A multilevel analysis of tobacco use and tobacco consumption levels in France: are there any combination risk groups?

Abstract

Background

Both the predictors for tobacco use and the determinants of the amounts of tobacco consumed by smokers should be taken into account when designing prevention programmes.

Methods

Using a sample of 12,948 individuals representative of the French population in 1999, multilevel models were used to carry out a comparative investigation for the predictors of tobacco use and the determinants of the amount of tobacco consumed by smokers.

Results

At the individual level, a combination of risks (higher risk of smoking and larger amounts of tobacco consumed by smokers) was found for males, for individuals with a low level of education and for divorcees. At the level of the area of residence, both the risk of smoking (odds ratio 1.07, 95% confidence interval: 1.01–1.12 for an increase by one standard deviation) and the amount of tobacco consumed among smokers (percentage variation +4%, 95% confidence interval: 0% – +8%) increased with the gross domestic product per capita.

Conclusion

This study justifies the combined use, in such analyses, of consumption levels for smokers in addition to the risk of smoking, in order to identify the profiles with the highest risk. It was possible to identify various groups with both a high risk of tobacco use and a high level of consumption among smokers, on the basis of individual (male, divorced or less educated) and environmental (living in a high GDP area) factors. The prevention efforts should thus be focused on such groups.

Tobacco consumption is a significant risk factor for cardiovascular diseases, lung, pharynx and oesophagus cancers, as well as chronic obstructive pulmonary diseases.¹ Consequently, awareness campaigns against tobacco consumption remain an essential objective of public health policies. Effective policies require the identification of populations most at risk, particularly in countries such as France where tobacco use is still prevalent, with 33.1% of smokers in the 12–75 year age bracket, in 1999.²

Numerous studies have analysed the risk-predicting factors for tobacco use while others have focused more specifically on consumption levels. However, few studies have tried to compare the risk-predicting factors for smoking with the determinants for tobacco consumption levels. Indeed, in addition to the number of years of tobacco use, which will not be considered here, risk profiles should involve both of these aspects simultaneously. These more complete profiles allow for better investigation of the following issue: are the risk determinants for tobacco use consistent with those for tobacco consumption levels? This approach allows one to determine if there are any population groups with a combination risk (significant tobacco use risk combined with high consumption levels for the smokers within such groups).

Many studies have observed a higher risk of tobacco use at the bottom of the social ladder,^3–20 but few have shown that more socially disadvantaged tobacco users also smoke more cigarettes.²¹ More recently, some health behaviour epidemiologists have been looking into contextual factors,^{4,8,9,22–29} mostly environmental factors, while simultaneously taking into account individual factors. This allowed them to separate the genuinely contextual effects from compositional effects.^{6,10,12,22,25,30–33} Some studies have shown that the risk of tobacco use increased with erosion of the socioeconomic level of the residential environment.^{6,7,12,13,30,31,34–38}

In this study, the residential environment was taken into consideration at various levels, simultaneously. On one hand, models involved the ‘area of residence’ of the individuals: these were the 95 administrative subdivisions of mainland France, which, in 1999, had populations between 75,000 and 2,500,000, with very diverse levels of economic development. Such diversity is likely to translate into differences in lifestyles, particularly with regard to tobacco consumption.^22,36 Furthermore, at a finer geographic level of differentiation, the degree of urbanization of the county of residence (French ‘commune’) was taken into account, by making the assumption that such a factor may have an impact on health-related behaviours. One of the objectives of this work was to determine if both county and area level effects on tobacco consumption persisted after adjustment for socioeconomic characteristics of the individuals.

These contextual analyses were carried out through multilevel regression models.^{10,33,36,39–42} Such models allow for more reliable estimates of the standard errors for the parameters involved, as they take into account the correlation between individuals from the same group, which is likely to persist after adjustment for individual and environmental factors.^{9,12,35,39,40,43}

The objective of this study was to compare the risk-predicting factors for tobacco use with the determinants of tobacco consumption levels, based on the hypothesis that factors at the individual, county and area levels can be combined to determine risk profiles. We have specifically tried to determine if there were any combination risk situations, i.e. a high risk of tobacco use combined with significant consumption levels, predicted by both individual and environmental factors.

METHODS

Data

Analyses were based on individual data collected in a phone survey carried out in 1999 by the INPES (the French National Institute for Prevention and Health Education).⁴⁴ A representative sample of the French population consisted of 12,948 individuals aged 16 to 75.

