Clinical examination for prognostication in comatose cardiac arrest patients

Abstract

Objective

To build new algorithms for prognostication of comatose cardiac arrest patients using clinical examination, and investigate whether therapeutic hypothermia influences the value of the clinical examination.

Methods

From 2000 to 2007, 500 consecutive patients in non-traumatic coma were prospectively enrolled, 200 of whom were post-cardiac arrest. Outcome was determined by modified Rankin Scale (mRS) score at 6 months, with mRS ≤ 3 indicating good outcome. The clinical examination was performed on days 0, 1, 3 and 7 post-arrest, and clinical variables analyzed for importance in prognostication of outcome. A classification and regression tree analysis (CART) was used to develop a predictive algorithm.

Results

Good outcome was achieved in 9.9% of patients. In CART analysis, motor response was often chosen as a root node, and spontaneous eye movements, pupillary reflexes, eye opening and corneal reflexes were often chosen as splitting nodes. Over 8% of patients with absent or extensor motor response on day 3 achieved a good outcome, as did 2 patients with myoclonic status epilepticus. The odds of achieving a good outcome were lower in patients who suffered asystole (OR 0.187, 95% CI: 0.039–0.875, p = 0.033) compared with ventricular fibrillation or non-perfusing ventricular tachycardia, but some still achieved good outcome. The absence of pupillary and corneal reflexes on day 3 remained highly reliable for predicting poor outcome, regardless of therapeutic hypothermia utilization.

Conclusion

The clinical examination remains central to prognostication in comatose cardiac arrest patients in the modern area. Future studies should incorporate the clinical examination along with modern technology for accurate prognostication.

1. Introduction

Neurological prognostication in comatose cardiac arrest survivors remains challenging. Despite advances in ancillary testing, the clinical examination continues to lend valuable insights to outcome. Levy et al. published their landmark study in 1985, using the rigorous definition of coma as complete absence of purposeful responses to the self or environment.¹ Subsequently there have been significant advances in treatment options, including therapeutic hypothermia (TH). Many prognosis studies included patients who were not truly comatose,² who commonly achieve good outcomes, thus weakening predictive accuracy.

Clinical signs that historically predict poor outcome include absent pupillary and corneal reflexes, and motor response of none or extension 3 days post-arrest.^3,4 We previously published our findings from 500 patients with non-traumatic coma from all causes.⁵ Herein we have modernized Levy et al.’s approach, focusing on the cardiac arrest population but staying true to the formal definition of “coma” so that the correct population is studied: patients in whom the prognosis is truly uncertain. Our hypothesis was that the clinical examination remains valuable for predicting outcome in comatose cardiac arrest patients, even with modern therapies and management that influence outcome. Prognostication in comatose post-cardiac arrest patients has evolved with advances in critical care and ancillary studies, and modern statistical methods based on the neurological examination provide important prognostic information.

2. Methods

We used the methodology employed in a previous study.⁵ In brief, from October 2000 to September 2007, 500 consecutive non-traumatic coma patients were enrolled in an IRB-approved, HIPAA-compliant, observational single academic center cohort study; 200 were comatose secondary to cardiac arrest. Waiver of consent was granted to enroll patients given the observational nature; consent was required only for survivors to subsequently assess outcome at 6 months. Patients were recruited from the Emergency Department and Medical, Cardiac and Neurosciences Intensive Care Units. Patients were identified within 24 h of arrest. We adhered to a strict definition of “coma” as defined by Plum and Posner⁶ as complete lack of awareness of the environment; no eye opening in response to noxious stimuli, or in situations of “eyes-open coma,” no closure in response to stimuli; sounds emitted could not reflect response to need/discomfort (most were intubated); no blink to visual threat; no grimace or purposeful movement to cranial or corporal noxious stimulation. Patients were excluded if not in coma as strictly defined, or comatose due to sedatives, trauma or environmental hypothermia. Standard post-arrest care included induction of TH within 6 h of ROSC. The target temperature was 32–34 °C, maintained for 24 h from the time of the initiation of TH. Controlled rewarming was performed no faster than 0.25 °C h⁻¹, and subsequently normothermia was maintained. The mean arterial pressure during TH was maintained between 70 and 80 mmHg, and normoxia and normocarbia were goal parameters. The components of the neuroprognostication work up included the clinical examination as well as electrophysio-logic, serum biomarker and neuroimaging testing at the discretion of the treating teams. The best examination on a given day, only as performed by a board-certified neurologist, was scored.

