 Research Insights Paying for Value: Progress and Obstacles Introduction Public and private policymakers are increasingly committed to us- Early “pay-for-performance” (P4P) efforts focused mainly on clinical quality and patient experience. Initially, different payers tended to ing provider payment as a tool to improve clinical quality, patients’ select their own priorities and metrics, attenuating the impact of the experience of care, and efficient use of resources. Since the turn of efforts on providers and making it difficult to assess their effective- the millennium, significant gains have been made in measuring and ness. The quality of physician care was particularly difficult to gauge incentivizing improvements in quality and patient experience. More because small sample sizes and heterogeneous patient populations recently, payers have also stepped up efforts to collect information thwarted statistically reliable evaluations. about costs and resource use and to devise payment strategies to encourage efficiency. The concept of value has emerged as a single Over time, however, payers have gravitated to core measures of metric that joins the dimensions of quality and efficiency and is process and outcomes such as hospital readmissions and clini- now a primary focus of payment policy in Medicare, Medicaid, and cal markers for high-prevalence conditions such as diabetes and private insurance. hypertension. In many cases, modest improvements in quality have resulted. A path-breaking development in physician quality assess- Background ment occurred in February 2016, when the Centers for Medicare & In January 2015, U.S. Health and Human Services Secretary Sylvia Medicaid Services and the private insurance trade group America’s Burwell announced that her agency’s goal was that, by 2016, 85 per- Health Insurance Plans announced consensus on the use of a core set cent of Medicare fee-for-service payments would be “tied to quality of physician quality measures. or value,” increasing to 90 percent by 2017. Burwell stated further that, by 2016, “alternative payment models” would be used for 30 For the most part, however, programs that combine quality measures percent of these value-based payments, increasing to 50 percent by and incentives for cost control and prudent resource use—paying for 2018. Her forecast reflected an expanding range of value-based pur- value—have not been in operation long enough to measure their ef- chasing programs for hospitals in Medicare called for in the Afford- fectiveness. Experimentation is increasing through approaches such able Care Act (ACA) and a comprehensive set of physician payment as shared savings and bundled payment, but many questions remain measures called for in the Medicare Access and CHIP Reauthoriza- about how best to define and measure value. tion Act of 2015 (MACRA). Genesis of this Brief: This brief is based on a meeting of federal policymakers and researchers that took place in Washington, D.C., on December 8, 2015. AcademyHealth convened the meeting as part of its Research Insights Project. Funding for the conference was made possible by Grant No. 5R13HS018888-06 from U.S. Agency for Healthcare Research and Quality (AHRQ). The views expressed in written conference materials or publications and by speakers and moderators do not necessarily reflect the official policies of the U.S. Department of Health and Human Services; nor does mention of trade names, commercial practices, or organizations imply endorsement by the U.S. Government. The Research Insights Project convenes invitational meetings, holds webinars, and produces reports and issue briefs to foster discussion of existing, relevant research evidence among policy audiences that need to implement health reform and develop new policy. Additional information and publications may be 1 1 found on the project’s website at http://www.academyhealth.org/programs/ProgramsDetail.cfm?ItemNumber=6752&navItemNumber=6787. Paying for Value: Progress and Obstacles In view of the many operational and implementation challenges payment models. As noted, past efforts to measure quality in smaller implicit in Burwell’s pledge, AcademyHealth, with the support of the practices have been persistently stymied by the problem of small Agency for Healthcare Research and Quality, convened an invitation- patient population samples that do not lend themselves to evaluation al meeting of researchers and agency officials, held in Washington, with statistical reliability. D.C., in December 2015, to discuss the state-of-the-art in ongoing value-based payment and delivery system initiatives and to identify These efforts also must address the challenge of adequate risk ad- high-priority research objectives to guide future rounds of experi- justment. Most existing risk-adjustment systems fall short of fully mentation. explaining patient health status and thus the likelihood of success for any course of treatment. “But my patients are