TL;DR
This paper details the user-centered design process for the Australian Cancer Atlas (ACA), a novel visualization tool for cancer incidence and survival data. It focuses on effectively communicating geostatistical model uncertainty to diverse audiences, including the public, researchers, and policymakers, through innovative visual methods like wave plots, V-plots, and map transparency.
What Problem Does This Paper Address?
Cancer atlases are crucial for understanding geographical health disparities and informing public health interventions. However, presenting complex geostatistical model outputs, especially their inherent uncertainty, to a broad audience is challenging. Traditional methods often fail to visualize uncertainty or do so in ways that are difficult for non-experts to interpret. This can lead to misinterpretations, where similar point estimates with different levels of uncertainty might be perceived as equally reliable, potentially hindering effective decision-making. The ACA aims to bridge this gap by providing a user-friendly platform that not only displays cancer data but also clearly communicates the uncertainty associated with these estimates to various user groups with differing data literacy levels and priorities.
Methods
The ACA project employed a rigorous 21-month user-centered design process, guided by Munzner's nested model of visualization design. Key methodological components included:
- Preliminary Research: A scoping study and literature review to understand the problem space and identify audience needs.
- Audience Focus Groups: Four focus groups were conducted with representatives from the general public, health practitioners, policymakers, and cancer patients/caregivers to gather insights into their information-seeking behaviors, awareness of geographical variations in cancer burden, and interpretation of disease maps.
- Iterative Design and Development: This phase involved digital prototyping in parallel with Bayesian model development. Six key milestones (M1-M6) were integrated throughout the process, with regular feedback sought from workshops, audience focus groups, and an expert external advisory group.
- Novel Uncertainty Visualization Techniques: The paper highlights three primary methods developed to communicate uncertainty:
- Wave Plot: Visualizes the empirical densities of estimates (e.g., Standardized Incidence Ratio - SIR) and their uncertainty, evolving from density plots and violin plots to a final design that uses a linear scale for the logarithm of estimates to ensure consistent interpretation.
- V-Plot: A scatterplot that compares an area's estimate to the Australian average on the x-axis and the probability (confidence) of the estimate being significantly different from the average on the y-axis, designed for straightforward interpretation of directional differences.
- Map Transparency: A feature that subtly reduces the visibility of areas with higher uncertainty by fading their color towards transparency, allowing users to intuitively gauge the reliability of displayed estimates.
Datasets
The study utilized data on 20 cancers (including all combined) obtained from all 8 Australian state and territory cancer registries. The smallest spatial unit for which data could be obtained, in accordance with data privacy requirements, was the Statistical Area 2 (SA2). The ACA version 1 used the 2011 SA2 boundaries, comprising 2,196 areas.
Key Insights and Ideas
- Uncertainty is Paramount for All Audiences: Contrary to initial assumptions, all target groups, including the general public, recognized the importance of understanding uncertainty for credibility and accurate interpretation, though their priorities and desired level of detail varied.
- Tailoring Visualizations for Diverse Data Literacy: The ACA's success hinges on offering multiple, complementary ways to visualize uncertainty, catering to users with different levels of statistical understanding. The wave plot and V-plot offer more detailed insights for researchers, while map transparency provides a more immediate, intuitive cue for the general public.
- User-Centered Design is Non-Negotiable: The iterative process, with continuous feedback from diverse user groups and an advisory board, was critical in refining the design and ensuring the visualizations were effective and avoided misinterpretation.
- Balancing Detail and Overview: Presenting data for a large number of small geographical areas (SA2s) while maintaining a national perspective was a significant design challenge. The "Estimate Overview" bar plots were developed to address this, allowing for detailed comparisons without obscuring the national map.
- The "Wave Plot" Name Reflects Process and Shape: The name "wave plot" emerged from the visual appearance and the underlying process of generating estimates from posterior distributions, emphasizing the iterative nature of the modeling and visualization.
Key Results
- The Australian Cancer Atlas (ACA) has been highly successful since its launch in 2018, attracting over 62,000 individual users from over 100 countries across all target audiences.
- The ACA has been replicated in other countries (e.g., Netherlands, New Zealand), demonstrating its global applicability and impact.
- The ACA has influenced policy, with the Australian Government inviting the ACA team to provide input for a Senate inquiry on equitable access to diagnosis and treatment for rare cancers.
- The design iterations, particularly the development of the wave plot, V-plot, and map transparency, were positively received by users and contributed to the atlas's effectiveness in communicating uncertainty.
Limitations and Future Work
- Late Model Development: Designing visualizations in parallel with model development meant that modeled estimates were not finalized until near launch, leading to some late design decisions (e.g., initially hiding the wave plot by default to reduce confusion).
- Spatial Pattern Emphasis: Emphasizing spatial patterns of small, geographically concentrated areas without obscuring the national picture remains a design challenge. Future work may explore new cartogram designs and bespoke map morphing.
- Temporal Trends and Risk Factors: Future versions of the ACA are planned to include temporal trends and cancer risk factors, requiring further visualization and design considerations.
- Further Cartogram Exploration: The authors mention ongoing exploration of new cartogram designs for visualizing ACA data, suggesting this as a potential avenue for future development.
Summary updated