This study investigated the differences in healthcare costs for children exposed to different modes of tobacco smoke using a bottom-up approach based on two German birth cohort studies. We found that children living in homes where smoking was reported (either inside or on patios and balconies) showed significantly higher medical costs than children not exposed to smoke. For children who are not exposed at home, but at other places, unadjusted medical costs were also slightly higher; however, these differences are not significant. These descriptive findings were confirmed by regression analysis. Thus, exposure to tobacco smoke is shown to have an effect on direct healthcare costs independent from socio-economic status. As the difference between the two in-home exposure groups (inside or on patios and balconies) is not statistically significant, attempts by parents who smoke to protect their children by smoking only on balconies or in gardens appear to be ineffective.
Measuring the burden of second-hand smoke in terms of implied healthcare costs rather than in terms of induced incidence or prevalence of diseases, has the advantage that impacts can easily be aggregated across different diseases. To indicate an impact from exposure to tobacco smoking, healthcare costs have to differ significantly, which is a more conservative approach than relying on a diagnostic label alone. Given that active smoking significantly increases the risk of age-related diseases such as heart and cardiovascular disease or cancer, one cannot expect comparable cost impacts of second-hand smoke on children with a mean age of 10 years. As the correlation between health problems and second hand smoke exposure in childhood is not trivial and still not completely understood, the excess cost approach is a possibility to capture all the differences between the analysed groups, but it does not allow for interpretation of causal pathways. The excess cost approach used in this study is aimed at identifying all excess healthcare utilisation due to either the condition under research itself (in this case smoke exposure) or any other disorder related to this condition, i.e. its consequences on health status. Thus, estimating the excess costs of exposure to second hand tobacco smoke in a data set comprising current utilisation of health care could potentially permit one to assess the total impact of this exposure on costs of care.
In addition to these strengths there are several limitations to this study. Both datasets provide information on healthcare utilisation for the 10-year follow-up. Unfortunately, it was only possible to include about 70% of the 10-year study participants due to missing data on healthcare utilisation or exposure to tobacco smoke. Like other retrospective cost assessments, this study was prone to recall error regarding utilisation of healthcare services. As far as recall abilities of parents are not associated with their child’s exposure to tobacco smoke, this should only influence the estimated average costs, not the differences between groups [42, 43]. Additionally, our results may not be fully representative for the German population with regard to health-related as well as socio-demographic characteristics. The fact that families participating in the two birth cohorts show above-average education and income levels  may lead to an underestimation of actual costs, but should not influence cost differences substantially between the different groups of exposure to tobacco smoke. In addition, all study regions in this study are relatively urban. As in more urban areas, the supply of healthcare may be more extensive, this may lead to an overestimation of the actual healthcare costs. However, the comparison with the probability of physician visits in the representative KiGGS-Study  shows a similar picture for most comparable specialist groups.
Furthermore, the magnitude of healthcare costs related to second hand smoke exposure might be underestimated due to the design of this study. Since children with low birth weight, defined as birth weight below 2500 g, were not included in this study and low birth weight might be a result of maternal smoking during pregnancy, this study might have indirectly excluded some children exposed to tobacco smoke in utero which were shown to have an increased risk of hospital admission as infants in an earlier study by Adams et al. . Moreover, it might be more likely that these children are also exposed after birth. Moreover, the excess cost approach may overstate the actual impact of parental smoking if there is unmeasured confounding that is associated with both, parental smoking and parental investment in child health endowments, e.g. a healthy diet.
Although the overall smoking prevalence in the German adult population was 29.9% in 2008/2009 , a previous study found that about two-thirds of all children aged 6–13 years were reported to live in a household with at least one smoker . One reason for this gap might be that the smoking prevalence is higher in younger adults (e.g., 37.9% in 18–29 year olds) . Furthermore, this gap might be explained by a possible higher concentration of larger families in lower-education smokers, as higher smoking prevalence was reported in people with lower levels of school education . However, in our study the percentage of children exposed to tobacco smoke at home (indoors or on patio/balcony) was closer to one half (55%).
