Hospital Mortality Rates: How Is Palliative Care Taken Into Account? J

914 Journal of Pain and Symptom Management Vol. 40 No. 6 December 2010

Special Article Hospital Mortality Rates: How Is Palliative Care Taken into Account? J. Brian Cassel, PhD, Amber B. Jones, MEd, Diane E. Meier, MD, Thomas J. Smith, MD, Lynn Hill Spragens, MBA, and David Weissman, MD Virginia Commonwealth University (J.B.C., T.J.S.), Richmond, Virginia; and Center to Advance Palliative Care (A.B.J., D.E.M., L.H.S., D.W.), New York, New York, USA

Abstract Context. Using mortality rates to measure hospital quality presumes that hospital deaths are medical failures. To be a fair measure of hospital quality, hospital mortality measures must take patient-level factors, such as goals of care, into account. Objectives. To answer questions about how hospital mortality rates are computed and how the involvement of hospice or palliative care (PC) are recognized and handled. Methods. We analyzed the methods of four entities: Centers for Medicare & Medicaid Services ‘‘Hospital Compare;’’ U.S. News & World Report ‘‘Best Hospitals;’’ Thomson-Reuters ‘‘100 TopHospitals;’’ and HealthGrades. Results. All entities reviewed rely on Medicare data, compute risk-adjusted mortality rates, and use ‘‘all-cause’’ mortality. They vary considerably in their recognition and handling of cases that involved hospice care or PC. One entity excludes cases with prior hospice care and another excludes those discharged to hospice at the end of the index hospitalization. Two entities exclude some or all cases that were coded with the V66.7 ‘‘Palliative Care Encounter’’ International Classiﬁcation of Disease, Ninth Revision, Clinical Modiﬁcation diagnosis code. Conclusion. Proliferation of, and variability among, hospital mortality measures creates a challenge for hospital administrators. PC and hospice leaders need to educate themselves and their hospital administrators about the extent to which these mortality rates take end-of-life care into account. At the national level, PC and hospice leaders should take advantage of opportunities to engage these mortality raters in conversation about possible changes in their methods and to conduct further research on this topic. J Pain Symptom Manage 2010;40:914e925. Ó 2010 U.S. Cancer Pain Relief Committee. Published by Elsevier Inc. All rights reserved.

Key Words Mortality, palliative care, hospice, hospital care, quality of care, measurement, coding

Introduction Address correspondence to: J. Brian Cassel, PhD, Virginia Commonwealth University Massey Cancer Center, Commercial and government payers and the 401 College Street, P.O. Box 980037, Richmond, VA general public are demanding information 23298-0037, USA. E-mail: [email protected] about health care quality and costs. Hospital Accepted for publication: July 8, 2010. mortality rates are viewed as a key indicator

Ó 2010 U.S. Cancer Pain Relief Committee 0885-3924/$ - see front matter Published by Elsevier Inc. All rights reserved. doi:10.1016/j.jpainsymman.2010.07.005 Vol. 40 No. 6 December 2010 Hospital Mortality Rates and Palliative Care 915

of quality, and several sources now provide the at best and unethical at worst. As one recent re- ability for the general public to look at risk- port phrased it, ‘‘Holding facilities accountable adjusted mortality statistics and other measures for patient mortality rates is not just unless rele- of quality and outcomes for a given hospital.1,2 vant factors associated with the patient’s care Although not exhaustive, a list of prominent en- are taken into account.’’11(p. 25) That study found tities measuring and disseminating hospital that do-not-resuscitate (DNR) status and pallia- mortality rates includes government agencies, tive care (PC) involvement contributed signifi- such as the Centers for Medicare & Medicaid cantly to the explanation of hospital mortality, Services (CMS); commercial entities, such as above and beyond the standard risk-adjustment U.S. News & World Report (a general interest model: ‘‘These results indicate that DNR and magazine), HealthGrades (a health care ratings PC designations can identify mortality risk at organization), and Thomson-Reuters (a health the margin, controlling for other observed risk care analytics firm); and health care consortia, factors already in the standard model ..For such as Premier, Inc., the Leapfrog Group, example the PC designation identifies patients and University HealthSystem Consortium whose risk of dying is between 9% and 57% (UHC). Hospital mortality is also a metric greater than predicted by the model..’’11(p. 31) used by CMS/The Joint Commission in their The underlying premise of the hospice care ‘‘core measures’’ and, thus, may be incorpo- and PC field is that death, except in catastrophic rated into hospital reimbursement incentive events, is a natural life cycle event; it may be de- e programs.3 6 layed, but it can never be entirely prevented.7,8 The presupposition of using hospital mortal- It is also a fact that many Americans turn to ity rates as measures of hospital quality is that hospitals for their end-of-life care. In this hospital deaths represent medical failure; the context, and given the findings reported re- more risk-adjusted deaths a given hospital cently by Kroch et al.,11 the primary goal of has, the lower its quality of care. Implicit in our study was to determine the extent to this ‘‘death ¼ poor quality’’ equation is the mes- which publicly reported hospital mortality sage that hospitals with more risk-adjusted rates take PC or hospice involvement into ac- deaths than other hospitals could have and count, describe how they do so, and attempt should have prevented the marginal difference to understand the reasoning behind their in mortality. These assumptions have been eval- decisions. e uated in several recent articles.7 9 Secondary to that, we sought to consider the A methodological assumption in hospital implications of these mortality rate analyses for mortality rates is that patient- and hospital- our field, specifically for hospital-based PC level attributes that are not indicative of quality teams and their hospice partners. We know but are correlated with mortality (e.g., age, first hand from working with numerous PC acuity of illness when entering the hospital, programs that there is widespread misunder- or disproportionate share of critically ill pa- standing among hospital administrators about tients) are adjusted for or factored out of the the meaning of hospital mortality scores and equation. Whatever hospital-level variability re- the possible or putative impact of hospice mains unexplained by those factors is assumed care and PC on those scores.12 to be a valid and trustworthy measure of hospital quality of care. This is the essence of risk adjustment and how it is used in equating mortality with hospital quality of care. Methods Others have examined whether these ap- We examined well-known national sources of proaches do an adequate job of factoring out hospital quality or performance data that in- hospital-level characteristics, such as tertiary clude mortality scores (based on hospital claims referral centers receiving large numbers of criti- data) as a critical part of their evaluations. We cally ill patients from other hospitals.10 In this identified more than 15 different entities that article, we focus on patient-level characteristics. calculate hospital mortality rates. We excluded If patient-level characteristics are not adequately benchmarking entities whose data are not gen- factored out, then attributing the remaining var- erally made available to the public (e.g., Premier iability to hospital quality of care is questionable Inc., UHC, and Thomson-Reuters Healthcare). 916 Cassel et al. Vol. 40 No. 6 December 2010

