International Journal of Epidemiology ©International Epidemiological Association 1990
Vol. 19, No. 2 Printed in Great Britain
When and Why Children First Start to Smoke A V SWAN*. R CREESER" AND M MURRAYt
Smoking is still regarded as the greatest preventable cause of disease in western society. In the UK tobacco still accounts for 15-20% of all deaths.' Adult prevalences are falling in the UK and elsewhere, but although there are some signs that these trends are starting to be reflected in the patterns of uptake in children thefindingsare not as encouraging as one might hope.2 On the other hand the apparently worldwide trend, over the last decade and more, for the prevalence of smoking in girls now to be greater than that in boys3 is a very striking phenomenon yet to be explained. There is a clear need to utilize all the evidence available on which children are most at risk, what factors are involved and when intervention is most needed. In contrast to almost all previous work in this field which concentrated on prevalence data, which confounds uptake and maintenance, this paper reports the results of a longitudinal analysis of the incidences observed in a cohort of over 6000 children in Derbyshire. The children were followed from early adolescence in 1974 when they were in the age range 11.7 to 12.7 years to young adulthood in 1984 when
they were 21.5 to 22.5 years old. The changing patterns of incidence with age and how they relate to the children's attitudes, circumstances and environment are investigated using 'survival data analysis' techniques.4 METHOD The MRC Derbyshire Smoking Study followed a cohort of over 6000 Derbyshire schoolchildren from their first year in secondary shcool in 1974 to school leaving age in 1978. The initial survey was in June 1974 when the children were aged between 11.7 and 12.7 years with a mean of 12.2 years. Each year the children answered a questionnaire in their classrooms about their social activities, attitudes to a variety of issues and their smoking behaviour. In 1981 when the children had left school and were aged 18-19 years they were sent a brief postal questionnaire. This was designed as a pilot investigation of whether a sufficient response rate could be obtained to justify a full postal follow-up survey of the cohort as young adults. Since the response rate was 79% of those contacted it was decided to mount the full follow-up survey. The members of the cohort had by then reached adulthood so it was necessary to re-design the questionnaire. This was done using the results of an interview survey of 80 members of the cohort. The resulting document included questions on work, social life and other activi-
'United Medical and Dental Schools, Guy's and St Thomas's Hospital Department of Community Medicine, London SE1 7EH, UK. "Thomas Coram Research Unit, London, UK. tCentre for Applied Health Studies, University of Ulster, UK.
323
Downloaded from http://ije.oxfordjournals.org/ at New York University on May 25, 2015
Swan A V (United Medical and Dental Schools, Guy's and St Thomas's Hospital, Department of Community Medicine, London SE1 7EH, UK) Creeser R and Murray M. When and why children first start to smoke. International Journal of Epidemiology 1990,19: 323-330. Most investigations of smoking in children focus on prevalence in which uptake and maintenance are confounded. This paper reports an analysis of pure incidence data in a cohort of over 6000 Derbyshire schoolchildren followed for ten years investigated using survival data analysis techniques. Over 70% of the cohort tried at least one cigarette before the end of the fifth year of secondary school. Some 40% identified themselves as regular smokers while at school. The risks of taking up regular smoking were higher if, at the age of 11.7-12.7 years, the children had smoking siblings, opposite sex friends, were dismissive of the health hazards and susceptible to peer pressure. More girls than boys in that age range spent time with opposite sex companions and in organized social activities which in turn were significantly associated with the risk of taking up smoking. Thus the earlier physical and emotional development of girls may help explain recent findings that adolescent girls are now more likely to smoke than boys of the same age. The greatest incidence of regular smoking occurred when the average age was increasing from 14.2 to 15.2 years. This has very clear implications for the timing of anti-smoking interventions.
