CCE Theses and Dissertations


A Bayesian Framework to Determine Patient Compliance in Glaucoma Cases

Date of Award


Document Type


Degree Name

Doctor of Philosophy (PhD)


Graduate School of Computer and Information Sciences


Sumitra Mukherjee

Committee Member

Michael J. Laszlo

Committee Member

K.V. Chalam


This dissertation develops a Bayesian framework to assess medication compliance in glaucoma patients. Bayesian Networks have increasingly become tools of choice in solving problems involving uncertainty in the medical domain. These models have been successfully applied to diagnosis applications. This research applied Bayesian modeling to medication noncompliance in glaucoma patients. Medication noncompliance is the failure to comply with a physician's instructions with regard to taking medications at specified times. If the patient is non-compliant, irrespective of the advances in medical field, the person does not benefit from medical intervention. A model-based decision support system using a Bayesian Network was developed to determine whether a patient was complying with the medications prescribed by the physician. The predictive ability of the model was investigated using the existing patient data. To assess research validity, the results obtained through the model were compared against a domain expert's evaluation of the patient cases. The results provided by the Bayesian framework agree with the information provided by the domain expert. The Bayesian model can be used to confirm an ophthalmologist's clinical intuition or to formulate a prescription strategy for a glaucoma patient. The model can be further refined using larger patient data sets and additional variables. A clinical decision support system can be developed using the refined model to prevent medical errors in glaucoma compliance process. Results from this study could potentially improve the decision making process, given the uncertain and incomplete data available to a physician. The Bayesian approach may be generalized to other applications where a decision has to be made based on incomplete and uncertain data sets.

This document is currently not available here.

  Link to NovaCat