Smoking-attributable hospitalisation costs in Europe: an econometric analysis of the European Health Interview Survey, 2019

Abstract

Introduction

Despite significant progress achieved in the past decade, the European Union (EU) tobacco epidemic remains one of the world’s most important. As the decline in smoking prevalence stagnated in the past years and little legislative progress has been made in tobacco control policies, we provide updated estimates of the healthcare costs of smoking in the EU.

Methods

The study uses 2019 data from the European Health Interview Survey, covering all EU Member States except for France, Germany, Ireland and Sweden, which are excluded because of data access and data quality reasons. We apply econometric methods to examine the differential use of inpatient and daycare hospitalisation services among smokers and never smokers. Based on the model estimates, we compute the smoking-attributable costs for these hospital services across Member States.

Results

We find that smoking is positively associated with inpatient and daycare hospitalisation. Across countries, smoking-related costs range from EUR 5 billion (Spain) to EUR 28 million (Latvia) for inpatient hospitalisation, and from EUR 550 million (Netherlands) to EUR 3 million (Malta and Slovenia) for daycare services.

Conclusions

The results of this paper show that smoking imposes significant healthcare costs across the EU. Importantly, the total healthcare costs of smoking are likely to be significantly higher than reported in this study, which only focuses on two types of healthcare services. These updated estimates underscore the need to reinvigorate tobacco control policies at national and EU level to reduce the health and economic burden of smoking and protect the health systems resources.

WHAT IS ALREADY KNOWN ON THIS TOPIC

• Monitoring smoking prevalence and the economic costs of smoking is important to inform tobacco control policies.

WHAT THIS STUDY ADDS

• The latest estimates on the healthcare costs of smoking in Europe are dated back to an epidemiological study conducted in 2009. This study provides updated estimates of the healthcare costs of smoking using econometric methods based on 2019 data.

HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY

• The results show that smoking depletes health systems of precious resources and highlight the importance of renewing efforts to reduce tobacco use.

Introduction

The adverse health impacts of smoking are well documented.¹ Numerous studies have also quantified the economic costs that smoking imposes on society, distinguishing between the indirect costs associated with workday losses due to morbidity and the direct monetary losses linked to healthcare and non-healthcare costs.^{2 3} Remarkably, the latest comparative estimates of the healthcare costs of smoking in Europe are dated back to a 2009 study commissioned by the Directorate General for Health and Consumers and are based on epidemiological methods.⁴ The epidemiological approach requires determining a list of diseases linked to smoking and calculating disease-specific smoking-attributable fractions (SAF) by comparing the fraction of deaths attributable to smoking through an epidemiological formula. The popularity of the epidemiological approach stems from the fact that it relies on aggregate level data instead of individual level data, which might not be available to the researcher. In the presence of detailed health survey data, econometric methods linking individuals’ smoking status with their healthcare use/costs can be implemented. Compared with epidemiological methods, econometric modelling allows for a more comprehensive analysis since the focus is on the use of healthcare goods and services, without restricting to certain smoking-related diseases. Since epidemiological methods require to first determine a list of diseases and then estimate the SAF for each particular disease, the estimated aggregate costs of smoking might be downward biased if such a list is not comprehensive enough.⁵ Additionally, econometric models are better suited to control for risk factors that might be correlated with both individuals’ smoking behaviour and health status, such as alcohol consumption, education, fruit and vegetable intake and physical activity.⁶ A review paper that compared different studies estimating the direct healthcare costs of smoking concluded that the econometric approach compares favourably with the epidemiological approach in the analysis of the economic burden of smoking.⁷

In this study, we provide estimates of hospitalisation costs attributable to smoking in the 27 European Union (EU27) Member States (MS) using individual-level data from the third wave of the European Health Interview Survey (EHIS) conducted between 2019 and 2020. The contribution of this study is twofold. From a methodological perspective, this research represents the first assessment of the healthcare burden of smoking in the European Union (EU) based on econometric methods applied to individual level data. Previous comparative estimates relied on epidemiological approaches and date back more than a decade, while more recent studies have been limited to individual countries. From a policy point of view, this paper provides updated comparative estimates of the healthcare costs of smoking in the EU. At a time of stagnating progress in the implementation and effectiveness of tobacco control policies, efforts to monitor the economic costs associated with tobacco use in Europe are particularly important. Despite significant progress achieved in the past decade, the EU tobacco epidemic remains one of the world’s most important, with 24% of adults currently smoking in the EU and a decline in smoking prevalence of only 1% between 2020 and 2023, according to the latest estimates from the Special Eurobarometer 539.⁸ While national-level measures, such as advertising bans or plain packaging, can contribute to reducing the burden of smoking, action at EU level is needed. Nonetheless, the revision of the 2011 Tobacco Taxation Directive (TTD) and 2014 Tobacco Products Directive (TPD), which were initially planned within the Europe’s Beating Cancer Plan (EBCP) for 2022 and 2024 to support the goal of achieving a tobacco-free generation, is long delayed.⁹ The findings of our study highlight the fact that, despite significant cross-country differences, the healthcare costs of smoking are substantial across the EU27, calling for urgent tobacco control measures to address them.

