نوآوری و انتخاب ریسک در تجدید ساختار بیمه سلامت اجتماعی

One important motive for deregulating social health insurance is to encourage product innovation. For the first time, the cost savings achieved by non-US managed care plans that are attributable to product innovation are estimated, using a novel approach. Panel data from a major Swiss health insurer permits to infer health status, which can be used to predict health care expenditure. The econometric evidence suggests that the managed care plans benefit from risk selection effects. In the case of the health maintenance organization (HMO) plan, however, the pure innovation effect may account for as much as two-thirds of the cost advantage.

The objective of managed competition in health care is to direct health insurers’ efforts towards product innovation rather than towards cream skimming in spite of premium regulation that creates incentives in favor of enrolling low risks. In order to neutralize this incentive, risk adjustment schemes have been introduced in Germany (Schneider, 1994), The Netherlands (van de Ven and Ellis, 2000), Switzerland (Spycher, 2002) and (in a somewhat different way) in the US (Greenwald et al., 1998). However, these schemes fail to fully bar risk selection because insurers always dispose of private information that goes beyond that contained in the risk adjustment factors (such as age, sex, and even diagnostic information). On the other hand, they easily punish an insurer for its innovation because new products typically attract mobile (and hence often younger) individuals. This suspicion is of particular relevance in the case of Switzerland, whose risk adjustment scheme controls for age and sex only (Beck and Zweifel, 1998). Therefore, an innovating insurer will not only be suspect of cream skimming but sanctioned by an increased payment into the scheme because of the age adjustment. Quite generally, the mix between product innovation and risk selection achieved under the influence of risk adjustment constitutes an important policy issue. The experience of a major Swiss health insurer is of particular interest in this context. In contradistinction to other countries (and most other competitors within Switzerland), this insurer not only writes policies with different annual deductibles but also offers a choice of managed care (MC) plans. Among these alternatives, the health maintenance organization (HMO) plan stands out with average treatment costs that are 62% lower than in the conventional fee-for-service setting. At the same time, HMO enrollees are some 7 years younger than enrollees of conventional plans. The objective of this contribution thus is to find out how much of the cost difference can be traced to risk selection (i.e. differences in socioeconomic characteristics and in particular latent health status) and how much of it remains attributable to innovation (i.e. differences in plan incentives, to be detailed in Section 4.1). Should it be found that most of the cost differential is due to innovation, then the MC alternatives may well constitute a Pareto improvement since the reservations formulated in the literature (see Smith and Betley, 2000) do not apply to the Swiss case. First, fees (and in particular hospital rates) cannot be individually negotiated by health insurers, which precludes cost shifting between MC and conventional service providers. Second, individuals are entirely free to choose between the alternatives considered, permitting their choices to be interpreted as an expression of revealed preference. Conversely, if risk selection effects should turn out to be dominant, then a case could be made for a mandated uniform health insurance system that is not fraught with the considerable loading for acquisition costs incurred for the marketing of differentiated insurance products. The traditional approach to the problem of controlling for risk selection in non-experimental situations has been to take the choice of plan into account in one of two ways, either by endogenizing this choice (Cameron et al., 1988) or by considering it as a sample selection mechanism calling for sample selection bias correction (Waters, 1999). However, the present database contains repeated measurements of the same individuals. This permits a novel approach, which consists of inferring latent health status from previous health care expenditure (HCE). This new approach is compared to the second alternative, viz. the sample selection bias correction, to check whether the two yield similar results. Section 2 is devoted to a literature review concerning the effects of MC plans on HCE and the question of whether risk selection may be the cause of the cost difference. Section 3 contains possible model specifications, which depend importantly on whether choice of insurance plan and of intensity of treatment are thought to occur simultaneously or in a sequence. In addition, the statistical determination of latent health from previous observations of HCE is described. In Section 4, the database and the variables used are defined and the estimation results presented. It turns out that accounting for plan choice through correcting for sample selection bias and using measurements of latent health result in very similar estimates. The main finding is that a major part of the cost differential in favor of MC can be traced to innovation, and Section 5 offers a discussion of the reasons for and possible criticisms of this result.

