Out-of-pocket expenditures for pharmaceuticals: lessons from the Austrian household budget survey

Comparisons between OECD member states reveal that the out-of-pocket share of total pharmaceutical spending (41 % in 2011) is more than twice as much as the out-of-pocket share of total spending on health services (18 % in 2011) [1]. The empirical knowledge on the determinants of out-of-pocket pharmaceutical expenditures (OOPPE) is limited. One reason for this fact is the lack of adequate routine data on pharmaceutical expenditures on the individual and household level. To understand the possible covariates driving OOPPE and to select a sound econometric identification strategy require close insights into the interaction between the relevant actors in the decision-making process on pharmaceutical consumption. Such analyses end in preferred specifications of indicators for pharmaceutical use/expenditures and of possible influential covariates. Available routine data sets on pharmaceutical consumption provided by socioeconomic (e.g., SOEP in Germany) and health-related surveys (NHANES in the US, EHES in selected EU countries, ATHIS in Austria) fulfill such claims only to a limited extent; specific information on OOPPE is missing. In the following article we study the determinants of OOPPE in Austria using cross-sectional information from the latest national household budget survey conducted in 2009/10 [2]. We give insights into the socioeconomic determinants of OOPPE, an undertaking that is new for pharmaceutical spending in Austria. The article benefits from the voluminous previous research work on out-of-pocket health-care expenditures based on microdata in general [3‐18] and on the scanty literature on OOPPE and respectively on self-medication [19‐25]. In the following we briefly discuss the previous research work on OOPPE and self-medication. In a two-part model, Leibowitz et al. [21] study cost-sharing effects of prescription drug expenditures using individual panel data from the Rand Health Insurance Experiment. They find that drug expenditures respond to the cost sharing faced by the consumers. They also observe very significant site, age and sex effects. Based on a cross-sectional data set on an individual basis from Spain, Figueiras et al. [23] identify sociodemographic factors associated with self-medication (use of non-description drugs) using multivariate Cox’s regression. They find that self-medication is more prevalent among women, singles, persons living in large agglomerations, and person with acute disorders and higher education levels. Grootendorst [20] uses individual data from the 1990 Ontario Health Survey on the self-reported use of prescription drugs to compare alternative econometric models (Poisson model, negative binomial model, two part models) to study the different determinants of drug use. Chang/Trivedi [22] develop a theoretical model of self-medication behavior of a utility-maximizing consumer. Empirically, they especially focus on the role of income and health insurance on self-medication. Chan/Trivedi use individual cross-sectional data from the World Bank’s Living Standards Measurement Survey of Vietnam 1997–1998 and apply an econometric framework similar to [20]. They find that self-medication is an inferior good at high income levels and a normal good at low income levels. Insurance coverage strongly reduces self-medication. Costa-Font et al. [19] study the determinants of drug consumption in Catalonia using official survey data on an individual basis. Econometrically, a two-part model is used. They find evidence that gender, health status and the existence of insurance are significant predictors for out-of-pocket pharmaceutical expenditures. Income and cost sharing are strongly associated with drug use but not drug expenditures. McLeod et al. [24] analyze the financial burden of out-of-pocket expenditures for prescription drugs based on cross-sectional national survey data from Canada. They focus on the household drug budget share and estimate Engle curves for different drug budget shares using predicted values from the kernel conditional quantile estimator. Tavares [25] derives a formal model of self-medication based on the maximization of a utility function that depends on consumption and on health status. Thereby Tavares especially focuses on the role of time. Individual data from the Portuguese 4th National Health Survey (2005) are used in a probit model with self-medication as a binary variable. The results show significant effects for age and gender. As far as waiting time is concerned, Tavares offers mixed results.

Summarizing these studies we observe important differences in the study designs, the data used and the observed results. The majority of the studies are based on individual data; only [24] uses data from the household level. All studies using individual data except [21] directly control for the individual health status. Reports [22] and [25] base their empirical estimation on a theoretical model of drug consumption (self-medication), while the other papers use different steps of reduced forms. Reports [20‐22] use two-part models in the econometric specification. Thereby, [20] and to some extent [22] present a detailed comparison and discussion of alternative model specifications. The clear focus of [19, 21] is price effects on the pharmaceutical consumption measured by the level of insurance coverage.

