Department of Public Health and Epidemiology, University of Birmingham, Birmingham B15 2TT, United Kingdom. Published online August 23, 2001

Transcription

1 Preventive Medicine 33, (2001) doi: /pmed , available online at on The Change-in-Stage and Updated Smoking Status Results from a Cluster-Randomized Trial of Smoking Prevention and Cessation Using the Transtheoretical Model among British Adolescents Paul Aveyard, M.P.H., 1 Emma Sherratt, Ph.D., Joanne Almond, M.Sc., Terry Lawrence, M.Sc., Robert Lancashire, B.A., Carl Griffin, M.Sc., and K. K. Cheng, Ph.D. Department of Public Health and Epidemiology, University of Birmingham, Birmingham B15 2TT, United Kingdom Published online August 23, 2001 The stages of change model underpins the British Gov- Background. The transtheoretical model (TTM) and ernment s creation of a network of smoking cessation computer technology are promising technologies for clinics [1]. However, as it is often used, the stages of changing health behavior, but there is little evidence change model represents little more than a commonof their effectiveness among adolescents. sense approach that recognizes that some people are Method. Four thousand two hundred twenty-seven not yet ready to change their behavior, so detailed dis- Year 9 (ages 13 14) pupils in 26 schools were randomly cussion of the strategies of change with them is misallocated to control and 4,125 in 26 schools were alloplaced. In contrast, the transtheoretical model (TTM), cated to TTM intervention. TTM pupils received three from which the stages are derived, makes specific prewhole class lessons and three sessions with an interdictions [2, 3]. Ten processes drive movement through active computer program. Control pupils received no special intervention. Positive change in stage and the stages, according to the theory. Use of the correct smoking status was assessed from a questionnaire comthe next stage. Nonuse of the correct process or use of process in the correct stage will facilitate movement to pleted at baseline, 1 year, and 2 years. Random effects logistic regression was used to compare the change in other processes that are helpful in preceding or later stage and smoking status between the arms. stages will hinder progress to the next stage. Results. Eighty-nine percent of the TTM group and In 1999, we reported the results of a clinical trial of a 89.3% of the control group were present at 1-year and computer program for adolescents [4]. The intervention 86.0 and 83.1%, respectively, were present at 2-year fol- consisted of three whole class lessons and three sessions low-up. The adjusted odds ratio (95% confidence inter- using a computer program during school year 9, when val) for positive stage movement in the TTM relative the adolescents were ages The computer proto control was 1.13 ( ) at 1 year and 1.25 (0.95 gram consisted of video clips of young people discussing 1.64) at 2 years and for regular smoking was 1.14 (0.93 their smoking, followed by questionnaires. The re- 1.39) at 1 year and 1.06 ( ) at 2 years. Subgroup sponses to the questions were used to derive the young analysis by initial smoking status revealed no benefit person s stage of change with respect to either acquisifor prevention or cessation. tion of smoking or cessation. Further questions related Conclusions. The intervention was ineffective to the other core concepts of the transtheoretical model: American Health Foundation and Academic Press Key Words: randomized controlled trial; smoking preeach questionnaire, young people using the program decisional balance, temptations, and processes. After vention; smoking cessation; adolescence. were given feedback about how their process use compared with others in their stage and what to do to move INTRODUCTION to the next stage, for example. On the second and third The stages of change model is the most used model occasions, students received feedback on progress since of behavior change in British health promotion practice. the last occasion. The control group received standard lessons on smoking, which is part of the English national curriculum. One year after the program began, 1 To whom reprint requests should be addressed. Fax: p.n.aveyard@bham.ac.uk. about 4 months after the last intervention in the TTM /01 $35.00 Copyright 2001 by American Health Foundation and Academic Press All rights of reproduction in any form reserved.

