Ongoing and recent research
Methodological work
(1) Bias in survival
analysis given patient heterogeneity in risk
(2) Bias in survival
analysis given informative censoring
Both of these themes were addressed
in my paper published in Statistics in Medicine Sept 2017 (McNamee R. How
serious is bias in effect estimation in randomised trials with survival data
given risk heterogeneity and informative censoring? Stat Med: 2017, 36;3315-3333).
The paper argues that heterogeneity
in risk is very plausible and should be the default assumption. Yet standard analysis
methods – such as those based on the semi-parametric proportional hazards
model - assume within group homogeneity and, in general, provide biased
estimates of treatment effects if this is not true. Fortunately – for past research findings - the
bias may be quite small, but it is not always. A formula is provided in the paper which enables
the size of the bias to be gauged.
The accelerated failure model with homogeneity
assumed is also shown to produce biased estimates. On the other hand, the accelerated
failure model with ‘frailty’ is unbiased, as is a simple G estimation
approach.
The problem is more severe when
there is informative censoring; the latter two methods are still biased in
these cases.
(3) A method for unbiased
estimation in survival analysis of morbidity data when there is informative
censoring by death.
A new method
of analysis is being investigated to address this special, but common case of
informative censoring.
Epidemiological work
Health effects of exposure
to ionising radiation in workers
I am a co-investigator in several
research teams –which are mainly led by the Centre for Occupational &
Environmental Health at University of Manchester but also at University of
Bristol – which are studying health effects of such exposures in workers around
the world.
|