Modelling prognostic factors in advanced pancreatic cancer

Area of Science:

• Oncology
• Biostatistics
• Epidemiology

Background:

• Pancreatic cancer is a leading cause of cancer mortality.
• Accurate prognostic models are crucial for patient stratification and treatment selection.
• Previous prognostic factor identification often relied on limited statistical methods.

Purpose of the Study:

• To compare standard statistical models with a novel nonlinear fractional polynomial (FP) approach for prognostic factor identification in pancreatic cancer.
• To improve the prediction of survival and selection of therapy by appropriately analyzing continuous variables.
• To identify novel prognostic factors beyond those previously reported.

Main Methods:

• Analysis of data from 653 pancreatic cancer patients.
• Comparison of standard statistical models (linear, dichotomized) with a fractional polynomial (FP) transformation approach.
• Multivariable analysis to assess the relationship between continuous covariates and survival.

Main Results:

• The FP model was most appropriate for analyzing prognostic factors in pancreatic cancer.
• Confirmed five known prognostic factors: albumin, CA 19-9, alkaline phosphatase, LDH, and metastases.
• Identified three novel prognostic factors: white blood cell count (WBC), aspartate aminotransferase (AST), and blood urea nitrogen (BUN).
• Demonstrated that simplistic statistical assumptions can obscure the impact of key factors like CA 19-9, alkaline phosphatase, AST, and BUN.

Conclusions:

• Fractional polynomial modeling provides a more accurate assessment of prognostic factors in pancreatic cancer.
• Appropriate statistical analysis of continuous variables is essential for identifying significant prognostic factors.
• The developed model can enhance individual patient risk stratification and inform future clinical trial design and analysis.