Quantitative methods for the assessment of systematic error in observational
studies: improving causal research.
Supervisors: Bianca De Stavola and Simon Cousens
Results from epidemiological studies often appear contradictory, leading
to some cynicism regarding medical research among the general public.
Many of these contradictions are attributable to the inadequate reporting
of the uncertainties that affect this type of research.
Uncertainty arising from random sampling variation is typically the
only uncertainty presented (in the form of confidence intervals). There
are, however, other sources of uncertainty which may dwarf the uncertainty
due to sampling variation. The extent of such uncertainties will vary
from study to study but in general they arise from:
(i) unmeasured factors in the study population that are related to
both outcome and exposure of interest ("unmeasured confounding");
(ii) non-random inclusion of subjects ("selection bias");
(iii) inaccuracies in measurements of outcome(s) and exposure(s) ("measurement
error")
(iv) absence of outcome/exposure data for some subjects ("missing
data").
The failure to consider properly these other sources of uncertainty
has two negative consequences. First, results of studies are presented
in a way which implies more certainty about them than is justified,
increasing the likelihood of subsequent "contradictory" findings.
Second, considering only uncertainty due to random sampling variation
leads to the perception that increasing the study size will be sufficient
to deal with uncertainty. Indeed data from different studies are now
often merged ("pooled") with the stated aim of obtaining "more
precise" estimates.
To improve understanding of research results and retain public trust,
we must provide more realistic assessments of the uncertainties involved.
Recently, new approaches to this problem have been proposed, but these
approaches are not immediately accessible to practitioners, as they
face difficulties that are both operational (e.g. how to specify the
likely magnitude of selection bias) and methodological (e.g. how to
combine this information with the observed data). The proposed PhD will
investigate three alternative approaches to improved quantification
of uncertainty; classical sensitivity analysis; Monte Carlo sensitivity
analysis and Bayesian bias analysis. The feasibility of these approaches
will be investigated in different contexts. The PhD should result in
practical guidelines on methods which provide robust outputs while being
transparent and accessible to a wide range of practitioners.
|