We excluded the Leapfrog Group, which limits The relative weight given to the mortality its mortality scores to several high-risk surgeries, rate, if a composite score for the hospital such as esophageal resection, pancreatic resec- is created. tion, and heart bypass surgery that rarely involve A brief description of risk adjustment is war- hospice care or PC. We excluded state-based ranted. Patient-level risk adjustment controls entities (e.g., Virginia Health Information). for how sick those patients were and other pa- We excluded entities that repackage existing tient characteristics, such as sex and age (to CMS ‘‘Hospital Compare’’ mortality data (e.g., our knowledge, none adjust for socioeconomic Commonwealth Fund). Four entities met our status at the patient level). Hospital-level risk- criteria for this review: CMS ‘‘Hospital Com- adjustment controls for whether the sickest pa- pare,’’ U.S. News & World Report ‘‘Best Hospi- tients tend to be seen in greater proportion at tals,’’ Thomson-Reuters ‘‘100 Top Hospitals’’ some hospitals rather than others. Determin- (a distinct entity from the Thomson-Reuters ing how sick the patients were is based on Healthcare benchmarking service), and the International Classiﬁcation of Diseases, HealthGrades (see Table 1 for brief descriptions Ninth Revision, Clinical Modiﬁcation (ICD- of these entities). 9-CM) diagnosis codes entered in the adminis- We accessed the methodological descrip- e trative or claims data for that hospitalization by tions from publicly available Web sites13 16 that hospital. for each of these four entities to ascertain How exactly these entities perform their risk the general approach to mortality rate analy- adjustment is a critical issue, but it is also a very ses. We coded the following key elements of complex one. The focus of this article is lim- mortality reporting: ited to the consideration of hospice care and The most recent mortality rate analysis PC and not on a detailed description or evalu- released; ation of these various risk-adjustment ap- The years of data that served as the basis proaches from a statistical perspective, which e for that analysis; can be found elsewhere.1,2,9,10,17 19 The source of the mortality data analyzed; In the context of relying on claims data as The number of specialties, diseases/con- the basis for creating mortality scores, hospice ditions, or procedures analyzed; care or PC involvement could be ascertained How much information about mortality is by various means and then interpreted in sev- released to the public; eral different ways. Prior hospice involvement The window of time for death relative to could be ascertained if the entity is using lon- hospital admission; gitudinal, patient-centered data (as opposed The general risk-adjustment methodology to anonymous data limited to the index hospi- used; talization). Another method of taking hospice Whether involvement of PC or hospice involvement into account is to determine care is the basis for exclusion or is incor- which cases were discharged into hospice at porated into risk adjustment; the end of the index admission. ‘‘Disposition

Table 1 Description of Entities Reviewed for Mortality-Reporting Methodology Entity Characteristics

CMS ‘‘Hospital Compare’’ Multidimensional, publicly reported data on hospital processes of care, outcomes (risk- adjusted rates of death and readmissions), and patient satisfaction. Web site allows comparison of three hospitals at a time. U.S. News & World Report Annual ranking of hospitals by specialty, using mix of Medicare data, information from ‘‘America’s Best Hospitals’’ American Hospital Association and other sources, and primary data (reputation scores based on specialist surveys). Web site shows detailed scores for every hospital. HealthGrades Annual ranking of hospitals by numerous conditions and procedures, based on complication and mortality rates. Web site allows comparison of all hospitals in a given area (city and state), one condition or procedure at a time. Thomson-Reuters ‘‘100 Top Annual rankings of hospitals generally and separately for cardiovascular care. Hospitals’’ Multidimensional, but details are not publicly reported. Vol. 40 No. 6 December 2010 Hospital Mortality Rates and Palliative Care 917

