IRVINE , CALIFORNIA Delivery by Hand The Honorable

-Questionnaire data obtained from 1,631 tenth quent smoking. That both cigarettes and SLT are tobacco grade students in 14 school districts in the State of Washington products and contain absorbable nicotine indicates strongly are used in this investigation ofthe relationship between the onset that an associated nicotine dependence may result in indiprocesses for smokeless tobacco use and smoking. Emphasized is viduals taking up one when quitting the other to maintain the use of time-to-event data on the ages of occurrence of six events in these onset processes . Concepts and methods for the a habituated nicotine level. statistical analysis of time-to-event data are demonstrated. The It is clear that the concurrent use of cigarettes and SLT occurrence of events in the smoking onset process are strongly are associated {3-6). The purpose of our investigation is related to increases in the subsequent onset rate for smokeless to use data on the ages at 'which young people begin to tobacco use. Compared with before initial smoking has occurred, smokeand use SLT to examine the relationship between the the onset rates for weekly smokeless tobacco use after initial smoking andSLTonset processes . Some basic concepts and ---smoking--has-occurred-are-2-03~P'G1}-and-6:7~-FPE:80ij---~netho$sforthe-stali'fti i lys>s o eo-even a a are times as large for males and females, respectively. Furthermore, both initial and weekly use of cigarettes contributes to the risk of subsequent weekly smokeless tobacco use. Conversely, the steps in the onset process of smokeless tobacco use are strongly related to increases inthe subsequent smoking onset rate. Possible implications for intervention in prevention of smokeless tobacco use and for further research are discussed.-NCI Monogr 8:6369, 1989 . The use of SLT among adolescents, especially snuff among adolescent males, has skyrocketed in recent years (1). This development has ominous health implications, because SLT contains known carcinogens and because a growing body of epidemiologic evidence indicates that its use carries the risk of various adverse health effects including oral cancer (2). Scientists are expending considerable effort to establish the circumstances and factors related to SLT use and its onset process among youth and to incorporate these findings in the designs of effective prevention programs. Determining the relationship between smoking and the use of SLT is important for their investigation of 1) smpking as a possible risk factor for die onset of SLT use and 2) the extent to which such use is associated with subseABBREVIATION: SLT = smokelesstobacco. 'Supported in part by Public Health Service grants CA-38269 and CA-34847 from the National Cancer Institute, and GM-24472 from the National Institute of General Medical Sciences, National Institutes of Health, Department of Health and Human Services. 'Division ofPublic Health Sciences, Fred Hutchinson Cancer Research Center, 1124 Columbia St, Seattle, Washington 98104. Addrsrs reprint re9u= to ArthurV . Peterson, Ph.D, Cancer Prevention Research Program UP603 . 'Department ofBiostatistics~ University ofWashington, Seattle. a We thank the entire data collection and management staff of the Hutchinson Smoking Prevention Project for their contributions to the study, Bobbie Nielsen for preparation of the manuscript, our colleagues who provided comments on the manuscript, and the personnel of the Participating school districts for their collaboration and cooperation with the Hutchinson Smoking Prevention Project NOTICE : This material may be profeded by copyright law (Title 17 U.S. Code) demonstrated The concept of a smoking onset process; i.e ., a series of events that describe an increasing level and/or frequency of cigarette use, has been advanced by Flay et al. (7), Leventhal and Cleary (8), Hirschman and co-workers (9), and others. Our description of such a process consists of 1) Specifying meaningful events of tobacco use and 2) determining transition rates between the events with data on the times (ages) at which the events occur. The investigations reported here are restricted to such events ; other important aspects of the smoking onset process, such as social influences, the environment, and motivation, will be added in subsequent investigations. We used data on the ages of occurrence of six tobacco use events. initial, tenth, and first weekly use of SLT, and initial, tenth, and first weekly use of cigarettes . METHODS Survey procedures--Tobacco use, including cigarette smoking and SLT, was assessed through a questionnaire administered in the classroom to entire grades of tenth-grade students in 14 rural and suburban school districts in the State of Washington in January 1986 . Through an informational letter to parents and by in-class procedures, parents and students were fully informed in advance and were given an opportunity to ask questions and to decline to participate. Thetobacco survey was part of a baseline assessment of tobacco use among students in school districts participating with the Fred Hutchinson Cancer Research Center in the Hutchinson Smoking Prevention Project, a long-term, randomized controlled trial in school-based smoking prevention. Of the total enrollment of 2,214 tenth graders, 1,91 .8 (87%) took part in the survey . Twelve percent were absent from class; 0.2% (parents) and 0.8% (students) declined participation. Data for all questionnaire items, pertinent in this investigation are available on 1,631 students. All results reported below are based on analyses of data from 1,631 students (840 males, 791 females). 