The Medicare Shared Savings Program. Avoiding Unintended Incentives In ACO Payment Models. Accountable Care Organizations

Transcription

1 By Rudy Douven, Thomas G. McGuire, and J. Michael McWilliams Avoiding Unintended Incentives In ACO Payment Models Rules For Setting ACO Benchmarks The ACO is a new model of health care delivery and payment in Medicare. 2 The Medicare ACO programs (the Shared Savings Program and the Pioneer model) rely solely on provider groups willingness to participate. This is in contrast to the Medicare Advantage program, in which beneficiaries must decide whether or not to enroll. As of early 2014 over 360 provider organizadoi: /hlthaff HEALTH AFFAIRS 34, NO. 1 (2015): Project HOPE The People-to-People Health Foundation, Inc. ABSTRACT One goal of the Medicare Shared Savings Program for accountable care organizations (ACOs) is to reduce Medicare spending for ACOs patients relative to the organizations spending history. However, we found that current rules for setting ACO spending targets (or benchmarks) diminish ACOs incentives to generate savings and may even encourage higher instead of lower Medicare spending. Spending in the three years before ACOs enter or renew a contract is weighted unequally in the benchmark calculation, with a high weight of 0.6 given to the year just before a new contract starts. Thus, ACOs have incentives to increase spending in that year to inflate their benchmark for future years and thereby make it easier to obtain shared savings from Medicare in the new contract period. We suggest strategies to improve incentives for ACOs, including changes to the weights used to determine benchmarks and new payment models that base an ACO s spending target not only on its own past performance but also on the performance of other ACOs or Medicare providers. Rudy Douven (R.Douven@ cpb.nl) is a health economist at the CPB Netherlands Bureau for Economic Policy Analysis, in The Hague, and at Erasmus University Rotterdam,bothinthe Netherlands. Thomas G. McGuire is a professor of health economics at Harvard Medical School, in Boston, Massachusetts. J. Michael McWilliams is an associate professor of health care policy and medicine at Harvard Medical School and Brigham and Women s Hospital, in Boston. The Medicare Shared Savings Program is intended to reduce financial rewards for higher spending in fee-for-service Medicare by providing incentives for accountable care organizations (ACOs) to reduce spending below targets set by Medicare. By basing an ACO s spending target for each three-year contract period on its own spending history in the previous three-year period, however, the program mixes some savings incentives with incentives to increase spending in some years in relation to fee-for-service Medicare. Indeed, if current rules remain in effect, we estimate that for every dollar increase in spending in the last year before an ACO starts a new three-year contract, the ACO will get back between $1.48 and $1.90 during the contract period, depending on whether it uses a one- or two-sided shared savings payment model options that this article explores in depth. The Centers for Medicare and Medicaid Services (CMS) is considering what rules will apply for the next phase of ACO payment systems in Medicare. 1 We contribute to this discussion by explaining key drawbacks of the current Medicare Shared Savings Program rules for setting ACO benchmarks and offering suggestions for potential regulatory changes or new payment models that could improve incentives for ACOs to achieve savings. January :1 Health Affairs 143

2 tions had contracted with Medicare as ACOs in the Pioneer program or the Shared Savings Program. Of these, 117 organizations started in the Shared Savings Program in In this program CMS sets a spending target for an ACO, referred to as the ACO s benchmark. In this article we concentrate on the Shared Savings Program the much larger of the two ACO programs and its rules for setting and resetting the ACO benchmark. A group of providers that decides to become an ACO can contract with Medicare for a three-year period. CMS sets one benchmark to apply throughout the entire period. The benchmark is a weighted average of spending for attributed beneficiaries over a baseline three-year period before the start of the contract. For example, for an ACO that started on January 1, 2013, spending for beneficiaries served by the ACO in the most recent year (2012) receives the highest weight, 0.6, and spending in 2011 and 2010 receives weights of 0.3 and 0.1, respectively. The declining weights are based on the idea that more recent spending is more predictive of current spending for ACOs. 4 The benchmark is then adjusted annually by a national inflation factor to establish the spending target in each contract year. It is also adjusted for yearto-year changes in the case-mix of patients served by the ACO. After each three-year contract, an ACO can opt to renew its contract with Medicare for another three-year period. CMS has not yet announced the formula for setting benchmark payments