Introduction to Clinical Research Design

1
Introduction to Clinical Research Design

• Lee E. Morrow, MD, MS
• Assistant Professor of Medicine
• Creighton University

2
Clinical Research Designs

• Descriptive
• Describe incidence of outcomes over time
• Case Reports
• Case Series
• Registries
• Cross Sections

• Analytic
• Analyze associations between predictors and
  outcomes
• Observational
• Cohort Studies
• Case-Control Studies
• Experimental
• Clinical Trials

3
Descriptive Studies

• Often a first step in research
• Doesnt always have a specific hypothesis to be
  tested
• Causality usually cannot be determined
• Examples
• Case Reports/Series
• Registries
• Cross Sectional Studies

4
Case Reports/Series

• Definition A single/series of patients with or
  without a disease or exposure of interest for
  whom data are collected in any fashion
• Sources Clinics, hospitals, disease registries
• Limitations Not randomly selected, bias due to
  selection factors inherent in the source, not
  representative of the population from which they
  are selected

5
Case Reports/Series

• Benefits Easy to do, useful for exploring
  relationships and/or generating hypotheses
• Key Point Associations seen in case series are
  highly likely to be biased and frequently do NOT
  hold up in more rigorous studies

6
Cross-Sectional Studies

• Definition A study based on a sample selected at
  one point or period in time

Risk Factor Present
Risk Factor Absent
Population
7
Cross-Sectional Studies

• Definition A study based on a sample selected at
  one point or period in time

Risk Factor Present
Risk Factor Absent
Sample
Population
8
Cross-Sectional Studies

• Definition A study based on a sample selected at
  one point or period in time

No Disease
Disease
Risk Factor Present
Disease
No Disease
Risk Factor Absent
Sample
Population
9
Cross-Sectional Studies

• If looking at a specified moment in time point
  prevalence
• If looking at a specified moment in time plus all
  new cases during the specified time period
  period prevalence
• If looking only at new cases during the specified
  time period incidence

10
Cross-Sectional Studies
Which cases are included in 7/1/02 point
prevalence? Which cases are included in
7/1/02-6/30/03 period prevalence? Which cases are
included in incidence?
1
2
3
4
5
6
7
8
9
7/1/02
6/30/03
11
Cross-Sectional Studies
7/1/02 point prevalence cases 1, 2,
8 7/1/02-6/30/03 period prevalence cases 1, 2,
3, 4, 6, 8, 9 incidence cases 3, 4, 6, 9
1
2
3
4
5
6
7
8
9
7/1/02
6/30/03
12
Cross-Sectional Studies
Assuming N100, calculate the 7/1/02 point
prevalence. Calculate the 7/1/02-6/30/03 period
prevalence. Calculate the incidence rate.
1
2
3
4
5
6
7
8
9
7/1/02
6/30/03
13
Cross-Sectional Studies
7/1/02 point prevalence 3 7/1/02-6/30/03 period
prevalence 7 incidence rate 4
1
2
3
4
5
6
7
8
9
7/1/02
6/30/03
14
Cross-Sectional Studies

• Limitations
• Exposure and outcome are assessed at the same
  time by the investigator (no temporality)
• Sample selection is not based on exposure or
  outcome
• Prevalence estimate is affected by duration of
  disease disease with longer duration is more
  likely to be detected
• Must consider at risk population only

15
Cross-Sectional Studies

• Benefits
• Easy
• Cheap
• Gives a snap-shot of exposure and outcome
• Good for hypothesis generation

16
Analytic Studies

• Involve a specific hypothesis that can be tested
  using a statistical model
• Involve assessing exposures as a predictor of
  outcomes
• Examples
• Observational Cohort Studies, Case-Control
  Studies
• Experimental Clinical Trials

17
Cohort Studies

• Involve following a group (cohort) of subjects
  over time
• Usually analytic but may be descriptive
• Was a treatment specifically initiated for
  evaluation?
• No Simple Cohort Study
• Yes Clinical Trial
• Randomized
• Non-Randomized

