Presentation (CTI)

OBJECTIVES: To provide a practical decision-tree framework to guide data scientists, biostatisticians and epidemiologists in selecting the appropriate type of data visualization for healthcare cost analyses depending on the context and objectives.
METHODS: A literature review was carried out to search for health economic studies reporting cost data. We examined reported methodologies and combined them with SNDS (French national health data system) database most common uses to produce a decision-tree framework. The decision tree is structured around binary questions and integrates key criteria: temporal analysis, number of cost categories and comparison groups, and level of granularity (cost distributions vs. global cost averages or totals). Additional criteria were used to refine recommendations, including cost hierarchy and target audience. All visualizations were created using a synthetic dataset designed to reflect real-world healthcare cost.
RESULTS: The decision tree guides readers through analytical scenarios starting with data temporality. For temporal analyses, line or bar charts are recommended to visualize trends in average costs, while stacked area charts and heatmaps display the evolving composition of multiple cost categories. Bubble plots are used when an additional numeric variable is involved. For static analyses, box plots and violin plots are appropriate for visualizing distribution across less than 10 groups, while ridgeline plots handle a larger number of groups comparisons. Waffle charts and tree maps represent global cost overviews effectively, with treemaps highlighting cost subtypes hierarchies. When comparing multiple groups (e.g., different cancer stages), lollipop, heatmaps, radar or waterfall charts are suggested depending on the number of groups and on the variability across cost categories.
CONCLUSIONS: Our work resulted in a decision-tree guide that helps users clarify their analytical questions before selecting the visualizations of healthcare costs that best align with their data and objectives.