2 314 AVEYARD ET AL. arm, we assessed the outcome with a paper question- individual may have made progress toward stopping, naire. For those who were followed up, the rate of smoking but not yet stopped. A second way would be to examine in the control group was 18.8% and in the interven- the smoking cessation rates a longer period after the tion group was 17.5%: the difference (95% confidence intervention finished. In this report, we report both intervals; 95%CI) was 1.3% ( 1.9 to 4.9%). Neither sen- stage of change at 1 year and smoking status and stage sitivity analysis for loss to follow-up nor adjustment for of change at 2 years, about 16 months after the final baseline smoking status and other potential confound- interventions to achieve both these aims. No trials of ers changed this. The intervention was ineffective. TTM-based interventions have reported outcomes at 2 One conclusion from this would be that the transtheo- years among adolescents. retical model is an invalid model. Another is that although Prochaska, one of the originators of the TTM and a the model is valid, the intervention was not codesigner of the intervention we tested, suggested that powerful enough, or perhaps badly configured, so that the intervention had not worked because the young it did not work. A third might be that the intervention people had not had enough of it [8]. Our previous analysis was working, but that our analysis did not detect this was on the basis of intention to treat, which meant because we had not waited long enough to see it. that the outcomes from adolescents who had no or few Stapleton et al. showed that nicotine replacement therapy interventions were included. Although intervention and advice to smokers produced a 16% difference rates were high, it may be that this method of analysis in the cessation rate in favor of the intervention when was obscuring evidence that more intervention was bet- assessed at 1 year [5]. However, 2 years later, this differ- ter and in this report we examine trial data on this. ence had declined to 9%. From the perspective of the TTM, this apparent decline in effectiveness is understandable. METHOD Since an intervention suitable for only those The full text of the previous report and additional in the preparation stage, a stage probably occupied by material are freely available at fewer than 20% of participants, was used, these individ- reprint/319/7215/948 [4]. In brief, we approached 89 uals will have made rapid progress to stopping. However, randomly selected schools from the West Midlands and the majority of participants for whom the inter- 53 (60%) agreed to participate. Twenty-seven schools vention was misplaced would not have made immediate were randomly allocated to the TTM intervention and progress, but would have worked through the stages 26 to the control. One school in the intervention group more slowly, catching up to those accelerated by the dropped out after randomization and before the study action-oriented intervention, tending to equalize the commenced. All Year 9 pupils (ages years) in quit rates in each arm. participating schools were invited to participate. Four Prochaska et al. s trial contrasts with this [6]. Partic- thousand one hundred twenty-five students (93% of all ipants were randomized to standard self-help materials on the school register) were enrolled in the intervention or a TTM-derived computerized intervention similar to group and 4,227 (91% of all on the school register) were that used in our trial with adolescents. At 6 months, enrolled in the control group by completion of the base- the difference in quit rates was 9%, but at 18 months, line questionnaire. This questionnaire was administered this difference was 14%, even though there were no by trained staff who read standard instructions, interventions between 6 and 18 months. This raises the and pupils completed the questionnaire under examination possibility that individuals in the TTM arm of our study conditions. The questionnaire measured smokpossibility were still working through the stages using lessons ing status, stage of change, and potential confounders, learned during the intervention, and examining the i.e., risk factors for smoking in the future. There were outcome in terms of smoking status at 1 year therefore no large differences between the groups in these predictors did not capture this. However, when Prochaska et al. s of smoking [4] at the baseline. Follow-up ocdid trial was analyzed assuming that those lost to follow curred 1 and 2 years after the program began, i.e., in up were smokers, then the difference between the arms Year 10, when pupils were ages 14 15, and in Year was small and not significant [7]. There is therefore 11, when they were ages The same personnel weak evidence currently that stage-based interventions administered an identical questionnaire in identical move people through the stages after the intervention fashion as at baseline, with identity numbers being ceases. Notwithstanding the paucity of evidence, however, used to link pupils through time. At 1-year follow-up, it remains possible that participants in our trial 7,444 (89.1%) pupils were present and smoking status were working through the stages of change because the could be allocated to 7,413 (99.6%) of those followed up. follow-up was, in some cases, only about 4 months after At 2-year follow-up, two control schools refused permission the final intervention. We may have assessed the outcome to administer the questionnaire because of concern too soon to see the effect. One way to examine if about the time taken. This was the main public examination this were so would be to examine differences in the year. These schools had 136 and 153 pupils enthis stage movements between the two arms at 1 year. An rolled in the trial. Not counting these in the percentage,

3 TTM-BASED SMOKING PREVENTION AND CESSATION 315 6,819 (84.6%) original participants were present at 2- describe their decisional balance. Three lessons and year follow-up. Smoking status was allocated to 6,782 three computer sessions were used because we felt this participants (99.5% of those followed up). Many (45.8%) was at the limit of what could be implemented in English absent from 1-year follow-up were present at 2-year schools if the intervention were implemented outabsent follow-up, suggesting that the main reason for loss of side of a trial. One commentator criticized the intervention follow-up was nonattendance at the particular lesson for being too intense to generalize beyond the when the questionnaire was administered. trial [9]. The Interventions Process Assessment Full details of how we assessed the process and the Our intention for control group schools was that they results were published in our previous report [4], but would receive no intervention. However, the content the method of process assessment was confined to and timing of lessons in England is controlled by the the version available on the World Wide Web only Government through the National Curriculum, so that ( Concerning all children receive smoking-related education by statthe whole class lessons, we did not obtain data on pupils ute as part of science at Key Stage 2 (ages 7 11) and attendance or reactions to the lessons. Like the com- Key Stage 4 (ages 14 16). In addition, all English state puter sessions, the lessons were integrated into the schools provide personal, social, and health education curriculum, so that attendance at the lessons is likely throughout all key stages, for which Government guidto have been similar to attendance at the computer ance, but not mandatory instruction, is given to teachsessions. Teachers reported on their delivery of the lesers. This guidance suggests that these lessons should son, how they felt pupils