at discharge’’ is a uniform data point in all specialties. HealthGrades provides up to three hospital billing data, and there are two different mortality statistics (admission sur- hospice-specific codes (inpatient hospice or vival, 30-day mortality, and 180-day mortality) otherwise). for each of 21 conditions and procedures. A third approach would be to determine Thomson-Reuters provides one report on hos- whether the V66.7 ‘‘Palliative Care Encounter’’ pital care generally and a separate report on ICD-9-CM code was present among diagnosis cardiovascular care specifically. codes submitted with the hospital bill. Because All four entities adjust their analyses of mor- the Medicare data files that these entities use tality based on patient characteristicsdand include only the primary diagnosis and first two entities also include some hospital-level nine secondary diagnoses,20,21 the V66.7 code risk adjustment as well. All four entities use would need to be submitted relatively high different methodologies for risk adjustment among the dozens of secondary diagnosis (Table 2). (As of March 2010, Thomson- codes typically submitted with complex cases, Reuters includes two different mortality analy- such as those ending in death. ses in its ‘‘100 Top Hospitals’’ program. Its own Simplified, hypothetical case vignettes were risk-adjusted mortality index of death within developed to illustrate whether each entity the index admission is described here and in would include or exclude a given case from Tables 2 and 3, but it also includes the CMS its mortality analysis. This gave us a means by ‘‘Hospital Compare’’ 30-day mortality as a sepa- which feedback could be obtained from these rate component of the overall hospital score.) entities’ methodologists on the accuracy of For most entities, the codes used in risk- our interpretation of their various approaches adjustment are those associated with the index regarding hospice care and PC involvement. admission, but in the case of CMS ‘‘Hospital One author (J. B. C.) sent draft versions of Compare,’’ the risk adjustment draws on condi- Tables 2 and 3 to the methodological contact tions documented in any inpatient or outpa- person at each entity, seeking confirmation tient setting up to 12 months before the index that the data elements and interpretations admission. The methodological details of these were correct. Representatives of all four enti- risk-adjustment approaches are beyond the ties (or their independent analytic groups) scope of this article but are accessible for e (Appendix) responded, and revisions were review.13 16 made accordingly. Hospice and Palliative Care Involvement As indicated in Table 2, currently, of these four entities, only CMS is accessing longitudi- Results nal, patient-centered data across multiple set- Our summary of the mortality rate methods tings and episodes for the year before the for the four entities is shown in Table 2. The index admission and is excluding cases based most consistent elements are that all four com- on prior (or on first day of admission) hospice pute hospital mortality rates using Medicare enrollment. data, the rates are risk adjusted, and mortality Exclusion of cases that were discharged is ‘‘all cause’’ (the cause of death is not neces- from the hospital to hospice is an approach sarily related to the disease or hospital care). used only by HealthGrades at this time, but Apart from those commonalities, there is those exclusions are applied only for medical wide variability across these entities concern- conditions, such as sepsis, pancreatitis, and ing most of the elements examined. For exam- chronic obstructive pulmonary disease. Hos- ple, they vary greatly in the number and kind pice cases are not excluded from their evalua- of conditions, procedures, or specialties ana- tion of procedural cases, such as heart bypass lyzed. CMS ‘‘Hospital Compare’’ describes surgery, valve replacement surgery, or gastroin- three conditions for which 30-day mortality is testinal surgeries and procedures. reporteddpneumonia, heart failure, and The V66.7 ‘‘Palliative Care Encounter’’ code heart attackdcompared statistically with the is used by Thomson-Reuters and HealthGrades national average. U.S. News provides 30-day as the basis for exclusion (again, for Health- mortality statistics for 12 medical and surgical Grades, this exclusion applies only to medical 918

Table 2 Methods Used in Mortality Analyses US News & World Report Thomson-Reuters Characteristic CMS ‘‘Hospital Compare’’13 ‘‘Best Hospitals’’14 ‘‘100 Top Hospitals’’15 HealthGrades16

Most recent (as July 2009 July 2009 March 2010 October 2009 of June 2010) Years analyzed July 2005eJune 2008 2005e2007 2007e2008 2006e2008 Data analyzed MedPar data (top 9 diagnosis and MedPar data (top 9 diagnosis and MedPar data (top 9 diagnosis and MedPar data (top 9 diagnosis and for mortality top 6 procedure codes from top 6 procedure codes from top 6 procedure codes from top 6 procedure codes from score index admission and prior year) index admission) index admission) index admission) Diseases or Pneumonia, heart attack, heart 12 specialties scored using Hospital level for their general 12 diagnoses and 9 procedures specialties failure mortality data and other report; separate report on criteria, with top 50 ranked cardiovascular diseases What the general A percentage mortality rate for Mortality index for each of these Top-scoring hospitals are listed, Ratings indicating mortality rate public can see each condition and ability to specialties for any given but detailed scores (e.g., are ‘‘best,’’ ‘‘as expected,’’ or compare any given hospital hospital, and reputation, other mortality) are provided only to ‘‘poor,’’ for each condition/

