Identifying Effective Biomarkers for Accurate Pancreatic Cancer Prognosis Using Statistical Machine Learning

Area of Science:

• Oncology
• Biochemistry
• Medical Diagnostics

Background:

• Pancreatic cancer (PC) has a low survival rate and is a leading cause of cancer mortality.
• Effective prognostic biomarkers for PC progression and post-surgical complications are lacking.
• Current literature has limited studies on comprehensive biomarker panels for PC prognosis.

Purpose of the Study:

• To investigate the role of serum biomarkers (CA19-9, CXCL-8, PCT) and clinical data in predicting PC progression.
• To identify an optimal biomarker panel for classifying PC patients into sepsis, recurrence, and other post-surgical complications.
• To evaluate the diagnostic accuracy of machine learning models using the identified biomarker panel.

Main Methods:

• Utilized a random-forest-based feature elimination method to identify key prognostic markers.
• Collected serum biomarkers including carbohydrate antigen 19-9 (CA19-9), chemokine (C-X-C motif) ligand 8 (CXCL-8), and procalcitonin (PCT).
• Employed machine learning classification models to assess the efficacy of the biomarker panel on independent test data.

Main Results:

• A combined biomarker panel demonstrated superior performance in classifying PC progression compared to CA19-9 alone.
• The developed panel achieved a maximum Area Under the Receiver Operating Characteristic curve (AUC-ROC) of 100%.
• Exclusively using CA19-9 yielded a maximum AUC-ROC of 75% for PC progression classification.

Conclusions:

• A novel, combined biomarker panel effectively diagnoses pancreatic cancer progression and post-surgical complications.
• This panel significantly enhances diagnostic accuracy for PC survivors, particularly in Egyptian patients.
• The findings highlight the potential of integrated biomarker strategies for improved PC management.