324
INTERNATIONAL JOURNAL OF EPIDEMIOLOGY
RESULTS Table 1 shows the proportions of those boys and girls who have never smoked, those who have tried at least once and those who have regularly smoked at least one or more a week. By the end of the school period the proportion of children who have at least tried smoking has reached 75% in the boys and 71% among the girls. At 12.2 years 15% fewer girls than boys have tried smoking, but this difference is halved by the time they are 13.2 years of age and the proportion of girls who have
experimented gradually catches up with that of the boys so at 16.2 years the difference is only 4%. The cross-sectional results for 1981 and 1984 cannot be used here, because the numbers who have never smoked are not identifiable from responses in those years. The prevalence of regular smoking taken as one or more a week while the children are at school and one or more a day when they are over 18 years of age, increases steadily in both sexes. The boys' prevalences are consistently above those of the girls, but the girls show signs of catching up after the age of 13.2 years when the gap narrows up to the age of 16.2 years. After that the difference between the sexes starts increasing again. Using the data linked longitudinally the lifetable appropriate for investigating the 'survival' of the members of the cohort as 'never smoked' was obtained (Table 2). Only 45% of boys and 60% of girls had never tried a cigarette at the start of the study so the 'cumulative survivals' start with those proportions. The uptake curves for experimental smoking in boys and girls are shown in Figure 1. Since the incidence was much the same in both sexes the curves remain more or less parallel with an estimated uptake of experimental smoking by the time their mean age was 22 years of 86% in the boys and 79% in the girls. Finally it can be seen from the estimated uptakes that only 2.9% of boys and 3.7% of girlsfirsttry a cigarette after their fifth year in secondary school (mean age 15.2 years). The equivalent life table for estimating the pattern of uptake of regular smoking is given in Table 3. In contrast to experimentation the prevalence of regular smoking was quite low in the 11.7-12.7 year olds: 6% in the boys and less than 3% in girls. The initial uptake was smaller, but increased progressively faster up to the fourth year of school when, on average the children were 15.2 years old. In both sexes the incidence in that year among those not previously classed as regular smokers was almost twice that of any other year: 24% in both boys and girls. In 1981 and 1984 regular smoking was defined as one or more cigarettes a day. The incidence between thefifthyear of secondary school and three years later in 1981 when the average age was 19.0 years was 6% in both sexes. Between the fifth year of school and 1984, when the members of the cohort had reached an average age of 22.0 years, the incidence estimated from the lifetable results was 8.8% in boys and 3.3% in girls. The girls started lower than the boys, but between the ages of 13.2 and 19.0 years had higher or similar incidences.
Downloaded from http://ije.oxfordjournals.org/ at New York University on May 25, 2015
ties and was designed to make the questions on smoking more relevant to young adults. The follow-up survey was finally mounted in 1984 when a 75% response rate of those not known to have moved was obtained. Further details on the study may be obtained from previously published reports.*•* Since experimenting with cigarettes and smoking regularly may be quite different behaviours the incidences of both were analysed. An individual was defined as an experimental smoker if he reported any smoking at all. Thus any response other than 'never smoked' meant that they had reached a 'terminal' event since the last response of 'never'. For the school years, up to 1978, regular smoking was defined as one or more a week. In 1981 and 1984, when the members of the cohort were entering adulthood, regular smoking was defined as one or more a day. Thefirstreport of regular smoking was taken to indicate the 'terminal' event of becoming a regular smoker in the period since the last response from that individual. The analysis uses the life table approach to analysing censored survival data to investigate the uptake of both experimental and regular smoking and how the pattern of uptake over several years is associated with a number of potential risk factors when the children were 11.7 to 12.7 years old. The incidences of the terminal events within each period are used to obtain the equivalent of actuarial lifetable survival curves. These curves display the estimated proportions surviving ie: The proportion who have not reached a terminal event, for each age which can be denoted by S. The uptake curves, which represent the proportions that have reached the terminal event are then obtained as 1-S which is the inverse of the survival curve. The differences between uptake curves for different categories of children are assessed in terms of the relative risks of uptake using Cox's regression approach to the analysis of life tables.9 The second part of the analysis uses this approach to assess the effects of the various factors on the uptake of regular smoking defined as one or more cigarettes a week.
