An Integrated Intelligent Framework for Public Health Informatics Based on Machine Learning Approaches

Abstract

In public health informatics domain, standard based interoperability frameworks with analytical capabilities on top remain a complex challenge. The increasing prevalence of communicable and non-communicable diseases all over the globe has made public health policy management and resource allocation a challenging task. Particularly in resource-limited locations, it becomes of high importance that appropriate policies and intervention programs be designed to target areas fairly and efficiently. Traditional diseases surveillance program are an effective way to collect such information. However, the one disease per program generates repeated data that remains under utilized for a potential broad view on public health status over a geographical area. The increasing number of public health surveillance programs for both communicable and non communicable diseases has increased our understanding of health status of population on one hand but on the other hand has also introduced challenges such as data redundancy, ineffective use of collected data and inability to understand multiple disease linkages with each other due to fragmented data. In addition, the spatio-temporal nature of disease data requires special attention in terms of application of machine learning algorithms for mining purposes. This thesis presents an HL7 based interoperability and analysis framework that is implemented as a disease data exchange model from multiple sources such as hospitals and laboratories to destinations including public health agencies for disease management processes necessary for research in public health decision making and disease prevention and control intervention programs. In particular, the thesis describes the framework in terms of architecture, components and algorithms for public health related data acquisition from multiple sources based on HL7 specification, data transmission using HL7 message transmission protocol, privacy preserving integration iv at a centralized data resource facility, disease management representation model and machine learning based data analysis models for disease management in public health context. We are able to demonstrate the use of machine learning algorithms for disease outbreak detection using model based approach that can be scaled as more techniques and algorithms evolve for spatio-temporal data analysis. The proposed framework in its adoption provides a very effective platform for generating alerts and alarms along with providing insights for better planning of healthcare related issues at national, district or at any level of administrative hierarchy. The presented results in thesis show that the disease outbreaks can be identified in near real time in its social and environmental context using multiple parameters and evidences. The model can then be used to inform decision-making by projecting the potential outcomes associated with different policy decisions. The strongest attribute of the presented work is its ability to incorporate any number of features of public health interest and flexibility of conducting different types of analysis as per the data available.