Study Design II

1
Study Design II
2
Weighing the Evidence
3
Non-experimental Studies

• Because of lack of randomization, blinding and
  placebo, prone to influence of extraneous
  factors
• Types
• Ecologic study/Case Reports/Case Series
• Cross-sectional study
• Case-control study
• Cohort study (follow-up study)

4
Case Control Study Overview
Exposed
Disease (cases)
Unexposed
Exposed
Disease- (controls)
Unexposed
5
Case Control Study

• Goal
• Compare the odds of exposure among the diseased
  (cases) with the odds of exposure among the
  non-diseased (controls)

6
CC Study Advantages

• More efficient for rare diseases
• More efficient for diseases with long latency
  periods
• Can study multiple exposures of the one outcome
• Decreased time and cost

7
CC Study Limitations

• Inefficient for rare exposures
• Temporal relationship can be cloudy
• Susceptibility to bias
• Usually difficult to estimate rates

8
CC Study Selection of Cases

• Definition of a case
• Idiopathic (of unknown origin)
• Homogenous diagnostic criteria (case
  identification)
• Diagnosis must be unrelated to exposure history
• Case selection
• Incident vs. prevalent cases
• Not necessary that cases be representative
• Sources of case identification
• Registries
• Medical Records/Insurance data/Admission logs

9
CC Study Control Selection

• Definition of a control
• Persons that would have been cases in the study
  had they developed the outcome of interest
• Comparability in selection forces
• Control selection
• Hospital based
• Population based
• Friend/Relative/Neighbor

10
CC Study Odds Ratio (OR)
Disease
( Cases) (- controls)
Exposure
(-) ()
OR (a d) / (b c)
11
OR RR?

• OR is typically an unbiased estimator of the RR
• OR IRR if incident cases
• OR RR if CIexp CIunexp are small (lt10)

12
CC Study Example

• OLeary et al. 2004
• Long Island, NY population of women
• Pesticide exposure and risk of breast cancer
• Incident cases of breast cancer from 1980-1992
• Identified from tumor registries
• Controls were randomly selected from general
  population (similar age, same race)
• Hazardous Waste Site estimated using mapping
  techniques

13
Case Control Study Example
Exposed (1 mile of Haz Waste Site) (n12)
Breast Cancer 1980-1992 (n105)
Exposed (1 mile of Haz Waste Site) (n11)
Breast Cancer - (n210)
Adapted from OLeary et al. Env Res. 2004 94
134-144
14
Case Control Study Example (cont.)
Br CaBreast Cancer HWHazardous Waste Site
15
Case Control Study Example (cont.)
OR (a d) / (b c) Therefore, women with
breast cancer were XX times XXXX likely to be
living within one mile of a hazardous waste site
than were women without breast cancer
16
Cohort Study Overview
Not Eligible
Outcome
Population
Exposed
No Outcome
Eligible
Outcome
Unexposed
Apply Inclusion and Exclusion Criteria
No Outcome
17
Cohort Study

• Goal
• Compare occurrence of the specified outcome
  across categories of the exposure

18
Cohort Study Types

• Prospective
• Exposure is determined at study onset and
  participants are followed forward in time for
  outcome assessment
• Retrospective
• Exposure and outcome have already occurred at
  study onset

19
Cohort Study Advantages

• Ability to assess multiple outcomes of one
  exposure
• Ability to directly estimate rates
• Temporal sequence is clear

20
Cohort Study Limitations

• Inefficient for rare outcomes
• Often very costly
• If retrospective
• Good records are necessary
• Loss to follow-up
• Especially problematic with prospective but also
  possible with retrospective

21
Cohort Study Example
Exposed (alcohol )
CHF 151 cases/50,944 p.yrs (2.96/1000 p.yrs)
Framingham Heart Study 10,333
6,289
Unexposed (alcohol -)
CHF 68 cases/10,654 p.yrs (6.38/1000 p.yrs)
Population
Sample
Adapted from Walsh et al. Ann Int Med. 2002
136 181-191
22
Experimental Studies

• Reduce external variation through 3 mechanisms
• Randomization
• Placebo (or more generally a comparator)
• Blinding

23
Randomized Clinical Trial Overview

• Select a sample from the population
• Measure baseline variables, randomize
• Apply the intervention in the experimental group
• Measure the outcome
• Analyze the results

