Introduction

The correlation between the scope of health insurance protection and the volume of health benefits claimed has been studied empirically for over 30 years (Reiners 2009). In this context, the term “moral hazard” refers to an individual covered by health insurance claiming more in health benefits than is necessary (Breyer et al. 2005). After closing an insurance contract, the insuree's behaviour may change to the detriment of the collective of insured persons (Sehlen 2004). Against this background, Pauly proposes that incentives should be offered that result in health benefits being claimed only in cases of genuine need. For example, this could be done in the form of cost participation by the insured party (Pauly 1968). Deductibles comprise one possible form of such self-participation.

The effect of deductibles on the claims for health benefits could be demonstrated in several studies. At the international level, for example, the RAND Study in California (Manning et al. 1987; Newhouse 1996) and the studies of the Swiss health service by Gerfin and Schellhorn (2006) and by Werblow (2002) were able to determine the influences on claims. In the meantime, studies have appeared that support these findings for the statutory health insurance system in Germany (GKV). Felder and Werblow (2006) investigated a voluntary tariff with deductibles offered by a GKV health insurance providing optional cover. In a recent study by Huber (2009)—which, however, used surveys from the years 2002 to 2003—significant correlations were found between deductibles and the volume of claims for health benefits. In order to ascertain the effect of deductibles, Swiss insurees (with deductibles) were compared with German insurees (GKV without deductibles).

The first results for the optional tariffs with deductibles available since 1 April 2007 as a result of the GKV Wettbewerbsstärkungsgesetz (Law for strengthening competition in statutory health insurance) were published by Hemken et al. (2010). The existing German studies by Felder and Werblow (2006) and by Hemken et al. (2010) all determine the effects on claims by comparing the behaviour of insurees with a deductible before and after taking up this deductible option with the behaviour of those who did not take up the option. Felder and Werblow (2006) compare the 2 years before taking up the option with the first year with a deductible; Hemken et al. (2010) compare the year before taking up the option (2006) with the first complete year with a deductible (2008). Both studies detected controlling effects deriving from the deductibles. Claiming behaviour over a longer period after taking up the option of a deductible has not yet been monitored. No research has been done on the medium- and long-term effects of deductibles in the GKV (Felder and Werblow 2006). There is an urgent need for research into the effects of the rules governing optional deductibles within the GKV (Schulenburg and Greiner 2007).

In addition, different levels of deductibles have not yet been evaluated in Germany. Studies performed within the Swiss health system have shown that the control effect increases with increasing size of the deductible (Gardiol et al. 2003; Werblow and Felder 2003).

Based on these results, this work adopts a new chronological perspective. Up to now, studies of deductibles in the European setting have never measured the effect for longer than 1 year after opting for a deductible. The present work investigates the medium-term effect of optional deductibles, defined as the effect from the second year after opting in. In addition, the effects of different amounts of deductible are determined. The research examines the following hypotheses:

  1. H1

    Insurees who have chosen tariffs with deductibles will claim fewer health benefits than members of a control group.

  2. H2

    The greater the deductible is, the greater the control effect and the lower the volume of claims for health benefits.

The analysis uses an extensive data set from the AOK (regional health insurance body) for Lower Saxony (AOKN). This article is organised as follows: the next section presents the tariff with deductibles and the data set, and describes the basic aspects of the matching procedure used in the study. The results of the two research hypotheses are then presented and, finally, a discussion and classification of the results of the study, taking account of the relevant limitations of this work.

Methods

Data set

This work investigates the optional deductibles offered by the AOKN, which were introduced on 1 April 2007. Table 1 presents an overview:

Table 1 AOK tariffs with deductibles (AOK Statute 2010)
Full size table

As seen in the table, there are seven different tariff levels. The optional deductible is essentially determined according to income.

The bonus to be paid for a calendar year is reduced if during the year, at the cost of the health insurance, prescriptions for drugs or remedies have been written by the SHI-authorised doctor, or the insuree has received hospital treatment. Visits to the doctor without subsequent services of this kind do not count towards the deductible. Items charged to the deductible are subject to fixed-price tariffs. The fixed amounts for the individual tariffs are given in Table 1. If in the course of a year the insuree has made no more than three visits to the doctor resulting in prescriptions for drugs or remedies, or has received a maximum of one hospital treatment, a bonus is always paid. If benefits claimed exceed this, the coinsurance amount is between €80 and €120, depending on the tariff level.

