Developing Evidence-Driven Strategies Using Data Mining for Improving Patient Satisfaction

The growing recognition of patient satisfaction as a quality indicator necessitates that health care providers focus on issues that will have most impact on patient satisfaction. The main objective of the study was to provide deeper insight into a development of decision rules for prioritizing efforts and resources in improving patient satisfaction.

We analyzed patient satisfaction surveys in five service areas: outpatient, inpatient, ambulatory surgery, emergency, test services. We used decision tree induction and association rule mining to predict overall patient satisfaction. A decision tree for each service was created and it was translated into a set of rules to facilitate developing action plans for improving patient satisfaction. We examined the data at two levels in selecting study variables: 1) overall mean scores of the question categories and 2) the score of each question in the question categories. We transformed the Likert scale variables to corresponding binary variables. To examine what causes a difference in the degree of satisfaction, we tested different splitting criteria for the binary variables. The decision trees were evaluated using cross-validation and independent test datasets. We also discovered co-occurrence patterns of the variables using association rule mining. The results of the study showed that patient satisfaction exhibited distinct characteristics according to the service areas, and the decision trees and co-occurrence patterns provided explicit decision rules for our prioritization process. Effective application of data mining methods would enable us to focus on specific patient needs that could maximize our efforts in enhancing quality of patient care.