Data and study design
We used data from the KORA (Cooperative Health Research in the Augsburg Region) F4 survey, a population-based study conducted in 2006-2008 in Southern Germany. The F4 study is the follow-up of the KORA S4 survey (1999-2001). In brief, KORA S4 randomly selected 6,640 adults of German nationality aged 25-74 in the city of Augsburg and the two adjacent administrative districts from population registries; of those, 4,261 participated in the baseline examination. Details about design, sampling xmethod, data collection and response rates have been described elsewhere [28, 29]. 3,080 individuals from S4 (72%) also participated in the follow-up survey F4, in which self-reported information on current smoking status, healthcare utilisation and non-productive working time was assessed in standardized computer-assisted interviews.
Participants were classified as current smokers if they smoked at least one cigarette per day at the time of the interview, as occasional smokers if they smoked usually less than one cigarette per day, as former smokers if they had smoked regularly or occasionally in the past and as never smokers if they had never smoked or less than 100 cigarettes in their lifetime.
Smoking status in F4 was compared with previous S4 information, and nine participants were excluded due to missing or implausible information on smoking status. Thus, the final study sample for this cross-sectional F4 analysis contained 3,071 subjects.
Information on age, sex, education (basic (≤9 years), secondary (10-11 years), higher (≥12 years)), alcohol consumption (based on WHO proceedings [30]: low risk alcohol consumption (average daily alcohol intake ≤ 20 g for women and ≤ 40 g for men), risky alcohol consumption (daily intake > 20 g for women and > 40 g for men)) and physical activity (active (regular sports in leisure time in summer and winter time for ≥ 1 hour per week), inactive (< 1 hour of sports per week)) were assessed in addition in F4.
A subsample of former smokers gave additional information on the date (n=1,164) and on reasons (n=791) for quitting in an additionally administered questionnaire.
Assessment of healthcare utilisation and cost components
Utilisation of medical services
As described previously, participants were asked to state the number of times they had visited a physician in the previous 3 months, subdivided for 15 ambulatory specialisations [31]. Additionally, the numbers of ambulatory hospital visits, visits to alternative physicians and physical therapy treatments in the previous 12 months were recorded, as were the numbers of inpatient hospital days (including days in intensive care units), and inpatient and outpatient rehabilitations. Finally, use of pharmaceuticals in the previous week was assessed, detailing name, national drug code, dosage, interval of intake, and prescription status.
Direct costs
For monetary valuation of health services, national unit costs were applied as recommended by the Working Group Methods in Health Economic Evaluation (AG MEG) [32]. These unit costs were updated to the year 2008 (Additional file 1: Table S1).
Costs per physician contact for each medical specialty and per physical therapy unit were updated from 1999 to 2008 using the rate of change in physician reimbursement per case [33]. The resulting contact values for physician visits vary from €18.3 for psychiatrists to €99.6 for radiologists, and are €26.1 for physical therapy. The number of physician contacts and physical therapies was multiplied with the corresponding contact value to determine costs. If participants stated that they had visited a physician in the previous 3 months but did not indicate the number of visits (n=7), one visit was assumed following a conservative approach. In a sensitivity analysis, the mean number of visits was imputed instead.
Costs for alternative physicians were requested directly. If participants stated that they had visited an alternative physician but did not specify their costs (n=45), the average costs per visit stated by the users was imputed (€50).
As the reason for hospitalisation was unavailable, hospital days and days spent in the intensive care unit were valued using mean costs per day as suggested by the AG MEG. Unit costs were updated from 2000 to 2008 as described by the AG MEG and using data from the Federal Statistical Office [32, 34, 35], yielding costs of €451 per hospital day and €1,293 per day in the intensive care unit. Costs per ambulatory hospital treatment were assessed as one physician visit according to social legislation. Based on data on average costs and length of ambulatory and stationary rehabilitation from the German Pension Fund [36], costs per inpatient and outpatient rehabilitation were estimated as €100 and €62 per day, respectively.
Utilisation of pharmacy only pharmaceuticals was estimated using participant’s information about name, national drug code and dosage of intake within the previous week, and costs were calculated from 2008 pharmacy retail prices [37] by subtracting mandatory discounts of manufacturers and pharmacies according to the Social Security Code for statutory health insurance. Non-pharmacy medicines, dietary supplements and vitamins were excluded.
Total annual direct medical costs were calculated by extrapolating physician and drug costs to 12 months, and summing the costs of all health services.
Indirect costs
To assess indirect costs with regard to production losses in those aged 65 years or younger, participants were asked whether disability benefits were obtained, and those with regular employment were asked how many days they had been absent from work due to illness in the previous 12 months. If participants stated a greater number of days of absence from work than the maximum number of 213 working days in 2008 in Germany, their days of absenteeism were restricted to 213 (n=6).
Following current guidelines, we used the human capital approach to calculate productivity losses in paid work for society [38]. As methodological discussion about the most appropriate approach is still on-going, we additionally applied the frictional costs approach within a sensitivity analysis [39]. Whereas the human capital approach assumes a perfect labour market, the frictional costs approach charges only 80% of losses of the human capital approach to avoid potential overestimation of indirect costs [32].
Annual labour costs published by the Federal Statistical Office [40] were used to value productivity losses per year of disability in accordance with AG MEG guidelines. Costs per day of work lost were calculated by dividing annual labour costs by 213 working days in 2008, yielding costs of €160 per working day. Costs due to unpaid work, further premature retirement and death, as well as intangible cost resulting out of pain or decrease in quality of life, were not considered.
Statistical analysis
Unadjusted analyses were performed regarding participants’ smoking status, healthcare utilisation and costs. To account for non-normality of cost data, 95% confidence intervals (CI) were estimated via a non-parametric bootstrap approach using a percentile method based on 1,000 replications.
The effect of smoking status on healthcare utilisation and work absenteeism was analysed in a two-step hurdle approach: In a first step we used multiple logistic regression models to calculate odds ratios of healthcare utilisation or inability to work as a function of smoking status, age, sex, education, alcohol consumption and physical activity. Secondly, for users only, factors influencing the number of healthcare service uses were analysed with generalised linear models assuming a zero-truncated negative binomial distribution with a log-link.
Finally, the effect of smoking status on direct, indirect and total healthcare costs was analysed, again adjusting for age, sex, education, alcohol consumption and physical activity. To consider the typically skewed distribution of costs, generalised linear models were used assuming a gamma distribution with log-link, where costs of €1 were assigned to participants with zero costs. This approach has been demonstrated to be a suitable method for healthcare costs [41, 42].
Appropriate tests were operated to confirm the choice of distribution and link function. Regarding total costs, the Modified Park Test supported the choice of the gamma distribution (p=0.39), and the Hosmer-Lemeshow-Test (p=0.61), the Pregibon Link-Test (p=0.63) as well as the Pearson Correlation Test (p=0.90) all confirmed the choice of the log link function.
Due to missing data on three participants for education, the regression analysis contained 3,068 observations.
Statistical analyses were performed using SAS software (SAS Institute Inc., Cary, NC, USA, Version 9.2), and p-values of 0.05 or less were considered to be statistically significant.
Non-linear relationships of variables were checked as well as possible interactions by using variable selection methods (PROC GLMSELECT with stepwise selection method), but no significant non-linear relationships and interactions were observed.