Methods

No ‘patient and public involvement’.

Data and sample

The writing of the report has been informed by the Consolidated Health Economic Evaluation Reporting Standards reporting guidelines.¹⁰ In this study, we explore the associative link between individuals’ smoking behaviour and the use of inpatient and daycare hospitalisation using the latest data from the 2019 EHIS covering the EU27. The EHIS is provided by Eurostat under strict data confidentiality agreements. In line with Eurostat’s data access regulations, no individual identifiers are included in the dataset. No additional ethical approval was required for this study. The EHIS targets the population older than 15 years of age living in private households and contains a rich set of information on respondents’ demographic and socioeconomic characteristics, health status, healthcare use and health determinants. We restrict the analysis to individuals aged 30 years and above, to account for the fact that the adverse health effects of smoking take years to develop.^{11 12} We also exclude individuals who reported being admitted to a hospital for an accident or injury.

The main outcome variables of the investigation are represented by the inpatient and daycare hospitalisation rate of the population under study in the year prior to the interview. The former refers to the number of overnight stays in a hospital and the latter to the number of times admitted as a day patient in a hospital. The distribution of such variables is presented in online supplemental table 1. Two aspects are noteworthy. First, since the EHIS does not distinguish the medical purpose of the treatment (eg, whether hospitalisation occurred to relieve symptoms, improve body functions or alleviate pain from a permanent deterioration of health), the outcome variables refer to treatments for both curative and rehabilitative care, and long-term care. Second, both variables include all possible financing schemes.

The principal independent variable of interest is represented by the smoking status of the respondents, which we code as a binary indicator. We classify an individual as an (ever) smoker if she currently smokes, even if occasionally, or if she is an ex-smoker. Hence, a person’s smoking status equals zero if and only if she has never smoked in her life. Table 1 presents the distribution of the different types of smoking behaviour in each country.

Table 1. Distribution of smoking variables.

