Introduction
Attempting to find traction on strategies or interventions aimed at bending the cost curve has been the Holy Grail of health care policy. The elusive hunt by both researchers and policy makers seeks intervention points where medical care is costly and amenable to cost improving changes while also improving, or at least not decreasing, quality. All-cause readmission to inpatient care has been an area of focus of recent policy attention and public measurement in the United States and a number of other countries [Centers for Medicare and Medicaid Services (CMS), in the United Kingdom by the National Centre for Health Outcomes Development (NCHOD), and in Australia by the Australian Institute of Health and Welfare (AIHW)]. Making readmission rates a policy target is related to growing attention on attaining value for patients, a combination of quality and cost. A key factor in making readmissions a target is that patients by definition have contact with a costly part of the health care system when admitted to the hospital. The admission then is a potential intervention point to impact cost and quality: all-cause readmissions would seem to raise cost and reduce patient welfare (Jencks et al., Reference Jencks, Williams and Coleman2009; Williams, Reference Williams2013).
Since health care regulatory systems set up by governments can focus their incentive schemes on health care providers, hospitals can be a source of health care system savings if total payment can be reduced by reducing readmissions. As such, readmissions have become a target of national policy in many countries, including the United States. The Patient Protection and Affordable Care Act stipulates that CMS has the authority to penalize hospitals for excess readmissions (HR 3590, 2010). As a result CMS has begun reducing reimbursements. Beginning in Fiscal Year (FY) 2013 all Medicare reimbursements will be reduced for those hospitals whose all cause 30-day readmission rates after an index visit for a specified subset of conditions (heart failure, acute myocardial infarction and pneumonia) exceeds the expected readmission rate for that hospital (DHHS, 2011). Similarly, the UK Payment by Results (http://www.dh.gov.uk/pbr) program, beginning in FY 2011–2012, evaluates all emergency admissions within 28 days and eliminates or reduces payments for these readmissions depending on the emergency/elective status of the index admission by condition. Note the arbitrariness of the choice of 30 days in the United States and 28 days in the United Kingdom as the relevant evaluation time frame for measure specification. The key similarity between the UK and US approaches is the focus on financial penalties imposed on hospitals for levels of readmission deemed inappropriate. However, financial penalties are a high powered incentive that we believe may or may not be warranted.
Generally, in government sponsored health care schemes, there is a hierarchy of incentives running from low powered incentives, such as data provided for quality improvement, through public reporting, to higher powered incentives, such as financial pay for performance. Use of all-cause readmissions has moved through that hierarchy relatively quickly in recent years to that highest powered incentive. The inclusion of the Hospital Readmission Reduction Program (HRRP) in the Affordable Care Act represents a movement toward high powered incentives to reduce hospital readmission. This has spurred a concomitant rise in research interest on this topic (Kangovi and Grande, Reference Kangovi and Grande2011; Joynt and Jha, Reference Joynt and Jha2013; Williams, Reference Williams2013).
For better or worse, readmission has now become a quality measure that is explicitly tied to high powered incentives. Long standing economic principal agent theory (e.g. Holmstrom and Milgrom, 1990), suggests that high powered incentives require strong confidence in the measurement and appropriateness of the target that anticipate and minimize unintended consequences. In addition, to be worthy of high powered incentives, quality measures need to be relatively immune from gaming, have strong face validity and importance, and be fair so that agents, such as hospitals in this case, have a reasonable expectation of being rewarded for proportional effort. Holmstrom and Milgrom (1990) in focusing on the impact of incentives in multiple activity organizations, such as hospitals, also emphasize that difficult to measure, yet important, concurrent activities can lead to use of lower powered incentives. Finally, the desired objective needs to be relatively clear, and the clarity of the objective is arguable when targeting readmission. We will build up to the current approach from a historical perspective concerning the use of readmission rates as a quality measure in hospitals, discuss contemporaneous policy issues, and attempt to focus a lens on the appropriate objective and incentives to achieve it.
The research literature is expanding rapidly and we are not choosing to conduct a full-fledged structured literature review. Existing systematic literature reviews [e.g. Fischer et al. (Reference Fischer, Anema and Klazinga2011), that reviews the European literature on hospital readmission rates] are limited in scope and are not focused on the external policy context. Yet, there is a lack of a unifying conceptual model that lays out the mechanisms by which one might expect the HRRP or alternative policies to have an impact on readmission in virtually all of this existing literature. The result has been conflicting views among policy makers and researchers alike on the value of such a policy. To construct a model that allows the system to move forward with a research and resulting policy agenda (as opposed to recycling old ideas under different jargon), we believe some historical perspective and international comparison of the issue is in order. Providing such a perspective and linking it to an intuitive and concise conceptual model should prove helpful in ensuring the research and policy communities are being realistic about organizational, behavioral and provider issues affecting the efficacy of quality measure systems to align the incentives to achieve the desired result. The point of this paper is to do a high-level review of the history of research and policy regarding hospital readmissions, lay out a conceptual organizational model of the potential role of the hospital in patient transitions, discuss the strengths and weaknesses of previous and proposed policies, and to set out a research agenda to improve the literature going forward.
