بیمه سلامت بهینه: مورد قابل مشاهده، بیماری شدید

We explore optimal cost-sharing provisions for insurance contracts when individuals have observable, severe diseases with a discrete number of medically appropriate treatment options. Variation in preferences for alternative treatments is unobserved by the insurer and non-contractible. Interest in such situations is increasingly common, exemplified by disease carve-out programs and shared decision-making (SDM) tools. We demonstrate that optimal insurance charges a copay to patients choosing the high-cost treatment and provides consumers of the low-cost treatment a cash payment. A simulation of the effect of such a policy, based on prostate cancer, indicates a substantial reduction in moral hazard.

Many health plans have carved out ‘disease management programs’ for specific diseases that have expensive treatment options, such as cancer, heart disease, and musculoskeletal conditions (LaPensee, 1997). These initiatives recognize that in many clinical situations, there are several alternative treatment paths that are medically acceptable. In some cases, shared decision-making (SDM) programs are employed to facilitate integration of patient preferences into the decision-making process.1 Virtually no theoretical work has been done to examine the optimal manner in which economic incentives might be incorporated into decision making in these situations. Examination of this issue is timely for several reasons. First, interest in facilitating (and managing) treatment choices in specific disease areas has been growing, as evidenced by growth in carve-out insurance designs and development of decision assistance tools for specific disease areas. Second, advances in medical and health services research have provided much greater information regarding the clinical and economic effects of treatment alternatives than previously existed. Third, many studies suggest that adoption and diffusion of new medical technology are the driving force behind rising health care costs Newhouse, 1993, Cutler and McClellan, 1996 and Chernew et al., 1998. New technologies typically present a choice between established treatment paths and those using the innovative approach. Patient preferences often are important determinants of treatment choices in these situations. Efficient allocation of resources is often not an issue of whether anybody gets the new treatment, but instead, it is an issue of who gets the new treatment (Cutler and Richardson, 1999). Cost sharing may thus facilitate efficient resource allocation in cases when the new technology is applicable to serious diseases (such as bone marrow transplantation for breast cancer, cryosurgery for prostate cancer, minimally invasive open heart surgery, or stenting for coronary artery disease). We allow contracting based on broadly defined disease state. Such contracting is consistent with the existence of disease-specific insurance carve outs and the use of SDM tools for allocating resources. We assume that within these disease states, expenditures have a lower bound corresponding to the cost of the least expensive, medically appropriate alternative; i.e., given the presence of the broadly defined disease state, all consumers will spend at least this minimum amount on medical care. Thus, there exists a non-discretionary component of expenditures (which may be quite large in absolute terms for certain serious disease states) to which it will not be optimal to apply any cost sharing because cost sharing would force the consumer to bear risk without any offsetting improvement in incentives. However, expenditures above the minimum are discretionary and result in considerable variation in spending because some patients opt for more expensive treatment paths. We emphasize heterogeneity of preferences for alternative medically acceptable treatments as the cause of spending variation. We explicitly allow the insurance contract to base copayments on the patient's choice of treatment option. More generally, this allows the insurance contract to make a cash payment to patients choosing the low-cost treatment options and charge a fee for patients choosing the high-cost treatment alternatives.2 Our model does not allow individuals to contract ex ante for specific procedures in the event that they experience specific illnesses. Such contracting would require individuals to become informed about the relevant treatment options for all diseases prior to realization of the illness shock. We view the transactions costs associated with acquiring such information as prohibitive. In our model, these costs are incurred after one's health state has been revealed and only for the relevant disease. Our model extends standard health insurance models Zeckhauser, 1970, Baumgardner, 1991 and Gaynor et al., 1997. Like these models, we recognize that the primary benefit of insurance is to mitigate risk (smooth income between the sick and healthy states). This goal is achieved by reducing the price enrollees must pay for treatment at the time of illness. Yet, reduction of patient cost sharing has the adverse consequence of encouraging excess consumption (Manning et al., 1987). As a result, insurance policies must weigh the risk mitigation benefits against the deleterious incentive effects.3 Our analysis indicates that in cases of severe, observable illnesses, for which there are multiple medically acceptable treatment paths, the optimal policy, given information constraints, would be characterized by a payment from the insurer to the patient if the patient chooses the low-cost treatment path and a payment from patient to the insurer if the patient chooses the high-cost treatment path. We term this policy the shared savings policy (SSP). Because of risk aversion, the SSP sets the spread in out-of-pocket costs to the consumer to be less than the difference in treatment costs. Allowing the insurer to share savings with patients receiving the low-cost treatment and charge patients receiving the high-cost treatment allows the insurance policy to preserve incentive effects while