EU MS	(Ever) smoker	Daily smoker	Occasional smoker	Former smoker	Mean years of smoking among former smokers
AT	47.08% (N=7256, missing=0%)	18.47% (N=2847, missing=0%)	4.89% (N=754, missing=0%)	23.72% (N=3655, missing=0%)	17.47
BE	39.97% (N=3049, missing=0.73%)	14.55% (N=1118, missing=0%)	4.22% (N=324, missing=0%)	21.13% (N=1607, missing=1.05%)	18.51
BG	43.22% (N=3105, missing=0.04%)	26.78% (N=1925, missing=0%)	6.57% (N=472, missing=0%)	9.86% (N=708, missing=0.04%)	17.35
CY	37.56% (N=2284, missing=0%)	20.14% (N=1225, missing=0%)	2.88% (N=175, missing=0%)	14.54% (N=884, missing=0%)	20.85
CZ	41.04% (N=3265, missing=0%)	18.15% (N=1444, missing=0%)	5.84% (N=465, missing=0%)	17.04% (N=1356, missing=0%)	16.49
DE	91.37% (N=10 582, missing=49.1%)	15.96% (N=3631, missing=0%)	4.53% (N=1030, missing=0%)	51.13% (N=5921, missing=49.11%)	NA
DK	49.68% (N=3075, missing=0.34%)	11.85% (N=736, missing=0%)	5.81% (N=361, missing=0%)	31.99% (N=1978, missing=0.43%)	20.49
EE	41.04% (N=1996, missing=0%)	17.95% (N=873, missing=0%)	4.81% (N=234, missing=0%)	18.28% (N=889, missing=0.02%)	16.50
EL	37.6% (N=3015, missing=0%)	22.63% (N=1815, missing=0%)	3.87% (N=310, missing=0%)	11.1% (N=890, missing=0%)	24.43
ES	42.9% (N=9443, missing=0.01%)	18.88% (N=4157, missing=0%)	2.09% (N=460, missing=0%)	21.93% (N=4826, missing=0.01%)	20.37
FI	71.69% (N=4228, missing=2.69%)	8.32% (N=504, missing=0%)	5.4% (N=327, missing=0%)	57.73% (N=3397, missing=2.92%)	3.17
HR	41.38% (N=2236, missing=0.39%)	22.25% (N=1207, missing=0%)	3.43% (N=186, missing=0%)	15.61% (N=843, missing=0.44%)	18.27
HU	39.95% (N=2083, missing=0.11%)	20.04% (N=1046, missing=0%)	2.22% (N=116, missing=0%)	17.68% (N=921, missing=0.21%)	17.29
IE	44.41% (N=3359, missing=0.01%)	14.26% (N=1079, missing=0%)	3.64% (N=275, missing=0%)	26.51% (N=2005, missing=0.01%)	17.43
IT	39.29% (N=17 294, missing=0.53%)	15.96% (N=7061, missing=0%)	4.98% (N=2202, missing=0%)	18.26% (N=8031, missing=0.62%)	20.36
LT	35.33% (N=1733, missing=0%)	16.86% (N=827, missing=0%)	4.38% (N=215, missing=0%)	14.09% (N=691, missing=0%)	13.62
LU	39.68% (N=1660, missing=1.6%)	10.49% (N=446, missing=0%)	6.59% (N=280, missing=0%)	22.36% (N=934, missing=1.74%)	14.80
LV	36.58% (N=2195, missing=0.02%)	20.91% (N=1255, missing=0%)	4.08% (N=245, missing=0%)	11.58% (N=695, missing=0.03%)	14.66
MT	37.68% (N=1641, missing=0.07%)	18.24% (N=795, missing=0%)	3.88% (N=169, missing=0%)	15.55% (N=677, missing=0.07%)	17.44
NL	49.11% (N=3712, missing=6.96%)	13.64% (N=1108, missing=0%)	5.2% (N=422, missing=0%)	30.58% (N=2182, missing=12.15%)	NA
PL	40.59% (N=7943, missing=0.51%)	18.46% (N=3631, missing=0%)	3.05% (N=600, missing=0%)	18.98% (N=3712, missing=0.55%)	16.53
PT	36.46% (N=5175, missing=0.13%)	12.1% (N=1720, missing=0%)	2.14% (N=304, missing=0%)	22.2% (N=3151, missing=0.13%)	20.95
RO	31.55% (N=5095, missing=0.01%)	17.05% (N=2754, missing=0%)	8.19% (N=1323, missing=0%)	6.3% (N=1018, missing=0.01%)	16.89
SE	38.73% (N=3667, missing=0.71%)	5.77% (N=550, missing=0%)	4.98% (N=475, missing=0%)	27.91% (N=2642, missing=0.74%)	14.91
SI	34.44% (N=3351, missing=0.29%)	15.11% (N=1474, missing=0%)	5.69% (N=555, missing=0%)	13.59% (N=1322, missing=0.29%)	15.28
SK	39.89% (N=2187, missing=0%)	19.62% (N=1076, missing=0%)	6.02% (N=330, missing=0%)	14.24% (N=781, missing=0%)	13.03

The table shows the prevalence of different types of smoking behaviour by country. For each variable, we report in parentheses the fraction of the sample with missing data. We define ever smoking as being a daily, occasional or former smoker. Germany is the country with the largest fraction of missing values (49%), due to missing answers on former smoking status. The last column shows the average number of years of smoking among former smokers.

AT, Austria; BE, Belgium; BG, Bulgaria; CY, Cyprus; CZ, Czech Republic; DE, Germany; DK, Denmark; EE, Estonia; EL, Greece; ES, Spain; EU MS, European Union Member States; FI, Finland; HR, Croatia; HU, Hungary; IE, Ireland; IT, Italy; LT, Lithuania; LU, Luxembourg; LV, Latvia; MT, Malta; NL, Netherlands; PL, Poland; PT, Portugal; RO, Romania; SE, Sweden; SI, Slovenia; SK, Slovakia.

In terms of geographical coverage, our analysis covers all EU27 except for France, for which access to the data was not granted; Germany, which was excluded because half of the sample missed data on former smoking status (coded as ‘non stated’ in the EHIS microdata); and Ireland and Sweden, which were excluded since more than 70% of the sample missed data on inpatient and daycare hospitalisation, again due to item non-response. In addition, Italy was excluded from the daycare model because of missing data on the dependent variable. At an individual level, observations with missing data were excluded from the analysis, rather than having their missing values imputed. The proportion of such missing data is generally small, as shown in online supplemental table 2. Nonetheless, given that household income exhibits higher item non-response in some countries, we also performed a robustness check excluding this variable from the models. The results, which are presented in online supplemental tables 11–14, remained substantively unchanged.

