A widely used method of monitoring patients kidney function is the
estimated glomerular filtration rate (eGFR), measured over time. The
eGFR tends to decrease naturally with increasing age, and to vary
unpredictably for other reasons, for example in response to changes in
a person's general level of fitness. Nevertheless, an unusually sharp
decrease is considered to be a useful predictor of kidney failure;
current guidelines suggest that a rate of change of 5% per year or more
is indicative of a need for specialist treatment. Measurements of eGFR
can be obtained from routine blood tests taken in primary care
settings. In this talk, we shall address the problem of using such
routinely collected data for early detection of incipient kidney
failure. Our data are obtained from two UK treatment centres. They
consist of an irregularly spaced time series of eGFR measurements on
each person together with a number of explanatory variables including
each person's age, any relevant co-morbidity and medication. We shall
formulate and fit a dynamic regression model for the data, in which the
rate of change of eGFR over time is modelled as a continuous-time
stochastic process B(t). |