Study Design I

1
Study Design I
2
Sequence in Investigative Strategies
Sources of suspecting the risk factor(s) Clinical
hunches Biologic theory Animal models Impressions
from trends Statistical association
Sequential strategies in etiologic studies
Case Reports/Case Series
Ecologic studies
Observational
Case-control studies
Attempts at establishing causality
Additional evidence Animal studies Lab
studies Other studies of underlying mechanism
Cohort studies
Risk factor experiments
Experimental
Attempts at establishing causality
or
Intervention trials
3
Weighing the Evidence
4
Hierarchy of Study Design

• When an area of inquiry is new, focus is often on
  descriptive and etiological studies
• As knowledge base grows, so does the complexity
  and the specificity of the research study designs
• With more complex study designs, causal evidence
  grows

5
Non-experimental Studies

• Ecologic Studies, Case Reports/Series
• Cross-sectional Studies

6
Ecologic Studies

• Unit of analyses is a group rather than an
  individual
• Group can be
• Organization
• Community
• State
• Country
• Exposure and/or Outcome are measured only in the
  aggregate

7
Ecologic Studies (cont.)

• ecologic variables are properties of groups,
  organizations or places
• Morgenstern, 1998
• Three types of ecologic measures
• Aggregate
• Environmental
• Global

8
Ecologic Measures Aggregate

• Summary measures
• Means
• Example Average fish consumption
• Medians
• Example Median number of children/family
• Proportions
• Example Prevalence of current smoking

9
Ecologic Measures Environmental

• Physical characteristics of place
• Example Air pollution level
• Have corresponding measures at the individual
  level which are often unknown/not measured

10
Ecologic Measures Global

• Attributes of group that have no corresponding
  individual measure
• Example
• Health insurance system of a country
• Smoking legislation
• Population density

11
Ecologic Studies

• Unit of analysis is the group not the individual
• Often cannot ascertain overlap of exposures
• i.e. you may know the average smoking rates in a
  population and the average alcoholic beverage
  consumption in a population, but how many people
  both smoke and drink?

12
Ecologic Studies Benefits

• Average measures of disease and exposure (or
  proxies) are usually readily available
• Relatively inexpensive
• Often useful in social level research
• Simple to analyze and present

13
Ecologic Studies Limitations

• Often results are difficult to interpret
• Lack of confounding information
• Ecological fallacy

14
Ecological Fallacy

• Occurs when group comparisons are applied to
  individuals
• Example Per capita fish consumption and heart
  disease-Japan versus US
• Average per capita Fish Consumption per year
• Japan71 kg
• United States3.65 kg

15
Japan vs. US-CHD Mortality
Adapted from Sekikawa et al. Heart.
200389255257
16
So What is the Public Health Message?

• Should the US eat more fish?
• Lead and mercury!
• Will the rates of heart disease actually
  decrease?
• Will there be economic repercussions?
• Is there something other than eating fish that
  puts the Japanese at lower risk for heart
  disease?
• Genetics
• Lifestyle factors
• Social Forces

17
The Ecological Fallacy.
18
Case Report/Series

• Describe the experience of a single patient or a
  group of patients with a similar diagnosis
• Case ReportIndividual
• Case SeriesSeveral Individuals
• Often based on clinical hunches
• 1/3 of all published articles

19
Case Report/Series (cont.)

• May aid in the identification of an epidemic/new
  disease
• Useful to generate a hypothesis for further
  investigation

20
Example Case Report/Series

• MedWatch System
• Reporting done by
• Healthcare practitioners (preferred)
• Patients
• Recent example Gatifloxacin and hypoglycemic
  events

21
(No Transcript)
22
Case Report/Series Benefits

• Useful for disease surveillance
• Useful to generate a hypothesis for further
  investigation
• i.e. clinical hunches
• Relatively inexpensive

23
Case Report/Series Limitations

• Only one/several patients
• No comparison (control group)
• Reports tend to trickle off after drug has been
  on the market for several years
• Potentially missing long-term side-effects
• Can be influenced by external factors
• Media
• Legal system

24
Cross-Sectional Study (Survey)

• Snap-shot of disease experience
• Ascertain exposures and/or outcomes at one point
  in time
• Typically only prevalent cases
• Incidence is often unknown or not reliable
• Population for survey is well-defined

25
Cross-Sectional Study Benefits

• Snap-shot of the health/demographics/resources
  of a population
• Hypothesis formation
• Multiple exposures and/or outcomes
• Relatively inexpensive compared to some other
  study designs
• Loss to follow-up not typically problematic

26
Cross-Sectional Study Limitations

• Temporal sequence can be clouded
• Beneficial for determinants that do not change
• Race/ethnicity
• Gender
• Hair and/or eye color
• Genetic Markers
• Subject to determinants of survival
• Relatively healthier population
• Prevalent not incident cases of disease

27
Cross-Sectional Study Analysis

• Typically analyzed as a case-control study (more
  later)
• Temporal sequence should always be considered in
  the interpretation of results
• Association rather than causal relationship

28
Cross-Sectional Study Example

• MEPS
• Medical Expenditure Panel Survey
• National Probability Sample
• Started in 1996 by AHRQ
• Longitudinal cross-sectional components
• Longitudinal
• Health care coverage
• Cross-sectional
• Diabetes Care Supplement (DCS)

29
MEPS Diabetes Care Survey (DCS)

• Part of MEPS 2001 survey (panel 5, round 3)
• Administered to sub-sample of patients previously
  identified as diabetic
• 1329 diabetic patients
• Response rate 86.4
• Self-administered Questionnaire
• Can be supplemented with other collected
  information including
• Age, gender, race/ethnicity
• SEP factors (education, income, etc.)
• Other health conditions and medications
• Insurance coverage

30
MEPS DCS Class Discussion

• What are the benefits/limitations of using the
  DCS to answer the following questions
  cross-sectionally?
• Are Type 1 vs. 2 diabetics more likely to get an
  eye exam annually?
• Are people who take oral medication (compared to
  insulin) more likely to have their A1C
  controlled?
• Do rates of A1C testing differ by race/ethnicity?