Patient adherence is key to successfully managing chronic disease and to helping the patients to maintain quality of life. Quarterly visits to health care providers are an important part of care management for patients with diabetes. The objective of this study was to determine if patient visit adherence could be predicted by a set of known variables enabling health care providers to individualize a care management plan based on the possibility of a patient not completing quarterly visits.
A secondary data set including 225 records was used for the study. The secondary data set included both clinical and patient reported variables. Data was collected from patient self-reported survey data and a retrospective chart audit conducted by knowledgeable health care professionals. There were a total of 39 variables in the data set. The dependent variable in the data set was visit compliance.
Five separate neural network models were initially developed as part of the variable reduction method. Both a ROC curve graph and variable importance chart were reviewed for each model. Variables that had low variable importance values were eliminated. Variables that were found to play an important role in the prediction and classification of patient compliance were included in the final model. Because the sample size of 220 was small, the final model was limited to no more than 15 variables. If the model included more than 15 variables, it would have difficulty discovering and learning the patterns that could exist. The final model was then tested for accuracy and validity.
This study suggests that a decision support model that could help a physician as well as a health care system manage this diabetic population could be a cost-effective solution to time and resource management. In addition, neural networks could be beneficial as a tool for attention focusing. Decision support systems incorporating models like the model developed during this study that are able to alert both patients and health care practitioners of high risk for certain diseases or attributes such as noncompliance could help a health care system care, assess, and diagnose the growing number of elderly patients accessing the health system on a regular basis. In addition tools like this can be a more cost-effective way of preliminary diagnosis that could or could not lead to additional and expensive diagnostic testing or variance in patient treatment plans.