Perioperative morbidity and mortality of cardiothoracic surgery in patients with a do-not-resuscitate order

Introduction

Advance directives have an increasingly important role in contemporary medical practice, and their use has come to affect perioperative decision-making more frequently than in prior eras (Burkle et al., 2013). A greater attention to bioethical considerations has spurred efforts to promote the discussion and use of advance directives in the surgical setting. The American College of Surgeons⁴ and the American Society of Anesthesiologists⁵ have opposed the practice of requiring DNR orders to be cancelled for a patient undergoing surgery, instead recommending that DNR patients may be candidates for an operation but that their specific wishes regarding which interventions would be appropriate in the perioperative setting should be discussed in detail with the anesthesiology and surgical teams. Specific guidelines exist to detail options for individualizing care and implementing DNR orders in the perioperative setting (Ewanchuk & Brindley, 2006; Truog, Waisel & Burns, 1999, 2005; Waisel et al., 2002).

Patient decision-making surrounding the risks and benefits of surgery and associated interventions (e.g. intubation, vasoactive medication use) in the perioperative period depend crucially on expectations about the likelihood of a successful operation, anesthetic, and postoperative recovery. However, outcome data are limited regarding the relationship between DNR status and surgical morbidity and mortality. While some studies have examined this relationship in a broad general surgical cohort (Kazaure, Roman & Sosa, 2011) and a colorectal surgical population, (Speicher et al., 2013) it remains incompletely evaluated in the setting of thoracic and cardiac operations, which in general involve more complicated recovery and carry greater overall risks.

We undertook the present analysis to better evaluate the relationship between DNR status and cardiothoracic surgical outcomes. We used a large administrative dataset with prospectively entered information on DNR status to assess the hypothesis that DNR status is associated with increased perioperative morbidity and mortality.

Materials and methods

The Stanford University Institutional Review Board granted an exemption from review because this research uses publicly available, deidentified data. The Public Discharge Data (PDD) file is a comprehensive public dataset of inpatient admissions consisting of one record for each patient discharge from a licensed hospital in California, provided by the California Office of Statewide Health Planning and Development (OSHPD) via the Medical Information Reporting for California (MIRCal) System. PDD records from years 2005 through 2010 were examined for this analysis.

Results

Patients with an active DNR order within 24 h of admission represented 3,129 (3.7%) of 85,164 admissions for thoracic surgery and 2,678 (1.1%) of 242,234 admissions for cardiac surgery during the study period. Matching resulted in a cardiac control cohort of 10,670 admissions and a thoracic control cohort of 12,290 admissions. Demographic and comorbid characteristics of the DNR and control cohorts are shown in Table 1.

Table 2 provides a comparison of outcomes between the DNR and matched control cohorts. The primary outcome comparison revealed high in-hospital mortality in the thoracic (25.4%) and cardiac (37.5%) DNR groups that was significantly increased compared to controls (p < 0.0001 for both). Many but not all measures of resource utilization and secondary outcomes were worse in the DNR cohorts (see Table 2).

Multivariate logistic regression performed to evaluate the effect of DNR status while controlling for baseline differences in patient characteristics and comorbidities resulted in models with acceptable area under the ROC curve (thoracic model AUC = 0.734, cardiac model AUC = 0.711). Results are presented in Table 3. DNR status remained an independent predictor of mortality in both models (p < 0.0001 for both). Other independent predictors (p value below the Bonferroni-adjusted significance level of 0.01) in the model for thoracic operations included arrhythmia, chronic kidney disease, and hypertension. Other independent predictors in the model for cardiac operations included coronary artery disease, congestive heart failure, chronic kidney disease, chronic lung disease, and hypertension.

Discussion

There are three primary findings of this study. First, we find it remarkable that such a substantial minority proportion of cardiothoracic surgical patients have an active DNR order in place at the time of the admission in which surgery occurs. The magnitude represented by the two study cohorts (3.7% and 1.1 of all thoracic and cardiac surgical patients, respectively) indicates that it is not an exceedingly rare event for DNR patients to be offered – and to accept – a major cardiac or thoracic operation. Second, the outcomes of those operations are startlingly poor, with 25% and 38% of the DNR thoracic and cardiac cohorts, respectively, dying in the hospital. Third, DNR status remains an independent risk factor for perioperative mortality when controlling for age, procedure, race, insurance status, and major comorbidities.

