Introduction

Advances in medical technology have generated multiple sophisticated life sustaining treatment (LST) options. These capabilities have undoubtedly saved or at least prolonged many patient lives, but their development may not always be matched by the prudence to apply them.

Effective decisions for LSTs and cardiopulmonary resuscitation are challenging when underuse can represent a life-threatening patient safety hazard[1] while overuse can have its own severe adverse consequences. Inappropriate decisions to withhold life-sustaining treatments have severe consequences,[2,3] and there is risk of patients labeled with “do not resuscitate” orders being misinterpreted to receive generally less aggressive care that they may benefit from. For this study, we focus on the complementary scenario. When both health professionals and the public systematically overestimate the efficacy of LSTs,[4] critically ill patients may receive burdensome high intensity treatments that do not effectively improve their quality or quantity of life.[5–7] Patients receiving LSTs who subsequently reverse or limit their goals of care with physician orders for Comfort Care or DNR can improve our understanding of the challenging LST decision making process and the effects of their application.[8–13]

Prior studies demonstrate differences in DNR orders based on patient characteristics, family, providers, and institutions.[10,12–24] Specifically, advanced age, female gender, and White race are associated with greater limitations, like DNR orders,[10,13,19,23] while physician projection of future mortality and recovery, age, patient wishes, severity of illness, and number of comorbidities are associated with withdrawal of life support.[25] Such differences can occur despite established guidelines, in part due to medical providers’ poor understanding of advance care planning documents such as living wills or physician orders for life sustaining treatments (POLST).[26,27]

Prior surveys suggest that physicians of different specialties have varying recommendations for end-of-life decision making.[14,15,28] Determining the factors associated with reversals in LST goals can improve patient and provider appreciation for the natural history and epidemiology of critical care to inform decision making on (continued) use of LSTs. In this study, we focused specifically on patients where life-sustaining critical care interventions were initiated, then had a subsequent reversal or limitation during the same hospitalization. We also evaluated the relevance of various modalities of life support and the potential variation in different specialty treatment teams in the decision-making process. The study was exempted as non-human subjects research by the Stanford IRB.

Methodology

Results

Baseline characteristics

During the study period, 10,590 subjects received LSTs. Of these, 433 already had Any DNR order at the initiation of LST, likely receiving them for critical care support before overt cardiopulmonary arrest (e.g., vasopressors for hypotension or mechanical ventilation for hypoxia). This leaves 10,157 patients in the study cohort who were Full Code at the entry point of initiating LST, with baseline characteristics summarized in Table 1. After initiation of LST, 1,669 (16%) subsequently had Any DNR order, 770 (8%) had Comfort Care orders, and 1,423 (14%) died, as summarized in Table a in S1 File and Fig 1.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Flowchart of patients receiving life sustaining treatments, code status orders, and in hospital mortality.

https://doi.org/10.1371/journal.pone.0190569.g001

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Summary of patient and clinical characteristics (N = 10,157).

https://doi.org/10.1371/journal.pone.0190569.t001

Treatment team specialty

Patients treated by the Medical ICU (HR 1.92 [95% CI 1.49–2.49]) and Hematology-Oncology (HR 1.87 [95% CI 1.27–2.74]) appeared were more likely to transition to Comfort Care, though the Hematology-Oncology association did not meet our stricter threshold for significance.

Patients treated by the Medical (HR 1.63 [95% CI 1.42–1.86]), Hematology-Oncology service (HR 1.63 [95% CI 1.33–1.98]), and Cardiac Care Unit-Heart Failure service (HR 1.41 [95% CI 1.15–1.72]) were more likely to change to Any DNR, while those treated by the Cardiovascular ICU (HR 0.32 [95% CI 0.25–0.41]), and Surgical services (HR 0.74 [95% CI 0.63–0.87]) were less likely.

Primary results are summarized in Figs 2 and 3. Supplementary Materials includes results for multiple additional variations of the Cox model for sensitivity analysis with respect to different data imputation strategies and inclusion vs. exclusion of different vital signs and lab result indicators of severity of illness.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Predictors of ICU patients transitioning to Comfort Care.

https://doi.org/10.1371/journal.pone.0190569.g002

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Predictors of ICU transitioning to any DNR (do-not-resuscitate).

https://doi.org/10.1371/journal.pone.0190569.g003

Discussion

Most inpatients entering life-sustaining critical care interventions were Full Code, which may reflect purposeful choices or limited prior advance care planning discussions (e.g., “default” selection that was not pointedly discussed with patients). Only 4.1% of patients had Any DNR order at initiation of LST. Among the Full Code patients, 8.0% subsequently had Comfort Care orders and 16% subsequently had Any DNR order. Of the 1,423 patients who died in the hospital, 714 (50%) died with Comfort Care orders and 1,161 (82%) died with DNR orders.

Prior literature suggests racial and cultural disparities in decision making,[10,19,21,23] though we found no significant differences for Hispanic, African American, Pacific Islander, or Native American patients changing to Comfort Care or Any DNR after controlling for other patient and clinical factors. The lack of significant disparities among racial minority groups suggests that differences are due to alternative characteristics, such as comorbidity, severity of illness, treatment modality, and team specialty. Similar to prior studies,[10,13,19,23] we found that advancing age and female gender are associated with decisions to limit life sustaining treatments, even after adjusting for other factors. Ongoing research and discussion are warranted to properly interpret and apply decision making in these cases to ensure both patient safety and informed consent.

