Laboratory

Statistical Demography

At a Glance Projects Publications Team

Project

Mortality at Extreme Ages and Potential Limits to Human Lifespan

Jutta Gampe, Marie Böhnstedt, Hein Putter (Leiden University Medical Center, Netherlands), Nadine Ouellette (University of Montreal, Canada); in Collaboration with Anthony Davison (École polytechnique fédérale de Lausanne, Switzerland), Holger Rootzén (Chalmers University of Technology, Gothenburg, Sweden), Léo Belzile (University of Montreal, HEC Montréal, Canada)

Detailed Description

The impressive gains in life expectancy made in recent decades directed scientific and public attention to lifespan measures beyond mean longevity. This includes the question of whether there is a hard limit to human lifespan or not.

Controversial claims have been made, and most of them are based on the statistical analysis of observed deaths at extreme ages. Being sparse by definition, the analysis of such data requires particular care, and the correct handling of observed extreme lifetimes is vital to perform adequate inference. Intrinsically linked to the question of a potential lifespan limit is the accurate estimation of the mortality trajectory at high ages. Ultimately, this trajectory determines whether the support of the lifetime distribution is bounded or unbounded.

This project addresses several aspects of statistical methodology in this context. One is the proper analysis of data on semi-supercentenarians (aged 105 or above) and supercentenarians (aged 110 or over) from the International Database on Longevity (IDL). Inference must respect the diverse observation schemes in different countries of the IDL; these lead to complex truncation and censoring patterns that, when ignored, would bias the results. Updated estimates from the IDL replicated earlier results: Mortality after age 110 is basically constant, although at a high level.

Combining methodology from survival analysis and extreme value modeling and using recent data sources from several countries, an extended study concluded that after age 109 human mortality most likely is constant, indicating an unbounded support of the lifespan distribution. Point estimates for the lifespan distribution typically indicate no upper limit, and when considering the involved uncertainty (95% confidence intervals), the lower interval boundaries are about 130 years, well beyond the largest age-at-death yet reliably recorded.

Death rates at high but not extreme ages have been found to increase more slowly than the exponential (Gompertz) trajectory found from mid-adult life onwards. Such mortality deceleration necessarily results in heterogeneous cohorts due to selection and would blend with a constant rate at the extreme ages. While theoretically compelling, it repeatedly has been contested empirically, not least because standard techniques tend to favor models without deceleration. We investigated the performance of statistical techniques for assessing mortality deceleration, including issues of study design, such as sample size and the age range covered by a dataset. We also have shown that model selection tools (Akaike’s Information Criterion) need to be corrected for the gamma-Gompertz model, a popular choice when investigating mortality deceleration. In addition, Focused Information Criteria for mortality deceleration were developed theoretically and their properties investigated in empirical studies.

Publications

Böhnstedt, M.; Gampe, J.; Putter, H.:

Information measures and design issues in the study of mortality deceleration: findings for the gamma-Gompertz model

Lifetime Data Analysis 27:3, 333–356. (2021)    

Gampe, J.:

Mortality of supercentenarians: estimates from the updated IDL

In: Exceptional lifespans, 29–35. Cham: Springer. (2021)    

Böhnstedt, M.; Putter, H.; Ouellette, N.; Claeskens, G.; Gampe, J.:

Shifting attention to old age: detecting mortality deceleration using focused model selection

arXiv e-prints 1905.05760. unpublished. (2019)