mortality rate with national score components, and total those purchasing the full report procedure; list of top-scoring al. et Cassel average mortality rate score hospitals across all conditions/ procedures Death within . 30 days from admission 30 days from admission Index admission Index admission Admit þ 30 days from discharge Admit þ 180 days from discharge Risk adjustment Two-level (hierarchical) modeling, Patient level only, based on Logistic regression incorporating Patient-level only methodology including both patient variables 3 M All Patient Refined both patient- and hospital-level and hospital-level clustering of Diagnosis-Related Groups variables severely ill cases. Uses patient (APR-DRG) data from year before admission and from index admission, for risk adjustment Hospice Excluded if Medicare Hospice Not excluded; not incorporated Prior or concurrent use not Prior or concurrent use not utilization Benefit enrollment in previous into risk adjustment excluded or incorporated into excluded or incorporated into considered? 12 months or first day of risk adjustment. Admissions risk adjustment. Admissions 2010 December 6 No. 40 Vol. admission; not otherwise ending in discharge to hospice discharged to hospice excluded excluded or incorporated into are survivors of the index in 12 diagnosis-based cohorts; risk adjustment admission (only index not excluded or otherwise admission analyzed, not 30-day incorporated into risk mortality) adjustment for procedure cohorts Palliative care Not excluded; not incorporated Not excluded; not incorporated Excludeddbased on V66.7 Excluded in 12 diagnosis-based considered? into risk adjustment into risk adjustment Palliative Care Encounter ICD-9 cohorts; not excluded or code otherwise incorporated into risk adjustment for procedure cohorts Vol. 40 No. 6 December 2010 Hospital Mortality Rates and Palliative Care 919

conditions and not procedures). This code is not among those incorporated by CMS into its risk adjustment22 and does not seem to af- fect the U.S. News & World Report’s risk adjustment. Apart from exclusion, these entities also have the option to incorporate PC or hospice care involvement in risk adjustment, but none are doing so currently. hospital score for thoseconditions/procedures. 21 Other aspects of care (e.g.,care intensive unit stafﬁng through Leapfrog Group) are also provided

Mortality rate comprises the entire Hypothetical Scenarios Hypothetical cases were developed to explore whether cases involving hospice care and/or PC involvement are included or excluded from the mortality analysis for each entity. All four scenarios involve an elderly patient with pneumonia, hospital entry, and death within 30 days of that hospital entry.

Scenario 1: Palliative Care Consultation. An elderly patient is admitted to the hospital with

reported, but their proprietary measure of index admission mortality is weighted attotal, 11% and of CMS’s ‘‘Hospital Compare’’ 30-day mortality is another 6% of the total pneumonia. On the seventh day of admission, No composite score for hospitals is he is seen by the hospital’s PC consultation team for help with symptom control and clari- ﬁcation of goals of care; this was documented with the V66.7 ‘‘Palliative Care Encounter’’ ICD-9-CM code in the claims data (and this happens to be positioned among the top nine diagnoses submitted to Medicare with the hospital claim). The patient dies in the hospital one week later. The patient was not enrolled in hospice at any point. CMS and U.S. News & World Report would

score include this death in their mortality analyses. Mortality is 32.5% of composite HealthGrades and Thomson-Reuters would exclude the case because of the V66.7 code. Note, only the top nine diagnosis codes are available to them through the Medicare Pro- vider Analysis and Review (MedPAR) data ﬁles used by all four entities.

Scenario 2: Hospital Discharge with Immediate General Inpatient Hospice Care Admission.An elderly patient is admitted to a hospital for pneumonia. On the fourth day of admission, measures include readmission rates, processes of care,patient and satisfaction)

No composite score (other he is enrolled in the Medicare Hospice Beneﬁt (General Inpatient [GIP] Care). Administra- tively, the hospital discharges the patient, and he is simultaneously enrolled in the Medicare Hospice Beneﬁt; his remaining care is billed under hospice, even though the patient never physically leaves the hospital bed.23,24 The pa- weight within hospital’s composite score

Mortality’s tient dies the following day. Medical records 920 Cassel et al. Vol. 40 No. 6 December 2010

Table 3 Inclusion/Exclusion Summary CMS ‘‘Hospital US News & World Report Thomson-Reuters Scenarios Compare’’ ‘‘Best Hospitals’’ ‘‘100 Top Hospitals’’ HealthGrades