We found that the modality of life-sustaining critical care interventions had varying associations with Comfort Care or Any DNR orders. Patients on either continuous renal replacement therapy or vasoactive infusions were more likely to receive Any DNR order to limit further escalation of aggressive life support interventions. Mechanical ventilation was not significantly associated with changing code status. This may be because patients are less willing to transition, or they have a physical barrier (intubation and sedation) to expressing revised goals of care. In the latter case, decision making responsibility then goes to the patient’s surrogate (typically family), who may be more hesitant to advocate reversals and limitations than the patient themselves if not previously addressed in documents like living wills. Alternative processes to best represent patient preferences rather than family member’s best guesses are thus an additional important area of development.[1]

Notably, we found that the treatment teams involved in patient care were associated with Comfort Care and Any DNR. Patients transferred to the Intensive Care Unit from Hematology and Oncology services were more likely to have Any DNR orders, while patients from Surgical, Cardiovascular and Cardiac Intensive Care Unit services were less likely. Decision variation is expected when different primary specialty teams reflect patients with different disease states (e.g. cancer), corresponding prognosis, and acceptable treatment goals (e.g. elective surgery). Yet, these variations are observed even after adjusting for patient comorbidities, admission diagnosis, and severity of illness indicators, which raises questions about differing practice patterns within various medical subspecialists. Clinicians and patients have different perspectives or abilities to recognize conditions amenable to intervention. For example, we may recognize metastatic cancer as an intractable terminable disease while believing there are always more interventions to salvage patients with cardiac and surgical problems. At the same time, evidence suggests that in some (surgical) settings, prematurely labeling patients “DNR” may result in worse outcomes perhaps from unintended “failure to rescue.”[34,35] This study alone cannot judge whether any of the observed practices are “appropriate” or not, but does illustrate that differences exist.

Supplementary Materials review the results of multiple sensitivity analyses with respect to different data imputation strategies and inclusion vs. exclusion of different vital signs and lab result indicators of severity of illness. The overall pattern of significant findings remained stable across different model variations, except for use of mechanical ventilation. In models excluding vital signs and laboratory results, mechanical ventilation is positively associated with subsequent DNR orders. This likely just reflects mechanical ventilation as a proxy for severity of illness as our primary model that included markers for acute illness did not show a significant association.

Limitations of this study include that data was collected from a single institution, which can restrict generalizability. Prior studies demonstrated that do not resuscitate physician orders vary according by region and are less prevalent in teaching hospitals than community hospitals.[23] The patient community surrounding the study institution generally represents a wealthier, more educated, and ethnically diverse population.[36,37] Our results identify patient and clinical factors associated with Comfort Care and Any DNR orders, but do not necessarily represent causal relationships. Similarly, given this retrospective study of structured electronic medical record data, additional contextual information such as living wills,[38] physician orders for life sustaining treatment,[39] and documentation of physician-family meetings were unavailable to provide qualitative rationale for the decisions in individual cases. Our study assumes that DNR orders electronically signed by attending physicians are “correct,” but does not account for the possibility of physicians misinterpreting patient preferences for life-sustaining interventions.[1]

While Comfort Care orders are a relatively explicit proxy for reversing LST decisions, Any DNR order after LST initiation may still be consistent with initial LST treatment goals (e.g., starting on vasopressors for low blood pressure, but later activating a DNR order to decline mechanical ventilation while still accepting ongoing vasopressors). Even in the case of subsequent Comfort Care orders, it is worth reflecting that while this may ultimately be a “reversal” of goals of care for ineffective LST, this does NOT entail that the decision to initiate LST was a “mistake.” As with all medical treatments, many trials of therapy do not ultimately produce a benefit. The initial decision to offer the therapy should be judged based only on information available at that moment, which is perfectly valid when we credibly expect a potential benefit that outweighs risks and harms.

What the results of this study can do is inform discussions around critical care decisions, similar to other prognostic risk scoring systems. Estimating patient risks for changing preferences can shift the balance of risks, benefits, and harms that are acceptable to patients when proceeding with trials of potentially burdensome, but life sustaining, treatments. Clinicians and health systems should take interest in identifying areas of care variation and in the characteristics, that may reflect critically ill patients who are more receptive to goals of care discussions, to help us deploy supportive and palliative interventions as early as possible in appropriate cases.

In summary, we found that Comfort Care and Any DNR orders subsequently occur for more than 1 in 13 patients who received life sustaining therapy, and are more common among older women. Patients on different forms of LST (e.g., continuous renal replacement and vasopressors) were more likely to reverse treatment compared to those on ventilator support. Different medical treatment team subspecialties are associated with varying resuscitation status. Awareness of these predictive factors can help inform complex decision-making processes to mitigate the risks of both underuse and overuse of critical care interventions.

Supporting information

Acknowledgments

Gustavo A. Chavez was supported by Stanford University School of Medicine Center of Excellence and Stanford University School of Medicine Medical Scholars Research Program. Jonathan H. Chen was supported by NIH Big Data 2 Knowledge Award Number K01ES026837 through the National Institutes of Health and National Institute of Environmental Health Sciences. Ilana Richman was supported by the US Department of Veterans Affairs Office of Academic Affiliations and Health Services Research and Development Service Research Funds. Dr. Periyakoil's time is supported in part by grants R25 MD006857, 5U54MD010724 and the Department of Veterans Affairs. The sponsors had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; and preparation, review, or approval of the manuscript, and decision to submit the manuscript for publication.