Our article contributes to the empirical research on OOPPE in several ways. First, it adds evidence from the perspective of the household and supplements the findings available from the individual level in the previous literature. Second, we use data from a health-care system that is based on Bismarckian principles and that holds a specific two-tiered institutional architecture of service provision and financing. Third, we keep in mind households’ decision-making process of either seeking professional health care or using self-medication, which leads to three types of out-of-pocket expenditures. We account for this fact by separating the available database into subgroups to analyze the different types of OOPPE separately. Finally, our source of information is the general household budget survey, while previous studies build on health-related surveys. Since national household budget surveys follow internationally agreed principals, our study also allows conclusions on whether budget surveys are an appropriate database to analyze the determinants of out-of-pocket health-care expenditures.

The remainder of the article is organized as follows. In “The policy setting of pharmaceutical consumption in Austria,” we present a brief overview of the main institutional characteristics of consuming pharmaceuticals in Austria. In “Databases and empirical approach,” we inform about the databases and derive conclusions for the empirical approach applied in the article. In “Econometric results and discussion,” we present the empirical results and discuss them. In “Conclusions,” we conclude our article.

In Austria authorities of the central state regulate basic dimensions of pharmaceutical consumption. They decide on the general preconditions and modes of market entry of pharmaceuticals, specifically on the separation between pharmaceuticals with obligatory prescription (8.026 pharmaceuticals in 2012; [26]) and over-the-counter (OTC) products (1.931 pharmaceuticals in 2012; [26]) and on pharmaceutical pricing. Thereby, the regulation of prices primarily focuses on maximum price margins of the wholesale firms and pharmacies, while factory prices are not regulated in Austria at this stage [27]. But this general regulation of market entry and prices primarily influences the provision of pharmaceuticals paid over the counter. There exists a second stage of public regulation of market entry and pricing conducted by the social health insurance system. Since social health insurance in Austria covers around 99.3 % of the whole population—excluding only marginal groups from public health insurance—this regulation has far-reaching consequences for pharmaceutical pricing and consumption [28].¹ Only pharmaceuticals included in the positive list of the Reimbursement Code are paid by the social health insurance system. Thereby, the Reimbursement Code includes pharmaceuticals with and without obligatory prescription [29].²

Pharmaceuticals that are part of inpatient treatments are free for patients with social health insurance coverage. Their costs are included in the DRG-based hospital remuneration system [29, 30]. Pharmaceuticals that are part of outpatient treatments provided by GPs/specialists having a contract with the social health insurance system are basically free if they are included in the positive list of the Reimbursement Code. Patients have to pay a prescription fee for every pharmaceutical prescribed. This prescription fee is an absolute amount of money (in the years of the household survey: 2009: €4.90, 2010: €5.00) with no link to the price of the pharmaceutical. If the price of the pharmaceutical is below the prescription fee, patients only have to pay the price of the pharmaceutical. Calculated over the total range of pharmaceutical consumption financed by the social health insurance system, the prescription fee leads to a cost sharing of approximately 13 % [31]. If patients consume medical services supplied by private doctors, pharmaceuticals are paid by the social insurance system on request.

As far as the prescription fees are concerned, two schemes influence the financial burden of individuals (households). An exemption exists from the prescription fee and a prescription fee cap. An exemption is granted without application (1) for retired persons who draw a small pension from a public pension plan, (2) for persons with notifiable communicable diseases, (3) for members of the alternative civilian service including their relatives and (4) for asylum seekers. On application, an exemption from the prescription fee is granted for insurance members (including co-insured household members) with a household net income below the threshold value of the basic income maintenance system. Since 2008 the exemption from the prescription fee is accompanied by a prescription fee cap at a 2 % share of the annual net income.