Background
Readmission: an early health services research perspective
By the 1970s, when research on the use of health services began to take off in earnest, Medicare had only been in place for a relatively short time. The focus of national policy in the late 1960s and early 1970s was concentrated more on Vietnam and then economic stagflation than it was on rising government health expenditures. The possibility that hospital readmission rates may be excessive in the United States did arise at this time, but this literature was almost exclusively focused on those patients with mental health conditions and a few articles addressed those patients suffering from paralysis. If one were to take this focus as an indication of the general thoughts on the use of the hospital at the time, it suggests that only someone with a mental illness would want or need to be in the hospital any more than was really necessary. Like many of the lessons in the past illuminated in this essay, we are suggesting that more reconsideration of mental or behavioral health impacts on choices made for readmissions by both patients and providers in the present environment is warranted as well.
The early 1980s saw the beginnings of a health services research focus on readmission for older adults related to their engagement in the Medicare program. In the United States, the initial interest was not so much to raise great alarm on the absolute rates or costs to the system, but interestingly it was more driven by quality concerns (Graham and Livesley, Reference Graham and Livesley1983). A select few studies examined whether some forms of outpatient and home care could prevent the readmission of those patients with certain chronic conditions such as diabetes (Fishbein et al., Reference Fishbein, Faich and Ellis1982) or chronic obstructive pulmonary disease (COPD) (Roselle and D'Amico, Reference Roselle and D'Amico1982). In retrospect, the takeaway from these studies and the trials conducted during this period was that using intensive outpatient treatment to avoid readmission had mixed results, efficacy was sometimes shown, but few if any studies demonstrated substantial cost effectiveness of readmission reduction efforts (Smith et al., Reference Smith, Weinberger, Katz and Moore1988; Townsend et al., Reference Townsend, Piper, Frank, Dyer, North and Meade1988; Weinberger et al., Reference Weinberger, Smith, Katz and Moore1988; Safran and Phillips, Reference Safran and Phillips1989).
As the decade progressed, concerns over a potential move to sicker and quicker discharges in response to the prospective payment system (PPS) were beginning to arise in the United States (Anderson and Steinberg, Reference Anderson and Steinberg1984). The earlier theorized impact of the PPS on readmission rates ultimately lacked empirical support (DesHarnais et al., Reference DesHarnais, Kobrinski, Chesney, Long, Ament and Fleming1987; Epstein et al., Reference Epstein, Bogen, Dreyer and Thorpe1991). Simultaneously the NHS was experiencing resource constraints, and concerns over shortening lengths of stay and their role in rising readmission rates also were emerging (Henderson et al., Reference Henderson, Goldacre, Graveney and Simmons1989).
Concurrent with these policy concerns over readmission, more data and computing power were becoming available to construct predictive models. The general finding in both the United States and the United Kingdom from most analyses of early readmission to the hospital was that readmission was in large part attributable to general deterioration by the patient resulting from the underlying disease that generated the initial hospital admission (Anderson and Steinberg, Reference Anderson and Steinberg1985; Gooding and Jette, Reference Gooding and Jette1985; Victor and Vetter, Reference Victor and Vetter1985; Andrews, Reference Andrews1986; Berkman and Abrams, Reference Berkman and Abrams1986; Williams and Fitton, Reference Williams and Fitton1988). There were different findings, though, in what role social factors played in readmission. The general division was between the results of large claims type analyses and those of smaller prospective and/or chart review studies. The former tended to find no impact of social factors (Holloway et al., Reference Holloway, Thomas and Shapiro1988; Holloway and Thomas, Reference Holloway and Thomas1989), while the latter found some social support factors did impact the probability of readmission (Fethke et al., Reference Fethke, Smith and Johnson1986; Evans et al., Reference Evans, Hendricks, Lawrence and Bishop1988; Williams and Fitton, Reference Williams and Fitton1988) and that more detailed health measures such as functioning scales might prove useful in identifying those at higher risk and in need of additional care (Berkman et al., Reference Berkman, Dumas, Gastfriend, Poplawski and Southworthe1987).
The predictive modeling attempts evolved into their logical sequelae. First among these were attempts to address whether and to what extent readmissions were avoidable. Early estimates, mainly from research in the NHS, were no better than the varied estimates we see today in studies in the United States, ranging from about 10% to 50% (Clarke, Reference Clarke1990; Gautam et al., Reference Gautam, Macduff, Brown and Squair1996; Parkes and Shepperd, Reference Parkes and Shepperd2000). The second sequelae, which follows from the first, was a growing interest in condition-specific readmission rates (Stanton et al., Reference Stanton, Jenkins, Goldstein, Vander Salm, Klein and Aucoin1985). This is because avoidability is a moving target that is highly dependent on the underlying condition's disease progression and treatment process (Hernandez and Granger, 2012). Heart failure was one of the conditions garnering an early focus in this period, ostensibly related to prevalence and disease progression that tend to lead to more and more hospitalizations (Gooding and Jette, Reference Gooding and Jette1985; Berkman and Abrams, Reference Berkman and Abrams1986). Interestingly, almost 30 years ago, Gooding and Jette found the 6-month readmission rate for Medicare HF patients to be 36%, which is quite similar to the readmission rate for patients with this disease today. Nearly three decades later, heart failure is still one of the conditions at the center of the policy debate surrounding readmission. Other conditions that became a focus of research, and thus policy targets, include other cardiac conditions such as ischemic heart disease and acute myocardial infarction, COPD and pneumonia, ostensibly owing to their high rank on the prevalence of admissions.
