Reader Comments

Post a new comment on this article

Concern over Inconsistency with Existing Literature Remains

Posted by sushah on 02 May 2013 at 16:33 GMT

We thank King et al for responding to our comments on their paper, and share their concerns that a similar data source should produce different conclusions. While they are right to compare the methods between their analysis and our results published in another journal (1), we think this deflects from the pertinent question regarding their methods which have produced surprising and inconsistent results compared to the existing literature.

However, as King et al request a specific response to their questions about our work, we note:

1. The time–dependent approach to bereavement analysis has been used extensively by other authors in the field in the UK and US (2-4)

2. As part of development of our study, we also carried out an age, sex and practice matched analysis, which is not subject to the age bias which King et al allege. These findings gave a similar one year effect of bereavement and re-assured us that our main analysis was valid. In our supporting analysis, matched controls who were still part of a couple at the index date (day of bereavement), can become bereaved during follow up. This contrasts with the methods described by King et al where selection of the control group appears to use information on cohabitee survival beyond their index date with the requirement that the cohabitee survived throughout follow up so the control could not become bereaved. We are concerned this may lead to systematic selection of controls with poorer survival from eligible control couples where one partner dies during follow up as the better surviving partner cannot be selected as a control. Exclusion of future bereaved individuals may lead inadvertently to a pool of potential controls with a higher mortality than the average married individual.

3. Our published analysis included all couples (with at least one member aged 60+ years), so, of course, it is true that bereaved individuals are on average older than non-bereaved individuals, as they would be in the general population. The concern from King et al, as we understand it, is that this age difference has not been accounted for adequately. Our analysis was adjusted for age and different approaches to entering age in our model (continuous vs single year categorical) made no difference to our findings. The Cox model assumes an exponential relationship between age and mortality, not linear as implied in their response. Crucially, our results showed an attenuation of the effects of bereavement within the first two years as is commonly reported in the literature. An artefactual effect due to inadequate adjustment for age differences between bereaved and non-bereaved groups would not reduce over time after bereavement. Lastly, the similar findings with the analysis matched on age and sex reassured us that the age adjustment was adequate and appropriate.

4. The concern over immortal bias is directly addressed by time-dependent analysis. This estimates the effect of bereavement based on the timing of bereavement and compares the mortality of bereaved couples appropriately with non-bereaved couple at the same point during follow up. As Suissa et al explain in the paper quoted in the author’s response, a time dependent analysis is the appropriate approach to data where the exposure occurs during follow up and addresses concern over immortal bias (5). A non-time dependent analysis may introduce immortal bias but this bias would inflate survival in the exposed (bereaved) group which would reduce any bereavement effect.


However, we reiterate that our key concern stands independently of our work. There is a large literature on the relationship between bereavement and mortality from numerous countries, using a range of data sources and methods, which have shown higher mortality in bereaved individuals. The studies are too numerous to cite individually but some are summarised in a review and two recent meta-analyses (6-8). We note that the paper by King et al fails to cite a number of the recent larger studies on bereavement and mortality or the meta-analysis published in PLOS One (7). The inconsistency with this large literature needs explanation and is crucially important for those experiencing bereavement and policy on bereavement.

Finally, King et al question the reason for our study if the bereavement effect is so well described. Our study did not aim to replicate this large body of work but to address specific focused questions about the nature of the bereavement effect.


1) Shah SM, Carey IM, Harris T, DeWilde S, Victor C, Cook DG. Do good health and material circumstances protect older people from the increase in mortality after bereavement. Am J Epidemiol. 2012;176(8):689-698.

2) Elwert F, Christakis NA. The effect of widowhood on mortality by the causes of death of both spouses. Am J Public Health. 2008;98(11):2092–2098

3) Hart CL, Hole DJ, Lawlor DA, et al. Effect of conjugal bereavement on mortality of the bereaved spouse in participants of the Renfrew/Paisley Study. J Epidemiol Community Health. 2007;61(5):455–460.

4) Boyle PJ, Feng Z, Raab GM. Does widowhood increase mortality risk? Testing for selection effects by comparing causes of spousal death. Epidemiology. 2011;22(1):1–5.

5) Suissa S. Immortal time bias in pharmacoepidemiology. American Journal of Epidemiology. 2008;167:492-499.

6) Stroebe M, Schut H, Stroebe W. Health outcomes of bereavement. Lancet. 2007;370(9603):1960–1973

7) Moon JR, Kondo N, Glymour MM, et al. Widowhood and mortality: a meta-analysis. PLoS ONE. 2011;6(8):e23465.

8) Shor E, Roelfs DJ, Curreli M, Clemow L, Burg M M, Schwartz JE. Widowhood and Mortality: A Meta-Analysis and Meta-Regression. Demography 2012;49(2):575-606



Sunil M Shah, Senior Lecturer in Public Health, St George’s University of London
Iain M Carey, Research Fellow, St George’s University of London
Tess Harris, Senior Lecturer in Primary Care, St George’s University of London
Stephen DeWilde, Senior Lecturer in Primary Care Epidemiology, St George’s University of London
Christina R Victor, Professor of Gerontology and Public Health, Brunel University
Derek G Cook, Professor of Epidemiology, St George’s University of London

No competing interests declared.