Identifying the Most Vulnerable Groups by Machine Learning during the 2009 Pandemic Influenza in Taiwan

Abstract

Background: Pandemics of influenza have always caused high disease burdens for emergency care and social panic. Accordingly, in order to minimize public health threat, scientists have been exploiting data collected during past pandemics to promote effective and evidence-based public health planning, resource allocation, and major decision making. Methods & Materials: In this respect, the traditional approach is to employ regression models to analyze the adjusted odds-ratio (aOR) to a particular risk factor. Nowadays, with the availability of high-volume clinical databases, it is of scientific significance and interest to employ machine learning algorithms to conduct comprehensive analyses on interactions among multiple risk factors and therefore identify specific groups that suffer extraordinary high risks. In this study, we aimed to identify those ILI patients who suffered extraordinary high risks of predisposing serious complications and fatality. The study was based on the ILI cases admitted to hospitals during the 2009 pandemic period (July 1, 2009 to June 30, 2010) in Taiwan. Results: Our first observation on the outputs of the decision tree algorithm was that age acted as the most important determinant for ILI-related prognosis but the risk factors for different age groups varied. We then proceeded to conduct in-depth analyses by generating decision trees with high specificity. The results showed that those middle-aged ILI patients with both diabetes (DM) and neurological diseases (N) [e.g. DM + N] suffered extraordinary high risks. On the other hand, those elderly ILI patients with renal (R) diseases accompanied with heart (H), cerebrovascular (C), or neurological (N) diseases, [RHC, RH, RC, RN, HN] required our special care. Conclusion: Our experiences with this study confirmed the power of machine learning algorithms. In the future, a real-time alert system can be developed to improve resource allocation and thus minimize fatal cases in epidemics/pandemics.