1. Palliative care consultation Include Include Excludea Excludea 2. Discharged to hospice Include Include Excludeb Excludec 3. Hospice revoked Excluded Include Include Include 4. Hospice GIP admission NAe NAe NAe NAe GIP Hospice ¼ Patient enrolled in the Medicare Hospice Benefit, admitted to a hospital setting (for, e.g., relief of acute symptoms that could not be managed in home setting), but remains under the Medicare Hospice Benefit.23,24 Explanation of exclusions are as follows. aExcluded if V66.7 ‘‘Palliative Care Encounter’’ ICD-9-CM code present (must be among top 9 diagnoses submitted). bTechnically, patient survived this acute admission; this entity uses admission mortality, not 30-day mortality. cCases excluded if discharged to hospice. dCases excluded if enrolled in hospice in previous 12 months or on first day of index admission. eNot acute hospital admission, therefore, not included in hospital mortality rate reporting. do not include a PC encounter (V66.7) code Medicare Hospice Benefit, dies after six days during the acute admission. in the hospital. Thomson-Reuters would consider this pa- In this scenario, the patient has entered the tient to have survived the inpatient admission, hospital, but his care is billed under the hos- because he was ‘‘discharged’’ to hospice even pice benefit; this is not seen as an acute admis- though he never left the hospital bed. The sion by any of the entities. In other words, this three other entities use 30-day mortality. Two utilization is not submitted to Medicare as an of them, CMS and U.S. News & World Report, acute hospitalization, and this death, although do not take into account the disposition at dis- occurring within the hospital’s walls, is techni- charge, essentially ignoring the involvement of cally not a hospital death. hospice at the end of acute hospitalization, The approach each entity would takedto in- and would include this death in their mortality clude or exclude such deaths in their hospital analyses. HealthGrades excludes cases that mortality analysesdis summarized in Table 3. were discharged to hospice, at least for medi- The only scenario that all four entities would cal conditions, such as pneumonia. agree upon is that a hospice admission to an inpatient setting is not an acute hospitalization Scenario 3: Hospice Revoked. An elderly patient and is not part of hospital mortality rate analy- is managed with home hospice services paid by ses at all (hospice care is hospice care, regard- the Medicare Hospice Benefit for six weeks. less of the setting in which it is provided). Of The patient becomes acutely ill at home, and the four entities examined, HealthGrades the family decides to admit the patient to the and Thomson-Reuters are most similar to hospital for life-prolonging treatment; he one another, and CMS and U.S. News are rela- revokes his Medicare Hospice Benefit. The tively similar to each other, in their inter- patient dies during the admission without pretations of these cases involving hospice further involvement of hospice care or PC. care or PC. Based on this scenario, three of the four entities would include this patient in their mortality-reporting analysis. Only CMS ‘‘Hospi- Discussion tal Compare’’ would exclude this case, because Many conceptual, methodological, and sta- the patient was enrolled in the Medical Hos- tistical issues are involved in determining pice Benefit in the previous 12 months. whether it is valid to use general mortality rates to differentiate hospitals in terms of quality of Scenario 4: General Inpatient Care Under Hospice. care. One such issue is whether mortality An elderly hospice patient with pneumonia is scores account for the inevitability of death, e admitted to the hospital under the Medicare for cases where death was inevitable.7 9,11 Hospice Benefit for GIP care23,24 to address Hospice involvement is, perhaps, the clearest difficult symptoms that were not manageable signal that the patient, family, and health at home. The patient, who remains on the care providers recognize that death is Vol. 40 No. 6 December 2010 Hospital Mortality Rates and Palliative Care 921

approaching and is inevitable. Involvement exclusion of such cases in the computation of of hospice is documented in at least two data hospital mortality rates. systems: the Medicare enrollment database With that said, a second thread was that and Medicare claims data. Only CMS ‘‘Hospital these entities differ as to whether the data Compare’’ recognizes prior hospice involve- available to them can indeed provide a clear ment (based on matching cases to the Medicare signal about goals of care during a specific ep- Enrollment Database), and only HealthGrades isode. Concerns were raised about the rela- recognizes, at least for medical conditions, hos- tively infrequent and inconsistent use of the pice involvement at the end of an admission V66.7 ‘‘Palliative Care Encounter’’ code (recall (based on the ‘‘disposition at discharge’’ field that these entities see only the top nine sec- in the hospital claims data). ondary diagnoses currently available in Medi- Involvement of hospital-based PC profes- care data sets). There are multiple obstacles sionals during the index admission would not in obtaining hospice enrollment or claims in and of itself be a clear signal that death is data and matching that with patients’ hospital- imminent for those patients, as PC is not mutu- ization data. The MedPAR data set includes ally exclusive with curative or life-prolonging a field indicating the time span from hospital- treatment of the underlying disease or condi- ization to death, making it useful for hospital tion. However, the description currently used mortality analyses, but does not include bene- for the appropriate use of the V66.7 ‘‘Palliative ficiary identifiers that would allow it to be Care Encounter’’ ICD-9-CM code does seem to linked to hospice claims or enrollment. Other be focused on terminal or hospice care,25,26 identifiable data sets available from Medicare and therefore, its use as the basis for exclusion for research projects are not obtainable by by Thomson-Reuters and HealthGrades does commercial entities for commercial purposes seem consistent. Two benchmarking entities, (see criterion #7 of CMS’s ‘‘Criteria for Review UHC and Premier, Inc., have reported else- of Requests for CMS Research Identifiable where that this code or other documentation Data’’ at http://www.cms.gov/PrivProtected of PC involvement makes a statistically signifi- Data/02_Criteria.asp#TopOfPage). Medicare’s cant contribution to the explanation of vari- Limited Data Set, on the other hand, which ance in observed hospital deaths.11,27 can be requested by commercial entities, Currently, the V66.7 ‘‘Palliative Care En- does not provide the means by which the counter’’ code appears to be geared toward time interval could be determined between end-of-life care (only) and not concurrent various health care utilization episodes (e.g., management of pain and other symptoms.25,26 hospice utilization and hospital inpatient ad- Although this is only part of what PC teams do missions) or between utilization and death. in hospitals, this interpretation of the code Third, these representatives made the argu- does speak to the issue of goals of care and ment that if they are going to use hospital mor- is, thus, relevant to the discussion here about tality rates, then they must be careful not to hospital mortality rates. exclude hospital deaths too liberally, which would pose both conceptual and statistical threats in evaluating hospitals on their mortal- Commonalities and Differences in Handling ity rates. As others have pointed out, a condi- Palliative Care and Hospice Care tion or procedure that generally has few Why do these entities computing mortality deaths in the numerator will not be a good rates have so much variation regarding the in- candidate for these analyses.9 volvement of hospice care and PC? Based on Fourth, mortality rate analyses attempt to our communications with the methodological control for patient- and hospital-level factors contacts at each entity (or with the third party that occur before hospital care (not after- actually conducting the analyses), there were ward); excluding cases that involved hospice four common threads regarding their inten- only at the very end of an admission may create tions and methods. First, all acknowledged an incentive for hospitals to use hospice as that a clear signal that the entire episode of a way to hide or cloak problems with their hospitalization was focused solely toward com- quality of care earlier in the admission. For fort care would be a cause for considering the that reason, CMS excludes cases with hospice 922 Cassel et al. Vol. 40 No. 6 December 2010