understood the material, and cover areas such as self-awareness, self-esteem, making how they felt pupils enjoyed the lesson. About half of healthy choices, the recognition of pressure from others schools gave these data, giving a mean score of about and where to get help, and the rules on the use of 4/5 for each of these. Concerning the computer sessions, tobacco in school (it is usually banned on school premwe measured pupil attendance, whether pupils took ises). Therefore all English state schoolchildren have long enough to read all the information presented, and education that might be regarded as smoking prevenpupils reactions to the computer session. Attendance tion. As an acknowledgment for the help teachers in data are reported in Fig. 1. Overall, 68.7% of baseline control schools gave us, we distributed three lesson regular smokers and 78.8% of baseline non-regular plans on smoking for personal, social, and health educasmokers used the computer three times as intended. tion, which were already available from teaching re- Approximately 70 80% of computer sessions lasted sources. Two of these consisted of quizzes and group long enough to read all the material presented, though work on the health effects of smoking, and one lesson this declined slightly from first to third use. Almost all concerned different methods of persuading someone to pupils rated the computer program easy to use and stop smoking, unrelated to the TTM. We have no data on nearly all thought it interesting on first use. About whether these plans were used nor how much smoking % thought the sessions valuable or useful, though related education participants in the control or interas with ratings of interest, this declined with repeated vention schools received during the 2 years of the trial use to around half to two-thirds on third use. Baseline or throughout their education, though it would vary smokers gave lower ratings than nonsmokers. No qualifrom school to school. tative or other data on process were collected. The hypothesis we were testing in this trial was whether the addition of a specific TTM-based intervention to the health education already provided by schools Outcome Assessment would protect young people from becoming smokers and In this report we examine the outcome of a positive help those who smoked quit compared with the unknown change in stage between baseline and 1-year follow-up effect of existing health education. It was there- and baseline and 2-year follow-up, which was defined fore a pragmatic trial. The TTM group received three in the protocol as a secondary outcome measure. Stage hour-long whole class lessons and three sessions on was defined using the algorithm described by Pallonen the computer throughout school Year 9. The computer et al. [10], although it used smoking status as defined program was described in the Introduction. Concerning below. Stage of change could not be allocated to 1,108 the whole class lessons, lesson 1 consisted of describing (13.4%) participants with known smoking status at the stages and using this knowledge to stage someone baseline, 745 (10.0%) participants with known smoking pupils knew. Lesson 2 concerned the pros and cons of status at the 1-year follow-up, and 511 (7.5%) partic- smoking (decisional balance) and an exercise on false ipants with known smoking status at the 2-year followup. beliefs about smoking. Lesson 3 consisted of exercises In separate test retest and parallel form assessbeliefs staging three fictitious letter writers and using this to ments, the (95%CI) for stage of change were 0.46

4 316 AVEYARD ET AL. FIG. 1. Flow of participants through the trial. ( ) and 0.52 ( ), respectively, indicat- made positive movements in the TTM and control ing only moderate reliability for stage. group, and the difference (95%CI) in those proportions. A positive change in stage between baseline and 1- Because cluster-randomized trials can result in imbalance or 2-year follow-up was defined as a movement to a of confounders despite large numbers of particor stage at which acquisition was less likely or cessation ipants, logistic regression was used to adjust for baseline more likely. We examined the proportions who had smoking status and other potential confounders.

5 TTM-BASED SMOKING PREVENTION AND CESSATION 317 The analysis adjusted for baseline smoking status is equivalent to examining the increase in the proportion that smoke. However, because smoking status was defined as never, tried smoking, ex-smoker, current smoker, and unknown smoking status, adjustment achieves slightly more than this. The other potential confounders adjusted for were age, sex, ethnic group, Townsend score quintiles (based on a census-derived score for the deprivation of the area of the participant s residence), and mother, father, sibling, and best friend s smoking habits. To account for the cluster randomization, for all these analyses, we used random effects logistic regression [11], with school as a random effect and all other variables as fixed-effects dummy terms. A binary outcome positive movement or no such movement obscures both negative movement and the degree of movement in stage. We created a variable to describe the degree of movement from the starting stage, using an arbitrary score, defined in Table 1. We calculated the difference (95%CI) between the mean score for the TTM and control groups. Again, we subsequently adjusted for baseline smoking status and the other potential confounders, calculating the adjusted difference between mean change scores for TTM and control groups. Our prespecified primary outcome measure was regular smoking, defined as regularly smoking at least one cigarette per week. This definition of regular smoking was used because this is the standard definition of adolescent smoking in the United Kingdom [12], is the basis for national prevalence figures [13], and is a definition also used in the United States [14]. Previously, we published data on smoking status at 1-year followup only [4]. Smoking status was provisionally defined by reference to responses to two questions. One question was very similar to the standard question used in the United Kingdom to define smoking status [12]. The question was Have you ever smoked cigarettes? The responses categorized number of cigarettes smoked in seven categories that ranged from Never tried through to One or more cigarettes per week. The second question was derived from an algorithm published by Pallonen et al. that is used to allocate stage of change [10]. The question stem is Which of these statements best describes your cigarette smoking now? The responses (abbreviated) are never smoked, tried smoking a few times, I am a smoker, and used to smoke regularly, but I have given it up. For smokers, responses to these two questions were checked against each other and with responses to four other questions: How long ago did you first start smoking and three questions on number of cigarettes consumed in the past 24 h, 7 days, and 30 days, respectively. For ex-smokers, responses to the two main questions were checked against each other and with responses to those questions on consumption in the past 24 h, 7 days, and 30 TABLE 1 Scoring System for Change of Stage Score Stage at follow up Acquisition recent Acquisition Acquisition Acquisition action or cessation Cessation Cessation Cessation Cessation Stage at baseline precontemplation contemplation preparation precontemplation contemplation preparation action maintenance Acquisition precontemplation Acquisition contemplation Acquisition preparation Acquisition recent action or cessation precontemplation a a a Cessation contemplation a a a Cessation preparation a a a Cessation action a a a Cessation maintenance a a a a This change is impossible and was not scored.