Growing medical costs were not as much of an issue in the early- to mid-1990s when there was a brief reprieve from relentless growth in medical costs (Glied, Reference Glied2003). This reprieve is generally attributed to an attempt by the US health care provider industry to lessen motivation to enact the stricter regulatory reforms proposed by the Health Care Reform Task Force by adopting managed care-based restrictions on utilization (Mayes and Lee, Reference Mayes and Lee2004). The focus quickly shifted in the later part of the decade after these reform efforts failed, managed care was deemed by the public as too great an intrusion on their choices, and these costs resumed their march upward. At this point, readmission gained a prominent focus as a potential source of cost savings and improved quality (Glied, Reference Glied2003). This revitalized concern over cost growth and simultaneous renewed interest in readmission is a coincidence not lost on some (Vaduganathan et al., Reference Vaduganathan, Bonow and Gheorghiade2013; McCarthy et al., 2013).
Contemporary US readmission policy
Bringing us up to the current policy landscape in the United States, that is similar though not identical to that of other countries, there has been a focus on the potential for quality improvements and cost savings in readmissions in the Medicare program (Fontanarosa and McNutt, Reference Fontanarosa and McNutt2013). Improved or sustained quality while reducing costs is viewed as part of a value equation to benefit Medicare beneficiaries and the fiscal sustainability of the program. The general understanding in the policy community, driving the policy incentives being imposed, seems to be that poor care coordination during and after the index hospital admission, combined with provider incentives and behavior promoting readmissions, generates these high and persistent rates (Berenson et al., Reference Berenson, Paulus and Kalman2012; Jenq and Tinetti, Reference Jenq and Tinetti2012). Yet, there are important counterarguments to this view that seem to get considerably less attention (Bhalla and Kalkut, Reference Bhalla and Kalkut2010; Kangovi and Grande, Reference Kangovi and Grande2011; Joynt and Jha, Reference Joynt and Jha2012; Hockenberry et al., Reference Hockenberry, Burgess, Glasgow, Vaughan-Sarrazin and Kaboli2013).
As we noted above, these service use patterns and comparisons to other similar problems in health system utilization are long standing, going back decades (McCarthy et al., 2013). Though CMS (Brennan, Reference Brennan2012) has argued that identifying Medicare beneficiaries at greater risk for readmission in advance can allow targeting of resources at these patients (Brock et al., Reference Brock, Mitchell, Irby, Stevens, Archibald, Goroski and Lynn2013), it is not clear that this can be done in a cost-saving way (Stauffer et al., Reference Stauffer, Fullerton, Fleming, Ogola, Herrin, Stafford and Ballard2011; Voss et al., Reference Voss, Gardner, Baier, Butterfield, Lehrman and Gravenstein2011). Furthermore, while randomized controlled trials (RCTs) show large effects, it may be that the individual level intervention models are not effective in a less motivated non-RCT patient population (Katz, Reference Katz2011). Moreover, these returning patients are often quite expensive, perhaps costing more than Diagnosis Related Group (DRG) average cost payment levels, and the most expensive patients seem to be those least amenable to current efforts to curb readmissions (Brennan, Reference Brennan2012).
Medicare's Recovery Audit Contractor (RAC) policy and readmission
Medicare's RAC program has generated additional questions and concerns for readmission rate policies. The RAC program was piloted as a demonstration in a few states by CMS in 2005–2007, then rolled out by 2010 nationally. The intent of the RAC is to address the growing concern that some short inpatient stays are the result of hospitals admitting patients who were not appropriate for hospital admission due to their relatively low levels of needed care. CMS contracts with private auditing firms whose job it is to monitor and audit Medicare patient admissions to hospitals and who are given authority to determine whether CMS should be recovering payments for inappropriate admissions. This program has generated over $10 billion in projected improper payments, most of which comes from admissions deemed to be medically unnecessary, and this is driving down short stay admissions (CMS, 2012). Concerns of fairness and unintended consequences with this program have plagued CMS, most recently leading to a myth busting frequently asked questions document (CMS, 2012) concerning how these medical necessity and coding errors are being identified by contingent fee contractors. This program has a direct bearing on how we think about hospitals gaming readmissions, because in theory the hospital would want to readmit patients who were likely to consume care that cost less than the DRG payment.