sicker,” physicians often Observation and comment at the meeting were not for attribution; assert when faced with subpar performance evaluations. Much work therefore, this Research Insights brief summarizes and paraphrases remains to be done to develop measures and the necessary data to the discussion and references relevant background from the health account appropriately for differences in the patients treated and the services literature and other sources. It roughly follows the meet- circumstances that determine treatment costs and outcomes. ing agenda, first considering selected examples of clinical quality measurement problems and payment issues associated with them; Finally, the ambitious goal of measuring provider performance by reviewing the development of patient experience measures; then observing patient outcomes is attenuated in a fragmented health looking more broadly at payment and incentive issues in general; system in which patients may see several clinicians, making it impos- and, finally, reporting on discussion at the meeting of unsettled ques- sible to attribute responsibility for ultimate results. Further, outcomes tions about payment for value, including definitions and goals as well may take years to materialize as patients frequently cycle from one as outstanding research needs. plan or provider to another. In addition, the process measures most widely used in the past decade’s P4P programs have been criticized for capturing only narrow sets of clinical indicators, at the risk of A New Physician Payment Environment—and Its inducing providers to focus their quality improvement efforts on Challenges what is easiest to measure—“teaching to the test”—at the expense of Although overshadowed by resolution of Medicare’s sustainable potentially more significant dimensions of care.2 growth rate problem, MACRA instituted a sweeping array of new performance requirements for physicians, reflecting what one E Pluribus meeting participant described as “a really radical shift” in physician The variety of responses to the above challenges has been as kalei- payment policy, albeit still on a platform of traditional fee-for-service doscopic as the health system itself. The AcademyHealth meeting reimbursement in most cases. The foundation was laid by previous participants considered examples that are poles apart in terms of legislation establishing a physician quality reporting system, first with their setting, scale, and sophistication but alike in their focus on small bonuses for compliance and then, as of 2015, with penalties for identifying shortcomings in quality and value and devising pathways failure to report. to improvement. Participants pointed to optimistic developments such as a group of 42 primary care clinics in New York City serving The ACA added a “value-based payment modifier” for large physician large Medicaid populations and, at the other end of the spectrum, an groups and charged the HHS secretary with developing such a system intensive care unit (ICU) in an elite urban teaching hospital. for all Medicare fee-for-service physicians by 2017. Accordingly, MA- CRA created a multifaceted “merit-based incentive payment system,” An innovative approach to risk adjustment showed promise in a including implementation targets for alternative payment models, to be small-practice program in New York in 2009–2010. The Primary effective January 1, 2019. The new system is intended to “consolidate Care Information Project (PCIP) furnished electronic health record and replace several existing incentive programs” and to encourage de- systems to support a tightly targeted P4P intervention to improve the velopment and participation in new “alternative payment models” such quality of antithrombotic prescribing, blood pressure and cholesterol as shared savings and bundled payment.1 control, and smoking cessation efforts. Clinics received incentive payments for each patient who showed measurable improvement in Many large physician groups and organized delivery systems are al- intermediate outcomes. ready well equipped to meet the new requirements. But, according to the latest available National Ambulatory Medical Care Survey (2012), After consulting with participating physicians, however, the PCIP nearly 70 percent of outpatient visits occur at practices of five or adjusted payments upward for patients who, according to three fewer physicians, which often lack the resources for full information simple indicators, were considered more difficult to treat successfully: technology adoption, care coordination, or capital to assume finan- whether they had comorbid conditions or whether they were unin- cial risk for their patients’ care costs as required by many alternative sured or on Medicaid, the latter as proxies for socioeconomic status, 2 Paying for Value: Progress and Obstacles a significant health risk factor in the physicians’ view. Improvements This openness to big data has direct applications to the