As about 45% of children included in the cohorts at birth did not participate in the 10-year follow-up, we are unable to rule out non-response bias. The proportion of people with high education was lower and the proportion of mothers smoking during pregnancy was higher in study dropouts than in participants. In addition, the proportion of children being exposed to smoke during their first year of life was higher in children who dropped out. As information on healthcare use and exposure to smoke was not available for all children at 10-year follow-up, we performed additional non-response analysis comparing these to the analysis group and found only moderate differences in socio-demographic characteristics.
These analyses are based on parent-reported tobacco smoke exposure but not on biological markers, whereas the validity is frequently criticised in relation to an underreporting of the exposure. However, a sub-study of these two birth cohorts showed a good agreement between self-reported smoking at home and measured residential air nicotine concentration as well as with cotinine measurements in children’s urine . In addition, measurements of cotinine levels in children’s urine are not very helpful when longer-term effects of second hand smoke exposure will be assessed, because of the short shelf-life of approximately 24 hours. We consider a major bias of our results by the parent-reported exposure to second hand smoke not as likely. However, the effect on excess costs is less clear. If smoking is underreported, cost of children not exposed may be overestimated, resulting in an underestimation of excess costs. Furthermore, the role of the length of exposure and transitions between exposure groups could not be considered adequately and should be further investigated.
Regarding estimation of direct healthcare costs, several assumptions were necessary that may have caused over- or underestimation of costs. We replaced missing values for the number of physician and therapist visits as well as hospital days using a multiple-step single imputation technique. This may lead to smaller variances in the estimates. We did not use multiple imputations, as this did not seem feasible in our multi-step approach. We based the assessment of cost components on the suggested values published by the AG MEG , which were updated to the base year 2007 to account for price changes. This method has several limitations, especially regarding price variation within healthcare sectors, which is described in more detail in an earlier study . As the reasons for hospital stays were not available, we used the same cost per day for all hospital stays. Actual reasons for hospitalisation might vary between groups with different exposure to smoke, which might influence the estimated cost differences. Moreover, we based cost calculation on weighted mean prices for Germany as these are assumed to reflect opportunity costs.
In infants, earlier studies found no significant association between second-hand smoking and physician visits [23, 47]. Levy et al. detected no differences in children’s overall Medicaid expenditures by presence of smokers in the household . One possible reason for this might be selection bias. That is, families that expose their children may also be less likely to take them to the doctor. A lower use of health services in adult smokers was found in earlier US studies [18, 49]. Other studies, however, did find an association with higher utilisation of hospitals [19, 22]. Yet, we found that in older children of about 10 years of age, second-hand smoke exposure is associated with a higher probability of physician visits as well as higher hospital costs. For the sum of the included cost components, the odds of positive costs, as well as the amount of costs, was significantly higher compared with children who were not exposed to smoke. Similar to our findings, van Reek et al. found increased likelihood of visiting a physician for children who were exposed to second-hand tobacco smoke 2 years before assessment . In a recently published study, Adams et al. observed no significant association of maternal smoking and admission to Neonatal Intensive Care Unit but a positive effect on the length of stay of exposed infants once admitted , which is in line with our findings for hospital utilisation in older children. Florence et al. report higher odds for respiratory expenses in children exposed to second-hand smoke in the USA . In Germany, there is a lack of evidence on the economic impact of second-hand smoke; however, Thyrian et al. estimate that tobacco smoke exposure at home is responsible for more than 14,000 hospital admissions in children. In line with our findings, they also found a longer duration of stay using a top-down approach . The aim of the present study was to fill the gap based on data from a large sample of children in Germany. Cost estimates are based on a bottom-up approach. In contrast to top-down studies, which are based on aggregated data from administrative statistics, this offers the possibility of adjusting the estimates for a broader range of confounding variables. Health insurance data might allow for more precise cost estimations, but they don’t include information on parents’ smoking status. As correlation between health problems and smoke exposure is not trivial, the presented analyses are based on the excess cost approach, attempting to capture all the differences between the analysed groups of smoke exposure and minimising the need for additional assumptions. Therefore, multivariate statistical analyses did not include adjustments for comorbidities .