A short introduction to epidemiology Chapter 4: More complex study designs

1
A short introduction to epidemiologyChapter 4
More complex study designs

• Neil Pearce
• Centre for Public Health Research
• Massey University
• Wellington, New Zealand

2
Birth
End of Follow up
Death other death lost to follow up
non-diseased symptoms severe disease
3
Study Design Options

• All epidemiological studies are (or should be)
  based on a particular population (the source
  population) followed over a particular period of
  time (the risk period)
• The different study design options differ only in
  how the source population is defined and how
  information is drawn from this population and
  time period

4
Incidence and Prevalence

• Incidence is the number of new cases of the
  condition over a specified period of time
• Prevalence is the number of cases of the
  condition at a particular point in time

5
Study Design Options
6
Chapter 4More complex study designs

• Other axes of classification
• Continuous outcome measures
• Ecologic and multi-level studies

7
Other axes of classification

• Continuous exposure data
• Timing of collection of exposure data
  (retrospective, prospective)
• Sources of exposure information (interviews,
  routine records, biomarkers)
• Level of measurement of exposure (individual,
  population)

8
Chapter 4More complex study designs

• Other axes of classification
• Continuous outcome measures
• Ecologic and multi-level studies

9
Continuous Outcome Measures

• Lung function in a cross-sectional study (a
  prevalence study is a cross-sectional study with
  a dichotomous outcome measure)
• Changes in lung function over time in a
  longitudinal study (an incidence study is a
  longitudinal study with a dichotomous outcome
  measure)

10
Continuous Outcome Measures

• Tager et al (1983), longitudinal study of
  pulmonary function in children aged 5-9 years,
  followed for 7 years
• Exposures maternal smoking
• Outcomes annual increase in FEV1 (this was 28mL
  lower in children exposed to maternal smoking)

11
Continuous Outcome Measures

• Roemer et al (1993), time series study of winter
  air pollution and respiratory health of children
  aged 6-12 years
• Exposures daily air pollution measures
• Outcomes asthma symptoms, medication use (e.g.
  wheeze was more common on days when particulate
  concentrations were high

12
Cross-Sectional Studies

• Particularly valuable for
• Non-fatal diseases
• Degenerative diseases with no clear point of
  onset (e.G. Chronic bronchitis)
• Examining effects on physiologic variables (e.G.
  Liver enzyme levels, blood pressure, lung
  function)

13
Cross-Sectional Studies Examples

• General household surveys (e.g. England and
  Wales, Spain, New Zealand)
• National Health and Nutrition Examination Survey
  (USA)
• International surveys (e.g. European Community
  Respiratory Health Survey (ECRHS), International
  Study of Asthma and Allergies in Childhood
  (ISAAC)
• Pre-employment surveys
• Studies in specific populations (e.g.
  occupational health research)

14
Cross-Sectional studies

• Disease is measured at one point in time
• Exposure may be measured at the same time and/or
  historical exposure information may be available
• May be difficult to know the temporal
  relationship between exposure and disease
• This problem is avoided in repeated
  cross-sectional studies

15
Study Design Options

• Case series
• Incidence studies
• Incidence case-control studies
• Prevalence studies
• Prevalence case-control studies
• Cross-sectional studies (with continuous outcome
  measure)
• Longitudinal studies (with continuous outcome
  measure)

16
Chapter 4More complex study designs

• Other axes of classification
• Continuous outcome measures
• Ecologic and multi-level studies

17
Ecologic Studies

• An ecologic study is a study in which one or
  more exposures (or confounders) is measured at
  the population level rather than the individual
  level

18
Reasons for Ecologic Studies

• Low cost and convenience
• Measurement limitations of individual-level
  studies (e.g. diet, air pollution)
• Design limitations of individual-level studies
  (e.g. lack of exposure variation)
• Interest in ecologic effects
• Simplicity of analysis and presentation

19
Levels of Measurement

• Individual measures, e.g. smoking status, income
• Aggregate measures, e.g. smokers, median family
  income
• Environmental measures, e.g. air pollution levels
• Global measures, e.g. smoking legislation, income
  inequality, GNP, type of health care system,
  population density

20
Levels of Analysis

• Individual level, e.g. average level of air
  pollution is assigned to each individual, and
  individual age, gender, ethnicity, smoking status
  are known
• Partially ecologic analysis e.g. some variables
  known for individuals (age, gender, air
  pollution) but others for the population
  (smokers)
• Fully ecologic analysis all information on
  exposure and disease only known for the population

21
Levels of Analysis

• Multilevel analysis
• First level individual level analysis within
  each group (population)
• Second level regression parameters from first
  level are modelled as a function of ecologic
  variables
• e.g. Humphreys and Carr-Hill (1991) used
  multilevel modeling to estimate the contextual
  effect of living in a poor area, controlling for
  individual income and other risk factors

22
Levels of Inference

• Individual (e.g. fat intake and breast cancer)
• Contextual (e.g. living in a poor neighbourhood)
• Ecologic (e.g. GNP, income inequality)
• The major problem is with cross-level inferences,
  e.g. using ecologic data to estimate the
  individual risk from fat intake

23
12 Month Period Prevalence of Asthma Symptoms in
13-14 Yr Old Children
24
The Ecologic Fallacy in ISAACIndoor Humidity
and Asthma
25
The Ecologic Fallacy in ISAACIndoor Humidity
and Asthma
26
Example of ecologic data
27
Example of ecologic data
28
Example With No Confounding by Group
29
Example With Confounding by Group
30
Example With Effect Modification by Group
31
Problems of Ecologic Studies

• Ecologic bias, in estimating effects at the
  individual level may result from
• Within group bias if there is confounding,
  selection bias or misclassification within each
  group then the ecologic estimate may also be
  biased
• Confounding by group the background disease
  rate varies across groups
• Effect modification by group the excess rate
  due to exposure varies across groups

32
Problems of Ecologic Studies

• The major problems of ecologic bias arise from
  attempts at cross-level inference, e.g. in
  studies where the intention is to make inferences
  at the individual level
• Nevertheless, ecologic studies have played a
  major role in the development, and to some extent
  in the testing, of epidemiological hypotheses
• Furthermore, some important risk factors can only
  be studied at the population level.

33
A short introduction to epidemiologyChapter 4
More complex study designs

• Neil Pearce
• Centre for Public Health Research
• Massey University
• Wellington, New Zealand