24
The Randomized Clinical Trial (RCT)

• Select a sample from the population
• The investigator selects a target pop.
  appropriate for the research question
• Usually this group is composed of persons with a
  certain set of characteristics (i.e. age, sex,
  disease)
• Example Patients ages 35-65 hospitalized in the
  previous week w/ unilateral lung pathology
• Determine an adequate sample size and plan the
  recruitment accordingly

25
RCT (cont.)

• Measure baseline variables
• Characterize the study population
• Demographics
• Clinical Characteristics
• Baseline Compliance???
• Randomize study participants

26
Randomization

• Purpose
• to prevent characteristics of the participants
  that are present at the beginning of the study
  from influencing experimental and control
  differences at the end of the study
• Should result in an approximately even
  distribution of confounding characteristics in
  experimental and control groups

27
Randomization Benefits

• Balance of known and unknown confounders
• Minimize investigator bias in allocation of study
  subjects
• Valid statistical tests

28
Example
Adapted from Martin DF, et al. A controlled
trial of valganciclovir as induction therapy for
CMV. NEJM 2002 346 1119-1126.
29
RCT (cont.)

• Apply the intervention in the experimental group
• Use an intervention that can be blindly applied,
  is specific, and is relevant to medical and
  public health practice

30
Blinding

• Purpose
• To prevent the unintended or co-interventions
  from influencing the outcome of the study
• Types of Blinding
• Single Blinding
• Participants are unaware of whether they are in
  the experimental or control group
• Double Blinding
• Both participants and observer are unaware of
  which group the participant is in
• Triple Blinding
• Participants, observer and data analyst are
  unaware of group assignments

31
Placebo

• Useful when no standard of care
• With standard of carecomparator
• Inert treatment intended to have no effect except
  other then the psychological benefit of offering
  treatment
• Facilitates blinding
• May be unethical

32
Compliance/Adherence/Persistence/ConcordanceXXXX

• Consider problems with XXXX in the design of the
  intervention
• Measure XXXX
• Pill counts, biological tests, etc.
• Use strategies to insure completeness of follow-up

33
RCT (cont.)

• Measuring the outcome
• Outcomes measured can be surrogates for actual
  phenomena of interest
• Statistical considerations
• Measure different aspects of the outcome of
  interest
• Measure potential adverse effects
• If possible, measure outcomes blindly
• Document loss to follow-up and attempt to obtain
  more information on these participants

34
RCT (cont.)

• Analyze the results
• Compare the two groups, i.e. check randomization
  procedure
• Determine whether the intervention was effective
• T-tests (means), chi-square (proportions) and
  multivariate analyses
• Sequential analysis
• Data is analyzed at intervals to determine
  whether a statistical difference exits. If it
  does, trial may be stopped early.
• Intention to treat analysis
• Incorporate individuals who were lost to
  follow-up
• Worst case scenario-all had no effect of treatment

35
The Pros and Cons of Experimental Design

• Advantages
• Can produce the strongest evidence for cause and
  effect
• Can be the only possible design for some research
  questions
• Experiments can potentially produce a faster and
  less expensive answer than prospective
  observational studies

36
The Pros and Cons of Experimental Design (cont.)

• Disadvantages
• Often too costly
• Example MRFIT Trial
• 10 years 120 million
• Ethical barriers
• Example Drugs in pregnancy
• Outcomes may be too rare
• Example Unusual side effects in new drugs
• Tend to restrict the scope and narrow the study
• Usually only one risk factor and one intervention

37
Example MRFIT

• Randomized Trial to test the effect of a
  multifactor intervention program on CHD mortality
• Study subjects randomized to intervention or
  usual care
• Intervention focused on 3 major modifiable risk
  factors
• Hypertension, smoking and elevated cholesterol

38
MRFIT Continued
RCT Steps 1) Select Sample 2) Measure baseline
variables 3) Randomize 4) Apply Interventions
5) Follow-up Cohorts 6) Measure outcomes/analyze
Intervention n6428
Disease/No Disease 17.9/1000
Men aged 35-57 years
12,866
Usual Care (control) n6438
Disease/No Disease 19.3/1000
Population
Sample