involvement only if that involvement occurred was present, only 41.1% had it positioned high in the year before or during the first day of acute enough to be included in the Medicare data hospitalization. HealthGrades takes a different sets used by the four entities reviewed here.27 approach to achieve the same end: Hospice in- Thus, the V66.7 code has been found to be an volvement is a cause for exclusion only for med- important factor in risk adjustment of hospital ical conditions and not for surgical procedures. mortality rates, but it is typically positioned too Both approaches try to navigate among the var- low to be included in the Medicare data sets ious hazards while preserving mortality rates as used for the hospital mortality rate analyses re- a useful and universal measure. viewed here. (Its low positioning may be because of the fact that it is not the kind of The Role of the V66.7 ‘‘Palliative Care secondary diagnosis that affects the assignment Encounter’’ Code of DRGs and, thus, does not contribute signifi- Based on our own research projects, we can- cantly to hospital reimbursement.) This speaks not disagree with the assessment that the V66.7 to the concern raised by others whether ‘‘coding ‘‘Palliative Care Encounter’’ code does not ap- depth’’dthat is, whether all relevant predictors pear consistently in the top nine diagnosis slots of mortality are actually included among codes submitted to Medicare for reimbursement. In used in analysesdcontributes significantly to our multihospital assessment of the financial bias.9,29 And it also speaks to the concern raised impact of PC, 3051 PC consultations were docu- by others11 that inadequately capturing and an- mented separately from the claims data, but alyzing patient variables, such as goals of care, only 558 (18%) of these had a V66.7 code would make it unjust to equate differences in among the top nine diagnosis positions in the hospital mortality rates with real differences in claims data.28 Similarly, Kroch et al.11 relied hospitals’ quality of care. on chart reviews to identify PC involvement, because the V66.7 code was not a sufficient means Summing Up the Conundrum to ascertain that. The following observations emerge from Although this may at first seem to provide this study: a rationale for all entities to ignore the code entirely, consider also that UHC has found To be a valid and ethical measure of differ- that the V66.7 code is a strong contributor to ences in hospital quality of care, mortality the prediction of hospital mortality for more rate analyses must take all patient factors than 50 disease-specific models,27 perhaps be- into account. cause it does not limit its analyses of mortality One such factor is the goals of care for rates to the top nine diagnoses (it uses up to a given hospitalization: was it focused on 99). Indeed, they indicated that it is a particu- comfort care rather than curative or life- larly strong predictor. prolonging care? In most of the diagnosis-related group (DRG)- DNR status and PC involvement have been specific models where PC (represented by the shown to predict mortality.11 V66.7 code) was included as a variable, it had When not limited to Medicare data limita- one of the highest positive coefficients, which tions, the V66.7 ‘‘Palliative Care Encounter’’ would result in a higher expected probability code has been shown to predict mortality. of mortality when coded. For two discharges in In most of the cases, the V66.7 code is po- the same DRG with exactly the same demo- sitioned too low to be visible to entities graphics and list of diagnosis and procedure relying on Medicare data sets. (Medicare codes except for the PC code, the expected plans to increase the number of diagnoses probability of mortality for the discharge with it accepts with hospital bills starting in the PC code would be higher than the probabil- January 2011,30 but it will be another ity of mortality for the discharge without that three to four years before the entities us- code (Jodi Neikirk, Clinical Data & Informatics, ing Medicare data sets will have several University Healthsystem Consortium, October years’ worth of expanded diagnosis data.) 24, 2008; personal communication). Impor- Most entities evaluating hospitals on mor- tantly, analyses of the UHC national data set re- tality rates rely solely on Medicare data, veal that among cases where the V66.7 code and some conclude that PC involvement Vol. 40 No. 6 December 2010 Hospital Mortality Rates and Palliative Care 923