Roughly 35 %³ of the population has signed contracts with private sickness funds, which predominantly offer additional coverage to services of the social health insurance system and/or improve the possibility to choose from a broader portfolio of providers/services within the system. But private health insurance does not play a significant role in financing pharmaceutical consumption. Only 0.2 % of the prescribed drugs and 1.7 % of the OTC products were paid by the private health insurance system in 2012 [33].

Having in mind the institutional setting of consuming pharmaceuticals in Austria, we are able to identify possible treatment paths in the health care sector that might lead to OOPPE (see Fig. 1). In the first step the patient has to decide whether to rely on self-medication or to seek professional health care [22, 25]. In Austria, self-medication accounts for approximately 20 % of total pharmaceutical consumption (outside the hospital) and is mainly financed out of pocket [27]. If the patient decides to use outpatient medical services, pharmaceuticals with and without obligatory prescription are consumed. If they are funded by the social health insurance system, the patient only has to pay the prescription fee. If they are not funded, the patient has to pay the price. On average, 80 % of the pharmaceutical consumption (outside the hospital) in Austria is based on a prescription, 88 % of this consumption is refinanced by the social health insurance system, 11.8 % is paid out of pocket, and 0.2 % is refinanced by private sickness funds [33]. Summarizing, we end up with three forms of OOPPE (see Fig. 1): (1) OOPPE as a consequence of self-medication (OOPPE type 1); (2) OOPPE as a consequence of consulting the professional outpatient health-care sector and consuming pharmaceuticals that are not included in the Reimbursement Code of the social health insurance system (OOPPE type 2); (3) prescription fees for pharmaceuticals prescribed by the outpatient health care sector and paid by the social health insurance system (OOPPE type 3).

Table 1 shows the descriptive statistics of the explanatory variables. Out of 5787 households, 1150 have non-zero expenditures for prescription fees with a mean per month of 34.47. In the case of OTC pharmaceuticals, the non-zero mean expenditures of the 1559 households sum up to 41.00. In the raw data, we observe substantial differences of OOPPE levels depending on the household structure, adults’ age, adults’ education level and the type of public health insurance. The differences are more pronounced for OTC pharmaceuticals compared to prescription fees.

Table 2 shows the econometric results of the TPM and GLM for OTC pharmaceuticals (OOPPE type 1 and 2). The probability for OTC spending is strongly influenced by the life cycle of the household. The signs of the coefficients are highly plausible; the size of the coefficients are partially unexpected. There is some evidence that the probability of OTC spending is lower in regions with a low degree of urbanization. As far as self-medication is concerned, the difference in the relative time costs of using professional health services compared to pharmacies could be an important covariate to explain this fact [22], but our data do not allow to test for this hypothesis. The positive relationship of the OTC spending with age—especially in the older age groups 45–65 and >65—is expected and well documented in previous empirical research. The education level increases the probability for OTC spending significantly. The health insurance characteristics of the household—either private or public—are of very limited influence on probability of OTC spending. This follows our expectations for several reasons. The general preconditions of consuming pharmaceuticals (e.g., pharmaceuticals included in the positive list, level of the prescription fee, exemptions from the prescription fee) do not differ among the different public sickness funds (GKK, BVA, SVA, SVB) compared in the sample. Differences might be caused only indirectly by differences in the socioeconomic characteristics of the different groups of publicly insured (e.g., opportunity costs of time when being ill, schedules of physician services remuneration). As already mentioned, private health insurance only plays a very limited role in financing pharmaceutical consumption in Austria. So we expect only indirect effects on the OOPPE-levels caused by, e.g., higher risk aversion of individuals with private health insurance or effects of the remuneration system of private health insurance on treatment behavior of health-care providers. Household income and female gender of the householder increases the probability of positive OTC spending. Finally, we observe that doctoral visits of a household member in the same period increase the possibility of OTC expenditures. According to the Box-Cox test (λ near zero), we use for the second stage an OLS model with a log-transformed dependent variable denoted as log OLS in Table 2. In contrast to the highly significant covariates of the first stage, the covariates of the second stage remain largely insignificant. The income elasticity of OTC expenditures is near zero (0.073), but the coefficient is insignificant. One interpretation of the results on the second stage could be that the probability of OTC consumption of a household systematically depends on several household characteristics, while the level of expenditures is highly stochastic in the short time perspective represented in our data.

