Biostatistics in Research Practice: Exercise: survival after retirement

This website is for students following the Health Sciences M.Sc. programme at the University of York.

Tsai et al. (2005) studied the age at retirement and long term survival of an industrial population: using a prospective cohort study. They wanted to know whether early retirement was associated with better survival. They noted that

“Some researchers had concluded that early retirement harms health, attributing this to illness before retirement or the change of life events associated with retirement. On the other hand, there is a widespread perception that early retirement is associated with longer life expectancy and that retiring later leads to early death. The possible health benefits of retirement, such as reduced role demand and a more relaxed lifestyle, have been postulated to improve longevity among people who retire early.”

The sample were past employees of Shell Oil who retired at ages 55, 60, and 65 between 1 January 1973 and 31 December 2003. They looked only at those who reached the age of 65 and they compared survival beyond that point.

The following figure shows the survival after age 65 of subjects who retired early at age 55 compared with those who retired age at 65:

Graph with 'Probability of survival' on vertical axis, 'Survival time (years)' on horizontal.  Two curved lines from survival time = 0, probability = 1.0, going down more and more steeply to time = 20, probablity = 0.38 and probability = 0.30 approximately.  Upper line is 'Retired at 65', lower line is 'Retired at 55'.

The authors say that there was a consistently lower probability of survival for employees who retired at 55 (173 deaths) than for those who retired at 65 (462 deaths). This difference was not, however, statistically significant (P = 0.09, log rank test).

Women accounted for about 11% (10% among early retirees and 12% among those retiring at 65) of the study population during follow-up. More than half of early retirees who reached 65 (57% who retired at 55 and 53% who retired at 60) were in the high socioeconomic group whereas less than half (44%) of those who retired at 65 were in this group

The authors used the Cox proportional hazards model to estimate the hazard ratios of death between the early and normal retirement groups, with adjustment for sex, calendar year of entry into the study, and socioeconomic group. After adjustment, employees who retired at 55 and were still alive at 65 had significantly higher mortality than those who retired at 65 (hazard ratio 1.37, 95% CI 1.09 to 1.73).

The authors concluded that retiring early at 55 or 60 was not associated with better survival than retiring at 65 in a cohort of past employees of the petrochemical industry. Mortality was higher in employees who retired at 55 than in those who continued working.

Questions

1. What kind of graph is this? What assumptions are required about the data? Do you think they are plausible?

Check suggested answer 1

2. Why do you think the curves become steadily steeper as survival increases?

Check suggested answer 2

3. What is a log rank test and what can we conclude from this one?

Check suggested answer 3

4. What is the Cox proportional hazards model, what are its assumptions, and why was it used here?

Check suggested answer 4

5. Why do they say that employees who retired at 55 and were still alive at 65 had significantly higher mortality than those who retired at 65, when no P value is given?

Check suggested answer 5

6. Why did they adjust for socioeconomic group? What would we expect the effects of this adjustment to be?

Check suggested answer 6

7. Why did they adjust for sex, which is very similar in the early and late retirement groups? What would we expect the effect of this to be?

Check suggested answer 7

8. Why did they adjust for calendar year at entry, i.e. the year at which subjects became 65 years old?

Check suggested answer 8

9. The authors concluded that retiring early at 55 or 60 was not associated with better survival than retiring at 65 in a cohort of past employees of the petrochemical industry. Why do think they put the conclusion in this way, rather than concluding that early retirement was associated with worse survival?

Check suggested answer 9

Reference

Tsai SP, Wendt JK, Donnelly RP, de Jong G, Ahmed FS. (2005) Age at retirement and long term survival of an industrial population: prospective cohort study. British Medical Journal 331, 995.


Back to Biostatistics in Research Practice index.

To Martin Bland's home page.

This page maintained by Martin Bland.
Last updated: 15 February, 2007.

Back to top.