CrossSectional and CaseControl Studies

1
Cross-Sectional and Case-Control Studies

• Jane Garbutt MB,ChB, FRCP(C)
• jgarbutt_at_dom.wustl.eduSeptember 10th, 2008

2
Cross-Sectional Studies
3
What is the Probability That A Child With AOM in
St Louis Needs Antibiotic Treatment for NSSP?

• Estimates for the prevalence of NSSP in children
  range from 2 to 90.
• Why?

4
What is the prevalence of NSSP in children in St
Louis with AOM?

• NSSP
• CDC 27
• SLCH 53
• Surveillance studies
• MEE 43 - 80
• NP specimens 20 - 90
• WU PAARC (2004)
• NSSP 12
• NSSP-A 2 (0.8 to 4)

5
Learning objectives

• For case-control and cross-sectional studies
• Be aware of sources of systematic error.
• Know main strategies to minimize systematic
  error.
• Know the strengths and weaknesses of these study
  designs.
• Be aware of potential uses of these designs.

6
Validity
Feasibility
7
(No Transcript)
8
Cross-sectional studies

• Study population - all or representative sample
• Measure exposure and outcome variables at one
  point in time
• Main outcome measure is prevalence
• Can compare prevalence in different subgroups
• P Number of people with disease x at time t
• Number of people at risk for disease x at time t
• Prevalencek x Incidence x Duration

9
E
E
D
D
E
E
D
D
E
D
10
Sampling Bias in Cross-Sectional Studies

• Is study population representative of target
  population?
• Often use a non-probability sample Convenience
  sample with non-consecutive enrollment
• Is there systematic increase or decrease of
  cases?
• Length-biased sampling. Cases are overrepresented
  if illness has long duration and are
  underrepresented if short duration (P k x inc x
  t)
• Is there systematic increase or decrease of
  exposure?
• Prevalence-incidence bias. Exposure does not
  alter disease risk, but alters disease duration.
• ?exposure in cases if mild disease eg, HLA-A2
• ?exposure in cases if rapidly fatal

11
Design Strategies to Minimize Sampling Bias

• Population-based sample
• Probability sample
• Non-probability sample
• Use consecutive enrollment if convenience sample
• Adjust for seasonal or other time-dependent
  trends

12
Measurement Bias in Cross-Sectional Studies

• Outcome
• Misclassified (dead, misdiagnosed, undiagnosed)
• Exposure
• Differential misclassification due to recall bias
• Time from exposure important indicator of
  accuracy
• Need etiologically relevant exposure. Use current
  exposure if
• Exposure is fixed eg, blood type
• Recent exposure correlates well with prior
  exposure
• Recall of prior exposure unlikely to be reliable
  eg, diet

13
Strategies to Minimize Measurement Bias

• Standardize measurement methods
• Train and certify assessors
• Refine the instruments
• Calibrate the instruments
• Automate the instruments
• Make unobtrusive measures
• For key variables, use data from gt 1 source
• Blinding of subject and observer
• Implementation of these strategies depends on
  importance of variable, potential effect of
  inaccuracy, feasibility and cost.

14
Cross-Sectional Studies Uses

• Planning
• Individual
• Population
• Hypothesis generating/screening
• Characterize within group change over time
• Describe distribution of variables in population
  eg, NHANES
• Assess success of randomization process

15
Cross-Sectional Studies - Strengths

• Study multiple outcomes and exposures
• Immediate outcome assessment and no loss to
  follow-up, therefore faster, cheaper, easier
• Can measure prevalence
• Hypothesis generating/screening for causal links
• Useful baseline assessment
• Generalizable results if population based sample
• Serial surveys eg, Census, NHANES

16
Cross-Sectional Studies Weaknesses

• Provide limited information
• Cannot establish sequence of events
• Not for causation or prognosis (inc, RR, AR)
• Look for biological plausibility in causal links
• Impractical for rare diseases if pop based sample
  (eg, gastric ca 1/10,000). Could use case-series
  to compare prevalence of RF in patients with rare
  disease eg, homosexuality and IDU in AIDS.
• Prone to bias (sampling, measurement)

17
Case-Control - Efficient Cohort Study
18
Case-Control Study
Study Onset
Time
Exposed
Cases
Unexposed
Exposed
Controls
Unexposed
Direction of Inquiry
19
Selection of Cases

• All cases or representative sample in specified
  population that develop outcome of interest
  during a specified time period.
• Case definition to identify homogeneous entity
• Define source population and time period
• Cases can be diseased, disabled, have a good
  outcome eg, smoking cessation, or have a
  side-effect.

20
Selection of Controls

• Purpose To assess if exposure of cases to RF is
  different from expected.
• Select controls from same source population as
  cases ie., would have been identified as cases in
  this study if they developed disease.
• Time eligible to be a control time eligible to
  be a case
• Can be control gt 1x in same study, but exposure
  data may vary.
• How many controls?