For the examination of hypothesis H1, the seven deductible levels offered by the AOKN were divided into three groups. Levels 1–2 were combined in the low deductible group, levels 3–5 in the medium deductible group and levels 6–7 in the high deductible group.

On the reference date 31 December 2008, a total of 8,765 persons insured by the AOKN had opted for a deductible. In order to enable a before-after comparison to be performed, the analysis included all those who had been continuously insured by the AOKN during the year preceding introduction of the tariffs (2006). Those opting for a deductible chose the starting date for the tariff individually. The analysis included those who had started the tariff no later than 1 January 2008 and remained with it until at least 31 December 2009. This exclusion criterion led to 5,009 persons insured with a deductible being included in the study.

The control group comprised 10,000 members and was selected by means of a random number generator from the AOKN's total insured collective of 2.2 m insurees. After elimination of duplicate data sets, the control group comprised 9,911 members.

The AOKN data set contains information about the insuree's socioeconomic variables and the benefits claimed in the years 2006 to 2009. The present study focuses on the data from 2006 and 2009. The matching procedure took account of the socioeconomic variables age, gender, income, status in profession (e.g. master or part-time employed), type of insurance (compulsory insurance, retired or voluntarily insured) and the insuree's identification number. The data for the benefits claimed contain the variables drug costs and drug prescriptions, hospital costs, days in hospital and hospital cases, and costs of remedies and prescriptions for these. Although the data on incapacity to work and statutory sickness benefit are also available, these are not included in the study since the AOKN tariffs with deductibles do not contain a control instrument for this area of benefits.

Matching

Gensler et al. (2005) provide an overview of the matching procedure to which we shall basically refer during what follows. In order to determine the effect of a tariff with a deductible, it is necessary to know how the individual or group would have behaved if they had not opted for a deductible. Needless to say, this comparison is impossible to make because only one of the alternative situations can ever occur, nor is it possible to ascertain how an insuree on a tariff with deductible would have behaved had they opted for a tariff with a deductible (Gensler et al. 2005). This problem of missing data is also known to be a fundamental difficulty of evaluation (Imbens 2004).

A direct comparison between the deductible group and a randomly selected control group from the collective of all insurees is not the solution, because the decision to opt in is influenced by self-selection and the result is therefore non-random (Pfeifer 2009).

The matching procedure enables non-random attributions in the test and control group and associated systematic differences in variables to be parallelised and thereby controlled. This is necessary in order to explain that part of the variance derives solely from the effect of the tariffs with deductibles. Without matching, differences in the benefits claimed would also depend on differences in the independent variables such as age, gender or claims for benefits during the year before the introduction of the tariffs with deductibles. The data set on which this study is based shows great differences in different variables in the comparison between the group opting for deductibles and the non-matched control group. Thus, an insuree opting for a deductible is 14 years younger on average, is male in 75% of cases and, during the year before opting for a tariff with a deductible, has claimed an average of €1,200 less in health benefits. The details are presented in Table 2.

Table 2 Selected variables before matching
Full size table

In order to remove these differences, a matched member of the control group (group with no deductible) was sought for every member of the group with deductibles. However, aligning every interfering variable makes the process very complex, and a trade-off is observed: the more interfering variables are included, the better the self-selection effect can be isolated—but, at the same time, the likelihood of finding no suitable matching partner increases for many members of the non-deductible group (Gensler et al. 2005).

Following Rosenbaum and Rubin (1983), a propensity score is used to reduce the complexity. This involves selecting the matched pairs by comparing their propensity scores. Propensity scores are defined as probabilities of selecting a tariff with a deductible. They are generally estimated using logit or probit models (Lechner 1998).

The nearest neighbour algorithm without replacemet is used. This method produces greater stability in estimating the effect of opting for a tariff with a deductible on the level of claims for health benefits (Gensler 2005). Each member of the group without a deductible was used only once as a matching partner.

The propensity score was calculated from the independent variables age, gender, income, status in profession, type of insurance, insurance ID and total claims for health benefits during the year 2006. The success of the matching was then tested by calculating correlations and by t-tests. The results of the t-test are presented in Table 3.