for the second round of ACO contracts, which are to begin in 2015 for ACOs that started in One- And Two-Sided Models Provider groups participating in the Medicare Shared Savings Program as an ACO have a choice of two risksharing arrangements. In both, an ACO s performance with respect to spending, measured by total per beneficiary Medicare fee-for-service spending for all services in a performance year, is compared in each contract year to the ACO s benchmark. Provider groups can choose either a shared savings and losses model (the two-sided model) or a shared savings model (the onesided model). In the two-sided model, ACOs face upside risks (that is, they receive bonuses if their spending is below spending benchmarks) and downside risks (that is, they incur penalties if their spending exceeds spending benchmarks). The maximum bonus or penalty is 60 percent of the ACO s cost savings or overruns in two-sided contracts. In the one-sided model, ACOs face upside risks only. That is, if an ACO s spending performance is sufficiently below the benchmark, CMS pays the ACO up to 50 percent of the spending difference. 3 The lack of downside risk implies that in each performance year an ACO will receive the traditional Medicare fee-for-service payments for services rendered even if it exceeds its benchmark. This may explain why 98 percent of the ACOs in the Medicare Shared Savings Program have opted for the one-sided model. 5 ACO program participation may be more attractive for provider organizations with historically high per beneficiary Medicare spending than for organizations with lower spending. This is because ACOs with high spending receive a higher benchmark and therefore may be able to generate greater shared savings, if the marginal cost of constraining use is lower when initial spending is high or, put another way, if they have more fat to trim. 6 Thus, there is an inherent trade-off between strengthening ACO incentives to generate savings and encouraging organizations, particularly inefficient ones, to participate. Unintended Incentives At first blush, it may seem that the Medicare Shared Savings Program can only reduce spending compared to the traditional Medicare fee-for-service program. If ACOs can achieve savings, there appears to be a potential win-win situation, with the gains from lower Medicare spending shared between the ACO and CMS. However, we show that unintended effects of the current model diminish incentives for ACOs to generate savings and may result in higher instead of lower Medicare fee-for-service spending. Since the most recent year of spending carries the greatest weight (0.6) in benchmark calculations, there are perverse incentives to increase spending in that particular year. Because current rules suggest that ACO benchmarks will be periodically rebased by CMS when a new ACO contract starts, care decisions in each contract period may affect the benchmark for the next contract period. The weight of 0.6 that CMS uses is so high that ACOs with high Medicare spending patterns in the year before the start of a new contract are rewarded by an elevated benchmark (instead of being penalized), which yields shared savings from CMS during the new contract period even if the ACOs have not reduced spending. Similarly, providers that achieve savings in the year before they enter or renew an ACO contract are penalized by a lower benchmark (instead of being rewarded) in the next contract period. The current benchmark rules thus increase incentives to encourage use under the fee-for-service payment system in traditional Medicare, instead of reducing those incentives. Exhibit 1 provides examples of how the current benchmark rules create weak incentives for 144 Health Affairs January :1

3 Exhibit 1 Marginal Revenues And Losses For At An Accountable Care Organization (ACO) From Performing One More Or One Less Hip Replacement, Using Weighted Spending For ACO s Cumulative annual base First contract period Second contract period marginal revenue, Payment model spending Benchmark One hip replacement more Fee-for-service a a a $10,000 a a a a $10,000 One-sided risk (share 50% savings) $1,000,000 b b 10,000 $1,006,000 $3,000 $3,000 $3,000 19,000 Two-sided risk (share 60% savings and losses) 1,000,000 b b 4,000 1,006,000 3,600 3,600 3,600 14,800 One hip replacement less Fee-for-service a a a 10,000 a a a a 10,000 Two-sided risk (share 60% savings and losses) One less replacement in ,000,000 b b 4, ,000 3,600 3,600 3,600 14,800 One less replacement in ,000,000 $4,000 b b 999, ,800 SOURCE Authors analysis. NOTES For the years and we show the marginal revenue for an ACO of performing one hip replacement more or less compared to annual base spending. We apply the annual weights from the Centers for Medicare and Medicaid Services of 0.1, 0.3, and 0.6 to the years 2013, 2014, and 2015, respectively, to produce the benchmark, or annual spending target. a Not applicable. b Cells are left blank because ACOs do not deviate from annual base spending of $1,000,000. ACOs to achieve savings and how the 0.6, 0.3, and 0.1 