Bayesian spatio-temporal modelling of tobacco-related cancer data in Switzerland

Tobacco use is the leading cause of preventable death worldwide. Each year the tobacco
epidemic accounts for 6 million deaths and costs hundreds of billions of dollars to the economy.
Cigarette smoking accounts for more deaths than AIDS, murder, legal and illegal
drugs, road accidents and suicide combined. Around 85–90% of all lung cancer deaths
are estimated to be attributed to active or passive smoking. In Switzerland, lung cancer
is the first cause of cancer mortality in men and second in women (after breast cancer).
Gender-specific smoking patterns differ essentially in time as well as in space. In the 19th
and the beginning of the following century, smoking was restricted to the male population,
finding its peak in the 1970s in most European countries. In the past, the image of female
tobacco use experienced an essential turn. In the middle of the 20th century the smoking
prevalence among women increased due to the changes in gender roles and the subsequent
effect on female smoking reputation. Before, female smoking had not been socially accepted.
After strong gender-related developments, female smoking was associated with
independence, emancipation and freedom. This movement was exploited to a great extent
by the tobacco industry by adjusting their marketing strategies regarding this new target
audience. In many developed countries the gap between gender and smoking prevalence is
closing since the last decades, as males are smoking less, while female tobacco smoking is
increasing steadily.
Information on spatial as well as temporal patterns and trends of a disease are essential
for health planning and intervention purposes. The Swiss Federal Office of Public Health
(FOPH) has launched the National Programme Tobacco 2008–2012 aiming to reduce the
proportion of smokers, targeting a decline of tobacco-related morbidity and mortality in the
country as a final result. Cancer mapping visualizes geographical and temporal patterns
and trends. Maps of estimated mortality serve as helpful tools to identify high risk areas
and therefore enable focused intervention planning at a higher geographical scale than the
national one.
Disease maps of crude rates can be non-informative and might even lead to misinterpretation,
as rare diseases or small populations might dominate the map and result in large
variability in the estimated rates. Distinction between chance and real difference of the
obtained variability is challenging. Spatial modelling of the rates enables the assessment
of covariate effects to explain observed patterns and highlight them by obtaining smooth
maps. Bayesian methods are the state-of-the-art modelling approach for spatio-temporal
analysis. They allow flexible modelling and inference and provide computational advantages
via the implementation of Markov chain Monte Carlo (MCMC). Model formulations
improve the estimates sparse, unstable rates by borrowing strength from their neighbours.
In addition, they allow risk factor analysis which takes into account potential spatial correlation.
Apart flexible modelling, Bayesian inference provide computational advantages
via the implementation of Markov chain Monte Carlo simulation methods.
This thesis aimed (i) to assess geographical differences and trends of age- and gender-specific
lung and all tobacco-related cancer mortality in Switzerland; (ii) to project tobacco-related
cancer mortality in Switzerland at different geographical levels accounting for spatial
variation; (iii) to develop Bayesian age-period-cohort (APC) models for projecting
cancer mortality data; (iv) to develop Bayesian back-calculation models to estimate age- and
gender-specific incidence from sparse mortality data; and (v) to develop models to
indirectly approximate gender-specific smoking patterns in space and time by unadjusted
and adjusted lung cancer mortality rates with non-smoking risk factors.