Statistical methods

For each measure of healthcare service, we use a ‘reduced-form’ econometric model⁶ and estimate the following two model equations:

$${\displaystyle {\displaystyle Pr\left(y_{it}>0\right)=f_1\left(X_i,Z_i,S_i\right)}}$$ (1)

$${\displaystyle {\displaystyle y_{it}|\left(y_{it}>0\right)=f_2\left(X_i,Z_i,S_i\right)}}$$ (2)

where $y_{it}$ is the healthcare utilisation outcome of interest (ie, inpatient or daycare hospitalisation), $X_i$ represents individual demographic and socioeconomic characteristics, $Z_i$ includes a set of variables affecting individuals’ health status and $S_i$ is an indicator variable for whether individual i is a smoker. Gender, age (30–40, 40–50, 50–65, more than 65), residential area, level of education (primary, secondary and tertiary), marital status, employment status, household size and households’ income quantile were considered main controls for the demographic and socioeconomic characteristics of the respondents. Among the health determinants, we included an indicator for practising sports at least 2 days a week, an indicator for being obese (body mass index (BMI) at least 30), one for being underweight (BMI less than or equal to 18) and a dummy variable for following a healthy diet, defined as eating vegetables and fruits at least 4 days per week. Finally, we also control for individuals’ level of alcohol consumption, distinguishing between people who never drank alcohol, people who quit drinking and people who self-reported episodes of heavy drinking (60 g of ethanol) on a weekly, monthly or yearly basis.

To capture cross-country heterogeneity, each model equation is estimated separately for each Member State rather than in a pooled specification with country fixed effects. By means of Equation 1, we model the probability that an individual received inpatient or daycare hospital assistance in the year prior to the interview. It is used to assess the impact of smoking behaviour on the likelihood of consuming the goods and services represented by the dependent variable on the left-hand side. It is estimated through a logit model. In Equation 2, we model the natural logarithm of the number of nights or days admitted to the hospital. It is used to investigate the factors that explain the length of stay for hospital inpatients and the number of admissions as day care patients. In this case, the estimation is restricted to those individuals who reported at least one night (inpatient) or at least 1 day (daycare) of hospitalisation in the prior year. It is estimated by ordinary least squares. Since the distribution of hospital nights and days is highly skewed, we apply a natural logarithmic transformation, which improves model fit and allows for coefficients to be interpreted as approximate percentage changes in hospitalisation days.⁶ The choice of the linear specification over Poisson or negative binomial alternatives, which are commonly used for count data, was motivated by its superior performance in terms of Aikaike Information Criterion (AIC) and Bayesian Information Criterion (BIC) information criteria. In addition, Ramsey’s Regression Equation Specification Error Test (RESET) tests did not indicate mis-specification of the log-linear model. Model diagnostics and estimation based on the negative binomial model are reported in online supplemental tables 8–10.

Given the models’ estimates, we compute the SAF using two fitted values. The first considers the predicted annual use of healthcare services for the actual categories of smokers in the sample, defined as the product between the predicted probability of consumption from the logistic model estimates and the predicted volume of consumption based on the estimates for Equation 2. Next, the same predicted annual use of health services is estimated for a hypothetical group of ‘never-smokers smokers’. This fictitious group is constructed by holding individuals’ observed socioeconomic characteristics and health determinants constant and changing only their smoking status to ‘never-smokers’. This ensures that the hypothetical group is identical to the counterfactual group in all observed covariates except smoking status. Algebraically, the SAF is then defined as follows:

$${\displaystyle {\displaystyle {\displaystyle SAF=\frac{\sum _S\left({\hat{y}}_i^S-{\hat{y}}_i^{S\to N}\right)}{\sum _S{\hat{y}}_i^S+\sum _N{\hat{y}}_i^N}}}}$$

where ${\hat{y}}_i^S$ is the predicted outcome of interest for the group of smokers, ${\hat{y}}_i^{S\to N}$ represents the predicted outcome of interest for ‘never-smokers smokers’ and S and N denote the size of the population of smokers and non-smokers, respectively. The numerator in the formula represents the excess level of the dependent variable attributable to smoking. The denominator represents an estimate of the aggregate consumption of healthcare services in the overall population.

To quantify the statistical uncertainty around the SAF estimates, we employ a non-parametric bootstrap procedure with 1000 replicates. For each country, individuals are repeatedly resampled with replacement from the dataset. Within each resample, both the logistic model and the linear model are estimated, and the SAF is recalculated using the procedure described above. This process yields an empirical sampling distribution of the SAF that incorporates the variability of the model coefficients as well as the non-linear transformation involved in the SAF definition. Standard errors were derived from the dispersion of the bootstrap estimates, and 95% CIs were obtained using the 2.5th and 97.5th percentiles of the bootstrap distribution.