weighting scheme is particularly problematic. In all of the examples, we assumed that a provider group has an ACO contract for and renews it for For simplicity, we assumed that the ACO s annual base spending is $1,000,000 in every example and that the ACO faces a benchmark of $1,000,000 during the first contract period, We considered the total financial consequences during the first and second contract periods of achieving a spending level in 2015 that was $10,000 above or below the benchmark in the first period. To obtain the benchmark for the second contract period, , we assumed that CMS set the benchmark for that period in the same way that it currently does for new ACOs. Specifically, we assumed that CMS multiplied spending for ACOs in the years by the declining weights of 0.6 for 2015, 0.3 for 2014, and 0.1 for One could think of the incremental increase or decrease of $10,000 in the first contract period as a discretionary decision to recommend (or not recommend) a hip replacement. In traditional fee-for-service Medicare, with no ACO, the provider group receives a fee-for-service payment for a hip replacement of $10,000 from Medicare (Exhibit 1). That amount is also the total cumulative marginal revenue for the provider group. ONE-SIDED MODEL: If the provider group uses a one-sided ACO payment model for the first and second contract periods, the financial consequences of providing the extra hip replacement in 2015 can be split up into four components. First, as in the example above, CMS makes a fee-for-service payment of $10,000 to the ACO for replacing the hip (Exhibit 1). Second, the ACO may share losses with CMS if the fee-for-service payment for the hip replacement raises spending above the ACO s first benchmark of $1,000,000. However, in a onesided ACO model there are no shared losses; therefore, in this case, the ACO is not penalized for performing the hip replacement. Thus, in 2015 the marginal revenue for the ACO is $10,000. Third, the hip replacement raises the benchmark for the second contract by $6,000 (using the CMS benchmark weight of 0.6 for the additional 2015 spending of $10,000). The higher benchmark of $1,006,000 will make it easier for the ACO to obtain shared savings in the second contract period. Indeed, fourth and finally, for performing this hip replacement in 2015 the ACO will be rewarded with shared savings for three consecutive years, from 2016 through 2018, if it simply reverts to the spending level it had before entering the second contract. The ACO s cumulative shared savings over the second contract period January :1 Health Affairs 145

4 in this scenario is $9,000: The shared savings percentage for one-sided models is 50 percent, so the ACO receives $3,000 (half of $6,000) for three years. In total, the ACO receives $19,000 from CMS to replace the hip. Put another way, for every $1 spent in the last year of its first three-year contract, the ACO is rewarded with $1.90 over the course of the second contract. 7 Notably, these effects for ACOs that are renewing contracts are similar for provider groups entering new ACO contracts. In sum, ACOs are paid very generously for any service provided in the year before entering or renewing an ACO contract. TWO-SIDED MODEL: Similar adverse incentives exist for two-sided models in the first ACO contract period. The cumulative rewards for performing the hip replacement are smaller because, in this example, there are shared losses as a result of performing the hip replacement in 2015, during the first period. In a two-sided model the hip replacement of $10,000 is a loss in relation to the first ACO benchmark. Thus, the ACO incurs a loss of $6,000 and is paid only $4,000 in 2015 (Exhibit 1). The cumulative shared savings effect in the second contract period, however, is larger because of the higher shared savings percentage: Instead of sharing 50 percent of savings, the ACO shares 60 percent, so it receives $10,800 ($3,600 for each of three years) instead of $9,000. In this example, the ACO receives $14,800 from CMS to replace the hip. Or, for every dollar spent in the last year of its first three-year contract, the ACO using the two-sided model receives $1.48 from CMS in the next contracting period. These are exactly the wrong incentives. The decision to replace the hip turns out to be very profitable for the ACO and very expensive for CMS. Perverse incentives also apply to any potential savings. A decision not to perform a hip replacement in 2015 (the first two-sided risk example in Exhibit 1) turns out to be very profitable for CMS and very expensive for an ACO, costing it $14,800. However, unequal weights imply that not replacing a hip could be financially attractive for the ACO in 2013, if the total costs for replacing the hip were higher than $5,800 (the second two-sided risk example in Exhibit 1). Existing ACOs or new provider groups entering their first ACO contract could gain from responding to these perverse incentives by inducing use in the year before starting a new contract. Because marginal profits are higher than feefor-service