Visual Analytics for Causal Reasoning from Real-World Health Data

Abstract

The increasing capture and analysis of large-scale longitudinal health data offer opportunities to improve healthcare and advance medical understanding. However, a critical gap exists between (a) -- the observation of patterns and correlations, versus (b) -- the understanding of true causal mechanisms that drive outcomes. An accurate understanding of the underlying mechanisms that cause various changes in medical status is crucial for decision-makers across various healthcare domains and roles, yet inferring causality from real-world observational data is difficult for both methodological and practical challenges. This Grand Challenge advocates increased Visual Analytics (VA) research on this topic to empower people with the tool for sound causal reasoning from health data. We note this is complicated by the complex nature of medical data -- the volume, variety, sparsity, and temporality of health data streams make the use of causal inference algorithms difficult. Combined with challenges imposed by the realities of health-focused settings, including time constraints and traditional medical work practices, existing causal reasoning approaches are valuable but insufficient. We argue that advances in research can lead to new VA tools that augment human expertise with intuitive and robust causal inference capabilities, which can help realize a new paradigm of data-driven, causality-aware healthcare practices that improve human health outcomes.