Columns 5 and 6 of Table 2 show the results of the GLM. We tested for the kurtosis of the log-transformed OOPPE that takes the value 2.99, which is very close to 3 and therefore justifies a log link function. As mentioned in the empirical approach, we performed a modified Park test. The corresponding estimates are λ = 1.55 (provisional OLS model with log-transformed dependent variable) and λ = 1.20 (provisional GLM model) favoring a variance proportional to the mean model. In the evaluation of the model performance, the variance proportional to the mean model clearly outperforms the alternative.⁵ We used this specification in our estimation. Our results reveal a significant effect of the household structure, adults’ age, adults’ education level and doctoral visits. The same is true with restrictions for the degree of urbanization. Income remains insignificant, which is also true for private and public insurance characteristics (exemption: BVA). Considering both model specifications for the analysis of OTC pharmaceuticals, the one-stage GLM predominately approves the findings of the TPM except for household income and the status of early retirement.

Table 3 shows the econometric results for the second form of OOPPE—the prescription fees (OOPPE type 3). We compared the performance of different econometric models using likelihood ratio tests, BIC, AIC and Vuong tests as model selection criteria. The zero inflated negative binomial model (ZINB) fits better than all other models. This is also true for the NB in the positive part of the hurdle model. Therefore, Table 3 only presents the results for the hurdle model—with the NB specification in the first step—and the ZINB model when focusing on the characteristics of the zeros.

The results of the hurdle model are shown in the left part of Table 3. The first part of the model is defined as logit and demonstrates the importance of the households’ life cycle. Especially households consisting of more household members (married couples, empty nest, full nest I and full nest II) increase the probability of having prescriptions significantly. Surprisingly, single parents have a significantly lower probability of non-zero prescriptions. Adults’ increased age (age groups 45–65 and 65–85), female householders and doctoral visits within the observation period increase the probability, while early retired householders and households who are insured at SVA decrease it significantly. The existence of private health insurance, income, education and the degree of urbanization have no effect on the probability of prescriptions. The second part of the model is defined as zero-truncated Poisson regression. Concerning the household structure, the results show that the log count of prescriptions increases significantly for unmarried and married couples, empty nests and adults with increased age (age groups 45–65 and 65–85) and decreases with income and doctoral visits of the affected household members.

The right part of Table 3 shows the results of the ZINB regression model. The splitting function (logit) reveals the covariates that influence the probabilities of true zeros. As expected, the coefficients of the covariates show a similar size but reversed signs compared with the first step of the hurdle model: e.g., the probability of a true zero in the prescription variable strongly decreases with age. Additionally, the existence of doctoral visits, female householders and children (full nest I, full nest II) decreases the probability of true zeros. In contrast, single parents and householders who are early retired increase the log odds of true zeros. The level of prescriptions (NB) sharply increases with age, is inverse to the degree of urbanization and decreases with income. The expected number of prescriptions for unmarried couples is 1.48 times the expected number of prescriptions for a single person I while holding all other variables constant. Furthermore, married couples, empty nests and households without doctoral visits in the observed time period have a higher expected number of prescriptions than the particular reference groups (see column 5 and 6 of Table 3). Public insurance characteristics and the existence of private health insurance remain insignificant in both estimation stages. The same is true for the level of education. Using AIC, we also tested the hurdle model and ZINB model [37] and found a better model fit for the ZINB model.

