We read with interest the article by Harper et al in the last edition of this journal (2011;1:306-309). By establishing a correlation between infant mortality and the increased mortality risk of the parents, the article speaks, albeit indirectly, to the possible links between emotional and physical states - a topic of important and growing interest. However, two points present themselves about the authors' analysis - the first is statistical, the second conceptual.

Harper et al are direct about possible confounding factors for which they were unable to control but which may nullify any causal link implied by the statistical correlation they identify (p308). However, given their use of census data, there is one possible confounding factor that they might have controlled for - age of parent at the time of birth. Harper et al imply that the small difference in mean age between the comparison and control group means age should not raise concern. However, if parent and infant mortality are more likely among both older and younger parents, the difference in mean age might be minimal or invisible, even though the casual implications of the chi-squared correlation would be weakened. Previous research suggests that such a two-sided confounding factor may exist for age. Young births, such as teenage pregnancies, have been linked to worse outcomes for both mother and child (see, for example, citations in Wellings and Kaye, 1999). Births to older parents are similarly more likely to face complications and older parents would naturally be closer to average life expectancies.

The results presented by the authors point towards this possibility. The 'bereaved' group manifests a wider range of ages than the 'control' group. The effect is most significant in the English data - the smallest difference in standard deviations exists for the 1981-1991 cohort, which is between 8.4 and 5.7. Since the English data manifest a higher relative risk, it is worth exploring whether age may be such a confounding factor. Year of birth might also be included in the analysis, to control for the possibility that both infant and parent mortality risks may be shifting over time.

The second concern is a conceptual one. If a ceteris paribus causal link exists between the death of one's children and an increased mortality risk for parents, it seems likely that this is linked to the trauma caused by the loss of a loved one. But emotional bonds are not all equal - on average, they might be expected to become stronger over time. From this perspective, the average relative risk of parent mortality after a miscarriage should be lower than death in childbirth, which would be lower in turn than death in early infancy and the death of a young child.

The Dutch study (Li et al, 2003) referenced by Harper et al bears out this principle - the relative risk of parent mortality in their study is higher for a child dying aged between 1 and 11 months than a child dying at less than one month. The analysis by Harper et al can be extended to explore this hypothesis. Parents suffering stillbirths and infant mortalities are currently combined into a single category - by splitting the two groups, differences might be observed. The accidental inclusion of parents whose child dies aged older than one in the 'non-bereaved' group would bias effects downward, but differences may still be observable.

For the Scottish data, ethical non-identification reasons and small sample sizes prevented such subgroup analysis (p307). However, the small sample size resulted by design - the choice of 20 birthdays or 5.3% of the population. Had the authors chosen a larger set of birthdays, it may have been ethically possible to analyse the two subgroups in greater detail. No such restriction applied to the English data.

The authors are to be congratulated on their research and we agree with them that further statistical and case study work is required. We hope that the suggestions are amenable to straightforward exploration on similar datasets to those used by the authors.

Dr Wing Chu, ST1 Doctor in Palliative Care; Christian Percy, Statistical Consultant

References Wellings, K. and Kane, R. (1999), 'Trends in teenage pregnancy in England and Wales: how can we explain them', J R Soc Med 1999;92:277-282

Conflict of Interest:

None declared