6 318 AVEYARD ET AL. days, plus two additional questions on intention to give The median consumptions of regular weekly smokers up smoking in the future, where the response I have at baseline, 1 year, and 2 years were 2.6, 2.9, and 7.6 already given up was expected. For never and occa- cigarettes per day, respectively. sional smokers, responses to the two main questions Using this new definition of regular smoking, we as- were checked against each other only. The computerized sessed the unadjusted and fully adjusted OR (95%CI) algorithm linking all these questions was complicated. for smoking at 1- and 2-year follow-up as described In brief, responses to the two main questions were above. These results should be viewed with some caution. checked against one another to code provisional smoking This variable was not specified in the protocol as status and against the other questions. The degree an outcome measure, but post hoc, after viewing the of inconsistency between questions was rated on a scale results. Adding extra outcome variables in this way of seriousness. If a response to only one of the main increases the chance that statistically significant differences two questions was provided, this was used. Serious inconsistency between control and intervention groups will resulted in unknown smoking status being arise by chance alone (type 1 errors). allocated. At the 1-year follow-up, 7,147 (96.4%) of those with known smoking status gave totally logically consistent answers. At the 2-year follow up, 6,579 (97.0%) Intention to Treat and On-Treatment Analyses of those with known smoking status gave totally logi- These analyses were based on the intention to treat cally consistent answers. In separate test retest and principle, with different assumptions about those lost parallel form assessments, the (95%CI) for regular to follow-up. For stage of change, we assumed, first, smoking, the primary outcome variable in this trial, that all those lost to follow-up had not made positive were 0.87 ( ) and 0.85 ( ), respec- stage movements, and second, that they had. For the tively, indicating very good reliability [15]. change score analyses, we assumed that their change We calculated the unadjusted odds ratio (OR) and score was the mean of those whose score could be calculated. 95%CI for regular weekly smoking at 2-year follow-up The final analysis was confined only to those for for the TTM group relative to the control group, and whom a valid change in stage could be calculated. Only from this derived the modeled percentage smoking in the latter is presented in these analyses, though fuller those groups. Again, we subsequently adjusted for base- versions with all the data are available from P.A. on request. line smoking status and the other potential confounders, calculating the adjusted OR (95%CI) for TTM and For regular weekly smoking, the assumptions about control groups. those lost to follow-up are as follows. We assumed that Both referees of the original submission of this report all those lost to follow-up were smokers, those lost were wondered whether the results would be different if a not smokers, those lost had the same smoking status different definition of regular smoking was used as an as at baseline (with unknown baseline smoking status outcome variable, and one referee suggested daily counted as smokers), and those lost had the same smok- smoking should be used. In response, we categorized ing status as at baseline (with unknown baseline smoking participants as either regular daily smokers or not, status counted as nonsmokers). We then confined meaning that consumption was on average at least one the analysis to all those who were followed up and for cigarette per day. This was done by reference mainly whom smoking status could be calculated and all those to two questions on average daily consumption in the followed up and who gave no inconsistent data on smoking past 30 days and number of cigarettes consumed in the status. Only the data for all those with known smokpast past 7 days. Where both responses were available, the ing status at follow-up are presented in this report, daily amount was a weighted average of the two, with though full tables are available from P.A. All these anal- the 30-day average counting double the 7-day average. yses were done without reference to how many interventions Where one was available, this was assumed the average trial participants received the intention to treat daily consumption. Where neither were available, the approach and sensitivity analysis for loss to followup amount consumed in the past 24 h was taken as the is recommended as part of that approach to analysis daily