Multi-state Markov models in cancer screening evaluation: a brief review and case study

Area of Science:

• Biostatistics
• Epidemiology
• Public Health

Background:

• Markov models are valuable tools for evaluating cancer screening programs.
• Assessing screening effectiveness requires accurate estimation of sensitivity and preclinical phase duration.
• Predicting mortality reduction necessitates models that account for disease progression and prognostic factors.

Purpose of the Study:

• To provide an overview of Markov models in cancer screening evaluation.
• To analyze the strengths and limitations of a three-state and a five-state Markov model.
• To discuss the impact of over-diagnosis on model parameter estimates.

Main Methods:

• A three-state Markov model was used to estimate screening sensitivity and preclinical phase duration.
• A five-state Markov model, including lymph node involvement, was combined with survival analysis.
• Data from French breast cancer screening programs were utilized to illustrate model performance.

Main Results:

• The three-state model's parameter estimates are highly correlated and sensitive to parametric assumptions.
• The five-state model has limitations due to unverified implicit assumptions.
• Over-diagnosis introduces bias in parameter estimates for both models.

Conclusions:

• While useful, current Markov models have limitations requiring cautious interpretation of results.
• More flexible models are needed for accurate cancer screening evaluation.
• Addressing over-diagnosis is crucial for unbiased parameter estimation and reliable mortality reduction predictions.