How Can We Bend the Cost Curve? Risk-Adjusting the Doughnut Hole to Improve Efficiency and Equity

Transcription

1 How Can We Bend the Cost Curve? Richard C. van Kleef Wynand P. M. M. van de Ven René C. J. A. van Vliet Risk-Adjusting the Doughnut Hole to Improve Efficiency and Equity An important goal of consumer cost-sharing in health insurance is to increase incentives for cost containment. A relatively new cost-sharing phenomenon is the : a gap in coverage starting at a predefined level of medical expenses. An important question is where to locate the starting point to achieve the strongest incentives for cost containment. We argue that the answer depends on an individual s health status. Using data from a Dutch insurer, this paper illustrates that using a risk-adjusted starting point results in both stronger incentives for cost containment and more equity than a uniform starting point. A major goal of consumer cost-sharing in health insurance is to reduce moral hazard by providing enrollees with incentives for cost containment. A relatively new cost-sharing phenomenon is the so-called : a gap in coverage that starts at a predefined level of medical expenses. This particular type of insurance design is currently applied in Medicare Part D prescription drug coverage and is potentially interesting for other health plans as well. 1 A simplified version of this concept is graphically shown in Figure 1, with d referring to the and s referring to the point where the begins. An important question for policymakers is where to locate the starting point to achieve the strongest incentives for cost containment. In this paper, we argue that the answer depends on the individual s health status, implying that we expect a risk-adjusted starting point to result in stronger incentives for cost containment than a uniform starting point. The article is structured as follows. First, we theoretically consider incentives for cost containment under a doughnut-shaped insurance design. We start from the arguments of Newhouse (1993) and the framework of Van Kleef, Van de Ven, and Van Vliet (2009), who discuss the incentives for cost containment under a first-dollar deductible plan. Secondly, we consider three technical requirements for implementation of a risk-adjusted starting point: availability of data, an adequate (risk adjustment) model for calculating the expected costs, and a measure for determining the starting point given the expected costs. Thirdly, we illustrate intuitively the concept of a riskadjusted starting point using administrative data from a Dutch insurer. We conclude that risk-adjusting the is not just expected to increase incentives for cost containment, but also to reduce the difference in out-ofpocket expenses between the healthy and the chronically ill. With respect to the latter, a Richard C. van Kleef, Ph.D., is an assistant professor; Wynand P.M.M. van de Ven, Ph.D., is a professor of health insurance; and René C.J.A. van Vliet, Ph.D., is an associate professor of statistics, all at the Institute of Health Policy and Management, Erasmus University Rotterdam. Address correspondence to Prof. Van Kleef at the Institute of Health Policy and Management, Erasmus University Rotterdam, P.O. Box 1738, 3000 DR Rotterdam, The Netherlands. vankleef@bmg.eur.nl Inquiry 48: (Winter 2011/2012) Excellus Health Plan, Inc. ISSN /inquiryjrnl_

2 Inquiry/Volume 48, Winter 2011/2012 Figure 1. Insurance schedule in the case of a (full coverage of medical expenses up to starting point s, no coverage in the interval s to s+d, and full coverage beyond s+d) risk-adjusted starting point can be considered more equitable than a uniform starting point. In the last section, we raise some important issues for discussion and further research. Doughnut Holes and Incentives for Cost Containment Compared to full coverage, cost-sharing reduces moral hazard by providing consumers with incentives for cost containment (Keeler and Rolph 1988). In the case of a, the strength of these incentives is expected to vary with the probability of reaching the gap and with the probability of exceeding the gap. For an explanation, we have to distinguish between the market price and the perceived price of medical care. With respect to the perceived price, we mean the price experienced by consumers, given their expected expenses in the contract period (Van Kleef, Van de Ven, and Van Vliet 2009). In the following example, Newhouse (1993) illustrates the relevance of this distinction in the context of a first-dollar deductible, which can be seen as a starting at $0 as applied in consumer-driven health plans in the