Hospital concerns and incentives being driven by these relatively large RAC investigations suggest that the vast majority of the all-cause readmissions being targeted already are not medically unnecessary. As CMS has noted (Brennan, Reference Brennan2012), the only longitudinal trend effects we are seeing are very small drops in the beneficiary rate of index admissions, which could be at least partly reductions being motivated by these RAC investigations. If these short stay index admissions are for individuals who on average are less likely to be readmitted from risk unaccounted for in the model, then one statistical artifact of the RAC program will be to increase risk adjusted hospital readmission rates among hospitals who reduce their short stay admissions in response to the RAC.
If anything, the rates of readmissions have been trending up very slightly, perhaps due to the RAC effect, but more likely also as a result of increases in chronic disease in the Medicare population. The relationship of readmissions to chronic disease and Hierarchical Condition Category (HCC) risk assessment to readmissions is quite strong. In 2010, only 2.6% of Medicare beneficiaries have HCC risk scores above 4.0, but these contribute over 20% of all Medicare readmissions (Brennan, Reference Brennan2012). But if these illness burdens are related to factors not adjusted for in the risk adjustment in the readmission rate measures being used, but these unadjusted factors are highly related to incentive related measured rates, it is unclear how hospitals or systems can respond (Hockenberry et al., Reference Hockenberry, Burgess, Glasgow, Vaughan-Sarrazin and Kaboli2013).
Observation services and readmission
Another issue affecting hospital treatment choices that is rising to prominence is the use of observation services. These are hospital-based outpatient services where patients are evaluated beyond the emergency room but are not yet admitted as inpatients. CMS reimburses for observation at a lower rate than inpatient services, and expects that patients treated in observation beds remain in this status for no longer than 48 hours while a decision on readmission or discharge can be reached. The contemporaneous interest in observation services has been driven by concerns over patient cost sharing, as time spent in observation is outpatient and incurs higher cost sharing and does not count toward CMS's 3-day inpatient qualifying stay for discharge to skilled nursing. However, CMS is currently addressing the 3-day inpatient qualifying stay issue. In addition, the Medicare Administrative Law Courts (ALJ) have consistently determined that hospitals that challenge the audit contractors and are unsuccessful in justifying inpatient payment are to be allowed to resubmit for outpatient payment. As a result of these determinations, CMS is now running a demonstration project that will automatically reimburse hospitals at the observation service rate for short stays that were determined inappropriate by the RAC as long as these hospitals agree not to challenge the RAC auditor's decision (CMS, 2012). Hospital response to this has been mixed, because some hospitals have been successful in turning over a large percentage (>80%) of the claims initially deemed inappropriate. While the RAC policy changes are not directly targeted at readmission, they have very direct implications for how we view readmission policy and any changes in trends that occur during this period and the attribution of these effects.
Current empirical evidence on both observation and readmission is thin; the relationship between these relatively recent RAC and observation service policies is theoretically quite direct (Dardick and Stein, Reference Dardick and Stein2012). Specifically, hospitals could use observation beds as a way to game readmission rates because observation is not an inpatient admission, and therefore by current definition would not be included in either the numerator or denominator of readmission rate calculations used by CMS. Given the existing concerns with how efforts to reduce unnecessary or inappropriate admissions would act to reduce both admissions (denominator) and readmissions (numerator), the interaction with observation service choices is clear. Current evidence suggests high patient satisfaction with these services, equivalent or improved clinical outcomes, and improvements in efficiency (Ross et al., Reference Ross, Compton, Richardson, Jones, Nittis and Wilson2003; Baugh et al., Reference Baugh, Venkatesh and Bohan2011). However, the equilibrium effects of the use of these services are not clear as the patient population treated in observation expands, given the relatively select patient population chosen to receive these services. In the extreme, observation could just add another layer of complexity to the already daunting challenges facing emergency department decision makers.
Behavioral health and readmission
Behavioral health issues (both substance use disorders and serious mental illness, of course, but also mild depression and anxiety) are key drivers of readmissions and other post-discharge costly health care utilization (Benzer et al., Reference Benzer, Sullivan, Williams and Burgess2012). As noted above, these issues were the genesis of the health services literature on readmission, and interest in this has been rekindled recently. These behavioral concerns add complications to efforts to reduce readmissions (Kartha et al., Reference Kartha, Brower, Saitz, Samet, Keane and Liebschutz2008) and needs for interdisciplinary and trans-disciplinary approaches between medical/surgical physicians and other health care providers to health care and coordination that have been difficult to achieve (Vest and Gamm, Reference Vest and Gamm2010). These and community psychosocial issues of homelessness, social supports, and community connections create challenges for hospital focused efforts that probably cannot be overcome in a single policy.
If behavioral and psychosocial issues are the key identifiable differences in readmission rates across hospitals and regions, then safety net hospitals and other urban hospitals overrepresented in patients experiencing these comorbid and non-clinical social issues will fall systematically behind. One of the few urban safety net hospitals with a below average readmission rate, Denver Health, still does not think it is a model for change in other safety net hospitals and suggests that much shorter readmission measures within less than a week may be better indicators of quality hospital care (Whitney, Reference Whitney2012). Yet, the system expenditures that worry policymakers and the payers like Medicare actually may go out much longer than 30 days. The distribution of readmissions spikes the second day after discharge, and then declines monotonically, and the 30-day cutoff is somewhat arbitrary (Dharmarajan et al., Reference Dharmarajan, Hsieh, Lin, Bueno, Ross, Horwitz, Barreto-Filho, Kim, Bernheim, Suter, Drye and Krumholz2013).
