Health Care, Access, Insurance
Impact of Big Data from wearable technology on Health Insurance
(A. James Clark School of Engineering (UMD) Office of Advanced Engineering Education Master's Student)
Objectives and Public Health Significance: In today’s world almost every individual has a health insurance either sponsored by their employers or held independently. Health insurance premium costs are determined based on several factors and rising health insurance premiums have always been a major concern. Wearable devices from various vendors are becoming increasingly popular these days with more and more people opting to possess one for the associated health benefits. In addition to health benefits, the big data collected through these wearable devices can make a huge impact on the current health insurance industry, by helping them understand the customer behavior, priorities and redesign core operations to suit the current trends and advancements in the health market. It can help them shape health plans, determine group pricing, assess risk factors, detect fraudulent claims and most importantly modernize the way they engage with their customers. This will enable them to make more informed decisions when designing ,suggesting and offering health plans to new customers and to build long term relationships with their existing customers by promoting better health and lower costs. Approach and findings: In this study, we have analyzed data from a wearable device from a set of users for a period of one year .A supporting brief survey was conducted to study the preferred health plans , awareness about wearable devices , health benefits and health insurance premium costs. Dataset extracted consisted of various health parameters such as Steps, Calories burned, Distance, Minutes Sedentary Minutes Lightly active, Minutes Fairly active, Minutes Very active, Activity calories, Weight, Fat and BMI.A supervised learning approach was applied to derive certain useful measures which can help determine ,if the individual was meeting the minimum standard requirement of physical activity , maintaining the energy balance needed in terms of the calories burned, the risk of weight regain based on the total number of steps that the individual covers on a daily basis, the risk of the individual developing metabolic disorders measured in terms of BMI, the risk of the individual developing heart conditions based on his physical activity. The next phase was to build a predictive model that can best help predict the probability of the outcomes. Predictive modelling involves creation, testing, evaluation and deployment .Dataset has been cleaned and prepared for building predictive models. The statistics of the dataset were studied before the dataset was divided in to training data and test data (33%).Predictors and the attributes to be fit in to the model were determined based on cross tabulations, co-relation and variance-covariance matrices generated. Three modelling techniques have been tried on the datasets namely Decision tree, Logistic regression and SVM modelling. SVM approach gave us the best results in terms of generalization performance. The model was evaluated by generating a confusion matrix, classification report, ROC and AUC metrics. Other approaches such as Principal component analysis have been applied to improve the performance of the model. This model can be effectively used and this study can further be extended to the next level by tracking other health parameters such as heart rate, sleep states, rates in the future.