Data-augmented multistate modeling for chronic disease processes using cross-sectional studies: application to HPV...

Area of Science:

• Epidemiology
• Mathematical Modeling
• Public Health

Background:

• Understanding human papillomavirus (HPV) infection and cervical precancer natural history is crucial for effective prevention strategies.
• Multistate models are valuable for characterizing disease progression but often require costly longitudinal data.
• Cross-sectional data are more accessible, especially in resource-limited settings, highlighting a need for methods adaptable to this data type.

Purpose of the Study:

• To develop a semi-Markov multistate model for analyzing HPV infection dynamics and cervical precancer progression.
• To enable the estimation of disease transition intensities using primarily cross-sectional data.
• To provide a flexible framework for inferring population-specific transition intensities influenced by local factors.

Main Methods:

• Developed a semi-Markov multistate model incorporating recurrent transitions between HPV-infected and uninfected states.
• Proposed a two-step estimation approach combining local cross-sectional data with external transition intensity information.
• Estimated biologically driven transition intensities from external data and incorporated them into an approximate likelihood for local inference.

Main Results:

• The proposed model effectively captures HPV infection dynamics and cervical precancer progression.
• The two-step approach successfully enables estimation from cross-sectional data by leveraging transferable transition intensities.
• Demonstrated the method's performance and practical utility through simulation studies and application to two distinct populations.

Conclusions:

• The developed semi-Markov multistate model offers a practical approach to understanding HPV natural history using accessible data.
• This method facilitates the estimation of population-specific transition intensities, crucial for tailoring cervical cancer prevention strategies.
• The approach holds significant potential for public health applications, particularly in resource-limited settings where longitudinal data collection is challenging.