U.S. (such as health savings accounts) and in mandatory health insurance plans in Switzerland and the Netherlands. As Newhouse (1993, p.81) explains: Consider someone with a first-dollar deductible of $1,000 who has to decide whether to visit his doctor for a market price of $40 on the first day of the contract period. If he would know for sure that his total expenses in the contract period will not exceed $1,000, the perceived price of the visit equals the market price. However, if he would know for sure that his total expenses will exceed $1,000 (i.e., if he would anticipate free care later in the contract period), the perceived price of the visit equals $0. Thus, although the market price is the same in both scenarios, the perceived price differs. Newhouse (1993, p. 81) argues that examples like this suggest a simple rule: The price a utility-maximizing consumer on a firstdollar deductible plan will use to decide whether a visit is worth its cost is the product of the market price and the probability of not exceeding the deductible amount. If the consumer in the example had had a 75% chance of exceeding the deductible, the perceived price of the visit would have been $10 (i.e.,.25 * $40). This rule implies that the in Figure 1 will hardly provide incentives for cost containment to consumers whose expected expenses (far) exceed that. A shift of the to a higher expenditure level changes the probability of exceeding it and, thereby, affects the incentives for cost containment. The important question is: In 314

3 Risk-Adjusting the Doughnut Hole which direction? On the one hand, raising the starting point reduces the probability of exceeding the, resulting in stronger incentives for cost containment. On the other hand, however, it also reduces the probability of reaching the, thus resulting in weaker incentives for cost containment. We argue here that for chronically ill individuals, the first effect dominates the second, at least up to a certain starting point. At the beginning of a new contract period, the probability of exceeding the depends on the expected expenses. In the absence of any knowledge about future expenses, this probability is the same for all consumers. In practice, however, it is obvious that expected expenses differ among individuals because of variation in health status. As a result, the probabilities of reaching and exceeding a certain also differ among individuals (Zhang et al. 2009; Ettner et al. 2010; Roblin and Maciejewski 2011). For example, the probability of exceeding a $500 that starts after an individual has reached $1,000 in expenditures (with full coverage kicking in again at $1,500) is nearly 1 for a diabetic with planned treatment costs of $3,000, but is much lower for a healthy person with no planned treatments. For the diabetic, the incentives for cost containment will be stronger with a starting at $2,500. For the healthy person, however, the opposite holds true because that person does not expect to reach an expenditure level of $2,500. The previous arguments reveal an important limitation of any uniform starting point: Incentives for cost containment will be weak for one of the groups, either the healthy or chronically ill. With a starting at $0 (that is, a first-dollar deductible), incentives are strong for the healthy and weak for the chronically ill, while the opposite holds true for a starting point of, say, $5,000. This implies that incentives for cost containment can be increased by risk-adjusting the starting point that is, differentiating the starting point according to individual risk characteristics. Technical Requirements for Implementation Certain technical requirements are essential to implement a risk-adjusted starting point. In the first place, individual-level data on medical expenses and risk characteristics must be available. In health insurance plans with riskadjusted capitation payments such data may already be available; in other plans it may be necessary to set up new data-registration systems. Secondly, an adequate (risk adjustment) model needs to be developed to predict individual health care expenses. In principle, various types of information can be taken into account, such as age, gender, prior hospitalization, and prior use of pharmaceuticals. Extensive literature on risk adjustment and prediction models show the pros and cons of different risk adjusters and different statistical methods (e.g., Duan et al. 1983; Manning and Mullahy 2001; Van de Ven and Ellis 2000; Van de Ven, Van Vliet, and Lamers 2004). With respect to functional form, it is obvious that the prediction model should perform well in the lower