Results

Table 2 shows the country-specific estimates and standard errors associated with individuals’ smoking status for each model related to inpatient and daycare hospitalisation. The smoking coefficients tend to be positive in most countries in both models for inpatient and daycare hospitalisation. Of the negative coefficients, only the logistic model estimates for Finland (FI) are statistically significant. For both types of healthcare services, the strongest smoking effects are found in the logistic model. Statistically significant estimates in the linear models for inpatient and daycare hospitalisation are found in only two (Spain (ES) and Lithuania (LT)) and one (Portugal (PT)) country, respectively. This evidence might be attributable to the evolution and treatment of smoking-related diseases. For instance, Miller et al¹³ argue that smoking-related illnesses that require inpatient care might deteriorate rapidly and require a shorter hospital stay compared with other diseases not related to smoking.

Table 2. Coefficient estimates and standard errors of the smoking indicator variable, inpatient and daycare hospitalisation.

Country	Inpatient hospitalisation		Daycare hospitalisation
	Logit model	Linear model	Logit model	Linear model
AT	0.106** (0.054)	0.029 (0.050)	0.121** (0.055)	−0.003 (0.024)
BE	0.188* (0.096)	0.110 (0.098)	0.254*** (0.093)	−0.008 (0.036)
BG	0.071 (0.101)	0.058 (0.067)	0.136 (0.106)	0.036 (0.047)
CY	0.471*** (0.120)	0.190 (0.125)	0.061 (0.166)	0.049 (0.081)
CZ	0.194** (0.080)	−0.067 (0.068)	0.145 (0.113)	−0.012 (0.048)
DK	−0.124 (0.217)	−0.126 (0.230)	0.228* (0.128)	−0.041 (0.049)
EE	0.029 (0.124)	0.168 (0.115)	−0.084 (0.146)	0.081 (0.067)
EL	0.091 (0.118)	0.025 (0.110)	0.252** (0.109)	−0.068 (0.046)
ES	0.209*** (0.066)	0.163** (0.068)	0.286*** (0.063)	0.018 (0.027)
FI	−0.301** (0.129)	0.055 (0.126)	−0.397*** (0.091)	0.001 (0.039)
HR	0.175 (0.128)	0.092 (0.135)	0.206 (0.126)	−0.027 (0.060)
HU	0.168 (0.104)	0.161 (0.107)	0.192 (0.190)	−0.130 (0.101)
IT	0.116*** (0.042)	−0.064 (0.040)	NA	NA
LT	0.301** (0.123)	0.207* (0.118)	0.318** (0.161)	0.022 (0.082)
LU	0.246 (0.151)	0.031 (0.146)	0.286*** (0.107)	0.031 (0.041)
LV	0.197* (0.112)	−0.081 (0.106)	0.102 (0.130)	−0.045 (0.068)
MT	0.162 (0.133)	0.026 (0.105)	0.261*** (0.101)	−0.028 (0.042)
NL	0.281** (0.116)	0.011 (0.155)	0.266** (0.104)	0.022 (0.040)
PL	0.197*** (0.064)	−0.036 (0.053)	0.265*** (0.085)	0.013 (0.036)
PT	0.334*** (0.095)	0.054 (0.088)	0.229*** (0.053)	0.042** (0.019)
RO	0.311*** (0.098)	0.070 (0.063)	0.080 (0.126)	−0.025 (0.050)
SI	0.044 (0.085)	−0.001 (0.086)	0.121* (0.067)	−0.034 (0.029)
SK	−0.072 (0.101)	−0.021 (0.075)	0.174 (0.167)	−0.053 (0.064)

*p<0.1, **p<0.05; ***p<0.01.

AT, Austria; BE, Belgium; BG, Bulgaria; CY, Cyprus; CZ, Czech Republic; DK, Denmark; EE, Estonia; EL, Greece; ES, Spain; FI, Finland; HR, Croatia; HU, Hungary; IT, Italy; LT, Lithuania; LU, Luxembourg; LV, Latvia; MT, Malta; NL, Netherlands; PL, Poland; PT, Portugal; RO, Romania; SI, Slovenia; SK, Slovakia.

The only counterintuitive result concerns the negative coefficient estimates in the logistic models for inpatient and daycare hospitalisation found in FI. But this might simply be a result of not being able to control for other behavioural variables such as alcohol consumption and sports frequency, which are not available for FI. While one would expect that the absence of these variables would bias upwards the coefficient on the ever smoking indicator (because of the positive correlation across unhealthy lifestyles), note that FI has a significantly larger fraction of ever smokers (72%) compared with the rest of the countries in the sample due to a relatively higher share of former smokers (58%). Quitting smoking is likely positively correlated with other healthy lifestyle behaviours, so for this country the pattern of correlation between the ever smoking variable and relevant omitted variables could well be switching the direction of the bias. Unfortunately, we cannot test this conjecture with the data at hand and therefore the results for FI must be interpreted with special caution, which leads us to exclude it from the subsequent analysis.