average. At baseline, the daily consumption could [16]. In all these analyses there should be no reason why not be assessed in 2 (0.2%) regular smokers, and 884 loss to follow-up or unreliable data would be associated (81.1%) regular weekly smokers were daily smokers. with the TTM or control group and differences in the Of those regular weekly smokers present at 1-year follow-up, ORs would raise doubts about the validity of the results. average daily consumption could not be as- For regular daily smoking, the analysis was confined sessed in 13 (0.9%), and 1,229 (87.0%) of regular weekly to only those who were followed up at 1- and 2-year as- smokers were daily smokers. Of those regular weekly sessments. smokers present at 2-year follow-up, average daily consumption One problem with the intention to treat analysis is could not be assessed in 3 (0.2%), and 1,424 that while it leads to a valid conclusion about the effects (90.0%) of regular weekly smokers were daily smokers. of policy, it is conservative and may obscure efficacious

7 TTM-BASED SMOKING PREVENTION AND CESSATION 319 therapy that could be effective in a different policy context. We addressed Prochaska s criticism of insufficient interventions [8] by comparing the ORs for those who experienced no, one, two, or three interventions. Evidence that more interventions would be more successful would be provided by data showing that those who had three interventions were less likely to be smoking than those who had none, one, or two. However, as failure to be present on the day of the intervention, which was how interventions were missed, is unlikely to be random with respect to smoking, this analysis is potentially biased. Data on whether this bias was potentially substantial can be gathered from comparison of the baseline characteristics of those in each of these categories, which we examined. Then we used random-effects analysis to examine whether changes in stage or smoking status at outcome differed between the control group and those with no, one, two, or three interventions, unadjusted and adjusted for baseline smoking status and all the above potential confounders. RESULTS Table 2 shows the outcome of change of stage at 1- and 2-year follow-ups. Slightly more people in the TTM group had made positive movements, with a difference (95%CI) of 1.2% ( %). However, this difference is no longer significant on adjustment for baseline smoking status and other potential confounders. Furthermore, the mean change score analysis suggests that young people in the TTM group made more negative movements with a mean difference in score (95%CI) of 0.05 ( ). At 2 years, however, this possible advantage seen in the TTM group had disappeared with no significant differences in the percentage making positive stage movements and almost identical mean change scores between the groups (Table 2). Using assumptions about the change in stage of those lost to follow-up described under Method showed no significant differences between TTM and control groups overall, for baseline smokers, or baseline nonsmokers (table available on request). Table 3 presents the results of smoking status at 2- year follow-up and is very similar to the corresponding table for 1-year follow-up, previously published [4]. There was a slight excess of smokers in the TTM group that persisted after adjustment, but this difference was not significant and was very small. Subgroup analysis on baseline smokers and nonsmokers shows no convincing evidence of effect modification. That is, there was no evidence that the intervention was effective for smoking cessation but not prevention, or vice versa. Using assumptions about the smoking status of those lost to follow-up described under Method did not alter these findings. There were no significant differences between TABLE 2 The Effect of TTM Intervention versus Control on Change of Stage for the Whole Sample and the Subgroups Baseline Weekly Smokers and Baseline Non-Weekly Smokers at 1- and 2-Year Follow-ups Unadjusted Fully adjusted Positive movement Mean change score Positive change Mean change score % positive % positive in control in TTM Difference (95%Cl) OR (95%Cl) Control TTM Difference (95%Cl) OR (95%Cl) Difference (95%Cl) Outcome at 1 year All participants % ( %) 1.19 ( ) ( ) 1.13 ( ) 0.03 ( ) Baseline smokers a % ( %) 0.99 ( ) ( ) 1.20 ( ) 0.12 ( ) Baseline nonsmokers b % ( %) 1.34 ( ) ( ) 1.35 ( ) 0.02 ( ) Outcome at 2 years All participants % ( %) 1.21 ( ) ( ) 1.25 ( ) 0.00 ( ) Baseline smokers a % ( %) 1.14 ( ) ( ) 1.40 ( ) 0.11 ( ) Baseline nonsmokers b % ( %) 1.29 ( ) ( ) 1.30 ( ) 0.01 ( ) a Regular weekly smokers at baseline. b Not regular weekly smokers at baseline.