and middle regions of the expenditure distribution. These are probably the regions where a would produce the strongest incentives for cost containment. Regarding the risk adjusters, criteria such as validity, measurability, and no possibilities for manipulation must be taken into account. The term no possibilities for manipulation means that consumers or other parties should not be able to influence risk adjusters, since this could lead to inappropriate starting points. For example, if prior costs were used as a risk adjuster, consumers could increase medical consumption once they exceeded their doughnut hole for that year in order to get a higher starting point in later years. Consumers might prefer a higher starting point because of lower expected out-of-pocket expenses. Given the expected expenses, a third requirement is choosing an adequate criterion for determining the starting point. The challenge for policymakers is to locate the so that the incentives for cost containment are maximized. In this respect, it is not just the probability of exceeding the that plays a role, but also the probability of reaching the. This can be illustrated by the simple doughnut-shaped insurance design in Figure 1. Compared to full coverage, the doughnut hole creates two kinks in the insurance schedule, resulting in the following three coverage intervals: full coverage up to starting point s, 315

4 Inquiry/Volume 48, Winter 2011/2012 no coverage in the interval s to s+d, and full coverage once a person s expenditures go beyond s+d. Increasing the starting point reduces the probability of exceeding the, which boosts the incentives for cost containment in the interval 0 to s+d. At the same time, however, an increase of the starting point level reduces the probability of reaching the doughnut hole, which decreases incentives for cost containment in interval 0 to s. The challenge is to find the point at which the first effect no longer dominates the second. In their framework of a shifted deductible for high risks, Van Kleef, Van de Ven, and Van Vliet (2009) propose the variance in expected out-of-pocket expenditures as the instrument for determining the optimal shift. Their argument is as follows. For any individual, the incentives for cost containment will be weak both with a high probability of exceeding the coverage gap and with a low probability of reaching the coverage gap. In the first case, the expected out-of-pocket expenses are close to the maximum, while in the second case they are close to zero. In both cases, however, the variance in expected out-of-pocket expenses will be small. As these probabilities (and thereby the out-of-pocket expenses) move away from these boundaries, the variance in expected out-of-pocket expenses will increase. Maximum variance indicates maximum uncertainty about reaching and exceeding the coverage gap, resulting in maximum incentives for cost containment. With this criterion, the will be located somewhere in the center of the expenditure distribution (Van Kleef, Van de Ven, and Van Vliet 2009). A drawback of this criterion, however, is its complexity. More practical instruments for locating the in the center of the expenditure distribution could be the mean or median of expected (out-of-pocket) expenses. An Empirical Illustration We illustrate intuitively why we expect a risk-adjusted starting point to result in both stronger incentives for cost containment and more equity than a uniform starting point. The data come from a Dutch health insurer and include individual-level information (N 5 50,970) on: 1) medical expenses in 1994 regarding inpatient care, outpatient care, and pharmaceuticals, among others, and 2) drug prescriptions in Based on the drug prescriptions in 1993, we distinguish among 20 health problems: psychosis, mood disorders, chronic obstructive pulmonary disease (COPD), inflammatory conditions, heart disease, thyroid disorders, high cholesterol, gout, peripheral artery disease (PAD), glaucoma, epilepsy, high blood pressure, diabetes, gastric disorders, inflammatory bowel disease, chronic pain, rheumatism, Parkinson s disease, cancer, and cystic fibrosis. People were assigned to one of these health problems if they had at least four prescriptions for a drug specifically meant for treatment of that particular problem. For example, those with at least four prescriptions for insulin were assumed to suffer from diabetes. Individuals with multiple health problems were classified in the category showing the highest average costs in This led to 14.51% of the population being classified as having at least one of these 20 health problems. We