Background Numerous studies have reported on the impact of comprehensive smoke-free laws on population health. Many early studies have ignored the potential effect of the long-term trend of the health outcome, and when included, subsequent studies have focused on either linear or non-linear trends. However, the choice of appropriate trend is not always straightforward. We illustrate this by investigating the short-term impact of smoke-free legislation in England, introduced on 1st July 2007, on myocardial infarction mortality.
Methods We investigate the impact of the legislation using weekly counts of all cases aged 18 years or older residing in England with a primary cause of death of a myocardial infarction (ICD–10 I21) between July 2002 to December 2010 (providing 5 years pre-legislative and 3 years and 6 months post-legislative data). We compare a number of models based on an interrupted time series design with a quasi-Poisson generalised additive model that adjusts for seasonality and long-term trends.
Results Myocardial infarction mortality shows a marked decline over the study period. We identify two competing models. The first shows evidence of a complex interaction between the introduction of smoke-free legislation and the long-term trend. We observe an initial statistically significant reduction in mortality (–8.5%, 95% CI –11.1% to –5.8%) coupled with a change in the long-term trend from a reduction of 4% over a six month period to a reduction of 3.5%. The second model fits a nonlinear trend and shows no significant smoke-free effect. Both models offer an almost identical fit.
Conclusion Investigating small effects in the presence of a pronounced long-term trend is complicated by the limitations of the available data. In particular, it is not clear whether we observe a gradual change in the long term trend or a discrete effect directly attributable to the legislation. The two models have near-identical fitted values and GCV scores, but have very different interpretation. We conclude that the data alone are insufficient to distinguish between the two models and warn that overly-simplistic analyses in such situations may result in misleading conclusions.
Statistics from Altmetric.com
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.