21
SOURCE OR TARGET POPULATION
No outcome
Outcome
Exposed to RF
Exposed to RF
C O N T R O L S
C A S E S
Not exposed to RF
Not exposed to RF
22
Assessment of Exposure Status

• Patient records (medical, occupational)
• Interview or questionnaire
• Direct measurement (blood tests etc)

23
Odds Ratio

• Measure strength of association of RF (exposure)
  and outcome with Odds Ratio (OR)
• Disease
• Yes No
• Risk Yes a b
• Factor No c d
• OR odds of exposure in cases a/c
• odds of exposure in controls b/d

24
Odds Ratio (cont)

• OR gt 1 indicates positive association between
  factor and outcome
• OR lt 1 indicates factor is protective
• OR 1 indicates no association
• Can calculate 95 CI for OR
• OR approximates RR if
• rare outcome
• sampling error is small for cases and controls
• Less precise than RR because smaller n

25
Odds Ratio (cont)

• Disease
• Yes No
• Risk Yes a b
• Factor No c d
• Risk Ratio (RR) a/a b c/c d
  
• If rare disease, a b? b and c d ? d
• Therefore, RR ? a/b ad OR
• c/d bc

26
Board Exercise
27
Sources of Error in Case Control Studies

• Sampling bias
• Cases
• Controls
• Measurement bias
• Risk factors
• Outcomes
• Confounding

28
Sources of Sampling Bias

• Cases
• Exclusion of undiagnosed, misdiagnosed, dead.
• Over representation of prevalent cases
• Available cases not representative of all cases
  because ascertainment bias, referral bias,
  diagnostic bias, detection bias
• Controls
• Difficult to define source population
• Exposure distribution may be different from
  source population eg, matching, friends

29
Design Strategies to Minimize Sampling Bias

• Use a population-based sample
• Cases from disease registry
• Controls - probability sample from same
  population (random digit dialing).
• Sample cases and controls in same way (same
  clinic, hospital)
• Factors associated with going to hospital are
  similar.
• Controls comorbidity /- assoc with exposure of
  interest
• gt1 control group (increases power,
  generalizability)
• Nested case-control design

30
Sources of Measurement Bias

• Outcome
• Misclassification
• Faulty Instrument
• Exposure
• Differential recall bias
• Wrong duration of exposure

31
Design Strategies to Minimize Measurement Bias

• Standardize definitions, instrument and process
• Train assessors
• Automate measurement process
• Use incident cases over defined time period
• For key variables, use data from gt 1 source
• Re-analyze data with more conservative
  definitions (degree of certainty of diagnosis)
• Use exposure period important for etiology
• Use data recorded before outcome is known
• Blinding of subject and observer

32
Blinding
33
Is g-o reflux a RF for esophageal cancer?

• Esoph Gastric Esoph
• adenoca adenoca squamous ca
• Heartburn, 7.7 2.0 1.1
• Reflux or (5.3-11.4) (1.4-2.9) (0.7-1.9)
• Both
• gt 1x/wk

Do you believe their results???
34
Are you worried about bias?

• How did they minimize sampling bias for cases?
• How did they minimize sampling bias for controls?
  
• How did they minimize measurement bias for
  outcomes?
• How did they minimize measurement bias for
  predictors?

35
Confounding

• Confounding variable is associated with exposure
  and a cause of the outcome eg., cigarette
  smoking, coffee drinking and MI
• Confuses interpretation of relationship between
  exposure and outcome.

36
Controlling for Confounding

• Design
• Restriction of study population to eliminate
  confounders.
• Matching (individual or group level)
• Analysis
• Stratification
• Adjustment (multiple confounders) eg, regression
  methods

37
Matching

• Case and control comparable on confounding factor
  that is not interesting, or not modifiable, e.g.
  age, gender
• Advantages
• Increases statistical efficiency
• Optimizes information/subject
• May control for confounding
• Disadvantages
• Loss of data, increased time, cost, complexity,
  irreversible.
• Special analysis eg, conditional logistic
  regression.
• Selection bias if matching criteria do not divide
  source population into mutually exclusive
  categories eg., friends
• If matching factor assoc with exposure, crude
  exposure rate in controls will be similar to
  cases. Results in selection bias, usually towards
  null (OR1).

38
Case-control studies - Strengths and Uses

• Can estimate strength of association of exposure
  with outcome by OR(RR)
• Useful to generate/screen causal hypotheses eg,
  Reyes syndrome and ASA
• Quick, cheap, feasible.
• Efficient if rare disease or long latency

39
Case-control studies - Weaknesses

• Cannot measure prevalence, incidence, RR
• Only one outcome
• SUSCEPTIBLE TO BIAS
• Sampling bias
• Are cases representative of patients with
  outcome?
• Are controls representative of patients from
  source population without outcome?
• Measurement bias - exposure (differential recall
  bias) and outcome.

40
Observational studies

• Case control Outcome to exposure
• Cross-sectional Exposure and outcome
• ARE USEFUL BUT PRONE TO BIAS
• Sampling Bias Population based sample,
  large sample (gt 1 control),
• matching.
• Measurement Bias Standardized definitions and
  processes, training, prospective data
  collection, blinding.

41
(No Transcript)
42
QUESTIONS???