Withdrawal of life-sustaining therapy in intensive care unit patients following out-of-hospital cardiac arrest: An Australian metropolitan ICU experience.

BACKGROUND
Withdrawal of life-sustaining therapy is a common phenomenon following out-of-hospital cardiac arrest. The clinical practices surrounding withdrawal of life-sustaining therapy remain unclear and warrant further inspection due to their reported impact on post-cardiac arrest mortality.


OBJECTIVES
To determine factors associated with withdrawal of life-sustaining therapy (WLST) in intensive care unit (ICU) patients following out-of-hospital cardiac arrest (OHCA).


METHODS
A retrospective review of ICU patients' clinical records following OHCA was conducted from January 2010 to December 2015. Demographic features, cardiac arrest characteristics, and targeted temperature management practices were compared between patients with and without WLST. We dichotomised WLST into early (ICU length of stay <72 h) and late (ICU length of stay ≥72 h). Factors independently associated with WLST were determined by multivariable binary logistic regression.


RESULTS
The study cohort included 260 post-OHCA ICU patients. The mean age was 58 years, and majority were males (178, 68%); 145 (56%) underwent WLST, with the majority undergoing early WLST (89, 61%). Status myoclonus was the strongest independent factor associated with early WLST (OR 42.53, 95% CI 4.97-363.60; p < 0.001). Glasgow Coma Scale (GCS) motor response of <4 on day three post-OHCA was the strongest factor associated with delayed WLST (OR 48.76, 95% CI 11.87-200.27; p < 0.0001).


CONCLUSION
The majority of deaths in ICU patients post-OHCA occurred following early WLST. Status myoclonus and a GCS motor response of <4 on day three post-OHCA are independently associated with WLST.


Introduction
Cardiac arrest is the single leading cause of death in Australia. 1 The Australian Resuscitation Outcomes Consortium (Aus-ROC) epidemiological registry 2 reported a crude incidence of 99.4 per 100,000 population of out-of-hospital cardiac arrest (OHCA) in 2015. 3 Despite an overall decline in the cardiac arrest trends between 2010 -2018, the annual incidences remains high at 57,000 cases per year. 4,5 Generally, about 60% of OHCA patients admitted to the intensive care unit (ICU) remain comatose 6 for more than 24 hours, of which two-thirds sustain hypoxic-ischaemic brain injury. 1,7 However, only a minority of these patients die from brain death secondary to neuronal damage. 8 Instead, a large proportion (about 80%) 9 of deaths occur following early withdrawal of life-sustaining therapy (WLST) post-OHCA. 10 WLST is a highly delicate and ethically complex process in uenced by multiple clinical and non-clinical factors. 11 These variables often make neuroprognostication and WLST challenging, requiring concerted clinical and imaging assessments alongside numerous family discussions to make a timely, uni ed decision. Standard practices based on the Australian Resuscitation Council 12 and the European Society of Intensive Care Medicine 13 prognostication guidelines advocate neuroprognostication after 72 hours of arrest and at the very least 12 hours following cessation of sedation. Prior practices of targeted temperature management further complicated this process as it obscured the optimal timing of clinical assessments due to the effects of hypothermia on the clearance of sedatives. 14 Additionally, family preferences, cultural beliefs and patients' own advance care directives add to the challenges of appropriate WLST.
The in uence of these multiple factors can lead clinicians to prematurely de ne a patient's outcome earlier than the guideline-endorsed 72-hour period. 15 A false impression of poor early neurological recovery inculcates the 'self-ful lling prophecy' notion, which skews clinical perception for a meaningful outcome, inevitably leading to early WLST. 16 17 Strict guideline compliance in such multifaceted, complex situations is challenging. Awareness of WLST factors may provide clinicians an additional insight into the prognostication process, thus avoiding common trappings that traditionally lead to higher incidences of withdrawal and mortality rates. This renewed perspective allows intensivists to manoeuvre the post-cardiac phase con dently without being in uenced by false-positive clinical ndings while concurrently acknowledging predictors that signal poor outcomes.
In order to better understand this crucial facet of post-resuscitative care, we aimed to identify factors that predicted withdrawal of life sustaining therapy in the early and late phases following out-of-hospital cardiac arrest. prognoses, such as brainstem re exes (pupillary, corneal and cough re ex) and decision to withdraw lifesustaining treatment]) by reviewing patients' case notes. We also obtained data on the severity of illness (Acute Physiology and Chronic Health Evaluation [APACHE II]), mortality (ICU and hospital, hours), and length of stay (ICU and hospital) from the ANZICS-APD.

De nitions
We de ned OHCA as cardiac arrest in (i) persons not occupying an in-patient hospital bed and/or (ii) persons in the Emergency Department (ED) who have not been formally assessed and admitted by the medical team 19 . WLST was de ned as the cessation of life-sustaining treatment in the form of termination of inotrope and/or vasopressors and/or ventilatory support 21 . Early WLST (EWLST) was de ned as the withdrawal of life-sustaining treatment within 72 hours following OHCA 8 . Conversely, late WLST (LWLST) was de ned as withdrawing life-sustaining therapy at or after 72 hours post-ICU admission 8 . Cardiac cause of cardiac arrest was de ned as cardiac arrest secondary to a primary cardiac-related disease, including myocardial infarction, arrhythmia, cardiomyopathies and cardiogenic pulmonary oedema 22 . Non-cardiac cause of cardiac arrest was de ned as cardiac arrest not related to cardiac disease, including but not limited to causes such as trauma, sepsis, exsanguination, drug overdose, metabolic disorders and endocrinopathies 22 . ROSC was de ned as the resumption of normal heart rhythm with a detectable pulse 19 .