ability to on all three indicators were better in participating clinics than in a measure and reward performance, especially when linked with control group and compared favorably with larger practices that set novel analytic techniques. Creating incentives to stimulate more similar goals.3 hospitals to evolve their ICU care as described above requires the ability to measure improvements in outcomes of care and preferably Physicians’ receptivity and responsiveness to incentive programs the cost of making those measurements would be low. However, may be diminished by distrust of formulaic risk-adjustment meth- given that ICUs vary in their patient populations, intensivists would odologies and concerns about whether practices will be penalized want outcome measures to be risk-adjusted and may expect all the for difficult-to-treat patients. Physicians worry that the expense of clinical data they have—from vital signs and blood tests to seizure practice change will not be covered by the incentives they hope to monitoring with electroencephalograms (EEG)—to be used to earn but can’t be sure of. Analysts’ takeaway from the PCIP experi- capture severity of illness. Recent research shows that big-data ap- ment demonstrated that physicians’ comfort level with and buy-in proaches, such as recording EEG data continuously, can accurately to program design can be a game-changer.4 The design spoke to identify patients at risk.7 In addition, improvements in natural lan- their concerns. Nevertheless, even after generous subsidies made guage processing (NLP) have led to the ability to identify risk factors available by the American Recovery and Reinvestment Act and the accurately from clinical text.8 Combining a big-data philosophy of outcome of programs such as the PCIP, some observers remain embracing the richness of the ICU data environment with innova- pessimistic about the pace of IT adoption—upon which efficient tions such as NLP offers new opportunities to adjust for risk more and effective quality measurement and risk adjustment may ulti- accurately so that clinicians are more confident that “my patients are mately depend.5 sicker, and the outcome measures adjust for that.”9 Big Data Even outside the ICU, in a comprehensive, IT-enabled system of Quality improvement in the ICU of an elite teaching hospital may clinical quality improvement, extensive data may now be captured also depend critically on human factors, but the process for identify- and organized on many dimensions of care without imposing ing improvement targets and forging interventions will look very a large collection burden on clinical staff. Many of these dimen- different from the small-practice environment. “Big data” has been sions could be adapted in quantitative detail for a P4P or global the watchword in a joint project of four academic health centers to payment program. harness large clinical and administrative databases that can pinpoint associations between environmental conditions in an ICU and the Patient Experience risk of patient harms. Real-time alerts on conditions associated with The overwhelming diversity of domains in clinical quality mea- harm—staff overload is prominent among them—in turn become a surement contribute strongly to the appeal of the more integra- management tool to mitigate risks on the unit floor. Workflows and tive results found in patient surveys on the experience of care and staff training are redesigned to emphasize prompt response to the self-reported functional outcomes—hallmarks of an era of patient- messages the data delivers. centered care. The Consumer Assessment of Healthcare Provid- ers and Systems (CAHPS) suite of patient experience surveys has Big data is not a new concept at Beth Israel Deaconess Medical rapidly gained widespread adoption in clinical management and Center (BIDMC), which uses granular clinical and demographic value payment systems, with Medicare a notable trend-setter. It is an data on its patients to manage hospital operations and business important tool for value-based purchasing in private insurance, and concerns. BIDMC, however, has still taken two years to organize the CAHPS survey results are increasingly available to consumers in available data flows into structured tools for the recent ICU safety public reporting programs. initiative. However, since heightening focus on preventable harms, the 650-bed BIDMC has reduced the number of harm events in its In 20 years of use, CAHPS has earned acceptance as a reliable and ICU from as high as 90 in the second half of 2008 to fewer than 25 valid source of information on questions for which patients are per six-month period throughout 2013 and 2014.6 a preferred source, particularly with respect to access to care and information, communication skills of clinicians, responsiveness of Nursing workload intensity was found to be the biggest driver of care providers, and receiving answers to medical questions. CAHPS elevated risk at