Managed competition in health care is an attempt to combine regulated premiums with competition in a way that should foster product innovation rather than inducing risk selection activities. Therefore, it is of considerable interest to find out about the extent to which innovations in the guise of different variants of managed care achieve cost savings thanks to improved incentives rather than risk selection. A literature review of attempts to control differences in health care expenditure for risk selection reveals that the health maintenance organization variant of MC features ‘true’ cost reductions of up to 39%, with the independent practice association and the preferred provider organization achieving much smaller reductions. This contribution benefits from a panel dataset prepared by a major Swiss health insurer that offers HMO, IPA, and PPO policies along with conventional ones, all individually contracted. It introduces a novel approach to the problem of controlling for differences in latent health and hence risk selection effects which profits from the availability of repeated observations. Specifically, the HCE of three consecutive years can be estimated using a Tobit random effects specification, yielding residuals whose time-invariant components reflect latent health status (Conclusion 1). When latent health is included in the Probit equation for the likelihood of initializing a treatment episode, only the HMO plan is found to have a lowering effect; however, all three MC variants compare favorably with the conventional alternative with regard to HCE given initialization (Conclusion 2). A comparison with the Lee (1983) sample bias correction approach reveals that its sample bias correction term loses statistical significance in the majority of cases once latent health is included in the HCE equation, suggesting that latent health does contain some extra information. However, estimated ‘true’ HCE reductions do not differ much between the two statistical procedures (Conclusion 3). The gross reduction in cost achieved by the HMO attains a high 62% in this sample, which may add to the suspicion that risk selection is at work. However, two-thirds of that figure remain as ‘true’ savings due to contractual innovation when the effects of risk selection are accounted for. Both the PPO and IPA variants feature lower gross reductions but also a smaller share attributable to contractual innovation (Conclusion 4). These results are subject to several qualifications. First, the high gross cost reductions observed may be due to the fact that the market share of MC has been too small (7%) to force adjustments in conventional medical care. It is an open question for future research whether the relative contributions of risk selection and contractual innovation will change in the event that conventional care and MC should converge in the future. Second, a more comprehensive set of observable socioeconomic characteristics of the insured might have resulted in different estimates of latent health status as well as of sample selection bias correction terms in the Lee (1983) approach. To the extent that the present estimates of these regressors contain measurement error, risk selection effects of MC on HCE may still be underestimated and incentive effects overestimated. However, the counter hypothesis that the incentive effects are all zero has little credibility because the ranking of the HMO, PPO, and IPA variants in terms of ‘true’ savings corresponds to theoretical expectations. The Swiss variants of the three MC plans indeed have characteristics that encourage changed physician incentives, such as a stable flow of patients in return for gatekeeping (IPA), remaining in the selected group of ‘reasonable’ providers (PPO), and capitation, salaried status, and utilization review (HMO). On the other hand, market power of the insurer is an unlikely reason for the cost savings in view of the small market share of these plans noted above and fees that are negotiated at the association rather than the individual insurer level. As a final note, this research also sheds some light on the possible side effects of risk adjustment schemes. As noted in Section 1, these schemes punish an insurer for launching a product innovation because this likely attracts mobile and hence younger individuals. Indeed, this study presents evidence that two-thirds of the cost reductions achieved by such an innovation may be due to risk selection in the IPA variant. On the other hand, up to two-thirds of these reductions were attributable to changed contractual incentives in the HMO variant. However, the financial sanction for an excessive share of young enrollees is the same regardless of plan. Moreover, cost savings must sooner or later be passed on in lowered premiums under the pressure of competition, causing the expected contribution to profit to be reduced, with no adjustment in the risk adjustment payment provided. These arguments point to the distinct possibility that risk adjustment schemes cause innovative plans not be launched or to be discontinued although their main effect would be cost and premium savings due to improved incentives rather than risk selection.