Journal of Public Health Advance Access originally published online on March 20, 2006
Journal of Public Health 2006 28(2):178-179; doi:10.1093/pubmed/fdl002
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Correspondence |
Reply
John Steward
Director WCISU
Gareth John
Senior Statistician Health Solutions Wales
Richard Roberts
Consultant in Public Health Medicine National Public Health Service for Wales
Sirs,
Our study (Roberts et al. 2003) addressed the issue of an alleged cancer cluster around a cement works in North Wales. Busby and Howard now raise the issue of the definition of the now obsolete Office for Population Censuses and Surveys (OPCS) area of residence (AOR) codes and the precise coding scheme used in the data extract they obtained from the former Wales Cancer Registry a decade ago in 1995.
The central part of our analysis used modern GIS techniques and high-quality postcoded data aggregated to 1981 census ward boundaries. The results showed no evidence of a cancer cluster around the plant. This conclusion is unaffected by any issues surrounding AOR which we discuss here.
The issue of AOR codes was raised in the paper solely to illustrate the dangers of using such obsolete areas for epidemiology and the risk of spurious clusters arising from inappropriate epidemiological analysis.
In our section on ‘Interpretation and application of our findings’, we clearly refer to this problem and question the value of any one using AOR codes for this sort of work. The Green Audit analysis by Busby and Howard, which was the source of the alleged cluster the paper addressed, was based on data tabulated by AOR. These AOR were broad bush planning measures designed in the 1970s, in use until the 1990s. Coders in hospitals allocated cases to an AOR from look-up lists. No digitized versions are readily available. As illustrated below, this process generated gross systematic errors by tending to assign to the nearest town, and therefore, using AOR will often generate spurious clusters.
Our Methods section stated that AOR 71EE ‘was most likely to refer to what the campaigning group study refers to as Mold’. Inference was necessary because the campaigning study authors declined to share with us the actual AOR coding scheme they used—had they done so, they could have cleared this matter up before publication.
It appears in the light of this correspondence 2 years later that the 1161 cases indeed relate to the five Mold wards in the older version used by Green Audit rather than the 12 wards as we thought from the 1992 codes. It is unfortunate that this AOR is coded 71EE in both the 1992 and earlier versions of the OPCS codes.
However, this still results in a spurious cluster. Of the postcodes assigned to 71EE, only 37% actually reside within Mold, and the others are scattered all over the area, and of the 1161 cases of all malignancies attributed to the five wards by Busby and Howard, only 578 were correctly assigned to that area and with another 12 in the surrounding area makes 590. The expected number from the agreed population at risk is 509.5. This gives a relative risk of around 1.2. This is considerably lower than that claimed by Howard and Busby and illustrates how such spurious clusters may arise when using obsolete AOR codes.
Use of AOR coded data and inappropriate analysis generates spurious clusters. Thus, in our view, all previous epidemiological work by Green Audit Wales using these old data should be disregarded. Postcode mapping as used by Welsh Cancer Intelligence and Surveillance Unit (WCISU) and Small Area Health Statistics Unit (SAHSU) is the appropriate methodology to use in the current state of knowledge. We stand by the scientific findings in this paper based upon this.
 |
Acknowledgement |
|---|
 TOP Acknowledgement |
|---|
We thank Mr Ceri White, Senior Statistician WCISU, for the detailed work to clarify this matter.
CiteULike 
Connotea 
Del.icio.us What's this?
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||