BIDMC. Pertinently, in a sister big-data program results are often correlated with measures of other dimensions of jointly conducted by Johns Hopkins University and the University quality and are frequently a component of P4P systems.10 of California San Francisco, the average time needed for unit staff to conduct a daily patient harms assessment fell from 14 to 6 minutes As with first-generation process measures of clinical care, limita- in the Emerge program, compared to a legacy EHR-patient order tions remain. Owing to expanding regulatory and reporting require- entry system. ments, the cost and administrative burden of administering several 3 Paying for Value: Progress and Obstacles versions of CAHPS has grown, along with the response burden on Value-based purchasing programs in the ACA and MACRA repre- patients and families. An overall response rate of 35 percent is now sent an extensive effort to factor quality and cost together. Private fairly typical, providing adequate but less-than-optimal reliability. insurers, too, have experimented with tiered networks based on Some physicians continue to be skeptical about the validity of pa- performance efficiency as part of “a range of increasingly sophis- tient ratings and worry that grievance-bearing patients will poison ticated approaches to combine indicators of cost and quality,” the their scores, although, in practice, complaints are rare and, in any NQF report says. Relentless increases in cost-sharing have also case, better captured by grievance procedures. Small practices made even insured consumers much more sensitive to cost issues struggle to manage collection. While essential for creating com- than in the past. As a matter of public concern, costs acquire grow- parable metrics for assessing performance, CAHPS standardized ing policy importance, as they do in competitive insurance markets closed-ended questions may fall short of capturing nuances in the such as the new state marketplaces, where price matters. interpersonal dimension of care. Incentivizing efficiency can be a tricky proposition. If a provider A complementary approach to eliciting information from patients learns how to deliver a given service of a given quality but us- about the quality of care uses open-ended narratives about their ing fewer resources than her peers do, or gets better at holding experience. Thus far, written comments from patients, when avail- down prices in her supply chain, who should reap the benefit? In a able, “are often seen by physicians as the most useful and meaning- capitated payment system, the answer is unambiguous: the savings ful form of patient feedback.”11 Patient narratives have the potential accrue to the provider organization, although Medicare Advantage to inform providers about how to improve their processes and requires the organization to plow the savings back into the plan in outcomes and how to supplement the quantitative information ob- the form of enriched benefits. Private market HMOs often do much tained through CAHPS closed-ended surveys. Even though studies the same to gain market share or otherwise burnish their product have found that consumers are not strongly influenced by CAHPS- and prestige. style performance scores, advocates for greater use of patient narra- tives point to the consumer appeal of open-ended formats such as If a payer wants to incentivize efficiency on a fee-for-service plat- Yelp and Angie’s List. Indeed, researchers have recently used NLP form, the choices are more complicated. In the Medicare Shared to analyze Yelp reviews of 1,352 hospitals by 17,000 consumers and Savings Program or the Physician Group Practice Demonstration, found that the information may provide a useful complement to payer and provider agree on a target cost and then share savings surveys such as CAHPS.12 if actual costs are less, and in two-sided risk contracts, losses are shared as well. Even after extensive experimentation, it is not yet That said, a proliferation of patient-comment websites has unloosed clear whether this formula can produce incentives sufficiently large the power of narrative on consumers with only random benefits, to cover the expense or compensate for the disruptions associated and may tend to crowd out objective measures in consumers’ with the practice redesigns usually necessitated by such programs. perspectives and to increase physician skepticism about the web- sites’ value. Champions of the narrative approach propose policy Another alternative is to design a method for assessing efficiency interventions to promote the use of more rigorous methods of nar- and pay an incentive premium for plans that achieve benchmark rative elicitation and to ensure their representativeness. They also scores or outperform their peers. Such an approach, however, argue that patient-reported information should be given a generous returns to the difficult questions of (1) whether providers should be share of influence in incentive