profits in the last year before a new contract, there is an incentive to treat more, not less. 8 Agenuine shared savings plan for ACOs should reward both CMS and ACOs when savings are realized. One way this might happen would be for providers to shift discretionary care to the last year before a new contract starts for example, by moving a procedure from December of one year to January of the next. Or providers may put programs to manage utilization on hold until a new ACO contract starts. Additional Features Our analysis implies that there is a testable hypothesis: Provider groups will exhibit unusually high spending patterns in the year before they enter a new ACO contract. If this happens, what are measured as savings during the new contract period may be the result not of lower costs during that contract but of intentionally high spending patterns in the last year before the contract started. Such responses by provider groups may be unlikely, however, because they run counter in purpose to the investments in care management and care coordination infrastructure that provider groups must make to be eligible for participation in the Medicare Shared Savings Program, and because systems implemented to generate savings may be difficult to turn off once started. Our calculations are based on simplified examples that do not account for many of the complexities in the rules for setting ACO benchmarks and shared savings (or loss) rates. For example, the risk-sharing rate depends on an ACO s performance on quality measures, and CMS updates benchmarks annually based on national spending growth and the complexity of the ACO s patient population. Furthermore, CMS sets minimum savings and loss thresholds outside of which ACOs are eligible to share savings or incur losses. These additional features of the payment model affect ACO incentives and payments, in some cases mitigating and in some cases exacerbating the problems we identify. However, they do not alter the main conclusions we draw namely, that perverse incentives exist that may distort ACOs population-level care performance deci- 146 Health Affairs January :1

5 sions, contrary to the intended goals for which ACOs were designed. Strategies For Improving ACO Incentives Modifying The Benchmark Weights We consider two alternatives to address the perverse incentives created by the unequal weighting of an ACO s previous years of spending when setting its future spending benchmark. A straightforward option is for CMS to change the weight distribution to equal weights across the three preceding years used to calculate the benchmark. To demonstrate the effect of this alternative, Exhibit 2 displays the calculations for all of the examples in Exhibit 1, but with an equal weight distribution (one-third for each of the three years). Equal weights would reduce the incentives to increase spending in the last year before a new contract: The incremental reward for performing an additional hip replacement (relative to a fee-for-service arrangement) in the year before a new contract is reduced substantially in the one-sided model (generating an additional $5,000 in revenue instead of the $9,000 in Exhibit 1). Equal weights would also make incentives consistent across the years of a contract period, as demonstrated by the fact that not performing a hip replacement in the twosided model produces the same cumulative revenue in 2013 as it does in 2015 (Exhibit 2). More important, equal weights would entirely negate the unintended incentive for ACOs in a two-sided model to increase spending. This is demonstrated by the two-sided risk examples in Exhibit 2. In all three of these examples. the sum of the shared losses and savings in both contract periods is equal to a fee-for-service arrangement of replacing one hip more or less. With equal weights, incentives are essentially equivalent to fee-for-service payments with a lag. When any other weight distribution is used, spending in at least one year is rewarded more than dollar-for-dollar in the subsequent contract period, which thus creates perverse incentives. It is Important to stress here that unintended effects will always be present in one-sided models and whenever a provider group enters a oneor two-sided ACO model for the first time. Because there are no shared losses in the years before the group enters the new contract, perverse effects remain for any given weight distribution. To further strengthen incentives for ACOs to generate shared savings in a given contract period, CMS could use more years before the new contract starts, to calculate the benchmark. For example, CMS could use equal benchmark weights of one-fifth for each of the five years preceding the new contract. Exhibit 3 displays the calculations for all of the examples in Exhibits 1 and 2, but with an equal weight distribution for five years. Under this weighting scheme, an ACO in a twosided risk contract would lose $2,400 less in revenue from not performing a hip replacement, relative to a fee-for-service arrangement (a loss Exhibit 2 Marginal Revenues