Demographic data collected included age, gender and comorbidities. We aimed to evaluate patients on the first calendar day of coma (day 0), but if impossible, the day 1 examination was used as the initial examination. Patients were excluded if they could not be examined on day 0 or 1. Examinations were by a board-certified neurologist on days 0, 1, 3 and 7, including eye opening, verbal and motor responses according to the Glasgow Coma Scale (GCS), corneal, pupillary, horizontal oculocephalic (OCR) and gag reflexes. Oculovestibular reflex (OVR) testing (“cold calorics”) was performed only if the OCR was absent or untestable; otherwise, it was considered present. Motor response was assessed via noxious stimulation. If patients could not be examined off sedation (rare except for TH patients), the previous day’s examination was carried forward. For patients undergoing TH, all were evaluated by a neurologist prior to the institution of TH to undergo a detailed neurological examination and to ensure coma. The most common sedative used was propofol, routinely held prior to examination. We also collected information on maximal daily temperature and whether nutritional support was provided.

Relevant ancillary testing recorded included neuroimaging, biomarker and electrophysiologic results. All treatment decisions, including the performance and timing of diagnostic testing, were at the discretion of the treating team.

A blinded mRS-certified neurologist determined outcome by mRS score at 6 months by a phone call or office visit. No patients were lost to follow up. The mRS is a validated scale assessing the degree of disability for a patient based on independence on a 0–6 scale, with 6 being death.⁷ If death occurred, the cause was categorized as withdrawal, unexpected cardiopulmonary arrest, or brain death. We dichotomized good mRS outcome at ≤3, reflecting the ability to ambulate and attend to bodily needs.

To build new prognostication algorithms, a classification and regression tree (CART) analysis was performed. CART is a non-parametric, decision tree procedure increasingly utilized in biomedical research.^8,9 Being a nonparametric method, CART does not make assumptions about the underlying distribution of the predictors. Missing values will not be dropped out of analysis, as CART uses a sophisticated algorithm to create ‘surrogate’ variables containing similar information to predictors used for analysis. CART therefore is impervious to outliers, skewed or missing data. Over-fitting was avoided by using the embedded cross-validation. The decision tree generated by CART is easy to interpret and apply by clinicians. More information can be found in our previous publication⁵ or elsewhere.¹⁰

Candidate predictors included the verbal response, pupillary reflex, OCR, OVR, corneal reflex, motor response, spontaneous eye movements and eye opening. We set the search intensity at 400 (more accurate) and chose class probability as the splitting rule to obtain valuable details of the data structure.¹¹ We set the cost at 20 if a patient with a good outcome was misclassified as having a poor outcome. Other options used the default values set in the Salford Predictive Modeler (SPM). We performed additional analyses to compare baseline characteristics of the patients with good and poor outcomes, and examined the association of other factors (e.g. temperature and nutritional support) with outcome at 6 months. Analysis was conducted using CART integrated in SPM (version 6.4.0.209, Salford Systems, San Diego, CA). All the other analyses were performed using SAS (version 9.2, SAS Institute Inc., Cary, NC). A p-value of <0.05 was considered statistically significant.