An international perspective
The United Kingdom has focused its policy slightly differently. The high powered readmission incentive they have chosen has been to stop automatic payments for emergency readmissions within 28 days. Their evaluation at that point is then similar to the Medicare RAC evaluation for medical necessity and less like the Medicare Value-Based Purchasing program for readmissions in the United States. Unlike the United States, a greater distinction is made between planned and unplanned admissions as they evaluate readmissions, so rates and procedures are not comparable across countries. The NCHOD has been publishing rates of all-cause and selected condition emergency readmissions within 28 days of discharge since 1998/1999. Interestingly, they exclude mental health readmissions (also maternity), though some have proposed creating combination measures that incorporate mental health (Lakhani et al., Reference Lakhani, Coles, Eayres, Spence and Rachet2005). The incentive that the United Kingdom uses built off of the NCHOD is payment denial for readmissions deemed to be emergency following elective admissions, and for such readmissions following non-elective admissions only paying up to a threshold relative to at least a 25% improvement in the historic NCHOD documented readmission rate for that hospital. Interestingly, measured readmission rates increased sharply from 2002 to 2005, as these measurement and payment policies were being implemented, before leveling off. The use of the emergency department as a vehicle for specialists getting a patient into a hospital bed and other coding issues, as well as a general worsening of case-mix, have been cited as reasons for these anomalous changes (RAND Technical Report, 2012; UK Department of Health, 2012).
Other countries are more accepting of a lower powered incentive approach to readmissions, using them as monitoring indicators rather than reporting and pay-for-performance indicators. For example, the Australian AIHW only publicly reports aggregate 28-day unplanned/unexpected readmission rates following selected surgical procedures and then privately reports rates to hospitals in ways that vary across the Australian states (RANDTechnicalReport,2012,http://meteor.aihw.gov.au/content/index.phtml/itemId/443711). In 2012, they changed the definition to call it an ‘episode of care’ measure since readmissions for this indicator are defined within 28 days from the end of the patient's surgical episode of care and unlike US and UK measures it has no direct pay-for-performance outcome. The effectiveness of such policies is in question, but note that the Australian policy is considerably lower powered than the US and UK programs.
Recent studies have also tried to compare readmission rates internationally taking a disease-specific approach (Kociol et al., Reference Kociol, Lopes, Clare, Thomas, Mehta, Kaul, Pieper, Hochman, Weaver, Armstrong, Granger and Patel2012). While readmission rates were higher for patients in the United States, the exact mechanisms leading to these differences, either because of insurance pressures to reduce length of stay leading to additional readmissions, or differences in baseline health of patients, are unclear (Andraws, Reference Andraws2012). However, other recent work suggesting long term reductions in length of stay in the US Veterans Affairs system was not associated with higher readmission would argue for the latter (Kaboli et al., Reference Kaboli, Go, Hockenberry, Glasgow, Johnson, Rosenthal, Jones and Vaughan-Sarrazin2012).
Reestablishing the theoretical role of the hospital in health research
Model of decision makers and readmission
Readmission is the result of a complex cascade of events, disease progressions, and decisions by patients and health care professionals. In Figure 1 we present our view of a US centric conceptual model of this system of interaction between patients and health care system decision makers. Recently conceptual models of readmission have simply tried to add in the fact that patient characteristics, both clinical and socioeconomic, play a role (Kangovi and Grande, Reference Kangovi and Grande2011). We have substantially expanded on this idea, and focused the lens on the decision points and decision makers, as readmission policy instruments ultimately attempt to encourage better decision making.
Model of patient flow in the community and hospital
The two decisions under the direct control of the hospital-based physicians and staff are represented by the solid black and white arched arrows. The arched white lines represent the discharge decisions. These decisions indicate that the inpatient was deemed healthy enough to return to home or to be sent to a long-term care facility and represents the start of the cycle of readmission. The solid black arrows represent an escalation of the intensity of care or expected time required to deliver needed care which is made within the hospital organization.
In contrast, the checkered and gray arrows represent patient decisions. The checkered arrows represent joint decision making by a potential inpatient and a health care worker external to the hospital for said inpatient to visit the hospital. In this respect, an expert is advising the patient that their need for further medical attention at the hospital is warranted. For patients in long-term care facilities, these decisions are informed both by patient need and sometimes unfortunately by financial motives of the long-term care facility (Mor et al., Reference Mor, Intrator, Feng and Grabowski2010). With respect to outpatient primary care, direct financial motives are less of a concern, though the malpractice liability regime may lead to a larger proportion of patients being referred to the hospital than is socially optimal. In addition, in some cases of outpatient primary care providers the advice to go to the hospital may be automated, meaning the patient calls the physician's office and a message is relayed that the office is closed, and that the patient should go to the hospital if they need immediate care.
