Mortality as a Measure of Quality: Title and subTitle BreakImplications for Palliative and End-of-Life Care

Mortality as a measure of quality has made a comeback. In 1986, the Health Care Financing Administration (now the Centers for Medicare & Medicaid Services [CMS]) released hospital-specific mortality rates to the public, but abandoned those efforts in 1993 given concerns about the validity of the comparisons. In 2002, the Agency for Healthcare Research and Quality released 15 administratively driven inpatient mortality indicators. Twenty-nine report cards now contain information about hospital mortality.¹ In addition, the Institute of Medicine in 2006 endorsed the inclusion of disease-specific mortality as 1 of the 2 outcome measures for consideration in developing a national system for performance measurement.²

The addition of mortality to the measures publicly reported by CMS and the looming reality of pay-for-performance have catapulted mortality back into national focus. In June 2007, CMS began reporting hospital-specific risk-standardized 30-day mortality rates for acute myocardial infarction and congestive heart failure with reporting of mortality rates for community-acquired pneumonia likely to follow (includes patients at least 65 years of age who die from any cause within 30 days of the admission date).³ As a result, hospitals have been co-opted into chasing the dizzying array of publicly available mortality measures to minimize the real or perceived effect this information may have on their reputation and market share.

Mortality has been criticized as a measure of quality for years and debates about methods of risk adjustment are almost clichéd. Considerable attention has been given to the unintended consequences of publicly reporting patient outcomes and mortality, including avoiding sicker patients, discounting patient preferences, gaming behavior, and diverting priorities from more worthy improvement projects.⁴

Absent from these discussions is a more fundamental question: what is the quality of care for the 1.16 million individuals who die each year in US hospitals (over 40% of all US deaths)?⁵ Deaths from errors and deaths after deliberate decisions not to pursue unwanted life-prolonging treatments are polar opposite outcomes. Conflating the 2 may have unintended consequences of undermining expert palliation of dying patients and rewarding overly aggressive treatment when it is not desired or beneficial.

Hospital mortality data obscure multiple factors that may contribute to short-term mortality at the individual level. Short-term mortality differences between hospitals may be due to variations in effective care, safe care, preference-sensitive care, and supply-sensitive care.

Effective care is “based on systematically acquired evidence to determine whether an intervention, such as preventive service, diagnostic test, or therapy, produces better outcomes than alternatives.”⁶ CMS-endorsed performance measures (effectiveness measures) correlate poorly with risk-adjusted mortality rates. More than 90% of the variation in reported risk-adjusted 30-day mortality rates for acute myocardial infarction, congestive heart failure, and pneumonia is not explained by the most evidence-based performance measures in use today.^{7 - 8}

Safety is freedom from unintentional injury. Safe care practices include a combination of structure and process measures embedded within the context in which clinical care is delivered.⁹ The evidence base to support safe care is weaker than effective care. However, when safe care is properly delivered, the likelihood of patient harm, including unexpected death, should be reduced. The National Patient Safety Goals endorsed by the Joint Commission on Accreditation of Healthcare Organizations and the 30 Safe Practices for Better Health Care endorsed by the National Quality Forum reflect the state of this science.

Although deaths from unsafe care occur, measuring deaths attributable to unsafe care has been a challenge.^{9 - 10} The difficulty in estimating preventable deaths has manifested in the debate and uncertainties regarding the highly touted estimate of 44 000 to 98 000 deaths attributable to medical error each year.¹¹ But even if deaths from unsafe care could be unambiguously measured and the high-end estimates of annual deaths due to medical error are correct (approximately 100 000 of the 1.16 million individuals who die each year in US hospitals), unsafe care would still account for less than 10% of all inpatient deaths.

Preference-sensitive care occurs when decisions require significant trade-offs among the available options, but there are no clear answers to the appropriateness of treatments either because of scientific uncertainty or different patient values.¹² As patients become sicker, care becomes less evidence-based and more preference-based. In severely ill hospitalized patients, these decisions may include balancing palliative and disease-directed treatments, withholding treatments of marginal potential efficacy (eg, mechanical ventilation, dialysis, chemotherapy, and hydration/nutrition), withdrawal decisions after treatments have been started, hospice referral for pure palliation, and determining whether end-of-life care will occur in the hospital or elsewhere.

Treatments provided to seriously ill patients are often inconsistent with patients' underlying preferences and reflect an unsettling amount of variation in preference-sensitive end-of-life care. For example, there is a 10-fold variation in the rates of early do-not-resuscitate (DNR) orders across hospitals and dramatic variation in the proportion of all intensive care unit deaths preceded by withdrawal of life support.^{13 - 14} These treatment choices can affect the location of death, length of life, and reportable mortality. Such decisions may be relatively more common in the 600 000 hospital deaths that occur each year (over 50% of all hospital deaths) in patients aged 75 years or older who are at increased risk for accumulating multiple chronic illnesses.¹⁵

Supply-sensitive care is driven by the capacity of the health care system relative to the size of the population it serves.¹⁶ Positive associations exist between the per capita supply of hospital beds and physicians, and frequency of use of acute-care services.¹⁷ Supply-sensitive care and the percentage of deaths occurring in hospitals also depend on the availability of alternatives (eg, hospice or palliative care) in the community.¹⁸

The number of hospice programs has increased from 1600 in 1990 to more than 4100 in 2005, and the number of patients cared for within hospice during that same time has increased from 200 000 to 1.2 million.¹⁹ The number of hospital-based palliative care programs has also increased from 632 (15% of hospitals) in 2000 to 1027 (25% of hospitals) in 2003.²⁰ The capacity to care for patients with advanced illness will continue to improve given the recent formal recognition of hospice and palliative medicine as a new subspecialty.

