CONCEPTS UNDERLYING STUDY DESIGN

1
CONCEPTS UNDERLYING STUDY DESIGN

• Nigel Paneth

2
(No Transcript)
3
DESCRIPTIVE VS ANALYTICAL STUDIES

• Descriptive studies often focus on a single
  variable such as a particular disease or health
  state. They frequently have no etiologic
  hypothesis, restricting their interest to disease
  frequency. A common question answered by a
  descriptive study in epidemiology might be How
  common is this disease in this place?

4
THE MAIN DESCRIPTIVE STUDY IN EPIDEMIOLOGYTHE
COMMUNITY SURVEY

• A community survey is a study which attempts to
  ascertain the frequency of a disease in a fixed
  geographic region (ideally) or in a group defined
  by a common membership, e.g. school-children.
  It usually assesses the frequency of disease in
  easily ascertained sub-groups, i.e. age, gender,
  geographic sub-units, ethnicity. etc. It is
  often the first step in looking at risk factors
  for disease.

5
ANALYTIC STUDIES

• Unlike descriptive studies, analytic studies
  in epidemiolgy must have a clear hypothesis
  (usually an etiologic hypothesis). Thus at least
  two variables must be specified in advance, an
  exposure and a disease or health state. A
  hypothesis might be What is the association
  between this exposure and this disease? But
  quantitative prediction is the key to good
  hypothesis formulation. So ideally we specify
  the size of the association for example I
  predict that this exposure will be have a RR of
  at least 2.0 with the incidence of this disease.

6
OBSERVATIONAL, EXPERIMENTAL, AND
QUASI-EXPERIMENTAL STUDIES
7
OBSERVATIONAL STUDY

• In an observational study, the exposure is not
  assigned by the scientist doing the study. For
  example, a study in which cigarette smokers are
  compared to non-smokers for lung cancer
  incidence.

8
EXPERIMENT

• In epidemiologic research
  (not necessarily in all science),
  an experiment is a study in which
  an exposure is assigned randomly

9
QUASI-EXPERIMENT

• In epidemiologic research, a quasi-experiment
  is a study in which an intervention is assigned
  by the investigator, but not randomly. For
  example, a new educational program is started in
  a classroom, and the results are compared to
  those found in a traditional classroom, composed
  of similar students.

10
ECOLOGICAL VS INDIVIDUAL STUDIES
11
ECOLOGICAL STUDIES

• In experimental studies, a study is defined as
  ecological, when the assignment is to a group
  e.g. assigning villages, not individuals, to the
  different arms of the study.
• In observational studies, the study is ecological
  when the unit of outcome analysis is a group
  e.g. the distribution of Lyme disease by county
  is compared to the distribution of tick species
  by county.

12
INDIVIDUAL STUDIES

• In individual studies, the unit of outcome in
  the analysis is an individual. An individual
  study, can however, have an ecological exposure
  (which is why observational ecological studies
  are defined as ecological by outcome. For
  example, an individual study of lung cancer may
  use an individual exposure (smoking by
  participant) or an ecological exposure (urban or
  rural residence)

13
THE ECOLOGICAL FALLACY

• 1. Statistical meaning The tendency of
  correlation coefficients to be larger when an
  association is assessed at the group level than
  when it is assessed at the individual level
•  
• 2. Common usage The tendency of ecological
  studies to imprecisely reflect reality as
  documented in individual studies.

14
EXAMPLE OF ECOLOGICAL FALLACY

• Countries with high salt intakes have higher
  levels of hypertension. But it has been hard to
  show that salt intake and hypertension are
  related in individuals. It may be that
  within countries, individuals are above or below
  some threshold of the relationship of salt
  intake to hypertension.

15
ATOMISTIC FALLACY

• This fallacy is the opposite of the ecological
  fallacy, and refers to the assumption that
  associations found at the individual level will
  necessarily be replicated at the group level.
•  For example, the major determinant of infant
  mortality is the birth weight of the baby. But
  it is not necessarily true that the differences
  between countries in infant mortality are due to
  differences in their birth weight distributions.
  (although in fact they often are)

16
KRAMER AND BOIVIN'S THREE DIMENSIONS OF RESEARCH
DESIGN

• 1. DIRECTIONALITY-The direction in which exposure
  and outcome are investigated from exposure to
  outcome, from outcome to exposure, or both
  simultaneously
•  2. SAMPLE SELECTION-Criteria used to choose
  study subjects based on exposure, outcome or
  other criteria.
•  3. TIMING-Relation between the time of the study
  and the calendar times of exposure and outcome
  historical, concurrent, mixed.

17

• DIRECTIONALITY is the key determinant of whether
  a study is cohort or case-control.
• A COHORT STUDY IS ONE IN WHICH SUBJECTS ARE
  INVESTIGATED FORWARD FROM EXPOSURE TO OUTCOME.
• A CASE-CONTROL STUDY IS ONE IN WHICH SUBJECTS ARE
  INVESTIGATED BACKWARDS FROM OUTCOME TO
  EXPOSURE.

18
SAMPLE SELECTION

• The sampling procedure is not the determinant of
  whether a study design is cohort or case-control,
  since one is not required to sample by exposure
  in cohort studies, or sample by outcome in
  case-control studies. Thus the way the sample is
  chosen does not determine the type of study.

19
SAMPLING AND STUDY DESIGN

• Sampling follows directionality in case-control
  studies, in that sampling of the outcome is the
  usual starting point. In cohort studies, however,
  the commonest strategy is sampling a population
  (e.g. Framingham, etc.) not sampling by exposure,
  although that is sometimes seen (exposure-control
  study).

20
TIMING

• Timing also does not determine whether a study is
  cohort or case-control.
• Cohort studies can be retrospective (historical)
  or prospective in relation to calendar time.
  Moreover, even in a prospective cohort study,
  baseline exposure information that is historical
  is commonly collected.

21

• Case-control studies are usually viewed as
  retrospective, although Kramer and Boivin view
  them as mixed, because the outcome status can be
  ascertained in the present (but doesnt have to
  be)
• Cross-sectional studies are almost always done in
  the present, but could be done in the past (i.e.
  through reviewing old records)

22
CONCEPT OF STUDY BASE

• The study base is the idealized source
  population from which the actual study population
  is selected. It is described in terms of the
  person-year experience of that source population
  in relation to exposure and disease. In some
  study designs, the study base is obvious, in
  others, it is not so clear. It is always useful
  to try to imagine the study base of each study.
  Deviations of the actual study population
  composition from the study base can help explain
  problems with the study findings.
•  We will return to this concept in more detail
  when we study case control studies.