Analytic Epidemiology

1
Analytic Epidemiology

• Determining the Etiology of Disease

2
Study Development Process
Descriptive Studies Data Collection and Analysis
Model Building and hypothesis formulation
Analyze results and retest
Analytic Studies for Hypothesis testing
3
Causality and Causal Relationships

• Must have statistical significance
• Association may be either positive or negative
  (if positive, the association is higher than
  expected if negative the association is lower
  than expected)
• Must try to rule out noise (assuring the
  comparison of apples to apples by controlling
  confounding factors)

4
Artifactual or Spurious Associations

• A false or fictitious association can result from
  chance occurrence or bias in the study methods
• Type 1error occurs from random fluctuation
• Through retesting, we can determine spurious
  relationships.
• Non-causal associations take place when a factor
  and disease are associated indirectly

5
Causal Association

• Strength of association
• Dose-Response Relationship
• Consistency of the Association
• Temporally Correct Association
• Specificity of the Association
• Coherence with Existing Information

6
Sources of Data

• Primary data information collected directly by
  the researcher
• Secondary data data that has already been
  collected and stored for analysis

7
Types of Surveys

• Administrative surveys, medical records, vital
  records and statistical data
• Telephone surveys
• Self-administered surveys
• Personal interviews

8
Measurement Issues

• Measurement is an attempt to assign numbers to
  observations according to a set of rules
• Variables can be categorical or continuous
• Intent is to translate observations into a system
  that allows assessment of the hypothesis

9
Types of Categorical Variables

• Nominal variables assigns name or number purely
  on arbitrary basis (e.g., race, sex)
• Ordinal variables measures assigned from
  (typically) a lesser to greater value
• Interval variables scale that assigns a number
  to an observation based on a constant unit of
  measurement
• Ratio assigns numbers to observations to
  reflect a true point

10
Improving the Survey

• How is measure administered?
• Has it been used on similar situations with
  success?
• Is measure understandable by those being
  surveyed?
• Is sample accessible and identifiable?
• Is special training required?
• What is length of time in measurement?
• Are results available in timely manner?

11
Reliability and Validity Issues

• Reliability the extent to which a measurement
  has stability and homogeneity
• Validity represents the precision to which the
  measure is truly measuring the phenomena being
  measured (measure must be reliable to be valid)

12
Reliability and Validity Issues

• Content Validity the extent to which the
  measure reflects the full concept being studied
• Criterion Validity assessed by comparing the
  test measure of the phenomenon

13
Sensitivity

• Sensitivity (Se) measures how accurately the
  test identifies those with the condition or
  trait, i.e., correctly identifies or captures
  true positives
• High sensitivity is needed
• When early treatment is important
• When identification of every case is important

14
Specificity

• Specificity (Sp) measures how accurately the
  test identifies those without the condition or
  trait, i.e., correctly identifies or excludes the
  true negatives.
• High specificity is needed when
• Re-screening is impractical
• When reducing false positive is important

15
Factors to consider in setting cutoffs

• Cost of false positives v. false negatives
• Importance of capturing all cases
• Likelihood population will be re-screened
• Prevalence of the disease (Pe)
• Low Pe requires high Sp, otherwise too many false
  positives
• High Pe requires high Se, otherwise too many
  false negatives

16
(No Transcript)
17
Determining SE and SP Rates

• SE TP / (TP FN)
• SP Specificity TN / (TN FP)
• False Neg. Rate 1 SE
• False Pos. Rate 1 - SP
• Positive Predictive Value
• TP/(TP FP)
• Negative Predictive Value
• TN/(TN FN)

18
SE and PE Example

• You need to test the validity of cervical smears
  pap smears to determine the presence of
  cancer of the cervix. Smears were taken from 120
  women known to have cancer of the cervix and from
  580 women who did not have cancer of the cervix.
  In the laboratory, the smears were read blind
  as positive or negative. Of the total 700
  smears, 200 were read as positive, 110 of which
  came from the proven cancer cases.

19
The Research Cycle
Theory
Operational hypothesis
Empirical findings
Observations and measurements
Statistical Tests
20
Types of Studies

• Cross-sectional prevalence rates that may
  suggest association (good for developing theory,
  but no causal association)
• Retrospective (Case-control) good for rare
  diseases and initial etiologic studies
• Prospective (cohort, longitudinal, follow-up)
  yields incidence rates and estimates for risk.
  Better for causal association.
• Experimental (intervention studies) strongest
  evidence for etiology

21
Considerations for Study Design

• Stage of hypothesis development
• Nature of disease
• Nature of Exposure
• Nature of study population
• Context of research

22
Cross-Sectional Studies

• Single point in time (snapshot studies)
• Risk factors and disease measured at the same
  time
• Determines prevalence ratios

23
Cross-Sectional Study Design
Cases
Exposed
Non-Cases
Sample Population
Cases
Non-Exposed
Non-Cases
24
Advantages and Disadvantages of Cross Sectional
Studies

• Advantages
• Gives general description or scope of problem
• Useful in health service evaluation and planning
• Baseline for prospective study
• Identifies cases and controls for retrospective
  study
• Low-cost

• Disadvantages
• No calculation of risk
• Temporal sequence is unclear
• Not good for rare diseases
• Selective survival can lead to bias
• Selective recall can lead to bias
• Cohort effect may be misleading

25
Prospective Study Desgin

• Disease free persons are classified on exposure
  at beginning of follow-up period then tracked to
  ascertain the occurrence of disease.
• Question of Study Do persons with the factor of
  interest develop or avoid the disease more
  frequently than those without the factor or
  exposure