18
Cohort Studies

• Prospective Cohort Studies
• Investigator defines sample and predictor
  variables before any outcomes have occurred
• Retrospective Cohort Studies
• Investigator defines sample and collects
  information about predictor variables after the
  outcomes have occurred

19
Prospective Cohort Studies
Is a given Risk Factor associated with a given
Disease?
The Present
Risk Factor Present
Risk Factor Absent
Population
20
Prospective Cohort Studies
Is a given Risk Factor associated with a given
Disease?
The Present
Risk Factor Present
Risk Factor Absent
Sample
Population
21
Prospective Cohort Studies
Is a given Risk Factor associated with a given
Disease?
The Present
The Future
Risk Factor Present
No Disease
Disease
Disease
No Disease
Risk Factor Absent
Sample
Population
22
Retrospective Cohort Studies
Is a given Risk Factor associated with a given
Disease?
The Present
No Disease
Disease
Disease
No Disease
23
Retrospective Cohort Studies
Is a given Risk Factor associated with a given
Disease?
The Past
The Present
Risk Factor Present
No Disease
Disease
Disease
No Disease
Risk Factor Absent
Sample
Population
24
Cohort Studies in General

• Strengths
• Powerful strategy for directly measuring the
  incidence of a disease
• Can examine multiple outcomes and multiple
  exposures
• Easier to establish temporal relationship
  improves inference for causality

25
Cohort Studies in General

• Weaknesses
• Attrition of the sample
• Level of exposure may change over time
• Inability to identify presence of confounders and
  effect modifiers
• Susceptible to follow-up bias there may be a
  difference in the exposure-disease relationship
  for those who follow-up and those who do not
• Cost and feasibility vs. representativeness
  general population sample vs. restricted cohort
  sample

26
Prospective Cohort Studies

• Strengths
• Allows opportunity for complete and accurate
  measurement of risk factors
• Uniquely valuable for studying the antecedents of
  fatal diseases
• End-point unknown can take a long time for
  sufficient number of cases to develop
• Observer bias
• Weaknesses
• Expensive and inefficient for rare diseases
• Observer bias

27
Retrospective Cohort Studies

• Strengths
• Much less costly and time consuming
• Observer bias
• Weaknesses
• Less control over the nature and quality of
  predictor variable data collected
• Incomplete data sets
• Observer bias, recall bias

28
Risk Ratios in Cohort Studies

• The Risk Ratio (RR) is the ratio of the incidence
  of disease in exposed persons to the incidence of
  disease in non-exposed persons

Cumulative Incidence in Exposed
RR
Cumulative Incidence in Non-Exposed
29
Risk Ratios in Cohort Studies

• RR calculation requires incidence data
• Used in cohort and intervention studies
• Not used in Case-Control

Diseased
-

a
b

a/(ab)
RR
Exposed
c/(cd)
c
d
-
30
Risk Ratios in Cohort Studies

• Is a measure of the strength of association
  between exposure and outcome does not imply
  causality

31
Case-Control Studies

• Compares people with disease (cases) to people
  without disease (controls) with respect to
  history of exposure
• If exposure is different between cases and
  controls, an association exists between exposure
  and disease
• Cases must represent the population of all cases
  while controls must represent the population of
  all non-diseased

32
Case-Control Studies
The Present
Population with Disease
Population without Disease
33
Case-Control Studies
The Present
Population with Disease
Risk Factor Present
Risk Factor Absent
Population without Disease
34
Case-Control Studies
The Present
Population with Disease
Select Cases
Risk Factor Present
Risk Factor Absent
Population without Disease
Select Controls
35
Case-Control Studies
The Present
The Past
Population with Disease
D/RF D/RF-
Risk Factor Present
Risk Factor Absent
Population without Disease
D-/RF D-/RF-
36
Case-Control Studies