325
CHILDREN AND SMOKING TABLE 1
Proportions of boys and girts by smoking category and calendar year 1974 12.2
1975 13.2
1976 14.2
1977 15.2
1978 16.2
1981* 19.0
1984* 22.0
Boys Girls Boys Girls Boys Girls
45.1 60.0 54.9 40.0 6.0 2.5
39.1 46.8 60.9 53.2 9.2 5.5
32.9 41.6 67.1 58.4 15.5 13.3
27.1 32 9 72.9 67.1 21.0 18.9
24.9 29.3 75.1 70.7 25.8 23.3
58.1 60.2 41 9 39.8 33 3 29.9
48.0 51.4 52.0 48.6 38.6 32.7
Boys Girls
3086 3225
3205 3689
2178 3355
2951 3165
2700 2915
2707 2903
2292 2368
Year Mean age (years) Never smoked Any smoking Smoke regularly (>1 per week)
No.
*ln both these years regular smoking means one or more a day and the never smoked category is used for those claiming to be non-smokers. Consequently this latter category will contain some experimenters.
TABLE 2
Life table with the first report of any smoking as the terminal event
Number Year
Mean age
ing have uptake curves the least elevated above those of the children without a smoking parent. For a rigorous assessment of the magnitudes of the effects of parental smoking and other factors separately and together it is necessary to use linear modelling methods for life table data. The variables included in the regression analysis cover their home environment, their social and other spare-time activities, their perceptions of peer pressure and their beliefs about the hazards of smoking. The children's responses under these headings have, in a number of cases, been grouped to form composite variables.5 Table 4 gives the distributions of the children according to their category of these factors when they were 11.7 to 12.7 years old.
interval
No. of terminal events
entering
No. lost to observation
Termination* rate
Survival rate
Cumulative survival
Cumulative uptake
Boys To—1974 —1975 —1976 —1977 —1978 —1981 —1984
Vol. 19, No. 2 Printed in Great Britain
When and Why Children First Start to Smoke A V SWAN*. R CREESER" AND M MURRAYt
Smoking is still regarded as the greatest preventable cause of disease in western society. In the UK tobacco still accounts for 15-20% of all deaths.' Adult prevalences are falling in the UK and elsewhere, but although there are some signs that these trends are starting to be reflected in the patterns of uptake in children thefindingsare not as encouraging as one might hope.2 On the other hand the apparently worldwide trend, over the last decade and more, for the prevalence of smoking in girls now to be greater than that in boys3 is a very striking phenomenon yet to be explained. There is a clear need to utilize all the evidence available on which children are most at risk, what factors are involved and when intervention is most needed. In contrast to almost all previous work in this field which concentrated on prevalence data, which confounds uptake and maintenance, this paper reports the results of a longitudinal analysis of the incidences observed in a cohort of over 6000 children in Derbyshire. The children were followed from early adolescence in 1974 when they were in the age range 11.7 to 12.7 years to young adulthood in 1984 when
they were 21.5 to 22.5 years old. The changing patterns of incidence with age and how they relate to the children's attitudes, circumstances and environment are investigated using 'survival data analysis' techniques.4 METHOD The MRC Derbyshire Smoking Study followed a cohort of over 6000 Derbyshire schoolchildren from their first year in secondary shcool in 1974 to school leaving age in 1978. The initial survey was in June 1974 when the children were aged between 11.7 and 12.7 years with a mean of 12.2 years. Each year the children answered a questionnaire in their classrooms about their social activities, attitudes to a variety of issues and their smoking behaviour. In 1981 when the children had left school and were aged 18-19 years they were sent a brief postal questionnaire. This was designed as a pilot investigation of whether a sufficient response rate could be obtained to justify a full postal follow-up survey of the cohort as young adults. Since the response rate was 79% of those contacted it was decided to mount the full follow-up survey. The members of the cohort had by then reached adulthood so it was necessary to re-design the questionnaire. This was done using the results of an interview survey of 80 members of the cohort. The resulting document included questions on work, social life and other activi-
'United Medical and Dental Schools, Guy's and St Thomas's Hospital Department of Community Medicine, London SE1 7EH, UK. "Thomas Coram Research Unit, London, UK. tCentre for Applied Health Studies, University of Ulster, UK.
