining personalized cancer screening protocols.