We report in online supplemental tables 3–6 the coefficient estimates associated with the other variables included in the model. The coefficient estimates are aligned with expectation, providing evidence in support of the model specification. We find that people belonging to older age groups exhibit higher inpatient and daycare hospitalisation rates, overweight and underweight statuses are strong predictors of both types of healthcare use, physical activity is associated with lower hospitalisation and being a former drinker is significantly correlated with higher hospitalisation. Finally, male individuals exhibit higher inpatient hospitalisation rates compared with women.

The SAFs are computed using the methodology described above based on the model estimates that are statistically significant. Since the computation of the SAFs hinges crucially on the coefficients for the ever-smoking variable, we set its value to zero in the cases where the null hypothesis of no statistical significance cannot be rejected. This means that, except for the cases of ES and LT in the inpatient hospitalisation model and PT in the daycare model, SAFs are determined by the logistic regression only. Sample averages are weighted using individual-level weights available in the EHIS data. The SAF of inpatient hospitalisation ranges from 4% in Austria (AT) to 17% in Cyprus (CY). The SAF of daycare hospitalisation ranges from 5% in AT to 13% in the Netherlands (NL). This variation across countries is partly due to the uncertainty of the model estimates, as reflected by the width of the bootstrap CIs. Nonetheless, in most countries, the lower bound of the CIs excludes the zero, suggesting that the estimated SAFs are unlikely to be explained by chance. Also, countries with higher SAF estimates, such as CY (SAF of inpatient hospitalisation is 17.1%) and ES (SAF of inpatient hospitalisation is 15.7%), also show relatively tighter confidence bounds (9.4%–24.8% in CY, 7.7%–23% in ES). Once the SAF is obtained, it is multiplied by the corresponding healthcare expenditures to obtain the smoking-attributable healthcare costs. Data on the healthcare costs by function come from the 2020 Eurostat Healthcare Expenditures Database. As shown in table 3, we find the smoking-attributable expenditures on inpatient hospitalisation range from EUR 5 billion in ES to EUR 28 million in Latvia. Smoking-related daycare expenditures are considerably lower, ranging from EUR 550 million in the NL to EUR 3 million in Malta (MT) and Slovenia (SI). The excess cost burden of smoking-attributable inpatient hospitalisation reflects the larger healthcare costs of inpatient services rather than differences in the estimates of the SAFs. A variety of factors, including disease severity, organisational aspects and resource costs, determine the allocation of resources across different types of healthcare services.¹⁴ Evidence across specialties suggests that the direct costs of inpatient treatment exceed those of daycare treatment.^{15 16} However, indirect costs (eg, travel and readmission costs) can partially offset this difference.¹⁷ The sum of the smoking-attributable inpatient and daycare costs gives the tobacco-related costs at hospital level. Smoking-attributable hospitalisation costs accounted for between 1.2% and 6.1% of total healthcare expenditures in countries where inpatient costs could be estimated. In countries where inpatient costs were not statistically significant, only daycare costs could be estimated, yielding much lower shares (0.1%–0.4%).

Table 3. Smoking-attributable healthcare expenditures in inpatient and daycare hospitalisation.