programs because of its potential for paid extra for what they should have been doing all along and (2) informing practice improvement.13 who should benefit from improvements in quality, cost, and value. In any case, it will be essential to define and somehow measure all Considering Quality and Cost Together of these terms and integrate them into a common framework. Thus, Without question, measuring quality poses several challenges, but an important part of the future research agenda will be developing it is still only the first step in determining value. As noted above, methods to measure value, just as methods to measure quality were pay-for-performance and public reporting programs have primarily needed to lay a foundation for pay-for-performance. helped the system correct problems of underuse, but the programs have not checked cost growth or provided incentives for efficiency, The NQF Report according to a 2014 report for the National Quality Forum (NQF) The NQF report analyzed 25 value-based purchasing programs that was recapped at the AcademyHealth meeting.14 Wide varia- involving both public and private payers. Some apply only to tion in the cost of a given service is evident both within and across specified services or provider types—specialties, surgery, hospitals, regions, giving rise to the inference that many variations may be primary care—while others pertain to all covered services and pro- unwarranted. viders. Quality measures vary widely among these programs. They address organizational structures as well as clinical process and 4 Paying for Value: Progress and Obstacles increasingly register levels of IT use, care teams, handoff protocols, gram’s first four years, according to a 2015 report from the Avalere and other characteristics of integrated care. Some also include pay- analytics firm. The report concludes that provider behavior may be ments to support the development of medical homes and account- changed with a combination of strong incentives: putting providers able care organizations (ACOs). Costs are frequently defined on a at risk for high costs and offering generous rewards for measured per episode basis but may also be measured in terms of resource quality.15 use, or globally, as per member per month or year. Limitations Authors Andrew Ryan and Chris Tompkins identified four basic Several limitations are associated with all the above approaches and types of approaches for combining cost and quality assessments models. To begin, adequate measures of overtreatment or excess re- to determine incentive payments. In the first approach, cost and source use are lacking.16 For large organizations with advanced data quality may be measured separately, with composite scores for each, systems, big-data solutions for this problem may be at hand. Even then grouped into performance levels for each—usually low, aver- so, detailed clinical information will be needed to reach conclu- age, or high—to determine payment adjustments. Medicare uses sions about appropriateness. Smaller organizations probably cannot this model for its physician value-based payment modifier. afford the type of risk-sharing that most of the alternative payment models envisioned by Secretary Burwell would entail, and are The second approach combines quality and cost scores into a single likely to stay on a fee-for-service basis for the foreseeable future.17 metric. Blue Cross Blue Shield of Michigan has adopted this model The AcademyHealth group was cautioned that no consensus exists for its hospital P4P program, and Medicare’s Value-Based Pur- on how to identify high-value performers with consistency in the chasing program for hospitals follows the same approach. In the absence of robust measures of long-term patient health outcomes. third approach, programs may establish hurdles or gates requiring providers to meet specified performance for quality or cost or both. Further, research has found only weak correlations between cost The Medicare Shared Savings Program is an example of this model. and quality of care, so this essential feature of performance is not It requires participating organizations to pass muster on measures yet understood. Consumers, though, may be apt to get this wrong of patient experience, care coordination, and clinical standards to and equate the two. At the same time, steep increases in cost- be eligible for a cut of savings. sharing are heightening consumers’ sensitivity to cost but leave them with only limited ability to obtain intelligible information The fourth approach does not attempt to combine cost and quality about what costs for their care will be, much less any comparative in a single framework but reports performance for each with a star information about the prices they are charged. It is not clear how rating system. It is used by the National Committee for Quality As- high-deductible plans marketed under the banner of consumer- surance and two Minnesota organizations: HealthPartners and the driven care can succeed in bringing consumers’ market power to pioneering Buyers’ Health Care Action Group Purchasing Initiative. bear in holding down costs or prices. The Alternative Quality Contract Much is still to be learned about how financial incentives and other An acknowledged leader among these value-based efforts and an influences affect provider behavior. Some have raised concerns example of effectively integrated cost and quality initiatives is the about how penalties and bonuses may “crowd out” intrinsic motiva- Alternative Quality Contract (AQC) in Massachusetts. The state’s tion, suggesting that the development of a supportive culture and Blue Cross Blue Shield plan has operated the AQC since 2009 with environment that nourish professionalism and cooperation may be about a dozen provider organizations that are ACO-like groups more important.18 In addition, behavioral economics has intro- ranging from large physician-hospital organizations to small-prac- duced fresh insights into loss aversion and the tendency of short- tice clusters. Contracting providers receive a risk-adjusted global term considerations to trump rational calculation.19 For provider payment for all their attributed enrollees, who numbered about organizations, weighing the cost of practice redesign against the 700,000 in 2015. The payment covers all services, and the provider uncertain future benefits of performance incentives poses uncom- organization is at risk for its contracted share of losses and gains— fortable choices, especially for small organizations that may not be two-sided risk that the Medicare Shared Savings Program (MSSP) able to absorb significant losses. does not yet require. Paying for Value: Progress and Obstacles The contracts run for three to five years in order to provide a stable This review of issues in measuring clinical quality, patient experi- platform for the provider groups. Real-time information-sharing ence, and efficiency documents significant gains on the path to between the parties is considered an essential supporting mecha- measuring value but highlights unsettled questions and outstanding nism, as it is in the MSSP, with which the AQC shares its basic research needs. A proliferation of clinical process measures, used approach. Spending growth rates declined by 50 percent in the pro- differently by different payers, threatens to overwhelm provider 5 Paying for Value: Progress and Obstacles organizations and blunt the effect of any single incentive program. quality measurement and payment for value to the goal of popula- Meeting participants expressed divided views about whether the tion health. But, as noted, ACOs are not a proven concept, and measure sets should be whittled down or whether to let a thousand some fear potential ill effects from further increases in provider flowers bloom and build an IT-enabled measurement infrastructure consolidation. to make the burden bearable. In the meantime, what has been accomplished in health care quality At the same time, some sensed that provider organizations per- and safety since publication of the Institute of Medicine’s 1999 To ceived an inevitable increase in accountability for both quality Err Is Human is considerable. Patient-reported outcome measures and cost and a growing focus on meeting evolving performance have the potential to counter-balance the centrifugal effects of pro- standards, such as they are. liferating clinical indicators. As slow as progress has been, IT adop- tion is gradually building a measurement infrastructure that could Nonetheless, the evidence is mixed on how closely process mea- help bring smaller practices up to speed with 21st-century quality sures are associated with patient outcomes, the gold standard for improvement technology. Experiment and experience are gradually system performance. Patient-reported outcome measures (PROM) bringing about a better understanding of provider attitudes and be- hold some promise as provisional measures, but reliable systems for havior and what it takes to change them. Policy conversations such constructing and collecting PROMs are in the embryonic stages of as the discussion at the AcademyHealth meeting described here are development. beginning to come to grips with imponderably difficult issues like the paradox of paying for efficiency and the question of how value Moreover, it is the assumed goal of the overall health system to for payers translates into value for patients and consumers. achieve optimal health outcomes for entire populations, not just for individuals. Integrated delivery organizations paid by capitation About the Author have assumed this responsibility, and ACOs are designed to do like- Rob Cunningham is an independent writer in Washington, D.C. wise. But ACOs are not yet a proven model and capitated HMOs Endnotes account for only a limited share of the insured population. 