Data analysis
The primary analyses were conducted on an intentionto-treat basis. Missing data were dealt with using various forms of imputation: the common one of assuming that they are smokers, but also assuming that they are quit,using their last known status, and analysing only cases with identified outcomes [15]. At 1 month, one case was excluded from the outcome analyses due to hospitalization. At 7 months, two participantswere reported to have died. This resulted in a final sample for analysis of 3529 at 1 month and 3528 at 7 months. Initial analyses compared the integrated and choice conditions separately. Where these two groups were equivalent, we conducted 2 ฅ 2 analyses (Table 1). We
also compared the four intervention conditions combined with the control group to determine an overall intervention effect. Logistic regression was used to examine differences by condition, controlling for demographics, recruitment source, cigarettes per day and baseline smoking status and, in subsidiary analyses, use of medication. To estimate more clearly the effectiveness of the interventions when used, the following methods were used. We estimate the proportion of the control group that would have taken up an intervention if offered, and the proportion that would not. Cessation rates for the latter are assumed to be the same as for the subset who did not take up the interventions in the intervention groups (Ir). We also know the observed cessation rate in the entire control group (Ct), and what proportion of participants took up an intervention (Pu). From this we can calculate the predicted cessation rate among those who would have taken up the intervention if offered (Eu). Eu =[Ct - Ir(1-Pu)]/Pu. We used the overall outcomes averaged across all interventions for this estimation.