323
Downloaded from http://ije.oxfordjournals.org/ at New York University on May 25, 2015
Swan A V (United Medical and Dental Schools, Guy's and St Thomas's Hospital, Department of Community Medicine, London SE1 7EH, UK) Creeser R and Murray M. When and why children first start to smoke. International Journal of Epidemiology 1990,19: 323-330. Most investigations of smoking in children focus on prevalence in which uptake and maintenance are confounded. This paper reports an analysis of pure incidence data in a cohort of over 6000 Derbyshire schoolchildren followed for ten years investigated using survival data analysis techniques. Over 70% of the cohort tried at least one cigarette before the end of the fifth year of secondary school. Some 40% identified themselves as regular smokers while at school. The risks of taking up regular smoking were higher if, at the age of 11.7-12.7 years, the children had smoking siblings, opposite sex friends, were dismissive of the health hazards and susceptible to peer pressure. More girls than boys in that age range spent time with opposite sex companions and in organized social activities which in turn were significantly associated with the risk of taking up smoking. Thus the earlier physical and emotional development of girls may help explain recent findings that adolescent girls are now more likely to smoke than boys of the same age. The greatest incidence of regular smoking occurred when the average age was increasing from 14.2 to 15.2 years. This has very clear implications for the timing of anti-smoking interventions.
324
INTERNATIONAL JOURNAL OF EPIDEMIOLOGY
RESULTS Table 1 shows the proportions of those boys and girls who have never smoked, those who have tried at least once and those who have regularly smoked at least one or more a week. By the end of the school period the proportion of children who have at least tried smoking has reached 75% in the boys and 71% among the girls. At 12.2 years 15% fewer girls than boys have tried smoking, but this difference is halved by the time they are 13.2 years of age and the proportion of girls who have
experimented gradually catches up with that of the boys so at 16.2 years the difference is only 4%. The cross-sectional results for 1981 and 1984 cannot be used here, because the numbers who have never smoked are not identifiable from responses in those years. The prevalence of regular smoking taken as one or more a week while the children are at school and one or more a day when they are over 18 years of age, increases steadily in both sexes. The boys' prevalences are consistently above those of the girls, but the girls show signs of catching up after the age of 13.2 years when the gap narrows up to the age of 16.2 years. After that the difference between the sexes starts increasing again. Using the data linked longitudinally the lifetable appropriate for investigating the 'survival' of the members of the cohort as 'never smoked' was obtained (Table 2). Only 45% of boys and 60% of girls had never tried a cigarette at the start of the study so the 'cumulative survivals' start with those proportions. The uptake curves for experimental smoking in boys and girls are shown in Figure 1. Since the incidence was much the same in both sexes the curves remain more or less parallel with an estimated uptake of experimental smoking by the time their mean age was 22 years of 86% in the boys and 79% in the girls. Finally it can be seen from the estimated uptakes that only 2.9% of boys and 3.7% of girlsfirsttry a cigarette after their fifth year in secondary school (mean age 15.2 years). The equivalent life table for estimating the pattern of uptake of regular smoking is given in Table 3. In contrast to experimentation the prevalence of regular smoking was quite low in the 11.7-12.7 year olds: 6% in the boys and less than 3% in girls. The initial uptake was smaller, but increased progressively faster up to the fourth year of school when, on average the children were 15.2 years old. In both sexes the incidence in that year among those not previously classed as regular smokers was almost twice that of any other year: 24% in both boys and girls. In 1981 and 1984 regular smoking was defined as one or more cigarettes a day. The incidence between thefifthyear of secondary school and three years later in 1981 when the average age was 19.0 years was 6% in both sexes. Between the fifth year of school and 1984, when the members of the cohort had reached an average age of 22.0 years, the incidence estimated from the lifetable results was 8.8% in boys and 3.3% in girls. The girls started lower than the boys, but between the ages of 13.2 and 19.0 years had higher or similar incidences.