• Strengths
• Shorter study period is possible
• Rare diseases are more easily studied
• Less expensive
• Multiple risk factors may be studied
• Particularly useful for studying new diseases
  about which little is known

37
Case-Control Studies

• Weaknesses
• Choice of appropriate controls is usually very
  difficult (selection bias)
• Cases and controls do not usually come from the
  same population (selection bias)
• May be difficult to assess whether exposure
  preceded disease (recall bias)
• Incidence rates cannot be calculated directly

38
Odds Ratios in Case-Control Studies

• The Odds Ratio (OR) provides an estimate of the
  Risk Ratio (RR) for Case-Control studies
• OR is a good estimate of the RR if the disease is
  rare (incidence lt10 per year in the
  population)
• Is a measure of the strength of association
  between exposure and outcome does not imply
  causality

39
Nested Case-Control Studies

• Select disease cases from within a cohort study
• Controls are selected from non-diseased cases
  within the same cohort, within the same time
  period as the cases develop
• If controls are randomly selected from within the
  cohort (i.e. includes diseased subjects in the
  case group and the control group) it is a
  Case-Cohort Study

40
A Few Words About Controls

• The most difficult aspect of Case-Control Studies
  is selecting appropriate controls
• Matching is often used to eliminate the effect of
  potential confounders
• Technically speaking, matching reduces the
  variance of the OR!
• Matching is difficult to do correctly and may
  paradoxically worsen analysis problems if done
  incorrectly
• Impossible to match for unknown confounders

41
Clinical Trials

• Definition A clinical trial is a scientific
  experiment involving human subjects which is
  designed to evaluate the effects of
  intervention(s) against a particular disease in
  order to elucidate the most appropriate care for
  future subjects

42
Clinical Trials

• Controlled or Uncontrolled
• Is there a concurrent comparison group?
• Randomized or Nonrandomized
• Are subjects randomly allocated to the control
  and experimental groups?
• Parallel Group or Crossover
• Parallel group implies each subject receives only
  one of the interventions
• Crossover implies each subject receives
  successively each of the interventions
• Hence the terminology RCT

43
Clinical Trials Randomization

• Participants are randomly assigned to Exposure
  or No Exposure
• Randomization refers to assigning subject to an
  intervention arm without regard for baseline
  characteristics
• Goal of randomization is to equalize all other
  exposures that may confound or bias the
  association between Treatment and Outcome

44
Clinical Trials Blinding

• Single Blinding examiners do not know treatment
  assignment
• Double Blinding examiners and subjects do not
  know treatment assignment
• Triple Blinding examiners, subjects, and
  statisticians do not know treatment assignments
• Blinding is not always possible

45
Clinical Trials

• Advantages
• Minimizes confounding and bias through
  randomization
• Allows clear assessment of temporal association
• Permits a test of causality between exposure and
  disease

46
Clinical Trials

• Disadvantages
• Ethical considerations of treatment or
  with-holding treatment
• Harms (drug side effects, emotional distress) may
  outweigh benefits
• Expensive and time-consuming
• Loss to follow up
• Non-adherence to group assignment
• Possible early termination
• Cannot always randomize an exposure

47
Quasi-Experimentation

• This is essentially a clinical trial without
  randomization
• Not possible to randomize patients being
  enrolled in a rare disease trial at a site which
  does not have access to a given intervention
• Not ethical to randomize patients with cancer
  who have already failed the chemo in one arm of a
  trial cannot ethically be randomized to that arm
• Uses statistical deductive processes to rule out
  threats to plausibility
• Causal inference is less strong

48
Factorial Designs
Intervention
-
X
a
b
Y

• This is essentially an attempt to evaluate
  multiple interventions concurrently
• Given costs and inconvenience of recruiting, this
  is particularly appealing
• Not a valid model if interaction, adds
  complexity, potential for polypharmacy, reviewer
  skepticism

Intervention
c
d
-
Cell Intervention a X Y b Y Placebo c X
Placebo d Placebo
49
Example 1 Design Type?