better smoothing income relative to a case in which insurers could not provide a payment to patients. We believe that there are significant clinical areas in which the intuition of our model might apply. Because health care expenditures are very skewed, the basic intuition of this model — that consumer cost sharing should not be applied to inframarginal expenditures — could have widespread applicability. Roughly 77% of full-time employees of medium and large establishments enrolled in non-HMO plans have maximum out-of-pocket limits less than US$2000 per individual and the most common coinsurance rate is 20% (Pemberton and Holmes, 1995). Thus, individuals with more than US$10,000 in total costs will face no cost sharing at the margin.4 Analysis of NMES data, updated to 1999, suggests that about 62% of expenditures occurs in individuals with over US$9044 in annual spending and 54% of expenditures occurs in individuals with over US$12,568 in spending (Agency for Health Care Policy and Research (AHCPR), 1999). Our model might not apply in all of these cases because in some of these cases, there may not be clinically acceptable medical alternatives with significant cost differences. Yet, in several major disease areas, such as ischemic heart disease and cancer, our model may be largely relevant. In these areas, the fiscal implications of the treatment choice may be very large. In 1991, over US$3.4 billion were spent on heart attacks for individuals over the age of 65 (Cutler and McClellan, 1996). Coronary revascularization for these patients, using either bypass surgery or angioplasty, entails substantial expense. Similarly, cancer patients are often faced with choices between surgery and less invasive treatment options. Moreover, the intuition of our model can be applied more generally in models of cost sharing. For example, concern over rising pharmaceutical spending has led to calls for ‘reference pricing’, in which consumers are charged for pharmaceuticals based on the cost of a ‘reference’ drug in each drug class Danzon, 1998 and Huskamp et al., 2000. Our model generally supports this framework because cost sharing is only applied at the margin (for expenditures above those associated with the reference drug). Further, our model suggests that in a reference pricing system, consumers of more expensive pharmaceuticals should not be charged the full incremental cost of their consumption because of the risk associated with illness and the risk that, ex post, their desire for the more expensive product will be high. The extent to which the SSP could be completely implemented depends on a variety of practical concerns and advances in medical and health services research. Transaction costs and lack of acceptable alternatives may prevent full implementation, though advances in health services research will expand the number of areas in which the full SSP could apply. In many cases in which it is impractical to fully implement the SSP, limited implementation based on the intuition of our model may be feasible. For example, copays could be waved in cases of heart attack and many cancers beyond those in early stages. We contend that many existing insurance contracts are designed exactly wrongly in situations of serious illness. They typically impose a deductible for low levels of expenditure followed by a coinsurance rate up to a cap (e.g., 20% of expenditures above the deductible up to US$2000 out of pocket) followed by full coverage until a maximum benefit is reached. In the case of many severe illnesses, expenditures will always exceed the point where coinsurance ends, and the maximum benefit is usually set sufficiently high that it will rarely, if ever, be met. This policy design creates a coverage profile in which the percentage of the cost paid by the insurer rises as expenditures rise. If illness is severe, the cost-sharing provisions apply when there is little discretion in expenditure levels and full coverage applies at the margin, exactly when economic efficiency would demand cost sharing. The efficiency loss associated with zero cost sharing at the margin is recognized by Feldstein and Gruber (1995). They argue that a policy of 50% cost sharing up to 10% of income would increase aggregate national welfare by US$34 billion a year. Our model demonstrates that if some of the cost sharing could be shifted to the margin, the welfare gains could be achieved with lower out-of-pocket expenditures (analogously, for the same out-of-pocket spending, greater efficiency could be gained). This paper is organized as follows. In Section 2, we set up the framework for our model. We allow insurance plans to be made contingent upon broadly defined disease type but not upon severity or willingness to pay. Too many patients (when compared with the social optimum) will select the higher cost treatment option when faced with the same standard copayment rate for each treatment. In Section 3, we show that to deal with this moral hazard problem, the optimal health insurance plan differs from the standard health insurance plan in several ways. The optimal health plan offers a cash payment to any patient who elects to undergo the low-cost procedure. It also decreases the copayment for the high-cost procedure and raises the premium. These changes better smooth income across sick and healthy states, and across treatment choices in the sick state. They also substantially reduce the moral hazard problem. Section 4 presents a simulation of the expected cost savings, relative to existing policies, resulting from implementing an optimally designed health plan into a disease management program for prostate cancer. In our example, we find that such a plan mitigates between 46% and 84% of the moral hazard problem. Finally, Section 5 provides concluding remarks, including a discussion of implementation issues related to adoption of the optimal insurance plan. Although several issues may hinder the literal adoption of the optimal plan, the model provides several insights regarding intermediate types of policy actions that may improve insurance design.