corrected for theincreaseinhealthcareexpensesfrom1994to 2008 by multiplying the 1994 expenses by a factor of For details about the data and the pharmacy-based classification see Lamers (1999). Since the definition of health problems is based on data from 1993, we consider the average costs per category in the following year as the expected costs in the following year for someone in that category. To generalize the illustration, 1994 is referred to as year t and 1993 as year t-1. Table 1 shows that for each of the 20 health problems the expected costs significantly differ from the group of individuals without any of these health problems. Let us assume that in year t all 50,970 individuals are given the same health plan, with a of 500 euros (1 euro5 $1.35 U.S.) starting at the mean of annual expenses and full health coverage outside the doughnut hole. Given the expenses of these individuals, this insurance design leads to average out-ofpocket expenses of 85 euros, a.19 probability of reaching the, and a.15 probability of exceeding it. 3 At the subgroup level, however, substantial differences can be observed: for each of the 20 health problems, the expected out-of-pocket expenses, as well as the probabilities of reaching/exceeding the doughnut 316

5 Table 1. Mean and standard deviation (S.D.) of total costs in year t, by health category, based on drug prescriptions in year t-1 (N=50,970) Health problem based on drug prescriptions in year t-1 Prevalence in year t (%) Mean of cost in year t (euros) Risk-Adjusting the Doughnut Hole S.D. of cost in year t (euros) None of health problems below ,109 3,869 Psychosis.30 2,468* 5,556 Mood disorders.47 2,743* 4,654 COPD ,113* 5,819 Inflammatory conditions ,405* 5,768 Heart disease ,720* 6,795 Thyroid disorders.45 4,042* 7,537 High cholesterol.29 4,024* 6,173 Gout.04 4,084* 6,772 Peripheral artery disease.37 4,488* 5,688 Glaucoma.34 4,761* 6,668 Epilepsy.47 5,083* 9,040 High blood pressure ,676* 9,981 Diabetes ,175* 8,388 Gastric disorders ,842* 10,208 Inflammatory bowel disease.11 7,056* 9,338 Chronic pain.05 8,167* 7,097 Rheumatism.39 11,366* 12,995 Parkinson s disease.12 11,910* 17,427 Cancer.13 12,441* 17,681 Cystic fibrosis.04 18,593* 18,750 Total ,674 5,450 Notes: Data are from 1993/1994. Monetary figures are in 2008 euros (1 euro5 $1.35 U.S). COPD5 chronic obstructive pulmonary disease. * Statistically significant from the group of individuals without any of the 20 health problems, with p #.001 (two-sided t-test). hole, significantly differ from the group of individuals without any of these health problems. This reveals two important consequences of a uniform starting point: 1) expected out-ofpocket expenses are higher for the chronically ill than for the healthy, a result that can be considered inequitable, and 2) perceived prices vary with health status, which implies differences in the strength of the incentives for cost containment. Given the probabilities of reaching and exceeding the, as shown in Table 2, and following the arguments by Newhouse (1993) and Van Kleef, Van de Ven, and Van Vliet (2009), we expect the incentives for cost containment to be low for both someone without any of the 20 health problems (due to a low probability of reaching the ) and someone with one of the health problems in the bottom rows (due to a high probability of exceeding the ). In between these extremes, we expect the incentives for cost containment to be relatively strong. Now let us assume a similar insurance design, but with the starting point of the varying according to the health categories. As argued earlier, different measures can be used to determine the starting point, such as the median, mean or variance of expected (out-of-pocket) expenses. For each category, we expect the to create the strongest incentives for cost containment if it is located somewhere in the center of the expenditure distribution. In this illustration, we apply a simple approach: for each category, we locate the starting point so that the expected out-of-pocket expenses for year t equal 250 euros (i.e., 50% of the amount left uncovered in the ). As shown in Table 3, this method leads to starting points ranging from 79 euros for individuals without any of the 20 health problems to 11,860 euros for those with cystic fibrosis. The inherent advantage of this approach is that it completely reduces the difference in expected out-of-pocket expenses among the health categories, which means that the risk-adjusted starting point can be considered as more equitable than the uniform starting point shown in Table 2. Another advantage is that for all health categories, the probabilities of reaching and exceeding the are around.50, 317

