Appropriate and inappropriate methods for investigating the “gateway” hypothesis, with a review of the evidence linking prior snus use to later cigarette smoking

BACKGROUND: The “gateway hypothesis” usually refers to the possibility that the taking up of habit A, which is considered harmless (or less harmful), may lead to the subsequent taking up of another habit, B, which is considered harmful (or more harmful). METHODS: Possible approaches to designing and analysing studies to test the hypothesis are discussed. Evidence relating to the use of snus (A) as a gateway for smoking (B) is then evaluated in detail. RESULTS—DESIGN AND ANALYSIS CONSIDERATIONS: The importance of having appropriate data available on the sequence of use of A and B and on other potential confounding factors that may lead to the taking up of B is emphasised. Where randomised trials are impractical, the preferred designs include the prospective cohort study in which ever use of A and of B is recorded at regular intervals, and the cross-sectional survey in which time of starting to use A and B is recorded. Both approaches allow time-stratified analytical methods to be used, in which, in each time period, risk of initiating B among never users of B at the start of the interval is compared according to prior use of A. Adjustment in analysis for the potential confounding factors is essential. RESULTS—REVIEW OF EVIDENCE: Of 11 studies of possible relevance conducted in Sweden, Finland or Norway, only one seriously addresses potential confounding by those other factors involved in the initiation of smoking. Furthermore, 5 of the 11 studies are of a design that does not allow proper testing of the gateway hypothesis for various reasons, and the analysis is unsatisfactory, sometimes seriously, in all the remaining six. CONCLUSIONS: While better analyses could be attempted for some of the six studies identified as having appropriate design, the issues of confounding remain, and more studies are clearly needed. To obtain a rapid answer, a properly designed cross-sectional survey is recommended.