The intuition behind these results is clear. Optimal co-insurance rates should vary with the elasticity of demand. In the case of discrete treatment alternatives, there is no discretion regarding expenditures below the cost of the least expensive treatment. Therefore, patients choosing this treatment path should be fully covered (any cost sharing just adds risk without any efficiency gains). The price sensitivity of consumption exists only over the incremental spending and therefore, only those costs above the costs of the cheapest treatment should be subject to cost sharing. The form that cost sharing should take in these cases is less straightforward. The use of rebates (shared savings) for individuals opting for the less expensive treatment alternative allows better income smoothing across sick and healthy states. At the same time, the rebate serves as a welfare-enhancing carrot that induces more patients to select the low-cost procedure in accord with their willingness to pay, mitigating a great deal of the moral hazard problem of health insurance. In fact, in the simulations, we saw that up to 84% of the moral hazard problem was resolved by the shared savings rebate. Moreover, because individual preferences are not observed, individuals can benefit if they receive some ‘insurance’ against the probability that they prefer the more expensive treatment. Because they cannot contract for specific treatments ex ante, the form this insurance takes is a reduced difference between patient expenditures for the less expensive and more expensive treatments, relative to the cost differential. In the optimal health insurance policy, the gap in price (to the patient) between the two treatments provides the appropriate balance between the incentive for efficient choice of services and risk aversion. Our results stand in stark contrast to the design of many existing insurance policies in which cost sharing is largely limited to relatively low levels of expenditure. We argue that, for certain conditions, insurance contracts, such as medical savings accounts combined with catastrophic coverage policies, are designed in a structurally suboptimal manner. In particular, a minimum level of medical care consumption exists for many verifiable disease states. Discretion in treatment choice only affects expenditures above this minimum level. As a result, patient cost sharing should only apply once the minimum expenditure has been met. Catastrophic plans provide full coverage exactly when it should not exist (at the margin) and provide partial coverage precisely when cost sharing provides no benefit (at expenditures below the cost of the least expensive, medically appropriate treatment alternative). Our paper is predicated on the idea that financial incentives are a valuable tool for guiding treatment decisions. We are generally sympathetic to critiques of this premise. The issues of social norms and ethics are important and worthy of more exploration. A variety of studies have addressed their role in guiding treatment choices. This paper is not intended as a normative study advocating for any particular insurance policy design. Instead, we view it as an analysis addressing how financial incentives might be structured if they were to be used. Kasper et al. (1992) raise the concern that the use of financial levers to guide treatment will enter the utility function. We recognize that it can be distasteful for individuals, when faced with stressful decisions that may have implications regarding their health, to be concerned with the financial ramifications of their care decisions. However, the key issue is not whether individuals are less satisfied if they are forced to pay for their preferred alternative. Of course, we would expect utility to drop if prices rise. The issue is whether being confronted with financial considerations lowers utility beyond the inherent disutility associated with paying. If this effect is sufficiently strong, the optimal policy design would suggest waiving cost sharing on inframarginal expenditures. Implementation of a SSP, or waiving copays in selected cases of serious illness, raises distributional considerations. It is important to remember that our analysis holds income constant so that the implications hold within each income class. To the extent that optimal policy design for less affluent individuals would differ from that for affluent individuals, it is certainly true that the parameters of any policy such as that which we outline should, to the extent possible, be set appropriately for each income class. This could be done by patient sorting in the insurance market. The SSP does not inherently charge patients choosing the high-cost treatment more than a standard insurance policy. In our second simulation, the SSP charges consumers of treatment B only US$31 more than in the standard plan. The difference is that the cost is responsive to behavior, not simply a tax on being ill. Several issues complicate the implementation of the SSP. First, for the outlined policies to be optimal, it is necessary that the patient be involved in the decision-making process and be well informed regarding the clinical consequences of each treatment alternative. Medical culture, pushed in part by the movement towards managed care and a variety of liability concerns, is increasingly involving patients in the choice of treatment path. For many diseases, physicians discuss treatment alternatives with patients informally. Frequently, patients are provided with literature that describes treatment options. More recently, researchers have begun to produce sophisticated decision assistance tools in media