And Losses For At An Accountable Care Organization (ACO) From Performing One More Or One Less Hip Replacement, Using Equal Weights For Spending ACO s Cumulative annual base First contract period Second contract period marginal revenue, spending Benchmark One hip replacement more Fee-for-service a a a $10,000 a a a a $10,000 One-sided risk (share 50% savings) $1,000,000 b b 10,000 $1,003,333 $1,667 $1,667 $1,667 15,000 and losses) 1,000,000 b b 4,000 1,003,333 2,000 2,000 2,000 10,000 One hip replacement less Fee-for-service a a a 10,000 a a a a 10,000 and losses) One less replacement in ,000,000 b b 4, ,667 2,000 2,000 2,000 10,000 One less replacement in ,000,000 $4,000 b b 996,667 2,000 2,000 2,000 10,000 SOURCE Authors analysis. NOTES For the years and we show the marginal revenue for an ACO of performing one hip replacement more or less compared to annual base spending. We apply the annual weight of one-third to each of the years to produce the benchmark, or annual spending target. a Not applicable. b Cells are left blank because ACOs do not deviate from annual base spending of $1,000,000. January :1 Health Affairs 147

6 Exhibit 3 Marginal Revenues And Losses For At An Accountable Care Organization (ACO) From Performing One More Or One Less Hip Replacement, Using Equal Weights For ACO s Cumulative annual base First contract period Second contract period marginal revenue, spending Benchmark One hip replacement more Fee-for-service a a a $10,000 a a a a $10,000 One-sided risk (share 50% savings) $1,000,000 b b 10,000 $1,002,000 $1,000 $1,000 $1,000 13,000 and losses) 1,000,000 b b 4,000 1,002,000 1,200 1,200 1,200 7,600 One hip replacement less Fee-for-service a a a 10,000 a a a a 10,000 and losses) One less replacement in ,000,000 b b 4, ,000 1,200 1,200 1,200 7,600 One less replacement in ,000,000 $4,000 b b 998,000 1,200 1,200 1,200 7,600 SOURCE Authors analysis. NOTES For the years and we show the marginal revenue for an ACO of performing one hip replacement more or less compared to annual base spending. We apply the annual weight of one-fifth to each of the five years to produce the benchmark, or annual spending target. a Not applicable. b Blank cells indicate that ACOs do not deviate from annual base spending of $1,000,000. of $7,600 instead of $10,000; Exhibits 2 and 3). This would be a true shared savings. In general, this alternative weighting scheme would penalize cost savings less heavily by slowing the decline of the benchmark for ACOs that became relatively efficient in the previous period. That would encourage the ACOs to participate in the second period. Although it might require a change in the law, modifying the benchmark weights could be a relatively easy way for CMS, in the short run, to remove the unintended incentives. Similarly, if it fell within CMS s regulatory purview, the agency could defer the rebasing of ACO benchmarks in the short term, with plans to rebase or otherwise reform benchmarks at an unspecified time in the future. This would strengthen incentives for ACOs to participate in a new contract because savings that were realized in the previous contract period would not be penalized by a lower benchmark level in the subsequent period. Yardstick Competition In the longer run, a more fundamental change to improve incentives for ACOs would be to introduce a form of yardstick competition. 9 The idea would be to base an ACO s benchmark not only on the organization s own past performance but also on the performance of other Medicare providers. This general approach is widely used in the Medicare program. Its advantage in this case is that an ACO would have stronger incentives to achieve and maintain greater efficiency because it would continue to be rewarded for spending below (or penalized for spending above) the benchmark, even if an ACO was unable to improve on its own past performance. Thus, already efficient providers would have an incentive to enter and remain in the Medicare Shared Savings Program. 6 A potentially major disadvantage of this approach is that inefficient providers would anticipate receiving a lower benchmark (at least, lower than would be the case with the current ACO models). That more challenging spending target might make such providers less likely to participate in the Medicare ACO programs. Conversely, efficient providers would receive higher benchmarks than under the current model and could thus be rewarded for simply maintaining the status quo. This potential problem could be addressed by grouping similar ACOs or non-aco providers together. The benchmarks could then be determined by giving equal weights to the past performance of providers in the peer group. The impact of an ACO s past performance on the benchmark would be diminished in proportion to the number of providers in the peer group. This approach would also introduce competition into the payment model because the ACO benchmark would depend on the performance of peer groups. An ACO that was more efficient than the average in its peer group would receive a higher benchmark than in the current ACO model, which would make it easier for the efficient ACO to realize savings. The more an ACO lowered its spending relative to that of its com- 148 Health Affairs January :1

