From exposure to death: how the calculation works
It is a study that has attracted considerable attention. As part of the Global Burden of Disease Study (GBD), a project led by the IHME (Institute for Health Metrics and Evaluation, University of Washington), whose estimates are notably used by the WHO when stating that “a given cause kills a certain number of people worldwide,” second-hand smoke was estimated to have been responsible for 1.66 million deaths worldwide in 2023. The figure has been widely reported by media outlets around the world. Here, we take a closer look at the methodology used to produce it.
Published on September 7, 2026, in the scientific journal The Lancet Public Health¹, the research is the work of more than 1,000 researchers. As is often the case with research on smoking, it was funded by Bloomberg Philanthropies and the Gates Foundation, two organizations also known for funding numerous studies critical of vaping.
Methodologically speaking, there is little point beating around the bush: this research is about as state-of-the-art as it gets for this kind of exercise. Its limitations do not stem from any lack of rigor on the part of the researchers, but simply from the fact that it is impossible to do much better. The aim was to determine the number of deaths caused by second-hand smoke over an entire year and across 204 countries. Since it is not possible simply to count all death certificates recorded around the world, modelling is required, and with it comes an unavoidable degree of uncertainty.
To arrive at their result, the researchers first estimated the number of people exposed to second-hand smoke worldwide: 2.71 billion, including 767 million children. This figure is itself an estimate based on numerous factors.
They then estimated the number of deaths caused by second-hand smoke from the number of people exposed. This was done using a standard epidemiological method that combines two pieces of information: relative risk (RR) and exposure prevalence.
Relative risk is derived from decades of previous research. For each disease associated with second-hand smoke — heart disease, stroke, lung cancer and others — it indicates how much more likely an exposed person is to die from the disease compared with someone who is not exposed.
By combining this relative risk with exposure prevalence (P), a standard statistical formula can be used to calculate the proportion of deaths from a given disease that is attributable to second-hand smoke. This is known as the population attributable fraction (PAF).
This percentage is then applied to the total number of deaths from that disease, which has already been estimated independently by the GBD programme, to obtain the number of attributable deaths. The calculation was repeated for every disease, in each of the 204 countries studied, and by age group and sex, before all the results were added together to produce the final figure: 1.66 million (95% UI 1.33–2.07). In other words, there is a 95% probability that the true number of deaths attributable to second-hand smoke in 2023 lies between 1.33 million and 2.07 million.
For this study, the researchers used a Bayesian uncertainty interval (UI) rather than a frequentist confidence interval (CI). The distinction between the two is subtle. A confidence interval (CI) means that if the same study were repeated an infinite number of times, 95% of the intervals produced would contain the true value.A Bayesian credible interval is easier to interpret: it simply means that there is a 95% probability that the true value lies within the reported range.
The limitations of the method
Of course, this type of study has a number of limitations. The main ones are as follows.
First, the researchers used what could be described as an “all-or-nothing” model. The intensity of second-hand smoke exposure was not taken into account, and each person was classified as either exposed or unexposed. As a result, a child whose parent systematically smokes at the dinner table was classified as exposed in the same way as a child whose parent only smokes by an open window or on a balcony.
Public places were also excluded. The research measured exposure only at home and in the workplace. Bars, restaurants and other public places where smoking is still permitted in some countries were not included.
In addition, the model applied exposure levels from 2023 to mortality rates from that same year. This poses a problem because some diseases associated with second-hand smoke, such as lung cancer, can take several decades to develop after exposure. The study therefore attributed the current disease burden to present-day exposure, even though for several diseases that burden actually reflects exposure in the past.
Finally, almost every new edition of the GBD changes its methodology. It is therefore not possible to directly compare the results of this study with those from previous years. That said, such a comparison would not have been possible anyway, since this is the first time second-hand smoke has been studied as an independent risk factor.
Sources and references
1. GBD 2023 Second-hand Smoke Collaborators. (2026). Global, regional, and national prevalence of second-hand smoke and attributable disease burden in 204 countries and territories, 1990–2023: A systematic analysis for the Global Burden of Disease Study 2023. The Lancet Public Health. https://doi.org/10.1016/S2468-2667(26)00168-4
