ISAAC Global rhinitis data

1

• Thank you for having chosen to view this
  PowerPoint Presentation.
• We would like to remind you that this work is the
  property of the author alone and cannot be used,
  in part or in whole, without their prior consent.
• ERS 2002

2
Slides for David Strachan

• ERS Workshop on Post-Genome Epidemiology, Cernay,
  France

3
The hazards of everyday life

• Most strong associations are already discovered
• Weak associations, widely spread, may add up to a
  lot of trouble, but are unconvincing for
  causality

Spurious associations True weak
associations (false alarms, scares) (focus for
prevention) Play of chance Bias Imprecise
measurements Residual confounding Strong
subgroup effects
4
Why study interactions?

• Antidote to determinism
• Genetic susceptibility, perinatal programming
• Lifecourse approach to disease causation
• More certain identification of remediable causes
• Interaction relative risks larger than overall
  relative risk
• Bias and confounding more easily excluded
• Guidance for public health policy
• Safety for susceptibles

5
The relevance of weak associations
Incidence

• Exposure RR PARF
• A- 20 1.0 0
• A 80 1.5 29
• A small excess risk widely spread adds up to a
  lot of trouble ...
• ... but is difficult to distinguish from bias and
  confounding

6
Effect concentration among susceptibles
Incidence
E F G H
A B
C D
Exposure RR PARF A- B- 1.0 0 A- B 1.0 0
A B- 1.0 0 A B 3.5 29
7
Pros and cons of studying interactions

• Pros
• Biologically plausible
• Improve specificity
• Increase strength in susceptible subgroups
• Less prone to bias and confounding

• Cons
• Scale dependent
• Problems of multiple comparisons
• Large samples often considered necessary for
  adequate power

8
Effect concentration relative risk (1)

• B- B All Cases
• Prevalence 80 20 for A
• A- 20 1.0 1.0 1.0
• A 80 1.0 3.5 1.5 374
• All 1.0 3.0
• Cases required for B 32
• Cases required for AB 242

80 power 5 significance Large
cohort Rare disease
9
Effect concentration sample size (1)
80 power 5 significance
10
Effect concentration sample size (2)
11
Multiple subgroups of susceptibles
Incidence
Exposure RR PARF A- 1.0 0 A B- 1.0 0
A B 3.5 29 A W 3.5 7 A X 3.5 7
A Y 3.5 7 A Z 3.5 7
12
Partial identification of susceptibles (1)
Incidence
W - unmeasured X - unmeasured Y - unmeasured Z -
measured
A B
Exposure RR PARF A- Z- 1.0 0 A- Z 1.0 0
A Z- 1.4 22 A Z 3.5 7
13
Partial identification of susceptibles (2)

• Z- Z All Cases
• Prevalence 95 5 for A
• A- 20 1.0 1.0 1.0
• A 80 1.4 3.5 1.5 374
• All 1.3 3.0
• Cases required for Z 162 (RR 3.0 / 1.3 2.3)
• Cases required for AZ 1471 (RR 3.5 / 1.4
  2.5)

14
Partial identification of susceptibles (3)
15
Effect concentration a summary
Incidence

• Susceptibles fully identified
• Increased RR in subgroup
• Power to detect interaction with sample sizes
  required to detect main effects
• Susceptibles partly identified
• Less marked increase in RR
• Loss of power to detect the effect concentration

16
Epidemiology in the 21st century

• Increased focus on effect modification, rather
  than association, as the basis for causal
  inference.
• Success is most likely in diseases where
  synergistic biological interactions are already
  suspected.
• Success in common multifactorial diseases is
  less likely and will depend on aggregating
  susceptibility.
• Functional traits (if they can be measured) may
  be more informative than genotypes for this
  purpose.

17
Prediction is an uncertain business...

• I think there is a world market for maybe five
  computers. (Chairman of IBM, 1943)
• Computers in the future may weigh no more than
  1.5 tons. (Popular Mechanics, 1949)
• There is no reason anyone would want a computer
  in their home. (Digital Corporation, 1977)
• 640k should be enough for anybody. (Gates, 1981)