Commentary

Mona Oshaka takes you through this paper and explains what it means

The report by Owen et al describes mortality in an area that was exposed to extremely high concentrations of aluminium in the water supply. These concentrations were the result of a single water pollution incident at a treatment works in Cornwall in July 1988.

Mortality in the area affected by the pollution was compared with mortality in a neighbouring area which was supplied by a different water company at the time of the incident. The authors chose to study overall mortality as the health outcome. The effect of aluminium exposure on hospital discharge rates is discussed in another paper by two of the same authors.

The authors call this a retrospective study; it is more usually called a retrospective cohort study. This study type entails looking at the health of a historically defined group of people. In my opinion, the study was initiated when the records of people living in the vicinity of the incident were flagged (see below). I would therefore term this study a cohort study, in which a group of people are followed forward in time and their outcome (death rate) measured.

The authors used a system of flagging records at the Office for National Statistics (ONS). ONS holds a list of all people who have ever been registered with a general practitioner in the United Kingdom. If a persons records are flagged for a particular study, the investigators are informed when that person dies and are provided with a copy of their death certificate. They are also notified if the person emigrates from the United Kingdom. In the water pollution study, most of the people who were exposed to polluted water were flagged, as were those people living in the neighbouring area. This allowed the calculation of death rates over a given period (in this case, 1988 to 1997) for each of the two areas.

But what if the two areas were not comparable? Maybe one area contained more elderly people than the other, so they were more likely to die anyway? The authors tried to account for this by a statistical process known as standardisation. Effectively, this allows the calculation of mortality rates, assuming that both areas have the same age and sex structure as a standard population. In this study, the authors used two standard populations: the first was that of England and Wales, and the second was that of Cornwall and the Isles of Scilly (the region where the incident took place).

Standardised death rates are interpreted as follows. The standard population is assumed to have a death rate of 100, and the rates in the population under study are relative to this. Therefore, a standardised mortality of less than 100 means that the death rate in the study population is less than that of the standard population. A standardised mortality of greater than 100 means that the death rate in the study population is higher than that of the standard population.

Each standardised mortality has a 95% confidence interval associated with it. This is because the rate is calculated from a sample of a notional population, and therefore has some degree of uncertainty associated with it. If the 95% confidence interval of a standardised death rate excludes 100 (the rate of the standard population), we can assume with 95% certainty that the rate in the study population differs “significantly” from the rate in the standard population.

Finally, the authors calculated a ratio of the standardised mortality in the exposed and the unexposed population. This allowed them to assess whether the death rates in these two populations differed. Again, 95% confidence intervals were calculated to assess whether any observed differences were statistically significant at the 5% level. This time, we look to see whether the confidence interval includes 1.0, which is the level of equivalence of rates.

The authors found that mortality in both the exposed and the unexposed populations was lower than that expected (see columns E and U in the table). This was the case whichever standard population was used (compare rows 1 and 2 in the table). This may be because the residents of Lowermoor and the adjacent area are healthier than the general population of England and Wales, and of Cornwall and the Isles of Scilly.

The ratio of death rates in the exposed and the unexposed populations is shown in the final column of the table (E/U). The ratio is greater than 1.0, irrespective of the standard population used. This shows that mortality in the exposed population is higher than in the unexposed population (and yes, that is clear from comparing the first two columns). The 95% confidence interval of the ratio of death rates includes the null value of 1.0. From this, the authors conclude that there was no statistically significant difference in mortality rates between the two populations. Essentially, they are therefore concluding that the observed differences in mortality between the two populations occurred by chance.

From the results presented in this paper, there seems to be some evidence to suggest that exposure to the water pollution is associated with higher mortality. The authors dismiss this, since their results do not reach what are conventional levels of statistical significance. What the 95% confidence interval tells us is that the exposed population have death rates somewhere between 3% less and 20% more than the unexposed population (these numbers are derived from the 95% confidence interval of 0.97 to 1.20).

In my view, such results deserve further investigation, especially what the people died of. It is unlikely that water pollution would increase all causes of death similarly, and a cause-specific analysis might shed some light on this.

Are there other analyses that would also be interesting? In reading this report, we would want to know how many people died in total. This will give us a feel about whether the results may have occurred by chance or not. Secondly, the authors only included deaths that occurred up to December 1997. For a paper published halfway through 2002, we wonder why the analysis was so out of date.

In summary, this paper provides interesting results relating mortality to a water exposure incident. The superficial analysis does, however, not allow the reader to come to a firm conclusion as to what health effects exposure to aluminium could have.