Modelling smoking prevalence with supercomputers

Dec 1, 2017 · 4 min read

Complex systems are an interdisciplinary subject of research and are used as the theoretical framework in a wide range of experimental fields such as economic and medical disciplines, as well as in mathematical and computer science studies.

A complex system is composed by a large number of elements that interact between each other.

The microscopic interaction among the elements gives rise to a global behavior of the system that can be very different, and at times even counterintuitive, with respect to the evolution rules of the single parts.

Leveraging on the complex systems approach, the ISI Foundation has developed a model that reproduces the adoption of and quit from the smoking habit in a population.

The model falls within the field of Global System Science (GSS) since it describes a large scale social system (in this case a whole country) to simulate its evolution in time.

Examples of Global Systems Science are the study of climate changes, financial stability and the description of new technologies adoption.

The simulations performed at the ISI Foundation enable the study of the process of adopting and giving up the smoking habit in Great Britain and have been performed using a High Performance Computing (HPC) infrastructure. In particular, the HPC supercomputers called Hazelhen at the HLRS in Stuttgart (Germany) and the Eagle at the University of Poznan (Poland) have been used to run the simulations.

The system description relies on an agent-based-model (ABM), i.e., each person in the population is represented as a computer variable and the behavior of each single element / agent in the system is simulated accordingly to the proposed model.

The ABM model is a modelling framework used in game theory, in computational sociology and in complex systems.

The model accounts for socio-economic indicators that measure the possibility to access tobacco products, thus simulating in a realistic way the behavior of the population of the country. In fact, the consumption of tobacco depends on several diverse factors such as social condition, age, economic factors etc. [ref. Psicol Soc (Bologna). Author manuscript; available in PMC 2014 Aprs 22. Italian.Published in final edited form as: Psicol Soc (Bologna). 2012; 2012(1): 7.30. Published online 2012. doi: 10.1482/36754 and others].

The HPC infrastructure allows for both a rigorous test of the model (by running a large number of different simulations) and for a detailed description of large socio-economic systems. These are two fundamental ingredients when dealing with GSS.

Indeed, GSS systems aim at the modelling of large population by precisely
describing each agent in the system and the interactions between agents.
However, the synthetic representation of the system under investigation quickly becomes large as the number of agents grows and cannot fit in ordinary computers.

By refactoring the simulation code making it HPC-compliant one can then share the system representation among many computing nodes. This results in a speed up the simulations and enables realistic description of large scale populations.

Finally, as the number of computing nodes increments, different scenarios can be tested at once by running simulations in parallel: this allows for a faster
calibration of the model and for a quick evaluation of the expected outcomes of different policies and interventions to be put in place.

The developed ABM model and the behavior of the population reproduced by the simulations have been validated against data of smoking prevalence in Great Britain from 1976 to 2016 (data from the National Health Service and the Office for National Statistics), showing a very good agreement between the empirical data and the simulations’ results.

The validation procedure against the statistical data has been carried out performing hundreds of simulations, with thousands of variables, on the HPC infrastructure.

After the validation phase, the resulting model closely reproduces the historical series of data, thus providing a prediction of the evolution of the smoking habit in Great Britain for the next years. Moreover, the model is able to forecast the smoking prevalence in a country based on the expected effectiveness of different policies regulating the tobacco consumption put in place by the government. Specifically, the smoking prevalence in Great Britain will have a decreasing trend in the next years. This trend can be either speeded up or slowed down by government interventions on the tobacco price or accessibility.

Enrico Ubaldi — ISI Foundation, Torino
Mario Scovazzi — CSP Innovazione nelle ICT, Torino

Originally published at coegss.

EXA — Future Global Systems

Reflections about the societal impact of High Performance Computing & Global Systems Science


Written by


CoeGSS is a European consortium of supercomputing centers, scientific institutions, businesses and NGOs providing support in the face of global challenges.

EXA — Future Global Systems

Reflections about the societal impact of High Performance Computing & Global Systems Science