Commentary

Mona Oshaka takes you through the paper and discusses its implications

The study investigated the number of children who died in 1998-9 as a result of drowning in the United Kingdom. The authors compared these results with the number of children who died as a result of drowning in the United Kingdom in 1988-9.

The authors obtained information on the number of children aged 0 to 14 who died from drowning in the years 1998-9. Their sources of information included national statistical offices, police forces, and a press cuttings agency. This is the same method that two of the authors used previously to describe deaths in the period 1988-9.

The authors were thorough in their searches, but were unlikely to identify all deaths from drowning, searching the sources that they did. The best approach would be to search all death certificates of children for the given period, although this may not be practical or ethically acceptable. Some attempt should have been made to indicate how complete their data collection was.

The authors counted the number of deaths, according to the location of drowning, in the two time periods (1988-9 and 1998-9). These are shown in the table columns titled “Observed.”

The authors calculated how many deaths would have been expected in 1998-9, if the risk of death in each location category had stayed the same over the 10 year period. In this calculation, the authors took into account the fact that there was a 6% rise in the child population over the 10 year period. These numbers are shown in the column titled “Expected.” These are not real numbers of children who died, hence the decimal points.

Then the authors compared the observed with the expected number of deaths in the second time period, to calculate the ratio (of observed to expected). If the observed and expected numbers were the same—that is, if no change in risk had occurred over the 10 year time period—the ratio would be 1.0. If the risk had increased, the ratio would be greater than 1.0, and if the risk had decreased, the ratio would be less than 1.0.

There is some degree of uncertainty involved in estimating expected numbers of death. Therefore, the authors calculated 95% confidence intervals surrounding the ratio of observed to expected numbers of deaths. If the 95% confidence interval excludes 1.0 (the “null value” of no change) we can be 95% sure that the estimated ratio was not just a chance finding. These are denoted in the paper as P<0.05—that is, the probability that the estimated confidence interval does not include 1.0 is under 5%.

The authors found that the expected total number of deaths in 1998-9 was 140.51, whereas only 104 deathswere identified (see the last row in the table). The value of 0.74 (104/140.51) indicates that the risk fell by 26% (1.00−0.74) over the period. The risk ratios, however, were not the same for each location. Drownings in garden ponds increased over twofold (ratio 2.03), whereas drownings in the sea fell by almost 50% (ratio 0.53). The authors noted that drownings abroad tended to happen in hotel or apartment swimming pools. Three children with autism (a behavioural disorder) drowned, whereas only 0.1 drownings would be expected in autistic children.

The “case definition” is not explicit from the description of the study methods—that is, the authors do not describe how they defined a case. Was it sufficient that a tabloid paper reported that a child had drowned, or were clinical records examined? Did the authors really identify all the children who had drowned? Insufficient “case ascertainment” may bias the results, particularly if the sources of data had changed how drownings were reported over the 10 year period. Differences in location of drowning probably vary with age. A thorough analysis (although restricted by the small numbers of deaths) would present results for different age groups.

The analytical approach used by the authors was suitable, given the assumption that these sources of data did not change over the 10 year period. If changes—that is, improved reporting of deaths by drowning—had occurred in the sources of data that the authors used, the differences in the numbers of deaths that the authors found could be explained by the method of analysis rather than real differences in risk of drowning.

An alternative approach would have been to describe trends in drowning over each year of the 10 year period. This would allow us to see whether the observed changes had happened gradually or suddenly. If a sudden change in risk of drowning had occurred, it would be interesting to see if this coincided with the introduction of legislation regarding, for example, barriers around ponds in public places.

The results from this study indicate that, overall, deaths from drowning have decreased over the 10 year period. The risk of drowning in garden ponds doubled over this time, whereas the risk at other sites fell. Since drowning is an entirely avoidable cause of death, it is important for preventive reasons that the locations of drownings are described. The authors suggest that garden ponds be covered or fenced (a practice implication) and highlight the need for further research in the area of drownings among children with autism (a research implication).