Taken alone, short-term mortality measures essentially treat death as a medical failure and reinforce avoiding death at all costs. As a result, short-term mortality measures do not acknowledge the naturalness of death, disproportionately reinforce treatments that prolong life, and potentially discourage palliative care. In addition, the absence of cancer-related mortality measures and the increasing number of mortality measures for noncancer conditions reinforces the erroneous assumption that patients with cancer are expected to die and patients with non-cancer diagnoses are expected to live.

If not soliciting preferences but always erring on the side of aggressive disease-directed treatment leads to improved mortality statistics by prolonging dying, good care may be penalized and worse care may be rewarded. For example, a hospital in Buffalo, New York, is reported as one of the 35 “worst hospitals” in the country because of its high congestive heart failure mortality rate during the year from July 2005 to June 2006, with a risk-adjusted mortality that was 4.9% higher than the national mean.³ The hospital has reviewed these deaths and found that 11 patients (about 40% of total congestive heart failure mortalities) were in a hospice program or in nonhospice palliative care–only treatment regimens based on patient preferences prior to death. If those patients were not included in the calculation, the crude mortality rate would have been 9% lower, which would have produced a result better than the national average (Edward A. Stehlik, MD, written communication, June 2007).

Thus, high-quality end-of-life care that respects patient preferences may be classified as worse quality of care by CMS. In addition, publicly reported mortality information may create perverse incentive for hospitals and clinicians and may undermine high-quality end-of-life care. Prior to the 30-day reporting period, clinicians may avoid difficult discussions of prognosis, discount patient preferences for less aggressive treatment, fail to explore important dimensions of suffering, and avoid discussions about treatment withdrawal even when treatment is futile and unwanted.

Hospitals and clinicians might not realize the nuances of the mortality measure they are trying to improve. For example, transferring a dying patient to hospice or another hospital will not improve the transferring hospital's CMS 30-day mortality, as mortality is assigned to the transferring hospital. On the other hand, the hospital that accepts a high-risk patient from another hospital may wrongly conclude that its CMS 30-day mortality measure may be adversely affected if the patient dies shortly after transfer. Incentives change dramatically, however, depending on how the mortality measure handles transfers, discharges to hospice, and length-of-outcome period (inpatient vs 30-day).

In contrast to the CMS 30-day mortality measures, the majority of the publicly available mortality measures have an inpatient reporting period and exclude transfers or hospice admissions, creating incentives of “transfer to another hospital” or “transfer everyone to hospice” before patients die. As health systems work toward improving public perceptions and respond to increasing scrutiny from payers and the public, pressures may increase to alter mortality rates by using maneuvers that might not relate to quality.

Mortality is a good quality measure for individuals with acute illness who are not supposed to die. Therefore, hospitals should monitor their progress toward implementing safe practices and be held accountable for deaths unambiguously related to unsafe care by publicly reporting deaths on the National Quality Forum's “never events” list.²¹ However, mortality is a poor quality measure for the majority of patients with multiple chronic diseases who are near the end of their life, and may be engaged in preference-sensitive decisions that result in an earlier (or less delayed) death. Therefore, hospitals should consider ways to implement and monitor their progress on a selected number of the 38 palliative care–preferred practices developed by the National Quality Forum.²²

A major challenge will be to better distinguish patients who are not supposed to die from those dying of advanced chronic illness. This distinction will be best informed not by studying decedents, but by following seriously ill patients over time and determining patient, physician, and system factors contributing to how and when individuals die, including the circumstances that the improved coordination, communication, and symptom control may improve quality of care, but have mixed effects on mortality. This is particularly relevant given the findings that more frequent use of acute care services during the last 6 months of life may be associated with overall worse health outcomes.²³ This longitudinal focus will demand more shared accountability across clinicians and settings with innovations in care coordination, financing, and performance measurement.

Mortality models that adhere to published “standards” have to consider several limitations.²⁴ For example, the CMS mortality measures are validated using 5- to 10-year-old medical record data, a time when practice patterns and capacity may have changed, particularly with regard to end-of-life care. In addition, the promulgation of standards reinforces the “case-mix fallacy” (ie, erroneously believing that unbiased comparisons between hospitals can be made with risk-adjusted data) without consideration of other factors that contribute to differences in mortality: preference-sensitive and supply-sensitive care.²⁵

Despite the abundant research on mortality, there are relatively dew published data on how patients who are hospitalized are actually dying, including the types of decisions and trade-offs they make (beyond DNR orders), and the circumstances that optimal preference-sensitive choices might lead to higher mortality. Good-quality deaths embody both informed decisions and optimal symptom control, whereas bad-quality deaths include uninformed decisions and poor symptom control. This research should include developing methods to measure the quality of such decisions and determining the best models of decision making when death is near. The urgency of such research is underscored by prior research showing that patients who are better informed are more inclined to want less aggressive treatment—a strategy that may improve quality while paradoxically increasing mortality.¹²