Downloaded from http://ije.oxfordjournals.org/ at New York University on May 25, 2015
ties and was designed to make the questions on smoking more relevant to young adults. The follow-up survey was finally mounted in 1984 when a 75% response rate of those not known to have moved was obtained. Further details on the study may be obtained from previously published reports.*•* Since experimenting with cigarettes and smoking regularly may be quite different behaviours the incidences of both were analysed. An individual was defined as an experimental smoker if he reported any smoking at all. Thus any response other than 'never smoked' meant that they had reached a 'terminal' event since the last response of 'never'. For the school years, up to 1978, regular smoking was defined as one or more a week. In 1981 and 1984, when the members of the cohort were entering adulthood, regular smoking was defined as one or more a day. Thefirstreport of regular smoking was taken to indicate the 'terminal' event of becoming a regular smoker in the period since the last response from that individual. The analysis uses the life table approach to analysing censored survival data to investigate the uptake of both experimental and regular smoking and how the pattern of uptake over several years is associated with a number of potential risk factors when the children were 11.7 to 12.7 years old. The incidences of the terminal events within each period are used to obtain the equivalent of actuarial lifetable survival curves. These curves display the estimated proportions surviving ie: The proportion who have not reached a terminal event, for each age which can be denoted by S. The uptake curves, which represent the proportions that have reached the terminal event are then obtained as 1-S which is the inverse of the survival curve. The differences between uptake curves for different categories of children are assessed in terms of the relative risks of uptake using Cox's regression approach to the analysis of life tables.9 The second part of the analysis uses this approach to assess the effects of the various factors on the uptake of regular smoking defined as one or more cigarettes a week.
325
CHILDREN AND SMOKING TABLE 1
Proportions of boys and girts by smoking category and calendar year 1974 12.2
1975 13.2
1976 14.2
1977 15.2
1978 16.2
1981* 19.0
1984* 22.0
Boys Girls Boys Girls Boys Girls
45.1 60.0 54.9 40.0 6.0 2.5
39.1 46.8 60.9 53.2 9.2 5.5
32.9 41.6 67.1 58.4 15.5 13.3
27.1 32 9 72.9 67.1 21.0 18.9
24.9 29.3 75.1 70.7 25.8 23.3
58.1 60.2 41 9 39.8 33 3 29.9
48.0 51.4 52.0 48.6 38.6 32.7
Boys Girls
3086 3225
3205 3689
2178 3355
2951 3165
2700 2915
2707 2903
2292 2368
Year Mean age (years) Never smoked Any smoking Smoke regularly (>1 per week)
No.
*ln both these years regular smoking means one or more a day and the never smoked category is used for those claiming to be non-smokers. Consequently this latter category will contain some experimenters.
TABLE 2
Life table with the first report of any smoking as the terminal event
Number Year
Mean age
ing have uptake curves the least elevated above those of the children without a smoking parent. For a rigorous assessment of the magnitudes of the effects of parental smoking and other factors separately and together it is necessary to use linear modelling methods for life table data. The variables included in the regression analysis cover their home environment, their social and other spare-time activities, their perceptions of peer pressure and their beliefs about the hazards of smoking. The children's responses under these headings have, in a number of cases, been grouped to form composite variables.5 Table 4 gives the distributions of the children according to their category of these factors when they were 11.7 to 12.7 years old.
interval
No. of terminal events
entering
No. lost to observation
Termination* rate
Survival rate
Cumulative survival
Cumulative uptake
Boys To—1974 —1975 —1976 —1977 —1978 —1981 —1984
