Small Area Studies

1
Small Area Studies

• Dr Patrick Saunders
• Head of Environmental Health and Risk Assessment,
  Health Protection Agency
• Assistant Director WHO Collaborating Centre for
  Chemical Incidents, University of Cardiff
• Honorary Senior Lecturer University of Birmingham

2
Introduction

• What they are
• The problems associated with them
• Guidance on when and how to conduct them
• My role
• advise the HPA
• produce guidance for the WHO
• published guidance in Oxford Handbook

3
Introduction

• Studies of the association between residential
  proximity to a potential source of pollution and
  increased levels of some disease(s)
• In response to concerns about an elevated level
  of disease AND/OR elevated level of exposure
• There are serious issues about the worth and
  appropriateness of such studies in the great
  majority of cases
• This is the case internationally-range of
  guidance has been developed

4
Issues

• Many sophisticated methods for investigation but
  they are all founded on same basic statistical
  principles and prone to error
• Type 1 error-appears to be an excess of disease,
  or to be an association between exposure and
  disease, when in fact no causal relationship
  exists
• Type 2 error occurs when no statistical
  association is found, but a causal relationship
  exists

5
Issues

• Randomness
• Exposure misclassification
• Plausibility
• Confounders
• Statistical power/techniques

6
Randomness
7
Statistical Issues

• An even more serious problem with some surveys
  of leukaemia is a tendency to draw the
  boundaries after seeing the data. It is
  essential to draw all boundaries (in space, time,
  age group and type of disease) before conducting
  the survey. Otherwise it is easy to create
  highly improbable incidences by choosing tight
  boundaries around apparent clusters. Studies
  by people who are not aware of the problem can
  lead to apparently frightening results. The New
  Scientist

8
Exposure misclassification

• Next biggest problem
• virtually no study uses actual exposure data
• difficult to obtain such data
• most use distance as simple proxy for exposure
• takes no account of the impact of weather,
  topography, technical factors
• can use modelling to refine
• Takes no account of personal behaviour

9
Plausibility

• Temporal
• Precedes exposure?
• Population movement-time of diagnosis not
  necessarily the same as exposure/illness
• Spatial-downwind or upwind?
• Biological
• Most diseases have several causes
• Is the condition plausibly associated with the
  exposure?
• Is the condition real and/or the correct one?

10
Confounders

• Other things could cause/influence the disease in
  question
• Age
• Gender
• Genetics
• Other exposures e.g. occupational
• Deprivation
• Smoking
• Drinking

11
Statistical Issues

• Multiple testing/95 confidence intervals
• Type 1 error is more likely if multiple
  statistical tests are performed. By definition,
  one would expect to find P lt 0.05 in 1 in 20
  significance tests, even if no causal
  relationship exists. If one performs 10 tests,
  then there is a 40 chance that at least one
  would be significant at the 5 level
• Small numbers
• Type 2 error may result from flawed disease or
  exposure data, or inappropriate study design. The
  commonest reason, however, is insufficient
  statistical powerthe cases are just too rare. It
  may be impossible, for example, to know whether
  three cases of a rare disease in a population of
  workers really is excessive

12
Guidance

• The reported experience of health agencies
  confirms that major associations between
  exposures and outcomes are rare. Minnesota, for
  example, has reported results from over 500
  investigations of clusters, six of which were
  full-scale investigations. In one instance, in an
  occupational setting, an important public health
  outcome was documented
• Saunders PJ, Kibble, AJ, Burls, A. 2006.
  Investigating alleged clusters. In Oxford
  Handbook of Public Health Practice, 2nd edition.
  (Pencheon D, Melzer D, Gray M, Guest C, Eds).
  Oxford University Press, Oxford.

13
Guidance

• If there is an exceedence of a health based
  standard-address it
• If there is a statutory nuisance-abate it
• If there is an excess of disease-investigate it
• Single site study only if there is plausibility
• environmental, biological, spatial and temporal