1. Hahn J, Blom K. H.R. 2: The Medicare Access and CHIP Reauthorization Act of 2015. WASHINGTON, D.C.:Congressional Research Service. March 26,2015. AcademyHealth discussants suggested that the structure of pro- vider organizations may have to be the focus of some future efforts 2. Berenson RA, Pronovost PJ, Krumholz HM. Robert Wood Johnson Founda- tion/Urban Institute, May 2013. Available at http://www.urban.org/research/ to improve performance among smaller entities, as the push for publication/achieving-potential-health-care-performance-measures. patient-centered medical homes would indicate. MACRA, it was 3. Bardach NS et al. Effect of pay-for-performance incentives on quality of care noted, gives physicians flexibility in how they approach quality in small practices with electronic health records: a randomized trial. Journal of improvement and should create opportunities for smaller prac- the American Medical Association, 2013;310(10):1051-9. tices to embrace change proactively. MACRA makes a ramp, in 4. See Roland M, Dudley RA. How financial and reputational incentives can be used to improve medical care. Health Services Research. 2015;50(6)Part one participant’s metaphor, but the landing strip for change will be II:2090-2115. alternative payment models, which are still under construction. An 5. Damberg CL et al. Measuring success in health care value-based purchasing awkward but germane coinage— “systemy-ness”—was proposed as programs: summary and recommendations. RAND Corporation, 2014. Avail- a measuring stick for such structural changes, presumably including able at http://www.rand.org/content/dam/rand/pubs/research_reports/RR300/ RR306/RAND_RR306.pdf. IT adoption, care teams, registries, collaborative care arrangements, and the like. Meeting participants concluded that unremitting 6. Sands K, Romig M, Dykes P, Schell-Chaple H. Redesigning care: a new play- book to improve quality, safety and patient-centered care. Beth Israel Deacon- experimentation is the necessary and inevitable direction in which ess Medical Center, Johns Hopkins University, Brigham and Women’s Hospital, the value imperative now drives the system. and University of California, San Francisco (slide presentation), 2015. Avail- able at http://www.healthforum-edu.com/summit/PDF/2015/SUM15%20 sands-romig-dykes-schell.pdf . Conclusion 7. Knight WA et al. The incidence of seizures in patients undergoing therapeutic The complexity of the measurement universe reflects the fractured hypothermia after resuscitation from cardiac arrest. Epilepsy Research, 2013 nature of the health care system and of the society that it serves. Oct;106(3):396-402. A twin problem looms large: the challenge of capturing long-term 8. Marafino BJ, Davies JM, Bardach NS, Dean ML, Dudley RA. N-gram support outcomes and attributing them appropriately to the responsible vector machines for scalable procedure and diagnosis classification, with ap- plications to clinical free text data from the intensive care unit. Journal of the provider or providers. Perhaps what is needed, some participants American Medical Informatics Association, 2014 Sep-Oct;21(5):871-5. said, is a way to assess outcomes in the aggregate, on a community- 9. Marafino BJ, Boscardin WJ, Dudley RA. Efficient and sparse feature selection wide scale, with rewards, penalties, and other incentives also for biomedical text Classification via the elastic net: application to ICU risk distributed to providers on a community-wide basis. ACOs are a stratification from nursing notes. Journal of Biomedical Informatics, 2015 Apr;54:114-20. step in this direction; they offer potential for harnessing the tools of 6 Paying for Value: Progress and Obstacles 10.Schlesinger M et al. Taking patients’ narratives about clinicians from anecdote 15.Seidman J. Payment reform on the ground: lessons from the Blue Cross Blue to science. New England Journal of Medicine, 2015;373(7):675-9. Shield of Massachusetts alternative quality contract. Avalere Health, March 2015. Available at https://www.bluecrossma.com/visitor/pdf/avalere-lessons- 11.Ibid. from-aqc.pdf. 12.Ranard BL et al. Yelp reviews of hospital care can supplement and in- 16.Damberg, Measuring success. form traditional surveys of the patient experience of care. Health Affairs, 2016;35(4):697-705. 17.Conrad, DA. The Theory of Value-Based Payment Incentives and Their Appli- cation to Health Care. Health Services Research,2015;50(S2):2057-2089. 13.Schlesinger M, Grob R, Shaller D. Using patient-reported information to im- prove clinical practice. Health Services Research, 2015;50(6)Part II:2116-54. 18.Berenson RA, Rice T. Beyond measurement and reward: methods of moti- vating quality improvement and accountability. Health Services Research. 14.Ryan, A, Tompkins, C., Linking quality and cost to measure efficiency. Na- 2015;50(6)Part II:2155-86. tional Quality Forum, August 2014. 19.Khullar D et al. Behavioral economics and physician compensation—promise and challenges. New England Journal of Medicine. 2015;372(24):2281-3. 7