Table 3 t-Tests after matching
Full size table

As shown in Table 3, there are no longer any significant differences in the control variables between the members of the group with deductibles and the members of the group with none. Comparability was established. The control group selected in this way is used for the present work.

The matching was possible using a program syntax for SPSS from Bacher (2002), adapted for the present work. This procedure confirmed that the assumptions on which the propensity score matching was based, namely:

  • SUTVA (stable unit treatment value assumption)

  • CMIA (conditional mean independence assumption)

  • Common support

are fulfilled (Pfeifer 2009).

After matching, the output set of data contained 2,789 insurees in each group (deductible and non-deductible). However, for a detailed study of the year 2009, the two groups could contain only persons who were insured with the AOKN for the whole of 2009. Moreover, members of the group with deductibles qualify only if they had opted for a deductible for the whole of 2009. After these additional limitations had been applied, a total of 2,511 insurees remained in each group and could be included in the study.

Results

Research hypotheses H1 and H2 are examined in the following, focussing initially on the medium-term effects of the AOKN tariffs with deductibles.

Medium-term effect of the AOKN tariffs with deductibles—study H1

In order to determine the medium-term effects of the AOKN tariffs with deductibles, the claims for health benefits during 2009 were examined. For this purpose, the total costs of health benefits claimed are compared between the group with deductibles and the matched control group. The total costs include the costs of drugs, remedies and hospital treatment. The study design ensures that for all insurees with deductibles, 2009 was at least the second year of their participation in this option. The large majority of participants had already opted for the tariff in 2007 and were therefore in their third year during 2009.

The normal distribution of the variable costs for 2009 was checked by means of the Kolmogorov-Smirnov test.

The value of asymptotic significance is found to be 0.000, whereby the variable “total costs 2009” does not exhibit normal distribution. The Mann-Whitney test must therefore be applied appropriately for testing hypothesis H1, that the claims for health benefits can also be controlled in the medium-term by means of deductibles (Bortz and Lienert 2008). The median total costs for 2009 is €2,865 for an insuree without deductible, and €2,158 for an insuree with a deductible. In each group, n = 2,511. The asymptotic significance is 0.000. Hypothesis H1 must therefore not be discarded.

The AOKN tariffs with deductibles also control the volume of claims for health benefits in the medium term. During 2009, the median of health benefits claimed by insurees opting for a deductible in the years 2008 and 2009 was €707 less than the median for insurees with no deductible. Research hypothesis H1 can therefore be confirmed.

A brief comparison with the results for 2008 should place the study results for 2009 in better perspective: the median total costs in 2008 are €2,119 for insurees with a deductible and €2,904 for those with no deductible. Here too, the asymptotic significance is 0.0000. It can be concluded that the difference between the medians for 2008 (€785) and 2009 (€707) has remained essentially stable.

Control effect of different sizes of deductible—study H2

The Kruskall-Wallace H-test is used to determine whether the application of higher levels of deductibles is able to control the total volume of claims for health benefits more strongly. The middle ranks of the groups are shown in Table 4. The median of the total costs for 2009 decreases with increasing size of the deductible—in other words, a higher deductible is associated with a lower volume of claims for health benefits.

Table 4 Kruskall-Wallis H-test
Full size table

The H-test yields an empirical chi-square value of 18.871 and is 2 degrees of freedom greater than the theoretical value. The null hypothesis can therefore be rejected. The significant results of the H-test demonstrate that the middle ranks of the different deductible levels do not exhibit an equal, central tendency. The Jonckheere-Terpstra test can be used to test statistically whether a higher deductible leads to a greater control effect—in other words, to determine the direction of the difference.

The result of the Jonckheere-Terpstra test refutes the null hypothesis, which assumes an equal distribution of the total costs for 2009. The amount of health benefits claimed by a group varies inversely with the group's corresponding level of deductible (low, medium or high). Research hypothesis H2 can therefore be confirmed.

Discussion

The present work examines two hypotheses concerning the effect of deductibles on the volume of claims for health benefits in the GKV. The individual results of this study are summarised once more in the following.

