P-4: Using Machine Learning Algorithms for Population Health Decision Support: A Case Example for Predicting Falls in the Elderly Population
Topic: Health Policy, Population and Public Health, eHealth, Geriatrics
McMaster University DeGroote School of Business, eHealth
According to the Public Health Agency of Canada, falls account for 95% of all hip fractures in Canada; 20% of these cases end in death. From as far back as 2007, the World Health Organization (WHO) has been able to consolidate evidence on risk factors of falls-related injuries and provide the Falls Risk Factor Model. However, the rate of falls-related injuries has not declined more than 10 years on. There are several risk factors of falls and related injuries. The purpose of our study is to reduce the number of statistically significant predictive factors used to predict a health problem, such as falls-related injuries. Our research analyzed the Canadian Community Health Survey Dataset (CCHS) 2014 and built a predictive model from WHO’s Falls Risk Factor Model. We provide the dimension reduction techniques that can be used to eliminate statistically significant risk factors of falls-related injuries to only a few distinct variables. This can have wide-ranging impact in publicly funded healthcare systems, for example by allowing population health decision makers to allocate tax dollars to only a few, priority risk factors rather than a wide variety of them. The highest accuracy reached from our model based on the Falls Risk Factor Model is 74% using regression and random forest, while the standard reference used reached the highest accuracy of 72% using random forest and support vector machine.