such as interactive video to facilitate patient input into the decision-making process. Second, our model focuses treatment decisions in a static framework. We have assumed that it unreasonable for patients to get both treatments, and that insurance would only pay for one treatment. In a dynamic context, it may be reasonable for patients to get both treatments either because their preferences changed or because their health status changed. For example, patients who opt for medical management of coronary disease may opt at a later date to have surgery. Choice of one treatment should not result in forfeiture of coverage for the other service indefinitely. It is therefore important that the cost sharing be set optimally over time and appropriate rules be set up to delineate when an individual can switch treatment choices and retain coverage. Similarly, within each treatment path, choices may exist which create variation in expenditures. For example, how long is the length of stay for surgery or what type of radiation therapy is used? This type of moral hazard should also be addressed, perhaps through guidelines or optimally designed cost sharing within treatment paths. Copayments at the initial treatment decision should reflect expected expenditures conditional on the treatment paths. Third, implementation of the SSP requires that the expected costs of each treatment path be estimated ex ante. In many cases, the immediate costs of alternative treatments can be calculated from existing databases. Yet, because the cost differences between alternative treatment paths may persist or decay over time, an accurate understanding of the long-term fiscal implications would be required to design optimal cost-sharing rules. Ideally, the difference in costs would capture the expected lifetime costs, based on optimal coinsurance rates at each decision point. In many cases, this is possible, at least to a first-order approximation, because cost containment concerns have motivated the economic evaluation of a variety of treatment alternatives. Nevertheless, one should recognize that because technology changes (in unpredictable ways), the lifetime cost estimates at any point in time will likely mis-estimate actual lifetime costs. Fourth, the optimal cost-sharing provisions may vary for different disease states. This would require contracts to include separate cost-sharing rates for each disease, and potentially for each treatment alternative. The transaction costs of constructing such contracts, or even informing prospective enrollees, might be great. However, insurers could implement an approximate copay rate for a whole set of diseases, exactly as they currently do. Deviations from these cost-sharing provisions could be made only for those diseases for which their expected benefits exceed the implementation costs. Finally, an additional implementation issue associated with the indiscriminant adoption of the SSP would generate incentives for individuals and their health care providers to attempt to identify illnesses which may qualify for a rebate. It is thus important that the SSP apply only when illness (and rough level of severity) is observable. In many cases, including prostate cancer, disease staging can allow exclusion of individuals with disease too mild to be realistic candidates for the more expensive procedure. Even so, the SSP incentives would result in increased costs to insurers due to incremental testing and identification of qualified patients. We cannot assess whether on balance, increased identification would be deemed cost-effective (as might be the case for mammograms for women over 50) or not cost-effective (as might be the case for increased prostate cancer screening). Even if ethical and implementation issues associated with the SSP prevent its literal adoption, our model provides useful insights. For example, in cases in which the minimum expenditure exceeds the point where cost sharing disappears, the patient contribution toward the inframarginal expenditures should be waived. This represents a relatively small deviation from the standard plan and would improve expected utility. Another alternative, which preserves financial incentives but may be more palatable than the SSP, would entail waiving the copayment only for patients choosing the low-cost treatment path. Patients choosing the high-cost treatment path would pay the same as they would under the standard plan. Such a policy would be analogous to reference pricing for pharmaceuticals (Danzon, 1998). This retains some incentive effects but, at the time of treatment, no patient would pay more than they would under the standard plan, regardless of their treatment choice. This would at least eliminate the illness tax on those patients with a relative preference for the less expensive treatment. If financial incentives are considered too unpalatable, in-kind transfers could be used. Anecdotally, we are aware of situations in which patients received in-kind transfers (home health visits) in exchange for accepting low-cost treatment (early hospital discharge). Despite the ethical and practical issues a SSP raises, it is probably true that some sensitivity to cost is better than none. The imposition of optimal coinsurance rates in cases of discrete treatment alternatives complements initiatives aimed at better informing patients of the clinical consequences of their actions. Patients who suffer from the ailments covered by these decision aids rarely share the financial benefit associated with choosing the lower-cost pathway. Incorporating a fiscal incentive can inject a valuable cost consciousness into the decision-making process.