Prior analyses from the American College of Surgeons National Surgical Quality Improvement Project have shown a high postoperative mortality rate in general surgical patients who are DNR (Speicher et al., 2013) and have suggested that excess mortality is due to a decreased willingness to pursue aggressive interventions in the postop period, (Scarborough et al., 2012) described as “failure to pursue rescue”, While this retrospective, observational study is unable to confirm the etiology of excess mortality in the DNR groups, the resource utilization implications of this hypothesis are profound. Tremendous financial and operational resources (including the labor of surgical, anesthesia, perfusion, nursing, and operating room technical staff, equipment and medication costs, postoperative intensive care and supportive services) are devoted to the types of cardiothoracic operations used to define our study cohorts. If “failure to pursue rescue” after making the decision to undergo a major surgical intervention plays a role in explaining the substantial elevation in postoperative mortality in this population, we believe it suggests unwise resource utilization. Perhaps even more importantly, thousands of DNR patients may be exposed to the discomfort and risk of highly invasive procedures that have a diminished prospect of a good outcome if they are not coupled with a range of certain aggressive postoperative interventions.

The major limitations of this analysis are the well-established limitations of a retrospective administrative database analysis, principally that of classification error (Jollis et al., 1993; Koch et al., 2012) The advantage of the PDD definition of DNR is that it avoids miscounting surgical patients who were not DNR at the time of admission or their operation but became DNR later in their hospital stay (e.g. after multiple postoperative complications). It remains possible that the PDD undercounts some patients who were DNR at the time of surgery (e.g. DNR order written on hospital day three, surgery on hospital day five) but if present, this classification error would only mitigate the relationship we observed.

Outcome information is similarly limited in an administrative database. Nonfatal complications are based on diagnosis codes, which do not include information on outcome severity, and are more likely to be undercounted (errors of omission) than overcounted, because of limited reliability of diagnostic codes for perioperative complications. Our decision to define inpatient complications using the absence of the “present on admission” variable may also result in undercounting of clinically meaningful deterioration – for example, a patient with compensated heart failure present on admission whose heart failure decompensates in the perioperative period would not be defined as “new CHF” in our methodology. We suspect these phenomena, among others, help explain the observation that death was more common than any of the secondary outcome measures in both DNR cohorts. For these reasons, we used mortality as the primary outcome, as we felt it represented a more reliable endpoint.

The variables used for matching were selected to create a comparison cohort that would represent a face value “peer group” for the study cohort, but matching is likely an imperfect strategy for risk adjustment. Multivariate analysis facilitated adjustment for many other potential confounders, but the observational nature of the study implies the possibility of residual unmeasured confounders. We would qualify the notion that DNR status “independently” confers worse prognosis after cardiothoracic surgery by pointing out that a DNR order itself is unlikely to be causally harmful. Our results do not demonstrate any direct effect of DNR status on outcomes, such as a difference in quality of care. DNR status, however, might be viewed as conferring increased risk precisely because it functions as a proxy for unmeasured or unquantifiable variables that affect perioperative risk.

DNR status may capture other elements of comorbidity and poor reserve that traditional measures of comorbidity (e.g. Charlson index) do not, such as frailty (Farhat, Velanovich & Falvo, 2012; Sündermann, Dademasch & Praetorius, 2011; Afilalo, Mottillo & Eisenberg, 2012). The public health, operational, and public policy utility of noting DNR status as a marker of surgical risk might lie in its ease of use: unlike comorbidity scores or frailty indexes, no effort, definition, or calculation is required to determine that a patient has decided to accept DNR status. It is plainly written in the medical chart, typically in a high-profile location. If one knows only a few details about a patient, one often knows their code status.

Of course, many patients without an active DNR order might benefit from efforts to improve the quality and frequency of advance directive discussions prior to any proposed surgical intervention and discussions of risk. The absence of a DNR order in a patient whose wishes would be consistent with one is unlikely to be protective. But the converse appears to remain true: patients who have had that discussion and decided against a course of aggressive resuscitation are easily identified as extremely high-risk surgical candidates.

Further investigations will help more clearly elucidate whether there is a causal relationship between DNR status and surgery, as well as help define the subsets of DNR patients that might be better identified as comparatively good operative candidates, for the purposes of improving counseling for patients and families and framing discussions of resource utilization and the ethical dimensions of end-of-life care.

Supplemental Information