Several aspects of the individual’s socioeconomic status were taken into account. The French nomenclature of Professions and Social Categories (PCS) was used to define the occupation of surveyed, non-retired individuals as blue collar workers, lower white collar workers, intermediate professions, upper white collar workers, farmers and craftsmen-shopkeepers. Three classes were defined to reflect the education level, with the intermediate category corresponding to individuals who completed secondary education. Four classes were distinguished to categorize household monthly per capita income (less than € 610, between € 611 and € 1,100, between € 1,101 and € 1,350, and above € 1,351 per person). Employment status was categorized as no activity, unemployment, State subsidized employments, full time and part time work. Marital status of individuals was defined as never married, married, divorced or widowed.

Individuals were considered to be smokers if their consumption was at least one cigarette equivalent per day. The average amount of tobacco consumed per day, in cigarette equivalents, was calculated for each individual, taking into account all forms of tobacco consumption, i.e. cigarettes, cigars and pipes.

At the area level, GDP per capita in 1996 was taken into account. These data, the most recent figures available at the area level, were provided by the National Institute of Statistics and Economic Studies. At a more local level, the size of the county of residence was used; i.e. rural county, small town with a population between 2,000 and 20,000, medium sized city with a population between 20,000 and 200,000, or major city with a population over 200,000.

Weighting coefficients were computed by the INPES at the individual level to ensure that the sample was more representative of the French population.

Statistical analysis

Age and area level GDP per capita were standardized (the standardized values ranged between −1.7 and 2.0 for age and from −0.8 to 4.5 for GDP). A weighted logistic multilevel model, with individuals at level one nested within areas at level two, was used to study the risk-predicting factors for smoking, while taking into account the correlation between individuals within the same area (the county of residence was not treated as a third level). Differences in tobacco consumption levels among smokers were analysed through a normal multilevel model, with the log of this variable taken as outcome (leading to a less asymmetric distribution). The intercept was introduced as a random effect and the individual factors were treated as fixed effects.

An individual factors model was first fitted to the data, separately for consumption risk and consumption levels (model 1). Since we adjusted for the same potential confounders in the two separate analyses, the corresponding results were interpreted concurrently. The size of the county of residence (French ‘commune’) was introduced as a fixed effect in a second step (model 2). The area level GDP was then included in a third and final step (model 3). We checked if the area level random intercept variations decreased when the size of the county of residence and the GDP per capita were introduced into the model. At each step, the Wald test in the logistic model, and the likelihood ratio test in the lognormal model were used to determine whether the variance of the random components was significantly different from zero.

The parameters of the multilevel models were estimated using the MLwiN 1.10 software (Institute of Education, University of London), with a second-order penalized quasilikelihood RIGLS estimation method. Odds ratios (OR) were computed, along with their 95% confidence intervals (CI), from the logistic model. The percentage variations of tobacco consumption levels (in cigarette equivalents) were calculated, along with their 95% CI, from the lognormal model.

RESULTS

The weighted sample included 49% of men. The average age was 43, 15% of individuals were non-retired blue collar workers and 10% upper white collar workers. Nearly 60% of individuals aged 25 or over had not completed secondary education, and 25% had graduated from university. Overall, 31% of the individuals smoked at least one cigarette per day, and smokers consumed an average of 14 cigarettes per day (standard deviation = 10).

According to the individual factors model, men were more likely to smoke than women (OR 1.44, 95% CI: 1.32–1.57) (table 1), and when they smoked, consumed larger amounts of tobacco (percentage variation in cigarette equivalents +8%, 95% CI: +2% – +15%) (table 2). The risk of tobacco use was higher among the least educated (OR low vs. high 1.50, 95% CI: 1.32–1.70), who also consumed larger amounts of tobacco when they were smokers (percentage variation +22%, 95% CI: +12% – +32%). Neither income nor occupational status showed any significant effect on either frequency or consumption levels. Divorcees were more prone to smoke than married individuals (OR 2.07, 95% CI: 1.74–2.45), and when they smoked, they also consumed more tobacco (percentage variation +24%, 95% CI: +12% – +38%). Craftsmen-shopkeepers smoked more frequently (OR 1.48, 95% CI: 1.09–2.01), and farmers less frequently (OR 0.45, 95% CI: 0.29–0.69) than upper white collar workers; never married individuals were more prone to smoke than married individuals (OR 1.36, 95% CI: 1.22–1.52). However, no association was found between these groups and the levels of tobacco consumption.

Table 1.