Objectives
The primary objective was to determine the factors associated with WLST in post-resuscitated OHCA ICU patients. The secondary objectives were to compare the demographic features, cardiac arrest characteristics, clinical attributes and targeted temperature management between patients with and without WLST.

Statistical analysis
The proportion of patients undergoing WLST was given as frequency (percentage). Continuous data were reported as mean with standard deviation (SD) or median with interquartile range (IQR) and were compared between WLST groups by the Student's t-test or Mann-Whitney U test, depending on the normality of the variable. Categorical data were reported as frequency (percentage) and groups were compared using the Chi-square test or Fisher's exact test, as appropriate. Initially, multivariable binary logistic regressions were used for all data to investigate the association between WLST and eleven a priori predictors, including ICU LOS (binary). Backward elimination was performed by eliminating the variables with the highest p-value one model at a time until all p-values were less than 0.5 (Harrell 2015) 23 .
Then, multivariable binary logistic regression was used for data relating to ICU length of stay <72 hours to investigate the association between WLST and ten a priori predictors by performing backward elimination once again. For data relating to ICU length of stay ≥72 hours (delayed WLST), brain stem re exes on days 2 and 3 post-cardiac arrest were removed due to collinearity, and then backward elimination was performed.
A Hosmer-Lemeshow goodness-of-t test was performed for all three nal models. A p value of ≤0.05 was considered to be statistically signi cant. All statistical analyses were performed using SAS, version 9.4 (SAS Institute Inc., Cary, NC, USA).

Derivation of the study cohort
During the study period, 346 patients with cardiac arrest were admitted to the ICU, including 77 with IHCA and 269 with OHCA. Overall, 17 patients were excluded because of lack of ROSC times, out of which 9 were from the OHCA cohort. The nal study cohort that was analysed consisted of 260 OHCA patients ( Fig. 1).

Patient characteristics
The mean age was 58 years (SD = 17), with most patients (156, 60%) being less than 65 years old. Most of the patients were male (178, 68%). There were 109 patients (42%) who survived hospital discharge. Of the 151 patients who died, 145 (96%) died following WLST, including 89 patients (61%) within the rst 72 hours of ICU admission-the 'early' cohort. Cardiac causes accounted for the majority of OHCA (141/260, 54.5%), with myocardial infarction as the leading cause. Asphyxia and pneumonia were the major noncardiac causes of OHCA (Table 1 and 2).
The non-shockable cohort accounted for 52% of the 260 patients enrolled in the study. Of these, 26% of PEA patients and 41% of the asystole group underwent WLST. In contrast, most patients from the shockable group (78/125, 68%) did not receive WLST (Table 3).
Out of the 260 patients in the study cohort, 161 (62%) received targeted temperature management, of which a target of 32-34°C and 34-36°C were given to 91 (35%) and 70 (27%), respectively. There was no association between WLST and targeted temperature management. However, the median duration of core body temperature >36°C in the rst 24 hours post-OHCA was higher in the non-withdrawal group compared to those sustaining WLST (10 (IQR 1-18) vs. 3.5 (IQR 0-12) hours; p < 0.001) (Additional File 1).
Although the overall proportion of WLST remained relatively the same over the study period, the gap between early and late WLST widened (Fig. 2). Interestingly, the trend of hospital death appeared to decline with an increased proportion of survivors, which is both unexpected and challenging to explain (Fig. 3).