26
Prospective Study Design
Cases
Exposure
Non-Cases
Sample Population
Cases
Exposure -
Non-Cases
27
Prospective Study Criteria

• Obtain Incidence data
• Obtain the incidence among the exposed A/AB
• Obtain incidence among the non-exposed to
  determine relative risk C/CD
• Determine Relative Risk A/(AB)/C/(CD)

28
Advantages and Disadvantages of Prospective
Studies

• Advantages
• Provides good assessment of temporal sequence
• Evaluate before onset of disease and watch for
  disease

• Disadvantages
• Selection bias
• Loss to follow-up
• Expensive

29
Retrospective Study Design

• Subjects are selected on the basis of disease
  status either cases or controls then classified
  on the basis of past exposure
• Question of Study Do persons with the outcome of
  interest (cases) have the exposure characteristic
  (or history of exposure) more frequently than
  those without the outcomes (controls)

30
Retrospective Study Design
Exposure Positive A
Cases
Exposure Negative B
Exposure Positive C
Controls
Exposure Negative D
31
Retrospective Study Method

• Compare the odds of exposure among the cases with
  the odds of exposure among the controls
• Odds of Exposure Among Cases A/(AC)/C/AC)
  or A/C
• Odds of Exposure Among Controls
  B/(BD)/D/BD) or B/D
• Get Odds Ratio or odds of expose among cases/Odds
  of exposure among controls (A/C)/(B/D)

32
Advantages and Disadvantages of Retrospective
Studies

• Advantages
• Less expensive than cohort (retrospective)
  Studies
• Quicker than cohort
• Can identify more than one exposure
• Good for rare diseases
• Well design leads to good etiologic investigation

• Disadvantages
• Selective Survival
• Selective recall
• Temporal sequence not as clear
• Not suited for rare exposures
• Gives an indirect measure of risk
• More susceptible to bias
• Limited to single outcome

33
Experimental Studies

• Uses an intervention in which the investigator
  manipulates a factor and measures the outcome
• Elements of a complete experiment
• Manipulation of data
• Use of a control group
• Ability to randomize subjects to treatment groups

34
Advantages and Disadvantages of Experimental
Studies

• Advantages
• Prospective direction
• Ability to randomize subjects
• Temporal sequence of cause and effect
• Can control extraneous variables
• Best evidence of causality

• Disadvantages
• Contrive situation
• Impossible to control human behavior
• Ethical Constraints
• External validity uncertain
• Expensive

35
Attributable Risk

• The rate of disease in the exposed group
  attributable to exposure.
• Relative risk measures the strength of the
  association
• Attributable risk identifies risk of the disease
  attributable to exposure or the proportion of
  incidence in exposed group attributable to
  exposure

36
Attributable Risk Calculation

• Begins with (Incidence in the Exposed Group) -
  (Incidence in the non-exposed group).
• Search for the proportion of AR

37
Attributable Risk Calculation
Incidence in the Exposed Group
Incidence in the Exposed Group
-
Incidence in the Exposed Group
38
Population Attributable Risk Requirements

• Incidence rate of disease among those exposed to
  a trait or characteristic
• Incidence rate of disease among those not exposed
  to the trait or characteristic
• The proportion of the population that has the
  trait or characteristic

39
PAR Example
Incidence lung cancer, smokers
of smokers in population
Inc. lung cancer, non-smokers
of non-smokers in pop.

EXAMPLE Using NV Rates
(28.0/1000)
(.32)

(17.0/1000)
(.68) 20.5
40
Attributable Risk Example
Incidence, total population
Incidence, non-exposed population
-
EXAMPLE Using NV Rates
(20.5/1000)
-
(17.0/1000) 3.5/1000
41
PAR
Incidence, total population
Incidence, non-exposed pop.
-
Incidence in total population
EXAMPLE Using NV Rates
(20.5/1000)
-
(17.0/1000)
3.5/20.5 17
20.5
42
Intervention Comparisons

• To demonstrate any therapeutic effect uses a
  PLACEBO
• To demonstrate improved therapy compare to
  CONVENTIONAL TREATMENT
• To demonstrate the most effective regimen compare
  DIFFERENT REGIMENS

43
Blinding in Experimental Studies

• The importance of blinding depends on the needed
  outcome. Less important if the outcome is clear.
• Non-blinded both subject investigator know
  the treatment allocation
• Single-blinded investigator knows, subject does
  not know
• Double-blinded neither investigator and subject

44
Sources of Bias

• During selection of participants
• Absence of blinding allocation can lead to
  differential classification
• Other sources of miscalculation
• Withdrawals, ineligible sources, loss to
  follow-up
• Premature termination

45
Selection Bias

• Cases and controls, or exposed and non-exposed
  individuals were selected is such that an
  apparent association is observed - even if there
  is no association.
• Biased selection - taking from a pool in which we
  know the risk is higher is selection bias.
• Small sample size or small response size

46
Information Bias

• Methods of information about the subjects in the
  study are inadequate and results show information
  gathered regarding exposures and/or disease is
  incorrect.

• Reporting bias
• Abstracting records
• Bias in interviewing
• Bias from surrogate interviews
• Surveillance bias
• Recall bias

47
Other Issues

• Confounding Variables
• To prove that Factor A is a result of disease B,
  we say that a third factor, Factor X is a
  Confounder if the following is true
• Factor X is a known risk factor for Disease B.
• Factor X is associated with Factor A bit is not a
  result of Factor A.
• Interactions