Editor,—In a recent paper published in this journal, Kposowa1 reported that divorced and separated men had a higher risk of suicide than married men. Men and women of other unmarried status reportedly did not suffer any excess risk of suicide in comparison with their married counterparts. We have some reservation about the findings, and wish to point out two methodological problems that may have affected many studies in this area.

In this study marital status was enumerated at the beginning of the study period. Marital transitions between the baseline survey and death or end of follow up were unknown to the researcher. The author did mention this issue at the end of the discussion section, but seemed to suggest that the problem would not have seriously affected the findings. The failure to capture marital changes would lead to a misclassification of marital status during the follow up period and at death. The follow up period is from 1979 to 1989. Martial status would have changed during the 11 year period. For example, among the elderly, the married person would become widowed; the never married person among the age group 25–34 would have been married during the period, etc. As such, the hazard ratios based on the initial marital status would be biased towards the null value. One study in the USA2 and one in the UK3 have emphasised the importance of this misclassification error. Furthermore, there were only 545 suicide cases with a rate of approximately 10.5 per 100 000, which could be very sensitive to the possible misclassification. In this study both the models with and without adjustment for baseline factors contradict previous research results that used the information of the marital status at death.4 5 We should be very cautious about the insignificant results.

Studies have demonstrated increased psychiatric disturbances in the first year of widowhood.6 7 Afterwards the widowed return to a usual level of psychological health. This strongly supports the plausibility that widowed people may have a higher risk of suicide shortly after the death of a spouse. Widowed persons available at the baseline survey were likely to be those who had survived the initial high risk period. The study design has a selection bias that favours the survivors. Much less is known about changes in psychiatric disturbances in divorcees. The same selection bias may exist in divorced people.

There has been a considerable number of studies on marital status and suicide since Durkheim's classic study.4 5 Unfortunately progress has been prevented by some common methodological problems that are often overlooked. Our first suggestion is that much more effort should be spent on updating the marital status of participants during a follow up period. A closely related suggestion is to study the suicide risk of participants married at baseline but who become unmarried during the follow up. This will avoid the selection bias in the studies of people divorced or widowed at baseline. Secondly, testing for non-proportionality should be routinely carried out and reported if the analysis is based on the Cox model. There are various methods for the testing, such as allowing break points in a hazard function and the use of Schoenfeld residuals. An initial hazardous effect that diminishes over time is partial evidence for the above mentioned problems. No information on the proportional hazard assumption was given in the paper under discussion.