Abstract

The quest to harness “big data” for better healthcare is driving new computational approaches to both discovery science and actionable analytics. Precision medicine for example, seeks actionable biomarkers of disease mechanisms. This approach is useful if the mechanisms uncovered explain a large proportion of the variation in clinical outcomes. Commonly, however, the real-world variation in outcomes involves many mechanisms, biomarkers and phenomarkers – some known, some yet to be known, and some unlikely to be known. Informaticians therefore seek to build computational approaches from the ‘middle out’, linking ‘bottom up’ computational biology with ‘top down’ (clinical) epidemiology. Professor Buchan will give examples of endotype discoveries made with biostatistical and machine learning approaches designed to extract more detailed phenomarker information from longitudinal (birth cohort) data. He will then take a similar approach to clinical and patient-reported data, putting the case that healthcare evidence needs to be generated in ways that more easily shrink onto local contexts. Looking toward a future of more predictive analytics around personal health information, Professor Buchan will consider the point at which an individual’s health ‘avatar’ might say “no” to a healthcare provider’s care pathway – a collision of equipoise – and how health systems might prepare for this. He will argue that current ‘low resolution’ healthcare maximises information utility for episodes of care, and that future data-intensive health systems could offer ‘high resolution’ approaches that focus on patient trajectories, supporting: more timely, preventive and tailored interventions; and continuous experimentation/adaptivity/learning.