• Investigators obtained lists of RNs age 25-42 in
  the 11 most populous U.S. states
• They mailed baseline questionnaires about diet
  and other risk factors
• Follow-up questionnaires were sent every 2 years
  for 20 years assessing additional risk factors
  and the development of disease outcomes

50
Example 1 Nurses Health Study

• Prospective Cohort Study
• Assembled a cohort
• Assessed baseline risk factors
• In the future assessed disease outcomes
• Repeated Cross Sectional Study
• Described changes over time in characteristics of
  the same study population

51
Example 2 Design Type?

• Investigators reported a 12-year old boy with
  adrenomyeloneuropathy
• Disease progression was markedly attenuated by
  treatment with a combination of oleic and erucic
  acids

52
Example 2 Lorenzos Oil

• Case Report
• A single patient
• Descriptive
• No specific hypothesis being tested
• Often not representative of the population at
  large

53
Example 3 Design Type?

• Investigators assembled 23 patients with
  adrenomyeloneuropathy from a national data base
• Randomized to two-years of treatment with oleic
  and erucic acids vs. placebo
• No statistically significant difference in
  disease progression, survival, etc.

54
Example 3 Lorenzos Oil II

• Randomized, Placebo Controlled Clinical Trial
• Prospective Cohort with an Intervention
• Cohort of subjects selected based on presence of
  disease of interest
• Followed prospectively over time
• A treatment was specifically initiated for
  evaluation
• Treatment was allocated randomly
• Treatment was compared to placebo

55
Example 4 Design Type?

• Investigators were interested in determining the
  prevalence of various pathology subtypes of
  inoperable lung cancer
• Through an institutional registry identified all
  patients diagnosed with a new lung cancer in the
  prior year (476 patients)
• Found that 38 were epidermoid, 28 were small
  cell, 18 were adenocarcinoma, 13 were large
  cell, and 3 were other types

56
Example 4

• Cross-Sectional Study
• A snapshot in time
• Subjects are included based on the designated
  point/period in time of interest
• Purely descriptive

57
Example 5 Design Type?

• Investigators recorded the smoking histories of
  1357 men with and 1357 age-matched men without
  lung cancer
• Risk of lung cancer is estimated to be 3.4 times
  greater for the smokers than for the non-smokers
• Doesnt imply causality replace smoked with
  carried lighter

Lung Cancer
-

857
457

Smoked
500
900
-
58
Example 5

• Case-Control Study
• Investigator selects people with disease (cases)
• Investigator selects people without disease
  (controls)
• Matching is often used is case-control studies
• The cases and controls are then compared with
  respect to a history of exposure

59
Example 6 Design Type?

• Investigators followed 40,000 British MDs for 10
  years
• Stratified by the number of cigarettes smoked
  each day at the start of the study
• Assessed annual death rate from lung cancer
• Found that the risk of death from lung cancer was
  32 times greater for heavy smokers as for
  non-smokers

60
Example 6 British Physician Study

• Prospective Cohort Study
• Investigator selects sample (cohort) and
  predictor variables before any outcomes have
  occurred
• The sample is then followed over time

61
Example 7 Design Type?

• Goal was to describe the natural history of
  thoracic aortic aneurysms and risk factors for
  rupture
• Investigators identified 133 patients diagnosed
  with aortic aneurysms
• Reviewed records to collect data on gender, age,
  size of aneurysm, risk factors for CV disease,
  rupture of aneurysms, and surgical repair of
  aneurysms
• 31 of aneurysms gt6 cm ruptured, 0 lt4 cm ruptured

62
Example 7

• Retrospective Cohort Study
• Identify a cohort based on past data
• Collect data on predictors from past data
• Collect data on outcomes from past and/or present
  data