8 320 AVEYARD ET AL. TABLE 3 The Effect of TTM Intervention versus Control on Regular Weekly Smoking for the Whole Sample and the Subgroups Baseline Weekly Smokers and Baseline Non-Weekly Smokers at 2-Year Follow-up Unadjusted Fully adjusted % smokers in control % smokers in TTM Difference (95%Cl) OR (95%Cl) OR (95%Cl) All participants % ( %) 1.06 ( ) 1.06 ( ) Baseline smokers a % ( %) 0.87 ( ) 0.96 ( ) Baseline nonsmokers b % ( %) 1.08 ( ) 1.07 ( ) a Regular weekly smokers at baseline. b Not regular weekly smokers at baseline. TTM and control groups overall, for baseline smokers, group who were present for all three computer sessions. or baseline nonsmokers (table available on request). There was no convincing evidence of subgroup differ- Table 4 presents the data for the added outcome vari- ences with baseline smokers or nonsmokers. However, able of regular daily smoking. For all participants com- there is evidence that the risk of smoking is higher bined, the ORs for smoking at outcome for daily smok- among those having fewer interventions than in the ing were similar to those for weekly smoking. At 1 year, control group or among those who had three interventhe adjusted OR (95%CI) for daily smoking was 1.12 tions. This is less clear when smoking status is exam- ( ) (Table 4) compared with 1.14 ( ) ined at 2-year follow-up. It is also apparent that adjustfor weekly smoking [4]. At 2 years, the OR (95%CI) for ment for potential confounding adjusts the excess risk daily smoking at outcome was 1.07 ( ) (Table downward, though does not eliminate this excess. Com- 4) compared with 1.06 ( ) for weekly smoking parison of the baseline characteristics (table available (Table 3). There was, however, a stronger suggestion of on request) of those who had one computer intervention benefit for baseline daily smokers. The adjusted OR with those in the control group showed that the one- (95%CI) for smoking at 1 year was 0.66 ( ), but intervention group was more deprived and more than at 2 years the effect was less strong at 0.92 ( ) twice as likely to have smoked at baseline, and over (Table 4). For weekly smoking at 1-year follow-up, the adjusted OR (95%CI) for baseline weekly smokers was half had mothers who smoked compared with 29% of 0.92 ( ) [4], and at 2 years it was 0.96 (0.64 the control group. The proportion deprived, smokers, 1.44) (Table 3). Thus the nonsignificant suggestion of and with parents that smoked in the group that rebenefit at 1 year for baseline daily smokers, which outthat of the control group and the one-intervention ceived two interventions was intermediate between come was added post hoc, is out of kilter with the other results. group. These characteristics were adjusted for, but it Table 5 shows the effect of the intervention on regular raises the possibility that those who were absent on weekly smoking by number of computer sessions reunmeasured the days when the intervention was delivered had other ceived. The risk of smoking was similar comparing differences from those who were present those in the control group with those in the intervention on those days. There was little evidence, however, that TABLE 4 The Effect of the TTM Intervention Relative to the Control Group on Regular Daily Smoking for the Whole Sample and the Subgroups Baseline Daily Smokers and Baseline Non-Daily Smokers at 1- and 2-Year Follow-up Unadjusted Fully adjusted % smokers in control % smokers in TTM Difference (95%Cl) OR (95%Cl) OR (95%Cl) Outcome at 1 year All participants % ( %) 1.06 ( ) 1.12 ( ) Baseline smokers a % ( %) 0.75 ( ) 0.66 ( ) Baseline nonsmokers b % ( %) 1.18 ( ) 1.21 ( ) Outcome at 2 years All participants % ( %) 1.06 ( ) 1.07 ( ) Baseline smokers a % ( %) 0.97 ( ) 0.92 ( ) Baseline nonsmokers b % ( %) 1.06 ( ) 1.07 ( ) a Regular daily smokers at baseline. b Not regular daily smokers at baseline.

9 TABLE 5 Regular Weekly Smoking at 1- and 2-Year Follow-up for Subgroups Defined by the Number of Computer Sessions Attended No computer sessions 1 computer session 2 computer sessions 3 computer sessions Unadjusted Unadjusted Unadjusted Unadjusted Control, a Fully Fully Fully Fully unadjusted, % smoking adjusted % smoking adjusted % smoking adjusted % smoking adjusted % smoking at follow- OR at follow- OR at follow- OR at follow- OR at follow-up up (95%Cl) OR (95%Cl) up (95%Cl) OR (95%Cl) up (95%Cl) OR (95%Cl) up (95%Cl) OR (95%Cl) Outcome at 1 year All participants ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) All baseline 74.9 c smokers b ( ) ( ) ( ) ( ) ( ) ( ) All baseline nonsmokers d ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) Outcome at 2 years All participants ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) All baseline 74.8 c smokers b ( ) ( ) ( ) ( ) ( ) ( ) All baseline nonsmokers d ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) a Control group is reference group, so OR 1.00 by definition. b Baseline regular weekly smokers. c No such participants. d Not baseline regular weekly smokers. TTM-BASED SMOKING PREVENTION AND CESSATION 321