Hypothesis H1 could be confirmed. Optional deductibles also control the volume of claims for health benefits in the medium term. Insurees who have opted for tariffs with deductibles claim fewer health benefits than do members of a control group.

Hypothesis H2 could also be confirmed. The higher the amount of a deductible is, the greater the control effect and the lower the amount of health benefits claimed.

The contribution to research made by the present work lies in the first-time verification of these two hypotheses with empirical data from the GKV. This study therefore demonstrates that the availability of tariffs with deductibles can guarantee a more efficient use of the GKV's resources.

On the other hand, tariffs with deductibles are rejected by a number of authors because of the possible occurrence of self-selection effects. Although self-selection effects certainly occur—healthy insurees choose tariffs with deductibles—these effects could be isolated to a large extent in the present study by means of the matching procedure. The remaining effects can be attributed with a high degree of probability to moral hazard. It should be mentioned that, despite the use of a comprehensive set of quantitative data, the qualitative reasons for an individual's choice of a tariff with or without a deductible remain unknown. This information could be acquired if the present data were extended by the results of a survey. Unfortunately, this extended study design is not possible because the present set of data is available to the authors only in anonymised form as a result of data protection legislation. Additional useful data would also be available if this survey could be performed prospectively for a new set of data. These surveys could ascertain the present and predicted state of health and also provide concrete information about individual decision behaviour.

Tariffs with deductibles have also been criticised for other reasons. Thus, Holst (2008) comments that it is not possible to distinguish the beneficial effects of the deductibles from the adverse effects. Adverse effects might include avoiding visiting the doctor, which could lead to considerable consequential costs: a neglected cold develops into pneumonia, for example. The design of the present study does not enable any differentiation between necessary claims for benefit and those that are unnecessary or of minor importance.

In addition, Huber (2009) states that deductibles have a particularly powerful influence on the behaviour of disadvantaged insurees (e.g. those with a low income, with a low educational level and inadequately aware of their own state of health). Here too, the failure to claim necessary treatment could lead to higher consequential costs. In contrast to Huber's study, the present work investigates optional deductibles, in other words, the insurees could freely choose whether to accept a tariff with a deductible or not. As a result, those opting for a deductible tend to be members of the privileged segment of the population. These privileged participants were compared with a matched control group and, in both the short and the medium term, claimed significantly lower amounts of health benefits after opting for a tariff with a deductible. The authors interpret this result to mean that privileged persons are capable of doing without unnecessary or not so important benefits and that moral hazard can be suppressed. Disadvantaged insurees generally do not choose these optional tariffs so that the effects observed by Huber are less relevant for the AOKN tariffs with deductibles. The resources saved in the case of privileged insurees could therefore be used to improve efficiency.

A further limitation in the investigation of deductibles is the frequent frequent assumption by health insurance providers that all other things remain equal; as a result, changes in behaviour are masked. Changes over time are then ascribed completely to the effect of a deductible (Reiners 2009). This limitation also applies to a lesser extent to the present study. The authors assume that the change in behaviour exerts the same effect on the insurees with deductibles as it does on the matched control group.

Moreover, propensity score matching provides no guarantee that all interfering variables are known and observable, and can therefore be accounted for in the study. Thus, Stukel et al. (2007) investigated the benefits of cardiac catheter examinations regarding the reduction in mortality and, by means of propensity score matching, determined a reduction in mortality of 50%. Comparable randomised clinical studies show benefits of no more than 8% to 21%. Stukel attributed this overestimation of the benefit by propensity score matching to the fact (among other things) that not all the interfering variables are known and/or observable. For example, the decision behaviour of the doctor cannot be taken into account. For reasons of risk aversion, doctors tend more to subject younger patients, and those with mild cardiac symptoms, to cardiac catheter examination (Stukel et al. 2007). Interfering variables of this type are not included in our study. However, it can also not be excluded that unknown interfering variables are present. The outcome of this would be that not all self-selection effects could be neutralised and the influence of the tariffs with deductible would then be overestimated.

In conclusion, it can be stated on the basis of the results of the present study that optional deductibles can represent a useful supplement to the full insurance protection provided by the GKV. Further, research is necessary, especially regarding the long-term effects of tariffs with deductibles. In the authors' opinion, the GKV regulation of deductibles will continue to be a focal point of research.