The gray arrows represent patient decisions reflecting some level of care is needed that are made in the absence of expert advice beyond that given at discharge from the initial hospitalization. These decisions can be autonomous, but for some patients, in particular older adults with complex comorbid conditions, they may be made in collaboration with a non-expert agent (family or friend). Current evidence suggests that the proportion of returns to the emergency department (ED), the pool of patients from which readmissions occur, is substantial (Vashi et al., Reference Vashi, Fox, Carr, D'Onofrio, Pines, Ross and Gross2013). As readmission is discouraged by strong incentives, attention will need to be paid to the effects on patient flow in EDs.
Gaming the system
Admittedly this is a simple model of readmission, but we believe it warrants careful attention in regard to the design of contemporary policies targeting readmission. But before we dig into how the current policy impacts the decisions in this model, it is worth considering exactly how a hospital could even manage to ‘game’ the PPS system to make money off of readmissions. To fix ideas, we consider two cases. One case we will call the nefarious behavior case, where the hospital is actively gaming the payment system. The second we will call the negligent behavior case, where the hospital is passively gaming the system. First, we turn our attention to the nefarious behavior case. In theory, at the time of discharge, represented by the white arched arrows, the discharging physicians and staff would have to have some indication of which patients were likely to be low cost should they return to the hospital. They then would have to provide some sort of differential care or instruction that led to a disproportionate share of these low cost patients returning to their facility for care. Not only that, but they then would have to send a signal to the Emergency Department physicians who would be making the decision to admit or discharge from the ED, that this subclass of patients was to be readmitted as part of a revenue maximizing effort. This sounds unlikely, but in fact part of the underlying arguments for readmission penalties is that hospitals could potentially be gaming the prospective payment system.
Now consider the negligent behavior case. In this case, the hospital physicians and staff at the time of discharge are treating patients fairly equally, though the time and potential financial cost leads to minimal effort to coordinate post-discharge care. In this case, some proportion of these patients will return to the hospital for reasons that might have been avoided with slightly more effort exerted at the time of discharge. However, for the hospital, this failure to coordinate care will lead to some of these returns being high cost and some being low cost, in a way that the hospital cannot predict. In this regard hospitals are making somewhat informed choices about the relative prospects of readmission, though it is highly unlikely that they can ‘game’ the reimbursement system by being negligent for the same reasons as in the nefarious case. Thus if we think the optimal rate of readmission is lower than the rate in the world where many hospitals operate under the negligent behavior hypothesis, then strong policies may be in order.
However, in reality, even without anything beyond the PPS, the physicians and staff have some incentive to try to identify patients who need to be discharged by might struggle over a narrow post-discharge window, i.e. to not be negligent. If they could effectively target this smaller population of patients, quality and efficiency might be improved. In considering the targeting issue, two things are worth noting (1) before the HRRP, the incentives embodied in a prospective payment system were already aligned to try to address the high readmission risk patients (though one might argue with the exact reimbursement levels) and (2) predicting who these high-cost readmitted patients will be within a subclass of patients has so far proven to be an elusive task (Kansagara et al., Reference Kansagara, Englander, Salanitro, Kagen, Theobald, Freeman and Kripalani2011).
Hospital reorganization and redirection: observation services
Hospitals historically have only directly controlled decisions that originate inside the organization. This boundary is represented by the dotted line and the decisions therein represented by solid black and white arched arrows. Increasing pressures on the hospital over the long term have led to a variety of changes in how hospitals behave. The introduction of the PPS in the United States induced decreases in length of stay (Lave and Frank, Reference Lave and Frank1990), and despite the concern at the time it was enacted readmission rates were unaffected (Epstein et al., Reference Epstein, Bogen, Dreyer and Thorpe1991). More recently, policies to reduce short inpatient stays were introduced. As mentioned earlier, the RAC program has been successful in recovering payments to hospitals deemed improper that may or may not be related to incomplete risk adjustment, and much of this was for short inpatient stays where patients were declared inappropriate for admission ex post. What has not yet been discussed in the literature on readmission, but would be of high interest, is whether the introduction of the RAC coupled with the PPS had already aligned the incentives for readmission rate reductions. Consider that the hospital benefits from a readmission financially only if the marginal cost of treating the readmitted patient is below the marginal revenue gained from the DRG payment either at the index visit or for the readmission. In effect, the lucrative readmitted patients would be those with very short stays requiring very low intensity of care. But these are the very types of stays from which the RAC auditors are clawing back payments, and thus hospitals are in effect being penalized for the low cost readmissions where they would have made a substantial positive margin. Conversely, there is little done to assist the hospital in the case of the readmitted patient who winds up being excessively costly, especially those where incomplete case-mix adjustment is a problem. Thus, even before the HRRP, the incentives were already aligned to reduce any potential nefarious behavior.