10 322 AVEYARD ET AL. the probability of making positive stage movements were present for all three, lasted a total of less than 40 varies by number of interventions (Table 6). min [4]. There were three supporting whole class les- sons lasting 3 h total. Throughout their education, the DISCUSSION young people in this study have had lessons and information about smoking, and they have lived in a society This study has shown that there was no evidence of surrounded by contrasting information about smoking. benefit at 1 year in change of stage nor any benefit The evidence from the mass of studies on smoking cessation appearing late after the finish of the intervention at 2- is that multicomponent interventions that inappearing year follow-up for either stage of change or smoking clude tobacco control in the community outside school status. There was no evidence either of any benefit are necessary [21], and these results only emphasize when the study group was confined to those who had this. three computer interventions. The groups were well A further reason the intervention may have failed is balanced with respect to measured confounders [4], and that it is intensely wordy. The computer intervention in any case we adjusted for these with little effect on gave participants feedback information about stage, the OR. Loss to follow-up was greater in the control temptations, decisional balance, and processes. This group, but only slightly so, and sensitivity analysis for concerned not only current status but also several dif- this did not change the results of the trial. Pupils in ferent strategies individuals should adopt to move to intervention schools may have been more willing to the next stage. We elected not to provide students writ- declare themselves nonsmokers when they truly were ten reports on this, because experience in the United not, because of the demand expectations arising from States was that many such reports were not valued. participating in the intervention. However, the level of In any case they gave confidential information about inconsistent data on smoking was similar in both arms smoking status and thoughts on this, which partic- of the trial, and the absence of apparent benefit from ipants may have feared sharing with their friends, the intervention makes this possibility unlikely. We teachers, or parents. We have no direct evidence that therefore conclude that neither confounding nor infor- participants could not remember these messages, but mation bias was concealing the effects of the intervention, staff in our department who tried the program could and it was in truth ineffective. not do so. Nevertheless, it should be noted that the Why was the intervention ineffective? Prochaska suggested participants enjoyed the intervention and thought it that more interventions would have been effec- was useful [4]. tive [6] and we cannot exclude this possibility. For the We have described in another currently unpublished prevention program, there is no evidence on which to report that stage was not measured as reliably as we decide how many sessions might be needed. We thought would have liked. On 60% of occasions, participants in three interventions would be what might be delivered stages other than acquisition precontemplation were in if this intervention ever became routine practice. For the same stage on retest as on test, even when this was the cessation program, the only relevant evidence measured almost immediately afterward. If stage was comes from a trial of a TTM-based expert system among measured unreliably, the stage-appropriate messages adults [17], in which one intervention was as effective were wrong. This problem would also mean that any as two, three, or six. In this population of adolescents, positive movements in stage would have been harder to 37% of regular smokers were in preparation stage [18], detect because of this random error. However, smoking rather than the 20% that is typical among adult populations status was measured very reliably, so this would not [19], and over 25% of regular smokers did stop have obscured benefit of the intervention, had it been [4]. There is therefore no reason to believe that this present. group was intrinsically more resistant to change than A final reason the intervention may have failed is adults and no reason to think, therefore, that a dose of that the TTM is not a valid description of either smoking three times that which was effective among adults acquisition or cessation among adolescents. Young would be ineffective. However, there was no benefit smokers characteristically change their smoking pat- even for those who had all three interventions. As a tern frequently, with long periods of abstinence or occasional computerized expert system based on the TTM is effective smoking [22]. Adults have a settled pattern and among adults [17, 20], but not among adolescents, chipping is rare [23]. The process of cessation among this might suggest that adolescent smokers do not go adults therefore may be different from that among through the same stages of change or that the drivers young people. As much of the data on young people and for movement through the stages are different. This in adults for the validity of the TTM is derived from possibility needs further research. cross-sectional studies [24, 25], definitive conclusion on Another possible reason the intervention was ineffective this possibility must await further data. is that no intervention of this type ever could be. Previous studies have shown that smokers are more After all, the computer element, even for those who likely to be absent from school on any given day than

11 TABLE 6 Positive Movement of Stage at 1- and 2-Year Follow-up for Subgroups Defined by Number of Computer Sessions Attended a 1 computer session 2 computer sessions 3 computer sessions Control, Unadjusted Fully Unadjusted Fully Unadjusted Fully unadjusted, adjusted adjusted adjusted % positive % positive OR % positive OR % positive OR movement movement (95%Cl) OR (95%Cl) movement (95%Cl) OR (95%Cl) movement (95%Cl) OR (95%Cl) Results at 1-year follow-up All participants ( ) ( ) ( ) ( ) ( ) ( ) All baseline smokers b ( ) ( ) ( ) ( ) ( ) ( ) All baseline nonsmokers c ( ) ( ) ( ) ( ) ( ) ( ) Results at 2-year follow-up All participants ( ) ( ) ( ) ( ) ( ) ( ) All baseline smokers b ( ) ( ) ( ) ( ) ( ) ( ) All baseline nonsmokers c ( ) ( ) ( ) ( ) ( ) ( ) a No model with 0 computer interventions would converge. b Baseline regular weekly smokers. c Not baseline regular weekly smokers. TTM-BASED SMOKING PREVENTION AND CESSATION 323

