There were two substantial changes in the estimation of second-hand smoke compared with GBD 2015. First, we estimated the proportion of a population exposed to second-hand smoke using information about household composition and smoking status from household surveys and censuses, rather than using questions that ask directly about exposure to second-hand smoke in surveys. Second, we modelled exposure using spatiotemporal Gaussian process regression (ST-GPR), borrowing strength across sexes and all ages, whereas in GBD 2015 we ran a DisMod model separately by sex and age. Further, we found significant evidence of associations between second-hand smoke exposure and two additional outcomes: breast cancer and diabetes, which were added to the list of risk-outcome pairs for second-hand smoke. More details on the estimation approach are presented in appendix 1 (p 98).
