is that controls should be representative of the popu- ... risk of death by natural
causes rather than because it ... dead controls may be different from living ones.
458
INTERNATIONAL JOURNAL OF EPIDEMIOLOGY
Trends and Patterns in Suicide in England and Wales From LAURA C RODRIGUES Sir—The recent paper by Charlton investigating risk factors for suicide compared cases—deaths from suicide—to deaths from natural causes.1 Is this a good choice of controls? The essence of a case-control study is that controls should be representative of the population from which the cases were drawn, in terms of the exposures of interest.2 All-cause mortality is related to many of the factors studied here, such as marital status3 and socioeconomic position.4 Therefore measures of these exposures in deaths do not reflect their prevalence in the living—this is particularly relevant for younger men, amongst whom death is very unusual. A risk factor may appear to be protective because it increases the risk of death by natural causes rather than because it decreases the risk of suicide. Is it likely that living in an area of high unemployment, low owner occupancy, or where there is a high proportion of single men living alone would reduce the risk of suicide—or is it more likely that this increases overall mortality? It is surprising that although Charlton mentions this design flaw in the discussion, he then ignores it and Epidemiology and Population Science Department, London School of Hygiene and Tropical Medicine, Keppel St. London WCIE 7HT. UK.
proceeds to discuss the findings as if he believes they are not biased. For example, Charlton suggests that being single reduces the risk of suicide in young men, but increases it in older men. It is more likely that being single is associated more strongly with overall mortality in men under 45 than in older men. This paper reveals nothing about the risk of dying—only about the chances of having 'suicide' rather than some other cause on the certificate. Suicide is increasing in young men. What we need now is reliable information to formulate prevention policies. At such a time it would be irresponsible to allow potentially biased results to go unchallenged.
REFERENCES ' Charlton J. Trends and patterns in suicide in England and Wales. Int J Epidemiol 1995, 24 (Suppl. 1): S45-S52. 2 Gordis L. Should dead cases be matched to dead controls? Am J Epidemiol 1982; 115: 1-5. 3 Ben Shlomo Y, Smith G D, Shipley M el al. Magnitude and causes of mortality differences between married and unmarried men. J Epidemiol Community Health 1993; 47: 200-05. 4 Black D, Townend P. Inequalities in Health: the Black Report. London: Harmondsworth Press, 1982.
Author's Response From JOHN CHARLTON Sir—Rodrigues rightly draws attention to the fact that dead controls may be different from living ones. I went to some effort to draw attention to this in the Discussion section of my paper,' the aim of which was to investigate how the data that are readily available on the death certificate can be used to shed some further light on the aetiology of suicides. This was a proportionate mortality study2 (similar to calculating PMR) comparing the risk of suicide death with the risk of death from natural causes. If the same factors increase deaths from suicides and deaths from natural causes, then lower odds ratios would be obtained than if living controls had been used. As pointed out in my Discussion, this Office of Population Censuses and Surveys, St Catherine's House, 10 Kingsway, London WC2B 6JP, UK.
Downloaded from https://academic.oup.com/ije/article-abstract/25/2/458/688411 by guest on 17 December 2017
needs to be borne in mind when interpreting the results. However high odds ratios would still indicate increased relative risk of suicide. In a previous paper3 we showed that marital status is likely to be the most important risk factor for suicide, particularly for men—being widowed or divorced carries the highest risk, but being single also carries a higher risk than being married. The effects increase with age, as shown in Table I, which provides more detail than the previous publication.3 Suicide is a rare outcome, and a case-control study involving living controls using the same variables would have entailed collecting identical data from living individuals and the relatives of the suicide cases—not an easy or inexpensive task, although possible given appropriate funding and ethical approval. Until such a major and costly study is done
