32 Chapter 3
2 Epidemiological studies rarely address pathogenesis – different ap-
proaches are needed to clarify the processes involved.
3 Sometimes an epidemiological approach gets it wrong. For example, a
number of observational surveys showed that women taking Hormone
Replacement Therapy (HRT) subsequently had fewer cancers, How-
ever, Randomised Controlled Trials then showed that HRT led tomore
cancers – it seems likely that the observational studies underestimated
(and therefore made insufficient allowance for) the extent to which the
women who chose to take HRT also tended to make healthier lifestyle
choices that reduced their chance of developing cancer. If so, this was an
example of observational results being misleading because of inadequate
adjustment for confounders.
4 In practice, epidemiological studies rarely include the evaluation of an
intervention.
Stages in an epidemiological study
1 Definethe population to be studied: catchment area; inclusion criteria;
everyone v. random sample v. stratified random sample. Examples in-
clude a random 25% sample of 3-year-olds living in a London borough,
and all 5–14-year-olds with new-onset severe head injuries in south-
east England.
2 Identifyindividuals who meet the criteria. For community samples,
identification is often via some sort of population register (for example,
for schools or immunisations). For rare disorders or risks, identification
is often via agencies that are particularly likely to be in contact with
relevant individuals, for example, doctors, special schools, voluntary
organisations. Use of multiple sources (known as ‘multiple ascertain-
ment’) is likely to identify more relevant individuals than use of any
one source. Even with multiple ascertainment, there is still the risk of
missing some affected individuals who have never been diagnosed or
sent to a special school. There is no straightforward way of estimating
the size of this problem.
3 Recruitidentified individuals. It is far easier to spot problems with
recruitment than problems with identification – problems with recruit-
ment show up as low participation rates. Ideally, all studies should com-
pare participants and non-participants on any available information.
4 Assesssubjects. There are two main possibilities: full assessment of all
subjects (known as a ‘one-phase’ procedure), or use of a two-phase
procedure:
Phase 1: use one or more screening tests (for example, parent,
teacher, or self-report questionnaires) to divide the sample up into
‘screen positive’ and ‘screen negative’ subjects.
Phase 2: fully assess a mixture of ‘screen positive’ and ‘screen nega-
tive’ subjects, sampling disproportionately more of the former (for
example, 100% of ‘screen positive’ and a randomly chosen 20% of
‘screen negative’ subjects). Inclusion of a random sample of ‘screen