epidemiological studies can be designed in various ways so as to collect new data
when designing a study two fundamental principles should be followed: the study should be comparative and should seek to avoid bias
associations between disease and exposures cannot be considered unless we compare exposed and non-exposed groups. For example, if we observed high numbers of diabetes cases amongst cinema goers could we assert a relationship?
two broad classes of study design:
observational studies: in which the investigators role is passive in that exposures are not manipulated. Examples include: studies of smoking and lung disease, coffee drinking and heart disease and processed meat consumption and disease risk.
intervention studies: in which the investigators role is active: groups are exposed to interventions of interest such as treatments. These are experimental studies: clinical trials.
intervention studies are the gold-standard in terms of causality confounding is handled by randomisation: forming comparable groups. Also, exposure is known to precede disease.
the major issue with experimental studies is one of ethics: it is rarely acceptable to force exposures on people
consider for example: how might we investigate the effect of traffic pollution on respiratory disease or of smoking on lung cancer?
observational studies predominate epidemiological research
the most commonly used designs in analytical epidemiology are the cohort and case control studies
cross-sectional studies: surveys also feature but since they provide information at a snap-shot in time (both exposure and disease) they are less useful and can only be used to measure disease prevalence
note that surveys are higher in the study design hierarchy than studies based upon routinely collected data since data collection can be planned and exposure and disease information relate to the individual
cohort and case-control studies and their methods of analysis are the primary focus in this module
study designs can be considered more broadly in terms of their hierarchy
the hierarchy is based upon strength of evidence for causality. Recall association does not infer causation
Unnumbered Figure: Link
the most useful type of observational study in terms of studying disease aetiology is the cohort study
definition
“A cohort study tracks two or more groups forward from exposure to outcome.” (Grimes & Schulz 2002)
groups are classified according to exposures (e.g proximity to major road network road, smoking, diet, alcohol consumption etc) and then followed through time in order to study the occurrence of disease (e.g cancer, heart disease, stroke etc) in the exposure groups
prospective cohort studies provide the strongest evidence for a causal relationship since groups are followed prospectively through time and thus exposure is known to precede disease
process-rationale: follow cohorts through time record disease occurrence compare exposure groups with respect to disease outcome interpret findings
(Woodward (1999), Example 1.2)
smoking and lung cancer: study by Doll and Hill
cohort: all British doctors registered in 1951 69% of men, 60% of women responded
expsoure ascertainment classify sub-groups according to their smoking habits
follow-up measure: death from cancer using automatic notification from death registrations
publication of 10-year follow-up results in 1964 (40-year follow-up in 1994)
advantages of cohort designs:
“best way to ascertain both the incidence and natural history of a disorder”
maintain temporal sequence
can examine multiple events after a single exposure; can study new or rare exposures; …
can measure risk. incidence rate and survival time
disadvantages:
large and costly
selection bias (observational study, prognostic factors)
not useful for rare diseases: take a long time to develop …
loss to follow-up
exposures can change during study period
confounding variables
further reading: Grimes DA, Schulz KF (2002) Lancet; 359:341–345.
definition
case and control groups are defined and selected according to their disease status: diseased/non-diseased
we then look backwards in time to ascertain each person’s exposure status (e.g exposed / non-exposed)
advantages:
efficient in terms of time and money (if frequency of exposure not too low)
can study rare diseases
disadvantages/issues:
choosing control group selection bias
obtaining exposure history recall bias
further reading: Grimes DA, Schulz KF (2002) Lancet; 359:431–434.
(Woodward (1999), Example 1.1)
smoking and lung cancer: study by Doll and Hill
cases: 709 lung cancer patients in 20 hospitals in London (1948/49)
controls: 709 non-cancer patients matched by gender, age group, hospital, and time in hospital
exposure: asked about smoking history
Question: possible source of bias?