Two-stage models for the analysis of cancer screening data

Area of Science:

• Biostatistics
• Epidemiology
• Medical Screening

Background:

• Analyzing disease natural history from screening data is complex when preclinical disease doesn't always progress.
• Existing models often assume a linear progression, which may not reflect reality.

Purpose of the Study:

• To develop statistical methods for analyzing disease natural history using screening data.
• To jointly estimate preclinical disease duration and screening test sensitivity.
• To accommodate models where preclinical disease progression is not guaranteed.

Main Methods:

• A two-stage model for preclinical disease is proposed (Stage 1 may progress to Stage 2; Stage 2 always progresses to clinical disease).
• A partial likelihood is developed for prospective screening data analysis.
• A conditional likelihood is proposed for retrospective data analysis.

Main Results:

• The methodology allows for joint estimation of total preclinical duration and screening test sensitivity.
• Special cases, including independent and limiting models for sojourn times, are considered.
• The methods were successfully applied to a cervical cancer screening case-control study.

Conclusions:

• The proposed methods provide a flexible framework for analyzing disease natural history from screening data.
• This approach enhances understanding of disease progression and improves the evaluation of screening test performance.
• The study demonstrates the utility of these methods in real-world epidemiological research.