NonRandomized Studies in Clinical Research

1
Non-Randomized Studies in Clinical Research

• Kelly K. Gurka, MPH, PhD
• Assistant Professor
• Division of Biostatistics and Epidemiology
• Department of Public Health Sciences
• University of Virginia School of Medicine

2
Learning Objectives

• Distinguish the basic attributes of select
  epidemiologic studies.
• Recognize the benefits and shortcomings of each
  presented study design.
• Understand potential threats to validity.
• Learn strategies for limiting bias during the
  design and analytic phases of a study.
• Appreciate the role that observational research
  plays in clinical research.

3
Randomized Controlled Trials
4
What Is Epidemiology?

• The study of the distributions and determinants
  of diseases in human populations.

5
Outline

• Observational Study Designs
• Descriptive
• Case series
• Cross-sectional
• Analytic
• Cohort studies
• Case-Control studies
• Types of Bias
• Selection bias
• Information bias
• Confounding

6
Study Designs
7
Epidemiologic Jargon

• Exposure Outcome

8
Common Sequence of Studies
Surveillance
Descriptive Studies
Analytic Studies
9
Descriptive Studies

• Characterize the exposure, outcome within the
  population of interest
• Describe the distribution of health outcomes

10
Specific to Clinical Practice

• Describe patients, at-risk populations, health
  care use
• Document natural history of disease

11
Descriptive Study Designs

• Case series
• Cross-sectional

12
Case Series

• Compilation of multiple case reports
• Cases are not compared to a control group
• Descriptive statistics
• Usually no statistical testing

13
Case Series Advantages

• Time efficient, less resource intensive
• Uses available clinical data
• Recognizes new diseases
• Rapid hypothesis generation
• Basis for analytic study
• Can launch a case-control study

14
Case Series Limitations

• Cases may not be representative
• No comparison group or underlying population
  represented
• Open to systematic errors

15
Cross-Sectional Studies

• Determines outcome and exposure status as they
  exist in a defined population at one specific
  point in time
• Describes the health of the population
• No follow-up occurs

16
Cross-Sectional Advantages

• Relatively time efficient
• Accurate estimate of population prevalence if
  sample is carefully selected
• Generates better hypotheses

17
Cross-Sectional Limitations

• Examines prevalent outcomes
• Depends upon incidence and duration
• Prevalent cases confound etiology with survival
• Access to limited cases
• Sensitive population may die or relocate
• Temporal relationship between exposure and
  outcome generally unknown

18
Descriptive Versus Analytic Studies

• Gray zones
• Many descriptive studies suggest hypotheses or
  causal associations
• Most analytic epidemiologic studies provide
  descriptive statistics
• Analytic studies unlikely to be fruitful in
  absence of high quality descriptive studies

19
Analytic Studies

• Utilize comparison groups to measure associations
  between exposure and outcome
• Identify determinants of health outcomes

20
Specific to Clinical Research

• Evaluate therapy in practice
• Post-marketing surveillance
• Lay the groundwork for intervention trials
• Outcomes research

21
Analytic Study Designs

• Cohort study
• Case-control study

22
Cohort Study
23
Cohort Studies

• Observational study
• Exposure is not randomly assigned
• Subjects are classified on the basis of exposure
  status
• Fundamental comparison is the incidence of the
  outcome of interest between the exposure groups

24
Cohort Indications

• Exposure assignment not possible
• Interested in multiple outcomes
• Exposure of interest is rare
• Population amenable to follow-up

25
Cohort Advantages

• Temporal relationship clear
• Less susceptible to some forms of bias
• Exposure ascertainment unbiased by outcome
• Allows direct measurement of incidence among
  exposure levels

26
Cohort Limitations

• No assignment of exposure
• Inefficient for rare diseases
• Loss to follow-up threatens validity
• More expensive than other observational designs

27
Case-Control Study
28
Case-Control Studies

• Observational study
• Exposure is not randomly assigned
• Subjects are selected on the basis of outcome
  status
• Cases are the same as in a cohort study
• Controls are a sample of the population from
  which the cases arose
• Fundamental comparison is the same
• Outcome experience among the exposed versus the
  unexposed

29
Case-Control Indications

• When RCT or cohort study not feasible
• Unethical
• Too expensive
• Too time consuming
• Interested in many exposures
• Disease is rare
• Well documented exposure data
• Readily identifiable base

30
Case-Control Advantages

• Handles multiple exposures
• Efficient for rare outcomes
• Efficient for outcomes with long latency
• Rapid assessment of outbreaks

31
Case-Control Limitations

• Inefficient for rare exposures
• Control selection challenging
• Historic exposure assessment difficult to
  ascertain
• Subject to bias

32
Sources of Bias
33
Types of Error

• Random
• Systematic

34
Systematic Versus Random Error
Sample size
From Rothman, KJ. Epidemiology, An Introduction.

35
Bias

• Systematic deviations in study findings from the
  truth
• Results from errors in the collection, analysis,
  interpretation, publication, or review of data

36
Bias

• Can arise at any stage in the design, conduct, or
  presentation of research
• Can originate from investigators or participants
• Occurs in both experimental and observational
  studies

37
Bias

• Selection bias
• Information bias
• Confounding

38
Selection Bias

• Systematic error resulting from participant
  selection procedures or factors influencing
  participation
• Result in an effect estimate different from that
  which would have been obtained from the entire
  population the study sought to characterize
• Occurs in all types of study, i.e. observational
  and experimental
• Cannot be corrected analytically, i.e. must be
  prevented

39
Select Types of Selection Bias

• Volunteer bias
• Sampling bias

40
Information Bias

• Systematic error made during data collection
  resulting in findings which differ from the truth
• Cannot be corrected analytically, i.e. must be
  prevented

41
Types of Information Bias

• Interviewer bias
• Recall bias
• Reporting bias
• Social desirability
• Publication bias

42
Confounding

• Refers to the mixing of the effect of an
  extraneous variable with the effect of the
  exposure and outcome of interest, i.e. the
  estimate of the effect of the exposure is
  distorted by mixing it with the effect of some
  covariate.

43
Necessary Properties of Confounders

• Both the exposure of interest and the confounder
  are associated independently with the outcome.
  In addition, the confounder is associated with
  the exposure of interest. The confounding
  factor, however, is not on the causal pathway.

44
Confounding

• Differs from selection and information bias
  because it can be evaluated and controlled to
  some extent in the analysis phase of the study

45
An Example
?
Maternal coffee consumption during pregnancy
Delivery of low birth weight infant
46
Example
Crude OR (170)(88) / (96)(90) 1.73
47
Smokers
Stratum-specific OR (160)(8) / (16)(80) 1.00
48
Non-smokers
Stratum-specific OR (10)(80) / (80)(10) 1.00
49
Evidence of Confounding

• ORcrude 1.73
• ORsmokers 1.00
• ORnon-smokers 1.00
• The association between coffee consumption and
  having a low birth weight baby is confounded by
  smoking. This is demonstrated by the lack of
  effect in each stratum.

50
An Example
Maternal smoking during pregnancy
?
Maternal coffee consumption during pregnancy
Delivery of low birth weight infant
51
Coffee Smoking
Crude OR (176)(90) / (90)(88) 2.00
52
An Example
Maternal smoking during pregnancy
?
Maternal coffee consumption during pregnancy
Delivery of low birth weight infant
53
Smoking Low Birth Weight
Crude OR (240)(160) / (24)(20) 80.0
54
Questions?