3. Results

A total of 34 patients (17%) did not have an examination on day 0. Table 1 presents the baseline characteristics by good/poor outcome. Good outcome (mRS ≤ 3) was achieved in 9.9% of patients; 5% achieved an excellent outcome (mRS 0 or 1). Of those who died in hospital, 152 (85%) died secondary to withdrawal of life-sustaining therapies, 20 (11%) progressed to brain death and 6 (3%) experienced a fatal second cardiac arrest. The initial cardiac rhythm correlated with outcome (Table 2). Asystolic patients had lower odds of good outcome compared to those with ventricular fibrillation (VF) or non-perfusing ventricular tachycardia (VT) (OR 0.187, 95% CI: 0.039–0.875, p = 0.033); a similar trend (not statistically significant) was seen for arrest due to pulseless activity (PEA) (OR 0.378, 95% CI: 0.134–1.070, p = 0.066). However, PEA and asystolic patients did not uniformly have a poor outcome; 6/80 (7.5%) with PEA and 2/52 (3.9%) with asystole achieved good outcomes, compared with 12/68 (17.6%) with VF/VT. There was a trend that patients who underwent TH were more likely to achieve good outcomes with initial rhythms of PEA, VF or VT; none of the 5 asystolic TH-treated patients achieved good outcomes (p = 0.56).

Table 1.

Baseline characteristics for patients by recovery status at 6 months.

Variable	mRS cutoff = 3
	Good recovery (n = 20)	Poor recovery (n = 180)	p
Age	60.2 ± 16.3	59.9 ± 16.5	0.94
Gender			0.77
Male	13 (65%)	111 (62%)
Female	7 (35%)	69 (38%)
Cardiac arrest			1
In hospital	3 (15%)	34 (19%)
Out of hospital	17 (85%)	146 (81%)
Coronary artery disease	4 (20%)	59 (33%)	0.24
Congestive heart failure	1 (5%)	36 (20%)	0.13
Chronic obstructive pulmonary disease	4 (20%)	40 (22%)	1
Diabetes mellitus	5 (25%)	48 (27%)	0.87
Renal insufficiency	2 (10%)	30 (17%)	0.75
Hepatic Insufficiency	0 (0%)	12 (7%)	0.61
Smoking history	3 (15%)	28 (16%)	1.0
EtOH history	3 (15%)	24 (13%)	0.74
Drug history	1 (5%)	25 (14%)	0.48

Table 2.

Patients achieving good outcome by initial rhythm.

	VF	VT	PEA	Asystole
Whole population	9/54 (16.7%)	3/14 (21.4%)	6/80 (7.5%)	2/52 (3.9%)
Hypothermia +	4/15 (26.7%)	1/3 (33.3%)	3/16 (18.8%)	0/5 (0%)
Hypothermia −	5/39 (12.8%)	2/11 (18.2%)	3/64 (4.7%)	2/47 (4.3%)

Hypothermia +, patients who underwent therapeutic hypothermia. Hypothermia −, patients who did not undergo therapeutic hypothermia.

We compared our results with Levy et al.’s algorithms (Fig. 1). A CART analysis then determined clinical signs related to outcome, resulting in new algorithms (Fig. 2). The motor response was often a root node (the initial node for splitting), and spontaneous eye movements, pupillary reflexes, eye opening and corneal reflexes were often splitting nodes.

Fig. 1.

Prediction of outcome in comatose cardiac arrest patients using Levy et al.’s algorithm. No patients achieved a mRS score of 4; therefore no separate column for this outcome is presented. Numbers in () represent 95% confidence intervals. Numbers in [] represent the number reported in the Levy et al. study. (A) Day 0, (B) Day 1, (C) Day 3, (D) Day 7.

Fig. 2.

Prediction of outcome in comatose cardiac arrest patients using CART analysis. Note that the number of patients is low on day 7, making the findings less reliable at this time point. OVR, oculovestibular reflex; eye opening = spontaneous eye opening. (A) Day 0, (B) Day 1, (C) Day 3, (D) Day 7.

