Title: Advanced Methods, Stubborn Biases: Two Case Studies on Causal Inference in Routinely Collected Medical Records
Abstract: Electronic health records offer unprecedented opportunities to answer clinical questions that randomised trials cannot, or have not, addressed, but their validity hinges on whether we can adjust for the reasons patients received one treatment over another. In this two-part talk, Dr Shishir Rao will present recent work examining the promise and limits of advanced causal methods, including deep learning and target trial emulation, when applied to routine clinical data. The first case study revisits reported harms of sodium-bicarbonate paracetamol in older adults, asking whether a widely publicised safety signal holds up under closer scrutiny, and what a careful look at indications can reveal about how such signals arise in the first place. The second turns to target trial emulation in patients with heart failure with reduced ejection fraction, comparing a range of adjustment methods against randomised benchmarks for beta-blockers and digoxin, alongside semi-synthetic experiments that examine their behaviour under controlled conditions. Together, the case studies raise questions about when routinely collected medical data can, and cannot, substitute for trials, and what it will take to close the gap.
_______________________________________________________________________________
Microsoft Teams meeting
Join: https://teams.microsoft.com/meet/361609817210153?p=bwaOamTINhqUB6dDrp
Meeting ID: 361 609 817 210 153
Passcode: cy3VZ6PH