Machine Learning for Personalized Medicine

Statistical Methods for real-time monitoring of health outcomes

by Peter Diggle

Electronically recorded health outcome data that accrue in real-time, both at community and individual levels, are becoming increasingly available, often at fine temporal and/or spatial resolution. The phrase "real-time" can mean different things (hours, days, months,...) in different contexts. I use the term loosely, to mean data that need to be analysed as they accrue in time, rather than at the end of a pre-specified time-period.

Statistical methods for analysing data of this kind necessarily depend both on the format and statistical properties of the data and on the objectives of their analysis, but will typically involve formulating and fitting stochastic models of temporal or spatio-temporal variation and making predictive inferences, parameter estimation or hypothesis testing being of lesser interest other than as means to an end.

In this talk I will describe several case-studies in this general area, covering monitoring progression towards end-stage renal failure, human and veterinary surveillance of gastro-enteric illness, and local-scale malaria prevalence mapping. 

>back to "Talks and Speakers"