Percentage of smokers and multivariate multilevel odds ratios of smoking status (model 1), France, 1999

	Percentage	OR (95% CI)
Age	–	0.55 (0.51–0.59)
Sex
Female	26.9	1.00
Male	34.7	1.44 (1.32–1.57)
Education
Highest	27.7	1.00
Medium	34.4	1.38 (1.19–1.60)
Lowest	29.8	1.50 (1.32–1.70)
Occupation
Upper white collar worker	29.5	1.00
Intermediate	32.6	0.97 (0.82–1.14)
Low white collar worker	36.4	1.09 (0.91–1.30)
Blue collar worker	41.7	1.08 (0.90–1.30)
Farmer	16.3	0.45 (0.29–0.69)
Craftsman-shopkeeper	40.8	1.48 (1.09–2.01)
Income
Highest	27.8	1.00
Medium high	32.9	1.03 (0.90–1.19)
Medium low	31.1	0.96 (0.85–1.07)
Lowest	33.8	1.03 (0.89–1.18)
Employment status
Full time work	35.1	1.00
Part time work	32.5	0.97 (0.84–1.12)
Unemployed	40.7	1.09 (0.91–1.30)
State subsidized job	45.0	1.07 (0.80–1.42)
Marital status
Married	25.0	1.00
Never married	42.2	1.36 (1.22–1.52)
Divorced	38.3	2.07 (1.74–2.45)
Widowed	13.2	0.93 (0.71–1.22)
Random components
σ2u0	0.014 (0.007)a

^ap<0.05

Table 2.

Average number of cigarettes smoked per day and estimated adjusted increases in amounts consumed (model 1), France, 1999

	Mean	Percentage variation	(95% CI)
Age	–	0%	(−4% – +5%)
Sex
Female	12.6	Ref.
Male	14.6	+8%	(+2% – +15%)
Education
Higher	13.5	Ref.
Medium	13.2	+3%	(−6% – +13%)
Lower	15.1	+22%	(+12% – +32%)
Occupation
Upper white collar worker	14.8	Ref.
Intermediate	13.9	+1%	(−9% – +12%)
Low white collar worker	13.5	−5%	(−16% – +6%)
Blue collar worker	15.6	+4%	(−8% – +17%)
Farmer	18.2	+25%	(−8% – +70%)
Craftsman-shopkeeper	17.3	+6%	(−12% – +28%)
Income
Highest	14.4	Ref.
Medium high	14.0	−2%	(−10% – +8%)
Medium low	13.9	−2%	(−9% – +6%)
Lower	13.5	+1%	(−8% – +10%)
Employment status
Full time work	14.8	Ref.
Part time work	13.1	−4%	(−12% – +6%)
Unemployed	14.8	+8%	(−3% – +20%)
State subsidized job	11.1	−4%	(−19% – +13%)
Marital status
Married	14.4	Ref.
Never married	12.5	+2%	(−5% – +9%)
Divorced	17.3	+24%	(+12% – +38%)
Widowed	14.1	+4%	(−14% – +27%)
Random components
σ2u0	0.007 (0.003)a

^ap>0.05

In the individual factors model, the area level random intercept variations were only significant for the risk of tobacco use, and not for consumption levels (bottom of tables 1 and 2). The size of the county of residence was then introduced into the models (table 3, model 2). The risks of smoking appeared to be lower for residents of medium sized towns (OR 0.89, 95% CI: 0.79–1.00) or rural counties (OR 0.87, 95% CI: 0.78–0.98), by comparison with residents of larger cities. For the risk of smoking, the area level unexplained variations decreased by 14% from model 1 to model 2, and were no longer significantly different from zero. No effect of county of residence was found on tobacco consumption levels for smokers.

Table 3.

Fully adjusted multilevel odds ratios of smoking status and estimated adjusted increases in amounts consumed for contextual factors, France, 1999

	Smoking status	Quantity smoked
	OR (95% CI)	Percentage variation	(95% CI)
Model 2
County of residence
Large city	1.00	Ref.
Medium sized city	0.89 (0.79–1.00)	+4%	(−3% – +13%)
Small town	0.97 (0.86–1.09)	−1%	(−9% – +6%)
Rural county	0.87 (0.78–0.98)	0%	(−7% – +8%)
Random components
σ2u0	0.012 (0.007)a	0.007 (0.003)a
Model 3
County of residence
Large city	1.00	Ref.
Medium sized city	0.92 (0.82–1.04)	+7%	(−1% – +15%)
Small town	1.00 (0.89–1.13)	+1%	(−7% – +9%)
Rural county	0.91 (0.81–1.02)	+2%	(−5% – +10%)
GDP per capita	1.07 (1.01–1.12)	+4%	(0% – +8%)
Random components
σ2u0	0.007 (0.006)a	0.006 (0.003)a

^ap>0.05

When the area level GDP was added to the models (table 3, model 3), residents of medium sized towns or rural counties no longer showed a risk of smoking significantly lower than that of individuals from larger cities. The risks of tobacco use appeared to increase significantly with area level GDP per capita (OR 1.07, 95% CI: 1.01–1.12 for an increase in GDP by one standard deviation). The introduction of GDP into the model for risk of smoking led to an additional 42% drop in area level random intercept variations. The level two residual variance decreased by 50% from model 1 to model 3. Among smokers, residents of areas with a higher GDP consumed larger amounts of tobacco (percentage variation +4%, 95% CI: 0% – +8 for an increase in GDP of one standard deviation).