Country	SAF inpatient hospitalisation	Smoking-attributable inpatient expenditures, million euro	SAF daycare hospitalisation	Smoking-attributable daycare expenditures, million euro	Total smoking-attributable hospitalisation costs	SAF total healthcare costs
AT	0.0431 (0.003 to 0.088)	605.1 (42.1 to 1235.5)	0.0525 (0.004 to 0.101)	10.2 (0.77 to 19.6)	615.3 (42.9 to 1255.1)	1.4%
BE	0.0695 (−0.002 to 0.138)	1082.7 (−31.1 to 2149.8)	0.0965 (0.028 to 0.175)	427.5 (124.1 to 775.3)	1510.2 (92.9 to 2925.1)	2.9%
CY	0.1713 (0.094 to 0.248)	117.5 (64.5 to 170.1)	NA	NA	117.5 (64.5 to 170.1)	6.1%
CZ	0.0739 (0.010 to 0.131)	448 (60.6 to 794.2)	NA	NA	448 (60.6 to 794.2)	2.3%
ES	0.157 (0.077 to 0.230)	5050.1 (2476.8 to 7398.3)	0.1229 (0.069 to 0.175)	315 (176.9 to 448.5)	5365.1 (2653.7 to 7846.8)	4.5%
IT	0.0437 (0.014 to 0.073)	1930.2 (618.4 to 3224.3)	NA	NA	1930.2 (618.4 to 3224.3)	1.2%
LT	0.114 (0.011 to 0.156)	113 (10.9 to 154.5)	0.108 (−0.002 to 0.207)	5.8 (−0.1 to 11)	118.7 (10.8 to 165.6)	3.2%
LU	NA	NA	0.0907 (0.024 to 0.154)	13.5 (3.6 to 22.9)	13.5 (3.6 to 22.9)	0.4%
LV	0.0587 (−0.004 to 0.124)	28.4 (−1.93 to 60)	NA	NA	28.4 (−1.93 to 60)	1.3%
MT	NA	NA	0.0868 (0.022 to 0.153)	3.3 (0.83 to 5.8)		0.2%
NL	0.1389 (0.034 to 0.248)	2244.9 (549.5 to 4008.2)	0.1312 (0.032 to 0.222)	549.5 (134 to 929.7)	2794.4 (683.5 to 4937.9)	3.1%
PL	0.0712 (0.025 to 0.115)	795.5 (279.3 to 1284.8)	0.106 (0.038 to 0.171)	81.2 (29.1 to 130.9)	876.6 (308.4 to 1415.7)	2.6%
PT	0.1106 (0.052 to 0.169)	439.8 (206.8 to 672)	0.068 (0.039 to 0.096)	126.4 (72.1 to 177.5)	565.5 (278.9 to 849.5)	2.7%
RO	0.0747 (0.028 to 0.120)	373.8 (140.1 to 600.4)	NA	NA	373.8 (140.1 to 600.4)	2.7%
SI	NA	NA	0.0363 (−0.003 to 0.076)	3.2 (−0.27 to 6.8)	0.0363 (−0.003 to 0.076)	0.1%

The table shows the SAFs and the healthcare expenditures for inpatient and day care hospitalisation by country. SAF estimates are reported with 95% CIs based on 1000 non-parametric bootstrap replications. Absolute denominator on healthcare expenditures are reported in online supplemental table 7supplementary table 7.

AT, Austria; BE, Belgium; CY, Cyprus; CZ, Czech Republic; ES, Spain; IT, Italy; LT, Lithuania; LU, Luxembourg; LV, Latvia; MT, Malta; NL, Netherlands; PL, Poland; PT, Portugal; RO, Romania; SAFs, smoking-attributable fractions; SI, Slovenia.

Discussion

Our estimates link individuals’ use of traditional tobacco products to inpatient and daycare hospitalisation, two main components of the total healthcare expenditures of the EU MS. Our findings indicate that smoking imposes significant costs on the healthcare systems of the EU MS, with smoking-attributable hospitalisation costs ranging from 1.2% to 6.1% of total healthcare expenditures in countries where inpatient SAF could be estimated. These results are in line with prior evidence. A recent report from the WHO suggests that smoking-attributable costs in the European region range from 1.2% to 8.9% of total healthcare costs.¹⁸ At national level, our estimates of the SAF of inpatient hospitalisation in Italy (4.4%, 95% CI 1.4 to 7.3%) are consistent with estimates by Possenti et al, who reported that tobacco accounted for 6% of hospitalisation costs in Italy.¹² Since daycare hospitalisation data were not available for Italy, our figures are likely conservative. Similarly, the estimates of Landovská,² indicating that smoking-related direct costs represented 2.9% of healthcare spending in the Czech Republic in 2019, are remarkably close to our figures for the country of 2.3%.

Importantly, the aggregate healthcare costs attributable to smoking are likely to be significantly higher than reported in these figures, since other healthcare goods and services that are consumed to prevent and treat smoking-related illnesses are not covered in the analysis. An estimation of the smoking-attributable costs associated with other healthcare functions, including outpatient visits, dental care, home-based care and expenditures for medical goods, is needed. In addition, updated estimates on the indirect burden of smoking (such as productivity losses) are also needed in the current policy debate.

