Bayesian Probabilistic Network Modeling of Risk Assessment for Patients with Melanoma

Author: Yan Xing, MD

Primary Advisor: Jiajie Zhang, PhD

Committee Members: Todd R. Johnson, PhD; Hongbin Wang, PhD

Masters thesis, The University of Texas School of Health Information Sciences at Houston.

 
53,600 Americans were diagnosed with melanoma in 2002. Nearly 60% of these patients required sentinel lymph node (SLN) biopsy techniques for precise staging. The 6th edition of the American Joint Committee on Cancer (AJCC) staging system for cutaneous melanoma incorporates a number of recently identified significant prognostic factors. This staging system is very complex and difficult to implement. Bayesian networks (BN) employ probability theory to reason under uncertainty, and have become an important formalism for medical decision support systems. Applying SEER data and literature-based estimates, two BNs were developed in Hugin Lite® v5.7, which integrate two demographic factors, three clinical findings, and six pathological features extracted by experienced melanoma specialists to generate patient-specific predictive probability estimates of stage and recurrence. The methods in the system's design, implementation, and evaluation are described. The models were evaluated against data from 40 Melanoma patients and a significant correlation in risk assessment was found. BN can be utilized as decision-support tools for counseling and educating patients with melanoma.