Clinical Outcome Prediction: Decision support at the right time, using the right timescale

Dr Matthew Sperrin, Mathematics and Statistics, Lancaster University

Tuesday 18 December 2012, 1015-1040
Management School Building

A clinical prediction model (CPM) estimates probabilities of outcomes based on the observable characteristics of an individual, possibly under different hypothetical interventions. A clinician-patient partnership may use CPMs to support their decision-making; for example, for a patient with a blocked coronary artery, a decision is made between angioplasty and coronary artery bypass graft. CPMs can also be used at a service level to target healthcare resources effectively, by determining population strata that will receive different care plans in management of long-term conditions.

'Using the right timescale' in CPMs is critical: this can be expressed as the deceptively simple question: 'when is time zero?' I will describe some recent work that demonstrates just how carefully time zero needs to be chosen.

The advent of electronic patient records raises opportunities and challenges for clinical outcome prediction. I will describe the early stages of multi-disciplinary work to exploit this. Decision support tools can, in principle, update in real-time to reflect the latest evidence, use information from a very large bank of patient and environmental characteristics, and randomise patients to different treatments when there is genuine uncertainty about which would lead to the best outcomes, in order to generate evidence. Robust, principled statistical methodology is needed to make this a reality.