(based on the limited incidence of it visi- Conclusion ble to them) does not help to predict Measuring hospital quality using mortality mortality. rates is a contentious topic. Confusion is inevi- Apart from the V66.7 code, these entities table when each entity measures mortality dif- have a variety of reasons (such as identifi- ferently. The interface of hospital mortality able data not being available to them and rate data and hospice or PC services is com- fear of creating new unintended conse- plex. Given the general level of confusion quences) why they do not incorporate hos- and interest in this topic in hospitals today, hospice utilization into their exclusion pice and PC professionals will benefit from criteria or risk adjustment. fully exploring the dimensions of patient- One way to undo a knot is to cut all the way centered quality care reflected by both mortal- through it, and several experts have recom- ity data and PC impact data and by sharing mended that general hospital mortality rates these analyses with hospital executives. PC (as measures of hospital quality) simply go leaders may also be able to make a strong case away.9 Not only is it unlikely that official public that PC outpatient clinics will do more to man- agencies, such as Medicare, will eliminate age patients appropriately with fewer inpatient e hospital mortality rates from their roster of admissions toward the end of life.31 33 metrics, it is even more unlikely that private At the national level, there is a need for bet- entities will cease and desist in this practice. ter research to model the statistical impact of Another approach would be for these enti- inclusion or exclusion of hospice- and PC- ties to limit mortality rate analyses to proce- involved cases, or the incorporation of those dures (such as the high-risk procedures that variables into risk adjustment. Would such the Leapfrog Group uses) or acute illnesses practices reduce or increase bias or inflate or (such as acute myocardial infarction) that do deflate mortality rates for hospitals with vary- not overlap much with end-of-life care for ing degrees of hospice and PC utilization? As chronic conditions. Again, there is currently a field, we could also simultaneously engage no external incentive for most of these entities the reporting entities in a dialogue to explore to limit their analyses in this way. how best to recognize that many patients use An incremental improvement would be for hospitals for their end-of-life care and that Medicare to capture all diagnoses submitted goals of care need to be factored into the equa- with hospital bills and not arbitrarily limit those tion that hospital mortality is synonymous with to nine or 25 diagnoses. This would be one way hospital quality. Such a dialogue may lead to to resolve the conflict inherent in our observa- revision of the inclusion and exclusion criteria tions mentioned earlier that, on the one hand, and risk-adjustment models used for comput- PC involvement does make analysis and inter- ing hospital mortality rates. pretation of mortality rates more meaningful,11 but on the other hand, the V66.7 code is usually positioned too low among secondary diagnoses to be available in Medicare data sets. Indepen- Disclosures and Acknowledgments dent of that, another solution is to encourage The authors were not supported by grants or hospitals to consider positioning manually the other financial support for this project, and V66.7 code higher so that it is visible for consid- they declare no conflicts of interest. eration by external entities evaluating mortality The authors wish to thank the methodologi- (or other metrics). cal contacts at the four entities who responded We would encourage all commercial entities to multiple queries, without whom this study performing these analyses to try to convince would have been impossible. Virginia Common- CMS to provide beneficiary identifier informa- wealth University (VCU) Medical Center tion in the MedPAR data set that would allow leaders, including Gordon Ginder, Mary Ann them to evaluate prior or concurrent hospice Hager, Dale Harvey, Marcos Irigaray, Laura involvement, and beyond that, to use 12 Kreisa, Cathy Mays, and Sheldon Retchin, com- months of prior inpatient and outpatient utili- mented on early drafts and presentations of this zation in adjusting for risk of mortality. project and collaborated on analyses of this 924 Cassel et al. Vol. 40 No. 6 December 2010