Thirty-nine patients underwent TH (Table 3); 28 survived to day 3. The pupillary reflex remained an accurate predictor of outcome, both on days 3 and 7; no patients with absent pupillary responses on either day achieved good outcomes, irrespective of TH treatment. Similarly, no patients with absent corneal responses on day 3 survived to good outcome. However, of the 74 patients with a motor response of none or extensor posturing on day 3, six (8.1%) achieved good outcomes; 3 underwent TH. Two of 61 (3.3%) patients with myoclonic status epilepticus in the first 24 h achieved good outcomes, one TH-treated.

Table 3.

Outcome of patients by hypothermia and clinical exam.

Hypothermiaa,b	Pupillary reflex	Day 3		Day 7
		Good recovery	Poor recovery	Good recovery	Poor recovery
+	Reactive	8 (36%)	14 (64%)	8 (47%)	9 (53%)
	Nonreactive	0 (0%)	6 (100%)	0 (0%)	2 (100%)
−	Reactive	12 (19%)	50 (81%)	12 (34%)	23 (66%)
	Nonreactive	0 (0%)	14 (100%)	0 (0%)	5 (100%)
p		0.0137		0.048
Hypothermia	Corneal reflex	Day 3		Day 7
		Good recovery	Poor recovery	Good recovery	Poor recovery
+	Present	8 (42%)	11 (58%)	8 (47%)	9 (53%)
	Not present	0 (0%)	9 (100%)	0 (0%)	2 (100%)
−	Present	12 (27%)	32 (73%)	12 (36%)	21 (64%)
	Not present	0 (0%)	32 (100%)	0 (0%)	7 (100%)
p		<0.0001		0.024
Hypothermia	Motor response	Day 3		Day 7
		Good recovery	Poor recovery	Good recovery	Poor recovery
+	Extensor or none	3 (16%)	16 (84%)	1 (11%)	8 (89%)
	Flexor or better	5 (56%)	4 (44%)	7 (70%)	3 (30%)
−	Extensor or none	3 (5%)	52 (95%)	0 (0%)	21 (100%)
	Flexor or better	9 (43%)	12 (57%)	12 (63%)	7 (37%)
p		<0.0001		<0.0001

Hypothermia +, patients who underwent therapeutic hypothermia. Hypothermia −, patients who did not undergo therapeutic hypothermia.

Elevated temperature (defined as ≥38.3 °C anytime during the first week) was not associated with poor outcome (p = 0.807). Patients who received nutritional support were more likely to have a good outcome (OR 11.2, 95% CI 3.83–32.7, p < 0.001).

4. Discussion

The American Academy of Neurology’s Practice Parameters (AANPP) for determining prognosis in comatose cardiac arrest survivors emphasize absent pupillary or corneal reflexes, or motor response of none or extensor posturing to noxious stimulation.⁴ The basis of these recommendations stems from data spanning several decades, with limited detail regarding examinations and including patients not strictly comatose. The clinical examination remains highly useful: it is reproducible, noninvasive, not subject to artifact, and highly reliable in the absence of confounders.

Given advances in neurocritical care (e.g. TH), reevaluation of all diagnostic testing should be performed periodically and systematically. A particular strength of our study is strictly emphasizing enrolling the appropriate population, including only those strictly comatose.^1,4 Patients who recover consciousness early after arrest commonly experience good outcomes.¹² Inclusion of noncomatose patients would have diluted our results. We were stricter than Levy et al., excluding patients with eye opening, moaning (thus responding to pain or need), or with motor responses of withdrawal or better. We were careful to exclude patients whose examinations were potentially influenced by sedatives.^1,13 Furthermore, Levy’s study of 210 patients included 22 patients with primary respiratory failure, and 38 patients with profound hypotension (without cardiac arrest) and “anesthetic accidents.” Patients with primary respiratory arrest or isolated hypotension commonly achieve good outcomes,¹⁴ and patients with anesthetic accidents clearly may have medications affecting their examinations.