Table 4 shows a number of predictions obtained from the full models. A 43-year-old married woman (mean age) with a high level of education and residing in a large city located in a low GDP area (the fifth percentile in the distribution of the GDP for the area of residence) had a 20% chance of being a tobacco user and consumed 9 cigarettes per day if she was a smoker. A man with an identical profile had a 27% chance of being a smoker, and, if an actual smoker, consumed 10 cigarettes per day, on average. If such a man was divorced, the probability of tobacco use was 0.43 and the average number of cigarettes consumed daily was 12. If instead of being highly educated, this man had not completed the secondary cycle, he had a 53% chance of being a smoker, and consumed an average of 15 cigarettes per day if he was a smoker.

Table 4.

Estimated probability of tobacco use and estimated consumption levels for a 43-year-old individual (mean age), France 1999

				Probability of smoking	Quantity smoked (cigarettes /day)
Woman	+ Married	+ High level of education	+ Low GDP	0.20	9
Man	+ Married	+ High level of education	+ Low GDP	0.27	10
Man	+ Divorced	+ High level of education	+ Low GDP	0.43	12
Man	+ Divorced	+ No Secondary education	+ Low GDP	0.53	15
Man	+ Divorced	+ No Secondary education	+ High GDP	0.60	17

When the same individual was moved from the fifth to the ninety-fifth percentile of the GDP distribution, the risk of tobacco consumption increased from 53% to 60%, and the average daily consumption from 15 to 17 cigarettes per day.

DISCUSSION

We set out to determine if any of the high risk groups were also combination risk groups (with a high tobacco consumption level for the smokers within these groups).

As with this study, several other studies showed that men faced a higher risk of smoking than women.^12,13,36,45 Lower consumption levels for female smokers vs. their male counterparts have also been established.¹⁰ Similarly, many studies concluded that the least educated smoked more frequently.^{4–8,11,12,14–18} However, not as many of these studies showed that smokers within this group consumed more tobacco. It is interesting to note that in both models, the socioeconomic gradient appeared only in terms of educational level, and not in terms of income or occupation. It has been suggested that a higher level of education could be associated with a greater health self-awareness,⁴ a better receptivity to prevention messages and better stress management.⁴⁵

Marital status also provided a condition of combined risk: divorcees appeared to be at a higher risk of tobacco use,^10,12 and when they were smokers, they consumed larger amounts of tobacco.¹⁰ It was not possible to distinguish between divorced individuals who lived alone and those who had a partner. However, divorcees, who are more likely to live alone than married individuals, may feel less pressure with respect to smoking in their own home.²¹ Such a factor may also offer a partial explanation for the higher risk of tobacco use among individuals who were never married.⁴⁶

At the level of the area of residence it was observed that, after adjustment for individual factors and for the size of the county of residence, the risk of tobacco use as well as the amounts of tobacco consumed by smokers increased with the GDP per capita. Since results were adjusted for the size of the county of residence, the effects of the area level GDP may not be due to the local environment of residence, but to the fact that, on a larger scale, a consumer society is better established in economically more affluent areas.

However, GDP is probably a proxy for a set of more complex mechanisms which remain to be identified.³² The use of administrative subdivisions in contextual analysis has often been criticized, especially when these are not the result of a fine division of territory and do not reflect the immediate environment of the surveyed individuals.³³ However, French ‘départements’, as defined previously, are well suited to an analysis of the effects of GDP, because local government institutions within them play an active economic role, through their public policies. One can always argue that the identified area level effects are the results of lack of adjustment at the individual level.^12,29,33 However, several socioeconomic variables were used to outline as many of the diverse individual situations as possible. On the other hand, the adjustment for local level factors was probably insufficient: only the size of the county of residence was taken into consideration. Ideally, other characteristics, such as the social and occupational make-up of its population should have been considered as well.

CONCLUSIONS

The significant spreads in consumption levels predicted by the model seem to justify the combined use, in such analyses, of consumption levels for smokers in addition to the risk of smoking, in order to identify profiles with the maximum risk. In this study it was possible to identify various groups with both a high risk of tobacco use and a high level of consumption among smokers, on the basis of individual (being male, divorced or less educated) and environmental (living in a high GDP area) factors. Prevention efforts should thus be focused on such groups.