Letters

Association between obstetric care and risk of suicide

BMJ 1999; 318 doi: http://dx.doi.org/10.1136/bmj.318.7192.1211 (Published 01 May 1999) Cite this as: BMJ 1999;318:1211

Study has methodological flaws

  1. D J Gunnell, Senior lecturer in epidemiology and public health medicine (D.J.Gunnell{at}Bristol.ac.uk),
  2. Andy Ness, Senior lecturer in epidemiology,
  3. Elise Whitley, Lecturer in medical statistics
  1. Department of Social Medicine, University of Bristol, Bristol BS8 2PR
  2. Department of Medical Engineering, F60 Novum, Huddinge University Hospital, SE-14186 Huddinge, Sweden
  3. Department of Obstetrics and Gynaecology, Karolinska Hospital, SE-17176 Stockholm, Sweden

    EDITOR—Jacobson and Bygdeman present intriguing data suggesting that the risk of suicide is influenced by birth trauma; they argue that this may explain recent increases in suicide among adolescents.1 However, deficiencies in the design and analysis of this case-control study should be addressed before the findings are accepted or possible mechanisms are considered.

    The significant findings in Jacobson and Bygdeman's paper are confined to a subgroup of men who committed suicide using violent methods, were born and died in the catchment area of Stockholm's forensic medicine department, and had siblings whose birth records were available. The 175 males for whom the significant associations were found probably represent 25% of all suicides. The authors' suggestion that the association is restricted to violent suicides might be tested by presenting separate risk estimates for violent and non-violent suicides. Furthermore, there is overlap in the cases included in this analysis and those included in a previous study which reported highly significant associations between the risk of suicide and perinatal factors.2 This overlap is likely to have influenced the current findings.

    In case-control studies, controls should be selected from the population giving rise to the cases.3 In Jacobson and Bygdeman's study, cases were restricted to the study catchment area but the same restriction was not applied to controls. Migration may be associated with socioeconomic and health related factors which in turn may confound the associations observed.4

    Finally, confounding factors are those associated with both the disease and the exposure under investigation. The statistical significance (or non-significance) of a factor as a predictor of a disease does not justify its inclusion (or exclusion) in a multivariable model. The important issue is whether inclusion of the factor influences the strength of the observed association.3 Factors likely to be associated with the risk of suicide and obstetric practice, such as year of birth, were not included in the final model, and so risk estimates may be biased.

    The epidemiology of suicide is complex, and it seems unlikely that the factors investigated in this study explain the high proportion of suicides. This view is supported by another investigation of this issue.5 Further analyses of these data are required to clarify the nature of the associations. These analyses should include both violent and non-violent suicides and make appropriate adjustment for confounding factors. The cases included in the earlier study should be excluded along with controls who no longer reside in the study area.

    References

    1. 1.
    2. 2.
    3. 3.
    4. 4.
    5. 5.

    Authors' reply

    1. Bertil Jacobson, Professor emeritus,
    2. Marc Bygdeman, Professor
    1. Department of Social Medicine, University of Bristol, Bristol BS8 2PR
    2. Department of Medical Engineering, F60 Novum, Huddinge University Hospital, SE-14186 Huddinge, Sweden
    3. Department of Obstetrics and Gynaecology, Karolinska Hospital, SE-17176 Stockholm, Sweden

      EDITOR—Gunnell et al raise several interesting questions, to which detailed responses with numerical data have already been made (on the 6 and 26 February) and posted on the BMJ's website.1

      Our first study showed that violent suicides were closely associated with perinatal trauma whereas non-violent suicides were not.2 Thus, we had no reason to include non-violent suicides in our new study. The “overlap in cases” results in conservative estimates—that is, if the old cases are removed then the estimated risk and attributable percentages increase. Consequently, we believe that it was correct not to exclude the old cases, especially since the methods of analysis and control were different: in the earlier study unmatched cases drawn from the general population were used and in the present study matched siblings were used.

      We thank Gunnell et al for pointing out that the migration of siblings is indeed relevant. Using currently available data, we found that the siblings who had had perinatal trauma had migrated from the study area about twice as often as had siblings who had had a normal birth. If siblings who have migrated are excluded from the regression analysis there is an increase in the risk of suicide associated with perinatal trauma and the absence of opiates given to mothers during delivery. Again, the results given in our article are conservative estimates.

      We agree with Gunnell et al's definition of confounding but we disagree about including the year of birth in the regression analysis. This variable was not significant but if it is forced into the analysis it enhances the estimated risk factor for suicide after perinatal trauma. It is methodologically incorrect to deviate from a postulated hypothesis and stated method of analysis by including a non-significant variable into the regression when it enhances one of the risk factors studied.

      Our study is not comparable with that of Neugebauer and Reuss; they did not use siblings as controls, and they therefore could not control as efficiently as we did for confounding by genetic, socioeconomic, and demographic factors.3 Besides, they did not analyse perinatal trauma in a way that makes a direct comparison with our results possible (we have asked them to supply such data).

      Thus, the proportion of suicides explained by perinatal factors is at least as high as that given in our paper.

      References

      1. 1.
      2. 2.
      3. 3.