Statisticial review
This is a statistical review and does not repeat the points made by another referee.
Points to which a response should be made
| 1 | This study is done in a large database that records 0.5 million births per year. There must be very many more births than this in the 18 countries listed. There are only just over 0.5 million births in the database for 13 years from 1985 to 1997. There must be a very strong selection process going on to obtain such a small number of records. From the description in reference 12 it is not at all clear what possible biases exist in the database, either over time or at any particular time. The system is voluntary and without any validation. While it is clearly a tremendous effort to create such a database, the results from it must be treated with extreme caution. The potential for bias in any analysis is so great that it cannot be used to make strong statements at all. |
| 2 | In the Introduction a contrast is drawn between studies that found some association and others that did not find an association. The results from such studies are not necessarily contradictory. The magnitude of any effects, together with a confidence interval should be given. It may then be seen that apparently contradictory results are merely due to lack of statistical power in at least one of them. The size of the erects and confidence intervals must be given. |
| 3 | It would appear that there has not been linkage between successive pregnancies in the same women, but that inter-pregnancy interval is based on the reproductive history for the current pregnancy. This may be unreliable. It may also be biased, especially if the mother dies during or after this pregnancy. How have other studies examined maternal death? Have they excluded deaths where the child also died? |
| 4 | It is not clear why the groupings of inter-pregnancy interval were decided. The period 18-23 months is much longer than the earlier ones. Was this done prior to examination of the data? |
| 5 | The method of analysis provides odds ratios, not relative risks of an adverse event. With quite high percentages the odds ratios exaggerate the relative risks of an adverse event. They should be reported as odds ratios. |
| 6 | Was the adjustment done using binary categorisation of all the factors used in the logistic regression? Were dummy variables used for the factors with more than two categories? Were factors that did not make any difference in the analysis retained in the model? |
| 7 | The findings for eclampsia (and pre-eclampsia) do not show any trend with inter-pregnancy interval. It is only the >60 month interval that shows a difference. History of hypertension also shows this step change. It would be helpful to have the numbers (or percentages) for eclampsia and pre-eclampsia stratified by history of hypertension. It is surprising that adjustment for this has not affected the odds ratio rather more than is suggested in Table 3. |
| 8 | The criticism of "mistaken analyses" (page 10) of other authors is not best done in the discussion of this paper. A letter to the original authors and the journal where those findings were published would be more appropriate. It is presumably because those authors matched on some factors that were reported as being associated with the outcome. |
| 9 | There is also no trend in the findings on maternal death. It would seem that those who have extremely short or long inter-pregnancy intervals may well be very different from other women in a variety of ways other than the interval between pregnancies. Although adjustment has been carried out for some measured factors, this adjustment will of necessity be imperfect. |
| 10 | The last paragraph of the paper assumes that the associations found are causal in nature. Given the limitations of the study, could this be an over-statement? |
SJW Evans