As a result of these pressures, many patients who historically would have received full inpatient care now frequently are held as outpatients receiving observation services. The use of this care has been recently expanding, but it bears mentioning that these types of services are not necessarily standard despite being reimbursed as if they were. Observation services can be provided in four contexts that likely have bearing on their relative effectiveness, though the research in this area is only beginning to develop. Observation services can be provided either in a geographically fixed unit usually under the supervision of Emergency Medicine physicians and staff, or in floating beds throughout the hospital. The other distinction within observation services is whether observation services are protocol driven or not. Experts in the field suggest that geographically fixed, protocol driven units are likely to be the most successful in terms of effectiveness, safety and efficiency because such units have clear accountability structures, regular monitoring, and decision rules for progression of care or discharge, though data limitations on hospital structure of these services has led to little systematic evidence (Cooke et al., Reference Cooke, Higgins and Kidd2003; Graff, Reference Graff2009).
From the perspective of the readmission question, the use of these services is interesting. Empirical evidence is needed to understand the drivers of the growth in and use of these services. We may observe decreases in readmission due to simple arithmetic rather than changes in the pattern of patient flow outside of the dotted area. To be readmitted, a patient must by definition first be admitted and then upon return to the hospital be admitted again. Routing patients to observation services at either visit will lower the readmission rate. In fact, in a recent final ruling on the RAC program CMS has declared that those hospitals with patients who had short stays deemed inappropriate will now be paid for observation for those patients. This is clearly one area the research and policy community interested in the readmission question needs to pay attention to going forward.
What role should the hospital play?
A confluence of other policies and reorganizations of care have been simultaneously developing in an effort to control costs and hopefully improve quality (Milstein and Shortell, Reference Milstein and Shortell2012). Accountable care organizations (ACOs) and the medical home movement are potentially going to expand the dotted line in the model to include both the outpatient primary care and long-term care components of the system. Rather than interacting with a fragmented system, patients making the decisions represented by the gray arrows will in theory be interacting with a unified organization delivering and coordinating the full continuum of care.
To the extent that ACOs and medical home creation efforts can be successful, readmissions may be slightly reduced. These models support coordination of care to improve transitions and to follow up with patients outside of the hospital. These interventions are potentially costly though. Further, home health was a large feature of the US system during the late 1980s and early 1990s, and readmission rates were at approximately the same level as today.
It will be an open question as to how much reduction in readmission we can expect with these interventions, and at what cost the reduction will come. Returning to the conceptual model, the gray arrows in and out of the home, once patients are discharged there, illustrates the relative dominance of patient life circumstances and patient decisions as determinants of location of care seeking (i.e. ED or outpatient clinics) and the condition they are in when they present. While the checkered arrows (connecting outpatient primary care to the hospital and long-term care to the hospital) represent the relative dominance of provider decisions in determinations of who goes back into the hospital. Despite substantial recent increases in the research around readmission, the determinants of the flows of patients into and out of the hospital are still not well enough understood for good policy making.
Of further concern is the potential link between improved (reduced) survival and increased (decreased) readmission rates. Laudicella et al., (Reference Laudicella, Donni and Smith2012) have presented a new model that relaxes assumptions of independence between mortality and readmissions to attempt to study the flows and outcomes on a more systematic basis. Their work strongly suggests that models that fail to take account of the relationship between mortality and readmissions are likely to result in substantial sample selection problems; however, other connections between follow-up outpatient care, ED visits, and long-term/rehab care also are likely to exhibit similar complex interconnections. However, direct examination of the question in the United States suggests no meaningful relationship between hospital readmission and short-term mortality rates (Krumholz et al., Reference Krumholz, Lin, Keenan, Chen, Ross, Drye, Bernheim, Wang, Bradley, Han and Normand2013). It has, however, reinvigorated concerns about using mortality, and by inference readmission, as a quality measure (Kupfer, Reference Kupfer2013).
Slightly better understood are the flows of patients within the hospital, with the possible exception of the use of the observation beds as an alternative to admitting patients or sending them home from Emergency Departments. These flows, represented by solid black arrows, are almost entirely controlled by providers and the health care system. The dotted line around the hospital is a bit simplistic in our model. Hospital administrators ostensibly set policies for the hospital as a whole that influence patient flows. Despite this, the Pauly and Redisch (Reference Pauly and Redisch1973) model of the hospital as a physician's workshop is relatively dominant in practice. This means that disciplinary management of patients by emergency medicine, cardiology, pulmonary medicine, and psychiatry tend to be operated in separated silos, to the detriment of coordinated holistic care for each patient. Thus one could imagine that even within the hospital organization, there would be a multitude of dotted lines.
One place where this dotted line is relatively strong, and has been growing stronger, is in keeping outpatient primary care separated from the inpatient hospital, even where it operates within a hospital system. The growth of inpatient internal medicine hospitalists as separate specialties means that relatively few outpatient primary care physicians manage their patients in the hospital any more, or even are aware of their admittance (Hoff, Reference Hoff2009). This has obvious negative consequences in the model of patient flow described earlier, for how can a physician work to provide outpatient care that mitigates the need for a return to the hospital if they are unaware the patient has had a hospital stay. This is where ACOs and the medical home concept are theoretically supposed to prove effective, but that remains to be seen empirically (Hoff, 2012).
