Assumed or presumed?

Mortality modelling and research is often critically dependent upon assumptions, but certainty over whether those assumptions are well-founded may come only with hindsight. Human beings are prone to a number of biases, and checking if an assumption is appropriate means looking beyond our preconceptions and searching for corroborative evidence.

I was reminded of the importance of evidence whilst reading some mortality research from Olshansky, Carnes and Cassel (1990). This paper attempted something inherently tricky — it tried to construct a method for estimating plausible upper limits to human life expectancy. This paper has been cited many times, including somewhat negatively in a landmark paper we discussed earlier. It had a number of significant points to make, especially about reducing focus upon life expectancy as a metric and instead concentrating on the growth in the elderly population and the burgeoning issue of age-related morbidity. This challenge occupies many developed societies today and will continue to do so in future.

The method proposed by the paper was to compare the mortality improvements necessary to move life expectancy upwards to the "average biological limit to life (age 85)" with those that would arise from hypothesised cures for the degenerative diseases of aging, including cancer, diabetes and heart disease. The baseline life expectancies were taken from the US complete life tables from 1985. The paper sought to demonstrate that even such relatively improbable progress as a complete cure for one of these diseases would be insufficient to move life expectancy to the target age.

To accomplish this the paper built upon a life table cause-of-death elimination method developed by the demographer Chin Long Chiang. The paper attempted to work around dependencies between causes by clustering diseases into categories. We've talked about the challenges of working with cause-of-death data previously and won't consider those further here. What actually caught my eye was a statement about the estimation method used to calculate the mortality reductions necessary to target a given life expectancy:

It should be noted that the reduction of q(0) was not allowed to decline below one-half of that observed in 1985. This constraint was imposed to conform with the likely biological reality that infant and child mortality cannot be reduced below five to six deaths per 1000 live births.

In search of Methuselah: Estimating the upper limits to human longevity.
(Science - 02 Nov 1990)

The authors were clear that they do not expect the life expectancy at birth to be sensitive to their assumptions on child mortality. However, this statement remains a striking insight into the kind of preconception that may feed into the creation of a research method, or worse, into the formation of public policy. This assertion of biological "reality" was not disproven in the US in the intervening years, but data from the World Bank shows it was potentially questionable in Japan around the time the paper was published, and has been contradicted in numerous other countries over the following decades.

The infant mortality disadvantage in a US context remains the subject of active debate, and may ultimately prove to be more socioeconomic than biological. However, in a broader sense, the moral of the tale seems to be that deep expertise in one field need not translate into well-founded opinion in another. In essence, proposing biological limits to human mortality and lifespan is probably best left to biologists!