7 petitors, the greater its cumulative rewards over multiple contract periods. Basing payments on cost performances on peer groups has worked well in Medicaid payments to psychiatric hospitals and psychiatric units in New Hampshire, accommodating systematic differences in casemix while maintaining incentives for cost-effective care. 10 Alternatively, an ACO s benchmark under the current rules could be blended with a local benchmark, such as the traditional fee-forservice Medicare spending average in the ACO s market or markets. To avoid discouraging inefficient organizations from participating in the ACO programs and providing unwarranted rewards for already efficient organizations, a blended payment model could be offered to ACOs that have demonstrated improvement during one or more contract periods in the Pioneer or Medicare Shared Savings Programs. That is, the current model could serve as a bridge to a model with stronger incentives to generate and sustain savings. Because an ACO s patients may differ significantly from beneficiaries served by other providers in its market, a blended payment would need to be risk-adjusted to minimize unfair penalties for ACOs that serve sicker patients and undeserved rewards for ACOs that serve healthier patients. Because many features of the Medicare Shared Savings Program have been set by the Affordable Care Act, incorporating yardstick competition into ACO benchmarks may require the development of new ACO models by the Center for Medicare and Medicaid Innovation. Conclusion A genuine shared savings plan for ACOs should reward both CMS and ACOs when savings are realized. We have shown in this article that the current model for setting benchmarks does not meet this criterion because of the perverse incentives built into its design. We propose several policy options that CMS could use to improve the Medicare Shared Savings Program, including balancing weighted years and introducing yardstick competition. These policies could help produce genuine shared savings as ACOs become an increasingly common payment model. When this article was written, Rudy Douven was a Dutch Harkness Fellow at Harvard Medical School and was supported by the Commonwealth Fund. Theviewspresentedherearethoseof the authors and not necessarily those of the Commonwealth Fund, its directors, officers, or staff. Thomas McGuire s research for this article was supported by the National Institute of Aging (Grant No. P01 AG032952). J. Michael McWilliams s workonthisarticlewas supported by a grant from the Laura and John Arnold Foundation. The views presented here are those of the author and not necessarily those of the Laura and John Arnold Foundation, its directors, officers, or staff. Douven thanks the Commonwealth Fund for providing an excellent research environment at Harvard Medical School. NOTES 1 Center for Medicare and Medicaid Innovation. Request for information: evolution of ACO initiatives at CMS [Internet]. Baltimore (MD): CMMI; [cited 2014 Nov 13]. Available from: 2 Centers for Medicare and Medicaid Services. Medicare Shared Savings Program: shared savings and losses and assignment methodology: specifications, version 2 [Internet]. Baltimore (MD): CMS; 2013 Apr [cited 2014 Nov 13]. Available from: Medicare-Fee-for-Service-Payment/ sharedsavingsprogram/downloads/ Shared-Savings-Losses-Assignment- Spec-v2.pdf 3 HHS.gov [Internet]. Washington (DC): Department of Health and Human Services. Press release, More doctors, hospitals partner to coordinate care for people with Medicare; 2013 Jan 10 [cited 2014 Nov 13]. Available from: pres/01/ a.html 4 Centers for Medicare and Medicaid Services. Medicare program; Medicare Shared Savings Program: accountable care organizations. Final rule. Fed Regist. 2011;76(2): Centers for Medicare and Medicaid Services. Fast facts all Medicare Shared Savings Program ACOs [Internet]. Baltimore (MD): CMS: 2014 May [cited 2014 Nov 13]. Available from: Medicare/Medicare-Fee-for-Service- Payment/sharedsavingsprogram/ Downloads/All-Starts-MSSP- ACO.pdf 6 McWilliams JM. ACO payment models and the path to accountability. J Gen Intern Med. 2014; 29(10): Note that marginal profits may rise to enormous proportions compared to profits in a fee-for-service environment. For example, if the cost of the hip replacement, including the new hip, is $8,000, then the ACO s profits are $11,000 ($19,000 minus $8,000). That is 5.5 times as much as the $2,000 profit in a fee-for-service environment (CMS s reimbursement of $10,000 minus the ACO s cost of $8,000). 8 McGuire TG. Physician agency. In: Culyer AJ, Newhouse JP, editors. Handbook of health economics. Amsterdam: Elsevier; p. 1: Shleifer A. A theory of yardstick competition. Rand J Econ. 1985; 16(3): McGuire TG, Hodgkin D, Shumway D. Managing Medicaid mental health costs: the case of New Hampshire. Adm Policy Ment Health. 1995;23(2): January :1 Health Affairs 149