Incentives to drive better decision making in these types of complex environments (and Figure 1 is a simplistic representation of those complexities in this one), while avoiding unintended consequences, has proven to be quite difficult in the history of health policy. Carrot-driven approaches aimed directly at the hospital have motivated behavior in ways that may have relatively fewer spillover effects than stick-driven approaches, but clarity about how either approach may or may not help reduce system level costs requires careful measurement and understanding of those spillovers. Risks to particular socially and economically at-risk populations and hospitals that serve these populations are particularly severe. While no one wants to give safety net hospitals a pass on performance goals, more carefully vetted models that take account of correlations and degrees of difficulty in dealing with particular patient populations are more likely to be more effective policy levers.
Alternatively, recent years have seen an explosion of care coordination efforts aimed mostly at expanding the influence of the hospital outside of its dotted line walls. These care coordination efforts may increase quality of delivered care for particular patients, but current evidence suggests they are highly unlikely to reduce system costs (Katz, Reference Katz2011). The most effective programs have made substantial changes in the organization of care that are difficult to implement and sustain, such as multi-disciplinary rounding within the hospital, redefinitions of the scope of practice of clinical and non-clinical staff, and use of innovative quality improvement methods (Ellrodt et al., Reference Ellrodt, Glasner, Cadorette, Kradel, Bercury, Ferrarin, Jewell, Frechette, Seckler, Reed, Langou and Surapaneni2007; Vina et al., Reference Vina, Rhew, Weingarten, Weingarten and Chang2009). Furthermore, an international comparison suggests that even in the systems the US seems to aspire to, there is little care coordination conducted by hospitals beyond their own walls, so why is it expected to work here (Schoen et al., Reference Schoen, Osborn, Squires, Doty, Rasmussen, Pierson and Applebaum2012)?
Moving forward
Do we care, or is it cost?
If one considers the history of the research on readmission, this renewed focus on readmission and these related policy efforts are yet another attempt at firing a magic bullet at the cost curve aimed at killing cost growth. This attempt is probably misguided at best and a futile waste of resources at worst. CMS’ goal to cut 20% of readmissions without dramatically increasing costs elsewhere is likely to be pure fantasy and will not achieve its stated goals. Indeed, recent efforts by Institute for the Future suggest that using the hospital as the focus of community wellness might not be the best source of change, if even potential welfare improving gains on a societal basis appear to be lying on the table to be realized (Future of Hospitals, 2013). Specialist and subspecialist physicians working in hospitals are expensive, community resources for promotion of health and well-being possibly are cheaper, but not if they are marshaled by, and supporting, these expensive in-hospital resources (Shortell et al., Reference Shortell, Gillies and Anderson1994; Shortell et al., Reference Shortell, Gillies and Devers1995).
From the perspective of trying to reduce cost, continued interest and deployment of programs that are driven by evidence from RCTs and involve the highest cost physicians are most unlikely to save any money, both since the time of these specialists and subspecialists is expensive and they tend to deploy particularly expensive hospital resources. In the same manner, strong incentives may prove futile in reducing readmission rates and exacerbate cost issues. System and payment redesign at this level is itself quite expensive with benefits that are not realized quickly and hospitals that fall behind on measures such as readmissions and lose resources from ‘stick’ incentives will have even fewer excess resources to mount these substantial redesigns.
This does not mean that there is no value in incentive programs to reduce readmissions. If we believe that the value of reducing readmission is important beyond how it might reduce cost, then developing effective care transition programs is important. The current incentives in the system are not necessarily congruent with this, but neither are some of the recently enacted and other proposed stronger incentives in the United States and elsewhere in the world.
There are a number of important research questions and policies that can help to address readmission issues more effectively. One approach to dealing with these issues at the patient level would broaden the focus of bundled payment currently inherent in the DRG hospital payment system to include care delivered outside the hospital. These bundled payment initiatives have been generated on both the supply side (e.g. ACOs) and on the demand side [e.g. Episode Treatment Groups, Prometheus]. Well thought out bundling approaches have promise, but how to define related episodes, severity and level of payment will be key factors in determining how successful they might be (Hockenberry et al., Reference Hockenberry, Burgess, Glasgow, Vaughan-Sarrazin and Kaboli2013). Otherwise, like capitated payments in the 1990s and current readmission reduction policies, these strong incentive approaches are simply going to shift more financial risk to hospitals; and the hospitals most likely to suffer are those who serve a disproportionate share of individuals with behavioral and psychosocial challenges. In this regard, one wonders if it would be better social policy to deal directly with these behavioral and psychosocial challenges rather than chasing after an elusive magic bullet.
Acknowledgement
Both Burgess and Hockenberry are funded investigators at VA HSR&D research centers (COLMR and CADRE). The analysis and opinions expressed in this manuscript do not reflect the official opinion of the Department of Veterans Affairs or the US Government or the NBER. They thank anonymous referees, Dr Joseph Restuccia and Dr Kim Rask for comments on earlier versions. All errors are our own.