12 324 AVEYARD ET AL. nonsmokers [26]. This study has confirmed this, by 5. Stapleton JA, Sutherland G, Russell MA. How much does relapse after one year erode effectiveness of smoking cessation treatshowing that the percentage of baseline smokers was ments? Long-term follow up of randomised trial of nicotine nasal higher in the group receiving one or two computer inter- spray. Br Med J 1998;316: ventions than in the other groups. Although this may 6. Prochaska JO, Diclemente CC, Velicer WF, Rossi JS. Standardbe explained if smokers were more likely to change ized, individualized, interactive, and personalized self-help programs school, many absent at 1-year follow-up were present for smoking cessation. Health Psychol 1993;12: again at 2 years. When this finding was originally detion 7. Lancaster T, Stead LF. Self-help interventions for smoking cessa- scribed [26], one explanation offered was that smokers (Cochrane review). In: The Cochrane library. Oxford: Update Software, 2000:Issue 4. may be made ill by their own smoking and this caused 8. Prochaska JO. Stages of change model for smoking prevention the absence. Further data from this study show that the and cessation in schools. Br Med J 2000;320:447. time sequence is that school absence predicts smoking 9. Reid D. Failure of an intervention to stop teenagers smoking. Br uptake, and not the reverse. Baseline nonsmokers who Med J 1999;319: were absent on one or more occasions when we attended 10. Pallonen UE, Prochaska JO, Velicer WF, Prokhorov AV, Smith to administer the computer intervention were more NF. Stages of acquisition and cessation for adolescent smoking: likely to be smokers at 1- and at 2-year follow-up than an empirical integration. Addict Behav 1998;23: baseline nonsmokers present on every occasion. This 11. Murray DM. Design and analysis of group-randomized trials. effect persisted despite adjusting for maternal smoking, New York: Oxford Univ. Press, paternal smoking, and other risk factors for smoking. 12. Bewley BR, Day I, Ide L. Smoking by children in Great Britain a review of the literature. London: Social Science Research Coun- The two alternative explanations are that experiencing cil, one or two interventions causes smoking or that these 13. Jarvis L. Smoking among secondary school children in 1996: young people were lying when they described their England. London: H. M. Stationery Office, baseline smoking status. The first explanation is im- 14. Leventhal H, Cleary P. The smoking problem: a review of the plausible, and the second unlikely. It is unlikely because research and theory in behavioral risk modification. Psychol the percentage of inconsistent data at baseline on smok- Bull 1980;88: ing status was low at less than 3% and was unrelated 15. Altman D. Practical statistics for medical research. London: to the number of interventions received. Chapman & Hall, Hollis S, Campbell F. What is meant by intention to treat analy- In summary, at 1- and 2-year follow-up, this trial sis? Survey of published randomised controlled trials. Br Med showed no evidence of benefit for those in the interven- J 1999;319: tion group in either the primary outcome, regular 17. Velicer WF, Prochaska JO, Fava JL, Laforge RG, Rossi JS. Inter- (weekly) smoking, or the secondary outcome, positive active versus noninteractive interventions and dose response movement in stage. Neither were there subgroup differa relationships for stage-matched smoking cessation programs in ences by baseline smoking status. Both the intention managed care setting. Health Psychol 1999;18:21 8. to treat and the on-treatment analyses confirmed these 18. Aveyard P, Cheng KK, Lawrence PT. Stages of change model for smoking prevention and cessation in schools. Br Med J findings that the intervention was ineffective. 2000;320:447. ACKNOWLEDGMENTS 19. Velicer WF, Fava JL, Prochaska JO, Abrams DB, Emmons KM, Pierce JP. Distribution of smokers by stage in three representative We are grateful to Professor Prochaska and his colleagues for help samples. Prev Med 1995;24: in the initial stages of this project. PMA hold the license for the TTM- 20. Prochaska JO, Velicer WF, Fava JL, Ruggiero L, Laforge RG, based computer interventions in the UK and trained the teachers to Rossi JS, et al. Counselor and stimulus control enhancements of deliver the lessons. We are grateful to them both. Sheila Hirst and a stage-matched expert system intervention for smokers in a Helen Evans provided support to this project and we are grateful managed care setting. Prev Med 2001;32: to them. 21. Lantz PM, Jacobson PD, Warner KE, Wasserman J, Pollack HA, Berson J, et al. Investing in youth tobacco control: a review of REFERENCES smoking prevention and control strategies. Tobacco Control 2000;9: Hodgson P. Smoking cessation in practice. London: Health Educaby 22. Goddard E. Why children start smoking. An enquiry carried out tion Authority, Social Survey Division of OPCS on behalf of the Department 2. Prochaska JO, Diclemente CC, Norcross JC. In search of how of Health. London: H. M. Stationery Office, 1990:1. people change. Applications to addictive behaviors. Am Psy- 23. Jarvis MJ. A profile of tobacco smoking. Addiction 1994;89:1371 chol 1992;47: Velicer W, Norman GJ, Fava JL, Prochaska JO. Testing 40 predicsmoking 24. Sutton S. A critical review of the transtheoretical model applied to tions from the transtheoretical model. Addict Behav cessation. In Norman P, Abraham C, Conner M, editors. 1999;24: Understanding and changing health behaviour: from health be- 4. Aveyard P, Cheng KK, Almond J, Sherratt E, Lancashire R, Lawliefs to self-regulation. Reading (UK): Harwood Academic, rence T, et al. A cluster-randomised controlled trial of an expert 25. Sutton S. Interpreting cross-sectional data on stages of change. system based on the transtheoretical ( stages of change ) model Psychol Health 1999;15:1 9. for smoking prevention and cessation in schools. Br Med J 1999;319: Charlton A, Blair V. Absence from school related to children s and parental smoking habits. Br Med J 1989;298:90 2.