issue at VCU. The UHC provided an opportu- 10. Rosenberg AL, Hofer TP, Strachan C, Watts CM, nity to present much of this material to UHC Hayward RA. Accepting critically ill transfer pa- members and collaborated with us on analyses tients: adverse effect on a referral center’s outcome and benchmark measures. Ann Intern Med 2003; of UHC data regarding mortality; the authors 138:882e890. thank Jeannine Finn, Michael Oinonen, Kathy 11. Kroch EA, Johnson M, Martin J, Duan M. Mak- Vermouch, and Julie Cerese. The authors are ing hospital mortality measurement more meaning- grateful to hospice and PC leaders who pro- ful: incorporating advance directives and palliative vided the impetus for this project and feedback care designations. Am J Med Qual 2010;25:24e33. on drafts of this manuscript or other presenta- 12. Cassel JB, Hager MA, Clark RR, et al. Concen- tions, including but not limited to Andrew trating hospital-wide deaths in a palliative care Billings, Meg Campbell, David Casarett, Judy unit: the effect on place of death and system-wide e Lentz, Dale Lupu, John Mastrojohn, Sean mortality. J Palliat Med 2010;13:371 374. Morrison, Judi Lund Person, Donald Schu- 13. QualityNet. Mortality measures overview. macher, Carol Sieger, Steven Smith, and Rodney Publicly reporting risk-standardized, 30-day mortal- Tucker. Finally, the authors sincerely thank the ity measures for AMI, HF, and PN. Available from http://qualitynet.org/dcs/ContentServer?c¼Page& reviewers at the Journal of Pain and Symptom Man- pagename¼QnetPublic%2FPage%2FQnetTier2&cid agement, whose thoughtful critiques helped ¼1163010398556. Accessed July 6, 2010. them make improvements to this article. 14. U.S. News & World Report. ‘‘America’s Best Hospitals’’ risk-adjusted mortality rate methodology. Available from http://static.usnews.com/documents/ health/2009-best-hospitals-methodology.pdf.Accessed References July 6, 2010. 1. Krumholz HM, Normand SLT. Public reporting 15. Thomson-Reuters. ‘‘100 Top Hospitals’’ risk- of 30-day mortality for patients hospitalized with adjusted mortality rate methodology. Available acute myocardial infarction and heart failure. Circu- from http://img.en25.com/Web/ThomsonReuters/ lation 2008;118:1394e1397. Thomson%20Reuters%20RAMI.pdf. Accessed July 6, 2. Hofer TP, Hayward RA. Identifying poor-quality 2010. (Visitors must ﬁrst register at http://www. hospitals: can hospital mortality rates detect quality 100tophospitals.com/research/). problems for medical diagnoses? Med Care 1996;34: 16. HealthGrades. Risk-adjusted mortality rate meth- 737e753. odology. Available from http://www.healthgrades. com/media/DMS/pdf/Americas50BestHospitals 3. Lindenauer PK, Remus D, Roman S, et al. Public Report2009.pdf. Accessed July 6, 2010. reporting and pay for performance in hospital quality improvement. N Engl J Med 2007;356:486e496. 17. Iezzoni LI, Ash AS, Coffman GA, See also editorial (Epstein, pp. 515e517) and corre- Moskowitz MA. Predicting in-hospital mortality: spondence: 356(17):1782e1784. a comparison of severity measurement approaches. Med Care 1992;30:347e359. 4. Mehrotra A, Damberg CL, Sorbero MES, Teleki SS. Pay for performance in the hospital set- 18. Ash A, Ellis RP, Pope GC, et al. Using diagnoses ting: what is the state of the evidence? Am J Med to describe populations and predict costs. Health Qual 2009;24:19e28. Care Financ Rev 2000;21:7e28. 5. Rosenthal MB, Fernandopulle R, Song HSR, 19. Blumenthal D, Weissman JS, Wachterman M, Landon B. Paying for quality: providers’ incentives et al. The who, what, and why of risk adjustment: for quality improvement. Health Aff 2004;23: a technology on the cusp of adoption. J Health Polit 127e141. Policy Law 2005;30:453e473. 6. Ryan AM. Hospital-based pay-for-performance 20. Research Data Assistance Center. Available from in the United States. Health Econ 2009;18:1109e http://www.resdac.umn.edu/. Accessed July 2, 2010. 1113. 21. Medicare Provider Analysis and Review (Med- 7. Holloway RG, Quill TE. Mortality as a measure PAR) File. Available from http://www.cms.gov/ of quality: implications for palliative and end-of- IdentiﬁableDataFiles/05_MedicareProviderAnalysis life care. JAMA 2007;298:802e805. andReviewFile.asp. Accessed September 29, 2010. 22. QualityNet. 2009 measures maintenance techni- 8. Shojania KG, Forster AJ. Hospital mortality: cal report: acute myocardial infarction, heart failure, when failure is not a good measure of success. and pneumonia 30-day risk-standardized mortality CMAJ 2008;179:153e157. measures. Available from http://qualitynet.org/dcs/ 9. Lilford R, Pronovost P. Using hospital mortality ContentServer?c¼Page&pagename¼QnetPublic%2 rates to judge hospital performance: a bad idea that FPage%2FQnetTier3&cid¼1163010421830. Accessed just won’t go away. BMJ 2010;340:c2016. July 2, 2010. Vol. 40 No. 6 December 2010 Hospital Mortality Rates and Palliative Care 925

23. Rosson CL, White JC. Inpatient hospice care consultation programs. Arch Intern Med 2008;168: (JPM patient information sheet). J Palliat Med 1783e1790. 2006;9:828. 29. Iezzoni LI, Foley SM, Daley J, et al. Comorbidities, 24. Centers for Medicare & Medicaid Services. Medi- complications, and coding bias. Does the number of care systems edit refinements related to hospice d diagnosis codes matter in predicting in-hospital mor- services JA6778. Available from http://www.cms. tality? JAMA 1992;267:2197e2203. gov/ConstractorLearningResources/downloads/JA6 778.pdf. Accessed September 29, 2010. 30. Centers for Medicare & Medicaid Services. Ad- 25. Centers for Disease Control and Prevention. In- ditional ICD-9 codes analysis and processing direc- ternational classification of diseases, ninth revision, tion (institutional claims only). 2010. Available clinical modification (ICD-9-CM). Available from from http://www.cms.gov/Transmittals/downloads/ http://www.cdc.gov/nchs/icd/icd9cm.htm. Accessed R648OTN.pdf. Accessed September 29, 2010. September 29, 2010. 31. Brumley RD, Enguidanos S, Jamison P, et al. In- 26. American Hospital Association. Official coding creased satisfaction with care and lower costs: results guidelines, 4th quarter. Chicago, IL: AHA Coding of a randomized trial of in-home palliative care. Clinics, 1996. pp. 47e48. J Am Geriatr Soc 2007;55:993e1000. 27. Cassel JB. Impact of palliative care reporting on publicly reported performance data. University 32. Brumley RD, Enguidanos S, Cherin DA. Effec- tiveness of a home-based palliative care program HealthSystem Consortium Webinar. Available from e https://www.uhc.edu/34895.htm. Accessed May for end-of-life. J Palliat Med 2003;5:715 774. 19, 2010. 33. Gade G, Venohr I, Conner D, et al. Impact of an 28. Morrison RS, Penrod J, Cassel JB, et al. Cost sav- inpatient palliative care team: a randomized control ings associated with hospital-based palliative care trial. J Palliat Med 2008;11:180e190.

Appendix

Methodological Experts Contacted at Each Entity

For CMS: Dr. Elizabeth Drye (Yale University), Angela Merrill (Mathematica-MPR, Inc.). For U.S. News: Dr. Murrey Olmsted (RTI International). For HealthGrades, Inc.: Kristin Reed, MPH. For Thomson-Reuters: Dr. David Foster, Jean Chenoweth.