Optimal statistical data analysis techniques have not been developed for smoking cessation studies with smoking outcome (abstinence versus smoking) as the response variable. The situation is exacerbated when grouping effects are present as when groups of smokers receive treatment together. It is proposed to test the suitability of a parametric limited failure population (LFP) survival analysis model and an accompanying bootstrap estimation procedure to model such data via a large-scale simulation study and analyses of prospective smoking cessation data currently being collected. The simulations will systematically investigate the properties of the LFP model and the proposed bootstrapping variant. Comparison of the analysis of interval follow-up data with that from continuous data will determine the extent of inefficiencies arising from the use of interval data. Analysis of results from varying numbers of follow-ups will determine how information about parameter values is lost by using a smaller number of follow-ups. An analysis will also be carried out on the optimal spacing of follow-ups when the number of follow-ups is specified in advance. The effects on parameter estimation of stopping the study at different times (such that many cases of recidivism are not observed) will be observed, as well as the impact of different subject drop-out rates. In terms of the presence of grouped data, the implications of various group sizes, numbers of groups, and degrees of correlated outcome within groups will be examined. This will yield information on the necessity and appropriateness of the bootstrap estimation procedure. The degree of bias associated with statistics that fail to take into account correlated outcomes will be studied. The analysis of prospective data will assist in determining the generalizability of the LFP model to actual smoking cessation data. The overall results should offer significant assistance to applied researchers in the design and evaluation of smoking cessation experiments.