63 Measures we used to enhance the accuracy of the responses to the questionnaire items included 1) administration of the questionnaire on an unannounced date; 2) procedures to maintain confidentiality and assurances about these to the parents and students; 3) classroom procedures designed to maintain and demonstrate confidentiality, including the use of study identification numbers and the handling of questionnaires by project data collectors only ; 4) explanation and collection of saliva samples from all participating students concurrent with administration of the questionnaire ; and 5) explanation of the data collection objectives and the important role of the students in achieving them. The questionnaire included items that assessed various aspects of current, past, and future intended use of cigarettes and SLT products. The wording for questions and multiplechoice responses for cigarettes was similar to that for SLT, so that differences in patterns for smoking and SLT use could be ascertained without confounding from differences in the nature or wording of the items. Analysis.-For binary data items (e.g., whether a certain level of past or current use of tobacco was achieved), simple proportions (prevalences) are reported . Data on time-to-smoking and time-to-SLT-use are analyzed by standard time-to-event statistical methods (survival analysis methods) that accommodate data on individuals for which the event (smoking, SLT use) has not occurred (censoring). Two time-to-event statistical methods are used: 1) Kaplan-Meier survival curves provide a descriptive display of time-to-event data (e g., age at initial smoking) obtained on a set of individuals. When data on time-to-event are completely available for all participants, then this curve (at any age t) is simply the fraction of individuals whose observed times-to-event are greater than t (e.g., a fraction of individuals who have not smoked at age t). The Kaplan-Meier curve can also accommodate the situation characteristic of time-to-event data, when, for some individuals, the age at initial smoking is not known, but only that no smoking has occurred y a aFrtam age e.g., e age a whil;M lion occurs). Mathews and Farewell (10), Lawless (11), Miller (12), and Kalbfleisch and Prentice (13) provide further descriptions of the Kaplan-Meier survival curve, including formulas for its computation and assumptions for its use. 2) We used the Cox regression method (10-14) to analyze the impact of the occurrence of one tobacco use event (e.g ., initial smoking) on the subsequent rate of onset of another event (e.g., weekly use of SLT). By such analyses, one can investigate directly the interrelationship between the smoking and SLT use onset processes. The Cox regression method models the onset rate X(t) for some specified event (e.g., weekly use of SLT) as a function of the follow-up time t (e.g., age). The model specifies that the onset rate X(t z) for any individual with expl4natory (regression) variables z1, z2, . . . zp is just the product of a "baseline" onset rate ko(t) and .a PETERSON ET AL function g(zp) of the covariates, often taken to be the exponential function g(zfl) = exp(z#): a(rz) = hu0) exp(zfl), where XD(t) > 0 is a completely unspecified baseline onset rate, z = (zl, . . . zp) is a regression vector consisting of the p explanatory variables, and p, A, . . . Pp) is a vector of regression coefficients to be estimated from the data. The Cox regression model offers a number of desirable features and improvements over more traditional methods that make it particularly helpful in investigations of onset processes, such as those of smoking and SLT use. 1) The age-specific onset rate A(t), a meaningful measure of smoking onset as a function of age, is modeled directly. 2) No assumption is made about the shape or magnitude of the onset rate as a function of age. It is data determined . 3) The quantities exp(P1), . . . . . for which estimates are readily obtained by the usual partial likelihood analysis, have the useful interpretation of rel. ative onset rates, e.g., the estimated smoking onset rate for prior SLT users relative to that for prior nonusers. 4) As in other regression models, the effect of other variables can be conveniently controlled by their inclusion as covariates in the regression model. 5) Unlike binary data methods, the model and analysis can accommodate censored data. 6) The model can be generalized in numerous ways for adaptation to a wide range of applications . . Used in this paper . i s a generalization that allows an explanatory variable to depend on follow-up time t. in our application below, we let z1 =zl (t) depend on the follow-up time (age) and define it to. be the indicator function for the occurrence of a specified prior event (e.g., the occurrence of initial smoking), taking the value 0 before the event occurs and 1 afterward . The quantity exp(f31 ) is interpreted as the relative onset rate (e.g., of weekly SLT use after the prior event of initial smoking compared with before). See the references above or a compete description o s m its generalizations and assumptions, and method of analysis. RESULTS Prevalence of Smoking and Smokeless-Tobacco Use Table 1 presents, first for cigarettes and then for SLT, the fraction of boys and girls who have ever used, currently use, and have attained certain events of the onset process. The percentage of males who have ever smoked cigarettes is about the same as have ever used SLT. More boys have dipped more than five pieces of SLT than have smoked more than five cigarettes (40.6% vs. 32.5%). Almost 70°k of the females have ever smoked cigarettes and about 31W, have ever used SLT; almost 6% of the girls have used more than five pieces of SLT. More boys are currently using SLT than cigarettes: 17.7 1A versus 14.4% (weekly use). Although 19.4% of the girls NCIMONOGRAPHS, NUMBER 8, 1989 TABLE 1.-Current use, lifetime use, and attainment of onset events: cigarettes andSIT