Structure-preserving integrated analysis for risk stratification with application to cancer staging

Area of Science:

• Biostatistics
• Medical Informatics
• Oncology

Background:

• Effective healthcare requires patient stratification into distinct prognostic groups.
• Current methods like the American Joint Committee on Cancer staging use established risk factors and outcomes.
• Analyzing individual patient data from multiple studies presents challenges due to heterogeneity and unbalanced data.

Purpose of the Study:

• To develop a statistical method for patient grouping using individual patient data from multiple studies.
• To enhance the lasso-tree method by establishing its theoretical properties.
• To improve cancer staging and patient prognosis stratification.

Main Methods:

• Utilizing a strengthened lasso-tree statistical method for patient stratification.
• Incorporating underlying order information in risk factors through lasso-tree parametrization.
• Leveraging individual patient data from multiple studies while accounting for data heterogeneity.

Main Results:

• The enhanced lasso-tree method demonstrates versatility in generating grouping patterns.
• The method effectively handles data heterogeneity and unbalanced structures across studies.
• Simulation studies and breast cancer data analysis validate the method's performance.

Conclusions:

• The proposed statistical method offers a robust approach to patient stratification for improved healthcare.
• This work strengthens the theoretical underpinnings of the lasso-tree method for cancer staging.
• The method facilitates more accurate prognosis assessment and tailored medical interventions.