When our data are applied to the Levy et al.’s algorithm (Fig. 1), many results are similar. It would be tempting to conclude that the neurological examination findings have remained static over 30 years. However, we would emphasize that this is likely due to antagonistic factors: (1) advances in critical care, including TH, have improved outcomes overall; counter-balanced by, (2) we included “sicker” patients than the Levy et al. Had we enrolled patients using the same approach, we would likely have observed more good outcomes using their algorithm.

We also observed differences when Levy et al.’s algorithm was applied to our data. The presence of spontaneous eye movements was consistently a splitting node by their algorithm. The utility of this variable in differentiating patients’ outcome is not as strong in our data. For example, on admission only 25% of patients (95% CI: 1–20%) achieved a good outcome if the spontaneous eye movement finding was roving conjugate or better, compared with 41% (95% CI: 22–61%) in Levy et al.’s data. Similarly, on day 3 if patients did not have orienting eye movements, 76% still achieved a good outcome compared with 18% in Levy et al.’s study. Thus, the value of spontaneous eye movements in predicting outcome in the modern age appears to be poor.

The use of TH may negatively impact the ability to determine prognosis in comatose patients in the early time period post-arrest.¹⁵ The two landmark trials of hypothermia for cardiac arrest stipulated that patients be in coma^16,17; however, the definition of coma was not the strict standard. The 2010 American Heart Association Guidelines for TH in cardiac arrest patients stipulate that patients “have a lack of meaningful response to verbal commands.”¹⁸ Strict adherence to definitions is necessary so that meaningful comparisons between trial populations can be performed.

Great emphasis has been placed on absent pupillary reflexes. Pupillary responses may be absent transiently immediately post-arrest and subsequently recover. The durability of absent pupillary reflexes as predictive of poor outcome at later time points (≥3 days) persists in our study; no patients with nonreactive pupils on day 3 post-arrest achieved a good recovery, regardless of whether or not they underwent TH. There is a report of one patient with absent pupillary light reflexes on day 3 who made a good recovery, but the data provided in this report are sparse.¹⁹ Two other studies identified no TH-treated patients with absent pupillary responses on day 3 who survived to good outcome.^20,21

Likewise, absent corneal responses in the TH era continue to be predictive of poor outcome. In our study, on day 3 post-arrest, 41 patients (39%) had absent corneal responses; all had a poor outcome, whereas 32% with present responses achieved a good outcome. Bouwes et al. reported 2 of 22 patients (9%) with absent corneal responses on day 3 post-arrest who achieved a good outcome,¹⁹ but other studies have demonstrated no patients with good recovery with absent corneal responses on day 3.^20,21

A very important finding of this study is that absent or extensor motor responses are no longer reliable predictors of poor outcome; 8.1% of patients with absent or extensor motor responses on day 3 achieved a good outcome. Rossetti et al. prospectively evaluated 111 patients treated with TH; the poor motor response on day 3 had a false positive rate of 24% for predicting poor outcome.²² Others have found false positive rates of 10–17%.^19–21

Myoclonus status epilepticus (MSE) is defined by the AAN Practice Parameters as “spontaneous repetitive, unrelenting, generalized multifocal myoclonus involving the face, limbs and axial musculature.”³ MSE is felt to portend a poor outcome, with a false positive rate of 0% (95% CI: 0–8.8%).⁴ However, several reports have suggested that MSE may not portend poor prognosis, with several patients achieving a good functional outcome.^23–25 Importantly, in our study, 61 patients had MSE as defined; 2 achieved a good outcome, one of whom had suffered PEA arrest and remained comatose with extensor posturing for 13 days before recovering.²⁶

The absence of bilateral cortical responses (N20 peaks) on SSEP strongly correlates with poor outcome.²⁷ However, prior studies were biased by high rates of early withdrawal of life support, up to 23% at 24 h post-arrest, leading to a self-fulfilling prophesy.²⁷ Unfortunately, SSEPs were performed in only a limited number of patients in our cohort.