6 Inquiry/Volume 48, Winter 2011/2012 Table 2. Applying a uniform of 500 euros in year t: expected out-of-pocket expenses and proportions of enrollees reaching and/or exceeding the, by health category (N=50,970) Health problem basedondrug prescriptions in year t-1 Uniform starting point of doughnut hole in year t (euros) Expected out-ofpocket expenses in year t (euros) Proportion of enrollees reaching in year t Proportion of enrollees exceeding in year t None of health problems below 1, Psychosis 1, *.39*.27* Mood disorders 1, *.44*.34* COPD 1, *.45*.34* Inflammatory conditions 1, *.45*.36* Heart disease 1, *.41*.33* Thyroid disorders 1, *.44*.39* High cholesterol 1, *.60*.47* Gout 1, *.55*.49* Peripheral artery disease 1, *.55*.45* Glaucoma 1, *.51*.40* Epilepsy 1, *.52*.43* High blood pressure 1, *.63*.49* Diabetes 1, *.74*.64* Gastric disorders 1, *.78*.68* Inflammatory bowel disease 1, *.66*.53* Chronic pain 1, *.76*.70* Rheumatism 1, *.80*.74* Parkinson s disease 1, *.93*.90* Cancer 1, *.93*.82* Cystic fibrosis 1, *.89*.78* Total 1, Notes: Data are from 1993/1994. Monetary figures are in 2008 euros (1 euro5 $1.35 U.S). COPD5 chronic obstructive pulmonary disease. * Statistically significant from the group of individuals without any of the 20 health problems, with p #.001 (two-sided t-test). suggesting more uncertainty about the expected out-of-pocket expenses and higher perceived prices during the contract period than with a uniform starting point. Moreover, a larger share of the population will close the contract period right in the area of the :.42 (i.e.,.77.35) with the risk-adjusted starting point (see Table 3) versus.04 (i.e.,.19.15) with the uniform starting point (see Table 2). Discussion In this article, we have argued and illustrated that regarding a in health plans, a risk-adjusted starting point is expected to improve both incentives for cost containment and equity compared to a uniform starting point. We note that our arguments are not just relevant for s, but also for other forms of consumer cost-sharing that are applied uniformly and in combination with a cap on out-of-pocket expenses. Two examples are first-dollar deductibles and capped coinsurance. As shown by Raebel et al. (2008) and Bayliss et al. (2010), these types of insurance design lead to variation in both expected outof-pocket expenses and the probability of reaching the cap on out-of-pocket expenses, which implies that these forms of cost- sharing can benefit from risk adjustment as well. To conclude, we raise some important points for discussion and future research. One issue concerns the criterion for determining the starting point. In our empirical example, we determined the starting points so that the expected out-of-pocket expenses for each category would equal 50% of the uncovered amount in the. An inherent advantage of this criterion is that it results in exactly the same expected out-of-pocket expenses for all categories, which can be considered as equitable. A disadvantage, however, is that this criterion does not necessarily maximize the incentives for cost containment. Although 318

7 Table 3. Applying a risk-adjusted of 500 euros in year t: expected out-ofpocket expenses and proportions of enrollees reaching and/or exceeding the, by health category (N=50,970) Health problem basedondrug prescriptions in year t-1 Risk-adjusted starting point of in year t (euros) Expected out-ofpocket expenses in year t (euros) Risk-Adjusting the Doughnut Hole Proportion of enrollees reaching in year t Proportion of enrollees exceeding in year t None of health problems below Psychosis Mood disorders 1, COPD 1, Inflammatory conditions 1, Heart disease 1, Thyroid disorders 1, High cholesterol 1, Gout 1, Peripheral artery disease 1, Glaucoma 1, Epilepsy 1, High blood pressure 1, Diabetes 2, Gastric disorders 3, Inflammatory bowel disease 2, Chronic pain 3, Rheumatism 4, Parkinson s disease 5, Cancer 4, Cystic fibrosis 11, Total Notes: Data are from 1993/1994. Monetary figures are in 2008 euros (1 euro5 $1.35 U.S). COPD5 chronic obstructive pulmonary disease. the probabilities shown in Table 3 indicate that the risk-adjusted in our illustration is located in the center of the expenditure distribution, there is no theoretical ground to argue that this maximizes incentives for cost containment. From the efficiency perspective, the maximum variance in expected out-ofpocket expenses may be a better criterion. A drawback of this alternative, however, is its complexity, and the fact that out-of-pocket expenses will probably not be (exactly) the same across categories. This implies that the choice for a criterion requires a trade-off among efficiency, equity, and simplicity. A second issue concerns the desirable level of differentiation that is, the number of risk classes to be distinguished. Some of the categories in Tables 1 to 3 may have similar cost patterns. For practical reasons, it may be preferable to merge these categories and to provide them with the same starting point. At the same time, however, these categories may to some extent be heterogeneous; this implies that further differentiating the starting point by including additional risk characteristics such as age, gender, and diagnostic information is expected to result in further improvements of efficiency and equity. On the other hand, further differentiation increases the number of starting points and thereby reduces the transparency of the insurance plan. This suggests that the level of differentiation requires a trade-off between efficiency and equity on the one hand and transparency on the other. A third issue involves the effect of risk adjustment on moral hazard, which is completely neglected in our simplified example. In practice, this effect is important when deciding the level of differentiation and the location of the starting point. In the case of substantially reduced moral hazard, for instance, expected health care expenses would decrease and the starting points in Table 3 would probably be too high. Additional empirical research is needed to obtain more insight into this effect. 319

8 Inquiry/Volume 48, Winter 2011/2012 A fourth concern is changes in the health status and the expected costs of individuals over time. For reasons of efficiency and equity, such changes require periodic updates in the riskadjusted starting points. The frequency of these updates implies a trade-off between efficiency and equity on one hand and administrative costs on the other. Like other applications of risk adjustment such as capitation to health plans it seems reasonable to re-estimate expected costs and to update starting points per contract period. A fifth issue is whether consumers can comprehend the concept of a risk-adjusted. Hsu et al. (2008) have shown that consumer knowledge about cost-sharing arrangements does not come naturally. To achieve incentives for cost containment, however, it is crucial that consumers understand how these arrangements work. This is evident in the findings of Hsu et al. (2008): Beneficiaries who were aware that their plan included a coverage gap were more likely to report any cost response compared with beneficiaries who were unaware of the gap. So, if consumers do not understand the concept of a risk-adjusted, some education may be desirable. With respect to this issue, it may also be preferable to start with a limited level of differentiation in the starting points. Another issue concerns the positive relationship between the expected value and variance of expenses. Comparing Table 1 with Table 3 shows that, in general, a higher variance in expenses leads to a smaller probability that a person will close the contract period somewhere in the. Feldstein (2006, p. 1607) has proposed a simple measure to expand the cost-sharing range. He wrote: Instead of a $5,000 deductible, the insurance policy might take the form of a 50 percent coinsurance on the first $10,000 of spending. Patients will still be protected against paying more than $5,000 out of pocket. But with a 50 percent coinsurance rate on $10,000 of spending, there would be fewer patients and fewer dollars that face no out-ofpocket costs. Following the spirit of this proposal, an interesting extension to a riskadjusted starting point for the might be a risk-adjusted coinsurance rate. For example, where a with a low starting point and 100% coinsurance may be preferable for healthy individuals, a doughnut hole with a high starting point and 50% coinsurance may be preferable for a chronically ill individual. With this extension, the maximum outof-pocket expenses are uniform, but the length as well as the intensity of the varies. A final issue involves the size of the doughnut hole. Although this article does not specifically focus on this topic, it should be mentioned that extensive coverage gaps may result in affordability problems and negative effects on health. Fung et al. (2010), for instance, have reported that the in Medicare Part D does not only result in lower total drug costs, but also in worse adherence compared with having no. In sum, we conclude that s, as well as conventional forms of cost-sharing, can benefit from risk adjustment in terms of