In 2009, exaggerated claims of “death panels” prevented Medicare from covering end-of-life discussions in ambulatory practice. Now, Medicare has proposed coverage of such discussions,1 and an increasing number of patients are presenting to the hospital with establishedgoals of care and end-of-life wishes. Nearly half of Medicare patients now die inhospice care, up from less thanaquarter in2000.2Quality measurement, however, has not kept up with the changing approach to end-of-life care. None of Medicare’s publicly reportedmortalitymeasures includesdo-not-resuscitate (DNR) status in riskadjustment, largelybecausesuchdataarenot routinely reported byhospitals, but also in part because of anunderlying assumption that the increasedmortality risk of such patients can be accounted for by comorbidity adjustment. In this issue of JAMA Internal Medicine, Walkey and colleagues3 demonstrate the flaws in this assumption for patients with pneumonia. Adding adjustment for DNR status within the first 24hours of hospitalization, and allowing such adjustment to vary byhospital (to accommodate the possibility that patients with DNR status at one hospital may not be as sick as patients with DNR status at another facility), markedly changed hospitals’ risk-adjusted mortality rates. Hospitals with high early DNR rates, which began with higher riskadjusted mortality rates than hospitals with few early DNR patients, had significantly lower risk-adjustedmortality once early DNR was accounted for. Although the resultsof this studyare relatively straightforward, the interpretation and implications are anything but. It is well established in studies of many different clinical conditions that earlyDNRstatus is an independentpredictorofmortality. If both earlyDNR status and subsequent outcomeswere purely a reflection of baseline prognosis and patient preference, itwouldbeappropriatetoexcludesuchpatientsfromqualitymeasures or to adjust risk for DNR status becausemortality in these patients is either unavoidable or congruent with patient preferences. Indeed, including such patients inmortality measuresmay be producing a perverse incentive for hospitals to resist patients’ wishes for less-aggressive end-of-life care. However, early DNR status and subsequent outcomes are also demonstrably dependent on hospital factors. Many studies4 have shownmarked hospital variation in DNR order rates that cannot be fully explained by patient clinical characteristics. Several studies have demonstrated that patients with early orders receive less-aggressive treatmentoverall, includingpotentially critical interventions, such as blood transfusions, cardiac catheterizations, cardiac bypass surgery, and operations for surgical complications.4,5 In fact, one study6 of patients with intracerebral hemorrhage found that hospitals with more early DNR patients provided so much lessaggressive care (eg, fewer intubations, craniotomies, ventriculostomies, and cerebral angiographies overall) that the odds of death for these patients vs non-DNR patients were higher than at other hospitals. However, this was not the case in the studyofWalkeyet al3 inwhichDNRpatientshadslightly lower mortality at hospitals with more early DNR patients. Other studies7 have shown that DNRpatients have highermortality at all levels of severity of illness, which in theory should not be the case since DNR status should be relevant only in catastrophic situations. Therefore, either excluding patientswith earlyDNRstatus fromqualitymeasures or risk adjustment for DNR status has the potential to obscure a tendency by hospitals to provide a lower standard of care for such patients even if that was not the patient’s explicit desire. There is no easy resolution to this conundrum. Disentanglingwhether the increasedmortalityofDNRpatients (and the increased mortality rates at hospitals with more of these patients) reflects careful attention to patients’ wishes for lessaggressive care or avoidable mortality from clinicians’ “failure to rescue” is impossiblewith existing data andmay never be feasible to determine. Qualitative studies might help, as might a more data-driven and patient-centered approach to DNRdiscussions.However,neitherof theseapproaches is likely in theshort term. In themeantime, the fact thathospitals’mortality rankingschangedmarkedlyafteradjustment forDNRstatus in the study ofWalkey and colleagues3 illustrates the substantial policy implications bothof the existingmeasurement system and of any potential change in mortality measurement to account for DNR status. The authors’ approach of including a randomslope for earlyDNRstatus to account for differential propensity among hospitals to label patients as DNR isnovel andmayhelp account forpatientpreferencewhilenot overly rewarding hospitals with a large fraction of early DNR patients (which, in this study, appear to be including lowerrisk patients in the DNR pool). Use of a random slope cannot, of course, account only for patient preference-relatedmortality risk over hospital quality-related mortality risk. Ultimately, thedecisionaboutwhethertoadjustriskforDNR status will need to be based more on philosophy than on science.AccountingforDNRstatus is likelytoencouragethehealth care system to pursue less-aggressive care for frail elderly patients,whichmaybemorepatient-centeredand improvequalityof lifebut isalso likely toresult insomedegreeofexcessmortality for DNR patients. However, continuing the status quo incentivizesamore-aggressiveoverall approach tocare,potentially saving some lives but also potentially causing some patients to undergo more interventions than they might otherwise have chosen. There is no right answer, but at least the implicationsof eachapproacharebecoming increasingly clear. Related article page 97 Do-Not-Resuscitate Orders and Hospital Mortality Rates Original Investigation Research
Read full abstract
Feb 25, 2024
David O Alao + 7
Open Access
