Worldwide, tobacco smoking is seen as the most urgent health risk behaviour requiring prevention today [1]. The list of diseases caused by smoking includes cancer, cardiovascular diseases, respiratory illnesses and many more [2]. Smoking significantly reduces health-related quality of life [3, 4] and is responsible for more than five million deaths worldwide every year [5]. Mortality and morbidity associated with smoking have considerable financial consequences for healthcare systems and economies [6, 7]. In Germany, smoking is still prevalent despite a reduction in recent years: 30% of the population aged 18 and older are smokers, 26% former smokers, while 44% have never smoked [8]. Previous studies have shown that smoking results in higher utilisation of medical services [9, 10], augmented work absenteeism [11–13] and increased resulting costs [14, 15]. Despite their shortened life expectancy [16], smokers have higher lifetime medical costs [17–19].

Several studies have explored the financial burden of smoking in Germany [20–26], reporting total smoking-attributable costs from €17.4 billion to €33.6 billion for different base years, with productivity losses accounting for as much as between 72-74% of total costs [20, 22]. However, these previous German studies all used a top-down approach. As this approach uses the concept of smoking-attributable fractions for mortality and morbidity known to be caused by smoking, the estimated financial burden of smoking may be inadequate if diseases and risks used for attribution do not fully reflect the structure of current smoking impact in Germany. For example, these studies used attributable mortality risks for attributing morbidity due to data limitations. Mortality risks may differ from morbidity risks and from the utilisation risk in the population at the time of sampling. Alternatively, the actual economic impact of smoking can be calculated by estimating excess costs of smokers compared to never smokers in a data set comprising current utilisation of individuals. This approach has the advantage of considering the entire spectrum of disease consequences associated with smoking and confounding factors [27].

The objective of this study was to assess differences in the utilisation of medical services and in direct and indirect costs from a societal perspective among current, occasional, former and never smokers using a population-based sample. To our knowledge, this is the first study in Germany to apply a bottom-up approach to estimate excess costs of smoking.

This study investigated the effect of smoking status on healthcare utilisation and direct and indirect costs in a population-based sample.

Total annual costs for current and former smokers were 24% and 35% higher, respectively, than for never smokers. Smoking was therefore associated with annual excess costs of €743 per current and €1,108 per former smoker. This was the first study to analyse costs of smoking in Germany using a bottom-up approach.

In all, our findings are in line with other international studies that have shown increased healthcare utilisation and costs in current and former smokers [6, 9, 12, 14, 15, 43]. Similar to a health survey from the US [9], we found an increased risk of inpatient visits and an increased number of physician visits in former smokers. Our findings that current smokers showed a lower probability of physician treatments compared to never smokers, but a higher number of treatments if they used physician treatments at least once, is comparable to a recent study which found a decreased likelihood of current smokers to use primary care services but slightly increased costs [44]. This pattern could be explained by the possibility of a healthy smoker effect or special attitudes of smokers which translate into denial of disease and delays in seeking healthcare [44].

Regarding cost, our results suggest that former smokers incur even higher costs than current smokers. Within the subsample of former smokers, we observed that those who had quit smoking in the previous 12 months caused considerably higher total costs than those who had quit more than 12 months previously (factor 2.37, p=0.0005). Other studies also have found that former smokers induce higher medical costs within a period after cessation or shortly before quitting [9, 14, 45]. This may be explained by arising health problems being the reason for smoking cessation. Nevertheless, subsample analysis showed that even former smokers who had quit more than 10 years ago cause higher total annual costs than current smokers (28% vs. 23% higher total costs compared to never smokers, p=0.002).

Further analysis showed that subjects who quit smoking due to existing medical conditions had higher total costs irrespective of their period of non-smoking. We also found that former smokers who quit in order to prevent future diseases showed substantially lower total costs, which could be explained by overall more health-conscious behaviour.

Up to now, only a few studies have examined the economic burden of cigarette smoking in Germany [20–23]. Using a top-down approach, these studies identified direct and indirect costs attributable to smoking of €17.4 billion for the year 1993 [22] and €21 billion for the year 2003 [21]. As this approach uses the concept of smoking-attributable fractions for mortality and morbidity known to be caused by smoking, health conditions where smoking may be one of several contribution factors are neglected. Furthermore these previous top-down studies used attributable mortality risks for attributing morbidity. Mortality risks may differ from morbidity risks and from healthcare utilizations probabilities. In addition, in focussing on mortality risks this method solely considers smoking-related fatal diseases and neglects non-fatal health consequences of smoking like e.g. osteoporosis or eye diseases like cataract and glaucoma. Therefore this approach is known to downward-bias cost estimates [27].

Compared to top-down approaches, calculating excess costs based on subject-level data has the advantage of considering the entire spectrum of disease consequences associated with smoking. In addition, the bottom-up methodology used in this study provides better adjustment for the actual impact of smoking on health conditions and for population characteristics as it considers other differences in smokers and non-smokers regarding education and other risky behaviours besides smoking.