Discussion
Based on our research, WLST factors can be broadly categorised into: neurological (status myoclonus, brain stem re exes and motor response of GCS) and non-neurological (age, ROSC time, non-shockable rhythm, APACHE II score) factors. Within the early cohort, we found a combination of neurological and non-neurological factors to be associated with WLST. The mortality following these withdrawals was due to cardiovascular instability and multiorgan failure. Conversely, in the late cohort, we noted GCS motor response <4 as the main factor for WLST, possibly due to sustained hypoxic brain injury.
In our study, most of the analysed patient cohort underwent WLST (145/260, 55%), with 61% having an ICU length of stay less than 72 hours. We noted WLST was responsible for 151/260 (58%) deaths in patients admitted following OHCA. Within that cohort, 96% of the patients died following early WLST. The signi cant mortality outcome con rms previously published evidence of the role of WLST as a common mode of death in OHCA patients. 8 In comparison to previous research by Albaeni et al. 8 in the United States in 2014, our study reported higher rates of WLST and mortality.
The key to explain this discrepancy required us to retrospectively review the growing research of withdrawal practices that occurred between 2006 and 2015. 13,27 Parallel with these developments were serial updates on therapeutic hypothermia, which at that time showed signi cant improvement in neurologic function and survival. 25,28 Despite these developments, the striking outcome in our study was that the majority of patients had life support withdrawn outside of current recommended guidelines.
Firstly, the high proportion of post-OHCA mortality from early withdrawal was likely a consequence of unstandardized withdrawal practices. Treating intensivists may have applied their own professional judgement and actioned what was thought to be the most appropriate treatment for the patient. Secondly, many of these patients received therapeutic hypothermia, which is known to confound early clinical assessments. 29 Unfortunately, the combined effect of the above factors led to the creation of the 'self-ful lling prophecy' which falsely predicts poor outcome, thus leading to high early withdrawal rates. This phenomenon explains the differences in our withdrawal rates, with the absence of brainstem re exes and status myoclonus driving this decision-making. We conclude that the lack of a protocolised withdrawal model with ever-changing recommendations on therapeutic hypothermia resulted in an array of individualised treatments rather than a standardised approach.
A resounding theme from previously published research, including ours, highlights one foundational concept-that WLST and neuroprognostication are inherently connected, 30 and the association resembles a cause-and-effect relationship. Repeatedly, multiple studies have shown poor neuroprognostication practices (cause) lead to higher rates of WLST 31 (effect). This relationship was illustrated in our study, wherein of 67 patients with absent brainstem re exes on day 1 post-OHCA, 93% underwent early WLST.
This trend continued on days 2 and 3 post-arrest, with 100% of the patients with absent brainstem re exes having their care withdrawn. This striking observation raises a crucial point: Survival outcome determined solely by neurological examination using brainstem re exes within 72 hours led to early WLST, with a high mortality outcome. 30 A common presenting feature following cardiac arrest is post-hypoxic myoclonus status epilepticus (MSE). This occurrence has historically been viewed as a marker of poor prognosis. 35 Our data con rms this with a strong association of withdrawal, especially within 72 hours of cardiac arrest. However, contrary to the perception of MSE as an 'agonal phenomenon', recent research 35 has shown some good outcomes following post-arrest myoclonus. These con icting outcomes raise an essential question: Should status myoclonus be viewed as a clinical sign of futility? Recent advances in phenotyping practices with electroencephalography have signi cantly in uenced patients' management and survival rates with status myoclonus. Good outcomes have also been associated with ICU length of stay of ≥8 days, 35 in contrast to our patient cohort with MSE, which had withdrawal of care within 72 hours. Hence, we believe that status myoclonus should not be viewed as a poor prognostic sign at face value until appropriately strati ed.
The challenges of WLST further continue with the incorporation of targeted temperature management. Having undergone large serial trials, the targeted temperature management practices continue to evolve from the proposed adoption of 33-36 o C 25 and, most recently, to normothermia 26 with the avoidance of fever. Although these hypothermic practices did not increase the likelihood of withdrawal in our study, we found a longer duration of exposure to hypothermia led to a higher incidence of WLST. This correlation con rms prior pharmacokinetics research on the delayed clearance of sedative medications due to hypothermia. 34 Overall, the clinical implications of these factors steer intensivists to form pessimistic clinical perceptions, thus leading them to arrive at hasty decisions on patients' survival outcomes. This 'selfful lling prophecy' is now a widely recognised conundrum that bridges poor and untimely neuroprognostication practices with withdrawal of life support. Acknowledging the presence of this intrinsic clinical bias is vital. Only then can these patients undergo rigorous objective evidenced-based assessments and outcomes determined on the merits of their actual neurological status.

Limitations
Our study has several signi cant limitations. The rst major limitation was its single-centre, retrospective design, which culminated in a small sample size. Second, the data collected from 2010 to 2015 may not be re ective of recent clinical practices. Third, our review did not include and thus lacked data on patients' advanced care directives. Additionally, it was evident from our clinical records review that there was a lack of protocolised management for WLST. Fourth, in versus out-of-hospital cardiac arrest de nitions remain controversial and lack universal consensus 36 . Different de nitions can potentially alter the sample sizes of both populations, thus risking misrepresentations of the case-mix. In order to prevent enrolment bias from misclassi cation, we delineated these two cohorts with a clear and sensible de nition of ours. Lastly, there were challenges with extracting information from the patients' case notes.
Some of the sought-after variables were not legible, inaccurately documented or missing entirely. These made the abstracting process di cult as the required information had to be sourced from other avenues, such as ICU charts, ED documentation and discharge summaries.

Conclusion
Early WLST is a signi cant contributor to OHCA deaths. From our analysis, APACHE II score, status myoclonus, absent brainstem re exes on day 1 post-cardiac arrest, non-shockable rhythm and prolonged ROSC time predict early withdrawal of care. Future research incorporating post-OHCA patients receiving normothermia and undergoing protocolised neuroprognostication, considering other variables such as advance care directives, family wishes and cultural variances, may more accurately capture the actual proportions of patients receiving withdrawal of care and their predictors in these current times of intensive care practice.     The graph above demonstrates the trend of total, early and late WLST from 2010 to 2015.

Figure 3
The graph above demonstrates the trend of outcomes at hospital discharge from 2010 to 2015.

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download.