The methodology used for the investigation is subject to some limitations. First, despite aiming to cover all the EU MS, this study only provides a partial analysis of the healthcare burden of smoking in the EU since five countries—namely France, Germany, Ireland, Italy and Sweden—are omitted from the analysis. While this limitation does not affect the validity of country-level estimates, it does not allow the estimation of the aggregate healthcare costs of smoking in the EU. Second, it should be noted that, even if we are able to control for some lifestyles that could confound the relationship between ever smoking and healthcare utilisation, our estimates are based on observational data and therefore cannot be interpreted as causal effects. This, however, is a common limitation in non-experimental studies and studies whose design does not grant the identification of treatment effects. Even in the absence of a clear causal interpretation, the direction of this relationship seems clear, with smoking being associated with higher hospitalisation rates across EU MS. A further concern is the potential reverse causation, whereby individuals quit smoking after developing an illness. Importantly, this risk is mitigated by the fact that individuals who quit due to illness are classified as former smokers (ie, ever smokers), rather than never smokers. Other potentially important limitations derive from measurement error in self-reported hospitalisations or smoking status. Since hospitalisations are the dependent variable in our econometric models, this would contribute to increasing the variance of our estimators. Measurement error in the smoking indicator would on the other hand lead to attenuation bias, meaning that the size of the impact of smoking on hospitalisations would be biased downwards, so our estimations may be considered in this regard lower bounds for the true effects. Finally, our study focuses on the use of traditional tobacco products, without covering the effects of consumption of other types of tobacco and nicotine products (such as smokeless tobacco products and e-cigarettes) nor the effects of secondhand smoke exposure. Existing evidence indicates that these forms of exposure to tobacco and nicotine use produce significant harm,^{19 20} suggesting that our figures substantially underestimate the overall hospitalisation costs attributable to tobacco and nicotine products use.

This paper provides updated estimates of the burden imposed by smoking on healthcare budgets, at a time when the European Commission has set as a priority to increase Europe’s competitiveness and the EU Member States are dealing with increasing pressures on their healthcare systems, as spending on health is increasing faster than Gross Domestic Product (GDP) growth in all EU27.⁸ Healthcare expenditure represents, for all EU economies, a considerable share of all government expenditure, ranging from 9.4% in Greece to 21% in Ireland.²¹ More evident are the healthcare labour shortages that many EU countries are facing, which have also become a priority for many policy makers at the EU and national levels.²² For these reasons, having a comprehensive overview of the sources of those costs becomes essential. Commercial determinants of health have been known to pose a considerable threat to our public healthcare systems,²³ and their quantification is a crucial ingredient for the design of public health policies geared to making European healthcare systems more resilient. Exposing the burden of the commercial determinants of health is especially important at a time when key legislative proposals at the EU level are delayed. In the context of tobacco control policies, some progress was made—by means of the TTD in 2011 and the TPD in 2014—after the publication of the estimates on healthcare costs in 2009. However, with the exception of Delegated Directive 2022/2100 regarding the withdrawal of certain exemptions for heated tobacco products, no further legislative progress has been achieved since then. The adoption of the revisions of the TPD and TTD, which were initially part of EBCP, is long delayed, with no publication of the proposal on the revision of the TPD yet made by the European Commission. A recent European Commission publication on the implementation of the EBCP highlighted the possible role of the tobacco industry in undermining progress,²⁴ which would contravene the principles embedded in Article 5.3 of the WHO Framework Convention on Tobacco Control. Policy stagnation is furthermore accompanied by a lack of transparency and openness about the legislative process and the reasons behind the delay, which affects not only the legislative adoption but also the publication of the proposals for such important disease prevention measures.²⁵

Meanwhile, in a context with marginal reductions in smoking prevalence—with the latest Eurobarometer in 2023 showing only a 1% decrease with respect to the 2020⁴—the consumption of new tobacco and nicotine products is growing fast. Member States have a degree of autonomy to address the burden of smoking to their healthcare systems with domestic-level policies, for instance, introducing plain packaging, applying point-of-sale bans and supporting quit attempts. However, other measures are best accomplished at the EU level. One example is the harmonisation of plain packaging and the removal of exemptions to include health warnings in tobacco and nicotine products packages, which should be addressed with a revision of the TPD. Another is ensuring that effective minimum tax rates are consistently used across the EU, which requires the revision of the TTD.²⁶ Such supranational regulations would duly address the fact that, in a single market, the policy stance regarding tobacco control of any Member State has spillover effects for its neighbours. Recently, a great opportunity to act was provided by the Commission to the Council of the EU with its proposal for a recast of the Council Directive on the structure and rates of excise duty applied to tobacco and tobacco-related products, the TTD.²⁷ The benefits of EU Member States adopting this proposal at the Council of the EU would reduce the number of smokers in the EU by 12.1 million people, equivalent to a 3.2% reduction in prevalence and leading to a EUR 6 billion reduction in direct healthcare costs.²⁸

Therefore, the results presented in this paper, while serving as a wake-up call for the need for urgent policy action, present relevant evidence to support policy makers in the adoption of tobacco control legislation. The healthcare burden of smoking-related diseases depletes health systems of precious resources. More importantly, this burden reflects the avoidable losses in health-related quality of life that smoking still imposes on Europeans. Finally, the results show that the burden of tobacco is a shared issue across all EU Member States and, due to the cross-border impact of the commercialisation of tobacco products within the internal market, action is needed at the EU level.

Data availability statement

Data may be obtained from a third party and are not publicly available.