efficiency and equity. Regarding some technical requirements and the issues raised previously, however, we recommend that policymakers start off simple when implementing risk-adjusted cost-sharing. For example, it may be preferable to begin by distinguishing only two risk classes: those classified as low risk and those classified as high risk. For reasons of efficiency and equity, any differentiation (even just two risk groups) is better than none. At a later stage when technical requirements have been met and consumers are familiar with the concept efficiency and equity can be further improved by extending the number of relevant risk classes. Notes 1 Recent proposals promise to incrementally reduce the in Medicare Part D up to 2020 (Shrank and Choudhry 2011). 2 We used 2008 as the reference year because at the time of this writing the actual costs of the benefit package in later years were not yet complete. The multiplying factor was calculated as the average costs for the benefit package of 2008 divided by the average costs of the (comparable) benefit package in

9 Risk-Adjusting the Doughnut Hole 3 For reasons of simplicity, we neglect the expected reduction of moral hazard caused by the. This implies that average outof-pocket expenses, as well as the probability of reaching and/or exceeding the, will be overestimated. This has no impact on the conclusions, since they are based on general cost patterns rather than specific numbers. References Bayliss, E. A., J. L. Ellis, T. Delate, J. F. Steiner, and M. A. Raebel Characteristics of Medicare Part D Beneficiaries Who Reach the Drug Benefit Threshold in Both of the First Two Years of the Part D Benefit. Medical Care 48(3): Duan,N.,W.G.Manning,C.N.Morris,andJ.P. Newhouse A Comparison of Alternative Models for the Demand for Medical Care. Journal of Business and Economic Statistics 1(2): Ettner, S. L., N. Steers, O. K. Duru, N. Turk, E. Quiter, J. Schmittdiel, and C. M. Mangione Entering and Exiting the Medicare Part D Coverage Gap: Role of Comorbidities and Demographics. Journal of General Internal Medicine 25(6): Feldstein, M Balancing the Goals of Health Care Provision and Financing. Health Affairs 25(6): Fung, V., C. M. Mangione, J. Huang, N. Turk, E. S. Quiter, J. A. Schmittdiel, and J. Hsu Falling into the Coverage Gap: Part D Drug Costs and Adherence for Medicare Advantage Prescription Drug Plan Beneficiaries with Diabetes. Health Services Research 45(2): Hsu, J., V. Fung, M. Price, J. Huang, R. Brand, R. Hui, B. Firemin, and J. P. Newhouse Medicare Beneficiaries Knowledge of Part D Prescription Drug Program Benefits and Responses to Drug Costs. Journal of the American Medical Association 299(16): Keeler, E. B., and J. E. Rolph The Demand for Episodes of Treatment in the Health Insurance Experiment. Journal of Health Economics 7(4): Lamers, L. M Pharmacy Cost Groups: A Risk Adjuster for Capitation Payments Based on the Use of Prescribed Drugs. Medical Care 37(8): Manning, W. G., and J. Mullahy Estimating Log Models: To Transform or Not to Transform? Journal of Health Economics 20(4): Newhouse, J Free for All? Lessons from the RAND Experiment. Cambridge, Mass.: Harvard University Press. Raebel, M. A., T. Delate, J. L. Ellis, and E. A. Bayliss Effects of Reaching the Drug Benefit Threshold on Medicare Members Healthcare Utilization During the First Year of Medicare Part D. Medical Care 46(10): Roblin, D. W., and M. L. Maciejewski Repeat Experience with the Doughnut Hole in Medicare Part D. When the Doughnut Hole Becomes a Tunnel. Medical Care 49(5): Shrank, W. H., and N. K. Choudhry Time to Fill the Doughnuts Health Care Reform and Medicare Part D. New England Journal of Medicine 364(7): Van Kleef, R. C., W. P. M. M. van de Ven, and R. C. J. A. van Vliet Shifted Deductibles for High Risks: More Effective in Reducing Moral Hazard than Traditional Deductibles. Journal of Health Economics 28(1): Van de Ven, W. P. M. M., and R. P. Ellis Risk Adjustment in Competitive Health Insurance Markets. In Handbook of Health Economics, A. J. Culyer and J. P. Newhouse, eds. Amsterdam: Elsevier. Van de Ven, W. P. M. M., R. C. J. A. van Vliet, and L. M. Lamers Health-Adjusted Premium Subsidies in The Netherlands. Health Affairs 27(3): Van Vliet, R. C. J. A., and R. C. van Kleef Herijkingsonderzoek vereveningsmodel (Re-estimation of the Risk Equalization Model of 2011.) Research report. Rotterdam, The Netherlands: Erasmus University Rotterdam. Zhang, Y., J. M. Donohue, J. P. Newhouse, and J. R. Lave The Effects of the Coverage Gap on Drug Spending: A Closer Look at Medicare Part D. Health Affairs Web Exclusive February 9 28(2):w317 w