Our study differs from the previous research in several respects. We use data from the household level, while the majority of the previous studies use individual data. Our study analyzes two different forms of OPPE, while the previous research concentrates on either self-medication or prescriptive drugs. Econometrically, our study is most similar to the studies of [20, 22]. The results of our study are to some extent in line with the previous literature [19, 20, 23‐25]. We confirm the effect of age on OTC pharmaceuticals and prescription drugs and the mixed results for income. In line with [23, 25], we find a positive effect of education on the consumption of OTC pharmaceuticals. In contrast to previous findings, we do not find an effect of private health insurance on the probability and level of OOPPE. The same is true for the type of public insurance. This is an indication that public coverage against the risk of pharmaceutical expenditures in Austria is high and homogenous. In line with the majority of the previous research—an exemption is [21]—we do not test for the price elasticity of pharmaceutical consumption directly, but control for the existence of different forms of insurance coverage.

We conclude our discussion section by pointing to possible limitations of our study. In our presentation of the empirical design of the study, we already pointed to several challenges. In line with [19‐21, 23, 24], we do not apply a theoretical model of pharmaceutical consumption, but test reduced forms. This means that our study to some extent has an explanatory character. Our data picture the level of households. This allows us to test for household characteristics, e.g., the household structure. But we should be aware of the fact that conclusions from the household level to the individual level should be drawn with caution. Although the decision to consume pharmaceuticals might be influenced by household characteristics, it remains an individual decision and mainly depends on individual characteristics, which are masked to some extent on the household level. This has important consequences for the reproduction of the decision-making process when consuming pharmaceuticals. Additionally, the covariates age and education have a different meaning on the household level compared to the individual level, which might lead to differences in the estimated effects. Finally, we want to discuss the suitability of data from general consumption surveys to picture the consumption of health care goods and to identify the covariates driving the consumption level and structure. General consumption surveys are well-established instruments of economic accounting and follow internationally agreed standards of reporting. This in principle qualifies them as an information source also for the consumption of health care goods and services. For several reasons, general consumption surveys offer high data quality concerning consumption expenditures and selected socioeconomic characteristics of the households. On the other hand, they only include rudimental information on socioeconomic characteristics of individuals (households) that are able to explain the utilization of health care services, e.g., the consumption of pharmaceuticals. Such characteristics are indicators for the health status, the need of long-term care and the individual disease profile over a longer time period. The empirical literature clearly reveals that the health status—e.g., measured by health expenditure in the past—is an important predictor of health-care expenditures and explains most of the variance in regression models. The missing of such information is an important source of unobserved heterogeneity and is also an indication for the low explanatory power of the used covariates in our estimations. The explanatory power of the covariates used in the previous literature, which controls for the health status—see especially [19, 20, 22, 25], is substantially higher compared to our results. In addition, general consumption surveys normally do not include information on the supply characteristics of health-care services (e.g., distribution of pharmacies and physicians, provision/utilization of public health care services), which might influence the utilization decision heavily. Finally, the information from consumption surveys is period based and does not allow a separation in different steps of the utilization process. Overall, we conclude that the information on health expenditures (e.g., pharmaceutical expenditures) from consumption surveys primarily should be used to study the covariates of pharmaceutical consumption/expenditure from a distributive point of view. They allow clear insights into the role of the household structure (life cycle of the household), the income and education in pharmaceutical consumption.

This article analyzes the socioeconomic determinants of OOPPE of private households in Austria using data from the household budget survey 2009/2010. Empirically, the data show substantial differences in the expenditures between households in different stages of their lifecycle. The characteristics of the data set (information from the household level, specification of the dependent variable, period based instead of illness episode based data, short observation period) pose several challenges to the choice of the estimation strategy. The advantages of the widely used TPM are no longer obvious. We react to this fact and compare and use different econometric approaches (TPM, GLM, Hurdle model, ZINB). Overall, we find that several household characteristics—especially the household structure, adults’ age, income, doctoral visits and adults’ education level—have strong effects on the probability and level of OOPPE. This is especially true for OTC pharmaceuticals, but to a reduced degree also for prescription fees. On the other hand, we do not find substantial effects of the type of public health insurance and the existence of private health insurance. The results of our study complete the picture of the covariates of OOPPE on the individual with evidence from the perspective of the household. The present study can help health policy decision makers to identify inequalities in pharmaceutical consumption and obtain insights into the covariates causing them.