Specific EEG patterns, including burst-suppression or electro-graphic status epilepticus, commonly associate with poor outcome, but with unacceptably high false positive rates, along with reports of recovery.²⁸ Recovery of normal EEG rhythms with time is common, with associated clinical improvement.²⁹ Recent findings suggest that EEG reactivity is predictive of good outcome.^22,28 We found no EEG patterns predictive of outcome in our cohort. Serum biomarkers, including neuron specific enolase (NSE), was put forth in the AANPP as an accurate predictor of poor outcome with levels >33 mcg L⁻¹. The value of NSE has recently been drawn into question in TH patients, with reports of levels >33 mcg L⁻¹ in patients who achieved good recovery.³⁰ NSE was tested in very few patients in our cohort, and no conclusions could be made from this.

We have previously published the results of CT and MRI findings from this cohort.^31–33 Reductions in CT Hounsfield unit values, globally and in specific regions, strongly correlate with poor outcome.³¹ Likewise, reductions in MRI apparent diffusion coefficient values correlate with poor outcome with high specificity.³² Changes that develop over time give insights to the nature of progressive injury following cardiac arrest.³³ Larger prospective studies are needed before CT and MRI can be applied in clinical practice.

There are several important limitations of this study. Given the observational nature, we could not dictate treatment decisions, including the use of ancillary testing or decisions to withdraw life-sustaining therapy. Some of the patients in whom life support was withdrawn may have achieved a good outcome had a more aggressive course been taken. This may be especially problematic when considering patients with clinical signs considered to be negative historically, as they are more likely to have less aggressive care. TH is being utilized commonly in patients who are not truly comatose as strictly defined, and those patients are likely to have a better outcome. Our study is one of a pure population of comatose patients. The number of TH-treated patients in this study was relatively small, and future studies will need to emphasize this population specifically. As a single center, the care of patients may have been subject to institutional bias. However, evaluations were performed by multiple examiners, minimizing this bias. We examined patients off sedation whenever possible, largely but not uniformly achieved. This was minimized by the most commonly used sedative being propofol, and an adequate time off sedation was a standard practice to ensure clearance and thus minimize residual effect. We did not prospectively collect data on severity of illness, such as APACHE II or SAPS scores, which might help control for severity of illness. Although the modified Rankin Scale was used for measuring outcome, other scales, such as the Cerebral Performance Category or Glasgow Outcome Scale, are quite suitable for this population. Although this study was performed in the 2000s, we believe it is still quite valid. Many patients are not candidates for therapeutic hypothermia, and yet still need accurate prognostication, and we included an analysis of patients who did undergo TH in our study. Since the advent of TH, there have not been any significant advances in neuroresuscitative care that would alter the impact of our results. We strongly advocate for a standardized approach to comatose cardiac arrest patients, including daily neurological examinations, EEG (specifically looking for reactivity and electro-graphic patterns that may require treatment), SSEP at 48–72 h, and neuroimaging in situations where the prognosis remains in question. An adequate observation period is encouraged when the diagnosis remains uncertain. Similar to our previous findings,⁵ it is difficult to quantify statistical uncertainty to the probability estimates. Nominally, a reasonable rule of thumb is that for events with 10% probability of occurrence, the standard error is about $30\%∕\sqrt{\mathit{n}}$. With 200 patients it is ~2%. CART is mainly an exploratory analysis. Future studies are needed to validate our findings.

5. Conclusion

The neurological examination remains central to prognosis determination in comatose cardiac arrest patients, including those who undergo therapeutic hypothermia. Absent pupillary and corneal reflexes remain accurate predictors of poor outcome. Physicians are encouraged to use all of the tools at their disposal, including the neurological examination, electrophysiology and neuroimaging, to accurately prognosticate for a given patient.