Disregarding issues of representativeness, the results of our study can be extrapolated to the whole population of Germany using data on national smoking prevalence [8]. Excluding costs of premature deaths, which were not measured in this study, smoking caused costs of €31.3 billion in 2008. Direct costs amounted to €17.9 as opposed to €8.23 billion in the top-down approach (inflated from €7.48 billion in 2003 using the consumer price index (CPI)) [21]. However, our results can only be based on the prevalence of smoking in a study population aged 32-81 years, which is lower than the national average used for projection. Taking these limitations into account, results strongly indicate that cost-of-illness estimates at the population level by the top-down approaches have been too conservative.

Compared with the most recent German top-down study, direct costs per current smoker derived from our bottom-up study exceed those of the top-down approach of €381 (inflated from €346 in 2003 using the CPI) [21] by nearly a factor of 2. This can be explained by the fact that top-down studies solely consider costs of diseases which are known to be associated with smoking or by different populations under study. Furthermore, using the bottom-up approach, costs for former smokers exceed those for current smokers by more than 45%. In contrast, using the top-down approach, costs for former smokers are smaller than those for current smokers, because all relative risks of smoking-related diseases for former smokers are smaller than those for current smokers. As former smokers in the cross-sectional KORA sample are about 8 years older than current smokers, they incur a higher probability of suffering from smoking-related disease with a delayed onset whereas health damage resulting of smoking may not yet have been revealed in current smokers.

A methodological issue, top-down studies base their calculation of excess cost on the entire adult population. In this bottom-up approach, the KORA population sample does not comprise younger age groups and might thus incur higher excess costs on average. As top-down studies do not consider age in the calculation of attributable healthcare utilisation and costs, an age-standardized comparison is beyond the scope of this paper. Differences in underlying age distribution are a restriction to this comparison of excess costs.

There are several limitations to our study. First, the cross-sectional design is susceptible to recall bias, as participants were asked to provide information retrospectively. By applying a recall time horizon of 1 week for pharmaceuticals, 3 months for physician visits and 12 months for more memorable services, the study design attempted to minimise this problem. This approach has been shown to be valid [46]. Nevertheless, recall error may have occurred, but it is unlikely to have affected the validity of our results because this may not have influenced the differences between the smoking groups [47].

Furthermore, as smoking habits were assessed by self-report without validation via biochemical tests, participants may have given socially desirable answers which are also prone to recall bias regarding smoking behaviour in the past. Therefore, the number of smokers may be underestimated, but any bias was minimised by comparing current information on smoking status with information from the previous survey.

Also, despite high recruitment efforts which positively affects representativeness [48], we cannot exclude selection bias. Whereas 75% of occasional, former and never smokers of the previous S4 survey were followed-up in the F4 survey, current smokers were underrepresented with a follow-up rate of 65%. This could be caused by higher death rates of current smokers or other unobserved factors, which could lead to systematic under- or overestimation of costs.

Monetary valuation of health services requires several assumptions that may cause under- or overestimation of costs. Also, unit costs were inflated to the year 2008. As the German healthcare sector has undergone considerable changes due to policy and law, real costs may have changed differently. Nevertheless, our updating approach represents a pragmatic approximation, and, although this may influence the amount of costs, differences between the smoking groups are unlikely to be affected.

Moreover, by using mean costs, differences in treatment intensity are not considered. Hospital costs in particular can differ significantly depending on the specialty department [32].

Drug costs were estimated based on participants’ information on intake and dosage. Using the defined daily dose methodology, as suggested by the WHO [49], leads to 5.4% higher costs with differences between smoking groups, but their ranking remains unchanged. As this approach assumes full compliance of patients, it was not considered.

Additional components of medical services such as nursing, medical aids and appliances and other medical therapies like logopaedia could not be included. However, the components considered in our study covered about 75% of total healthcare costs in 2008 [50]. Furthermore, direct non-medical costs, such as time and travel expenses, costs of premature death and intangible costs could not be considered. Although this probably led to underestimation of total costs for society, it is unclear if relative differences between smoking groups were affected.

Moreover, we cannot exclude confounding through differences in unobserved behaviours. The effect of smoking could, therefore, be under- or overestimated, for example by not adjusting for other risk behaviour patterns or socio-demographic characteristics, which could differ systematically between the smoking groups. In order to minimize potential confounding we controlled for alcohol consumption and physical activity. The inclusion of these factors hardly influenced the results concerning smoking status. Nevertheless, the levels of physical activity and alcohol consumption could also be the result of underlying health conditions which cannot be analysed in cross-sectional data. E.g. the finding that high risk alcohol consumption is associated with lower direct medical costs has to be explored in detail elsewhere.

Finally, the cross-sectional design of the KORA study gives a snapshot of medical utilisation and its costs, but does not allow longitudinal analysis using a life-cycle approach and therefore any accounting for the possibility of shorter life expectancy of smokers. Several studies have shown that lifetime medical costs of current and former smokers are increased compared with never smokers and outweigh potential cost savings due to shorter life expectancy [17–19].

In conclusion, our results emphasise that smoking incurs high direct and indirect costs, and therefore imposes a significant burden on society. The economic impact of this public health problem highlights the importance of efforts aiming to discourage smoking. Although further research is needed to examine dose-effect relations of smoking, the timing of quitting on societal costs and results from longitudinal studies, policy makers should strengthen their efforts towards prevention, for example in realising the WHO Framework Convention on Tobacco Control [51] which suggests measures to reduce tobacco demand and therefore prevents smoking and its negative economic consequences.