Life tables (cont.)
What are life tables used for?
Mortality life tables are used to calculate risks of dying from any age to any older age. This could be for academic studies or for commercial use such as life insurance. Knowing these risks allows you to predict how many people are likely to die in certain periods of time. This has use in clinical trials for estimating expected deaths, and also for demographic projection of populations. They are also used for calculating life expectancy.
Life tables more generally can be used for any process involving duration data - data which identifies a start point, followed by a duration of time until there is an end event. Some examples are given below It is important to remember that life tables are used to summarise and make sense of this sort of data and as such can be applied to any duration data.
- duration of breastfeeding
- time to re-offending on release from prison
- time to first sex ("virginity table")
- birth spacing
- use-continuation of contraceptives
- cancer survival
Probation records
A criminologist could use a life table to make sense of data on lengths of time after a person is released from prison before re-offending occurs. In this case the start event is release from prison and the end event would be a first re-offence. The lx column would represent the numbers free from re-offending, from a radix of, say 1000, by numbers of months or years since release. The concept is exactly the same as for mortality and the life table would be constructed in the same way. In this example the duration would be measured in units of time, not of age.
Use continuation of contraceptives
The start event here would be starting to use contraceptives. In a study this would likely be the date of first prescription for pills, or date of insertion of IUD or implant. The end event would be a decision to stop use. In the case of an IUD this would be relatively easy to document as it would be date of removal. Duration would be in weeks or months, but not units of age.
This study would have some difficulties to overcome:
- For many forms of contraception you may not know a date of stopping use – the user will just stop and you will not see them again.
- If the idea is to investigate length of use related to side-effects how would you deal with durations where the reason for stopping is not side effects but because another child is wanted? This would be an example of censoring – where there is an end event but it is not one that you wish to count. The life table can cope with this but this is dealt with in a later session.
Cancer survival
The start event would be a diagnosis of cancer; the end event would be death. These analyses are routinely done by Cancer Registries internationally as many countries register cancer diagnoses and dates of death are usually known. An interesting consideration is whether you include a non-cancer death (e.g. car accident) or not. The period of exposure would always be included – it would bias a study if that were excluded - but whether all events (deaths) are used in the analysis, or only cancer deaths, will be the subject of discussion. Not including non-cancer deaths would be an example of censoringi.e. they would be included but as censored events rather than events of interest.
Very commonly in cancer studies all deaths are analysed as full terminating events and then death rates are compared with general population life tables to find the excess death rate of the cancer. This type of analysis is called relative survival.