Automatic Classification of Patient-Generated Messages from a Patient Portal
Cronin, Robert Michael
People with questions about their health have been seeking answers from online resources. Patient portals, electronic applications that allow patients to interact with their healthcare providers, have had increasing adoption because of consumer demand and governmental regulations. Secure messaging is one of the most popular functions of patient portals, allowing individuals to communicate with their healthcare providers. We sought to automate the classification of the contents of patient-generated portal messages into categories from a consumer health communication types taxonomy. We developed classifiers using a simple rule-based approach and three machine learning methods, Naïve Bayes, logistic regression and random forests, using a bag of words and natural language processing techniques. We evaluated the ability of these classifiers to identify a single communication type within a message with area under the receiver-operator curve (AUC) and the ability to predict all communication types in a message with a Jaccard Index. Using these approaches, secure messages were automatically classified into communication type categories with good accuracy, with AUCs over 0.75 and Jaccard indices over 0.85 for the best performing classifiers. The performance of the different approaches to classification varied by communication type. As adoption of patient portals increases, automated techniques may be needed to manage growing volumes of secure messages. Automated classification of secure messages through patient portals may aid in triaging secure messages, connecting patients to needed resources, and understanding the nature of communications and types of care delivered within patient portals.