OBJECTIVES: Health insurance claims is a feature rich data that has been studied to analyse health outcomes. Machine Learning has been used extensively to segment members and identify the causes of insurance claim amounts. Topological Data Analysis (TDA) is a method robust to noise that helps extract meaningful insights from the shape of data.

METHODS: Mapper is a TDA based visualisation algorithm that captures the shape of the data by first creating covers for the data, then clustering it and finally forming a graph by connecting overlapping clusters. This helps uncover connections that are often missed by classical algorithms or manual grouping and aggregations. Mapper is also more robust to noise and can track features over multiple scales. This is used to visually investigate the member features and their relationship to insurance claims. Overlapping clusters also give us information about features that may lead to transition to and from various clusters. Thus, mapper gives a way to create interpretable models.

RESULTS: The results show four specific kinds of member profiles based on claims. Members with higher claim amounts tend to have more children, poor smoking habits and high BMI. It was also observed that a very small number had lower claim amounts. Such members had a significantly better BMI, no smoking habit, and fewer children. The members with high insurance claim amounts in the lower age group were seen to be smokers. It was also observed that two disconnected components were formed which had similar claims groups, but clustered by region.

CONCLUSIONS: This proof of concept was done on seven member features. However, this methodology may be scaled up with minimal computational requirements, and can efficiently segment members based on a vast number of features and instances.