This editorial refers to ‘A practical risk score for early prediction of neurological outcome after out-of-hospital cardiac arrest: MIRACLE2’†, by N. Pareek et al., on page .
Over 90% of patients suffering cardiac arrest die, most before arriving at a hospital. Efforts to ensure optimal care in the ‘chain of survival’, including community response, emergency medical services care, and hospital care, can substantially improve outcome, but mortality remains high. There are few situations in medicine more satisfying than the complete recovery of a previously healthy patient with cardiac arrest. Yet there are many patients who die after days or weeks of aggressive care, when knowing that would be the outcome could make decision-making for the family and care team more appropriate. To date, limited data are available to help clinicians navigate this complex scenario.
Many factors are associated with outcome after cardiac arrest. The Utstein Style established data elements to classify out-of-hospital cardiac arrests into categories that relate to outcome.1 For example, patients with arrests with cardiac aetiology, witnessed by bystanders, and with shockable rhythm have more favourable outcome than those without these features. Older age and comorbidities, and presence of cardiogenic shock are other major contributors to risk. The remaining challenge is how a clinician should interpret these inter-related factors to make the most accurate risk assessment for each patient.
The clinician treating cardiac arrest is faced with one of the most important decisions one can make: given that most patients will die, when is the likelihood of recovery so small that ongoing care is futile? Unfortunately, the studies that have tried to define this are small and they are inadequate to provide the level of certainty which providers and families would desire in this setting. Guidelines2–4 and professional society statements5 provide some recommendations, for example to continue life support for at least 4 or 5 days after arrest treated with targeted temperature management before declaring no chance of recovery. They also acknowledge that a multifactorial approach is needed, and uncertainty remains when assessing risk of death.3 Moreover, there is the issue that risk assessment is dynamic and thus our approach must be customized to the time of assessment: three such important time points are the time of resuscitation, the time of hospital presentation, and several days later (Figure 1).
Determining if prognosis is so poor after cardiac arrest that limitation of care should be considered.
Pareek et al., in this issue of the European Heart Journal,6 provide important results of a study to help fill the knowledge gap for one of these time points, that being at the time of hospital admission. Their derivation data consisted of 373 patients who were alive at hospital presentation to a heart attack centre in London, UK, who prior to arresting had good neurological function and lacked severe comorbidities that would lead to life expectancy <6 months. Using seven independent risk factors, the investigators developed a simple point score to estimate the chance of death or poor neurological function at 6 months. The seven independent risk factors were unwitnessed arrest, initial non-shockable rhythm, non-reactive pupils, older age (60–80 years, 1 point; >80 years, 3 points), changing intra-arrest rhythms, blood pH <7.20, and epinephrine administration (2 points). Four of these—older age, shockable rhythm, pH, and epinephrine use—were also factors in the Cardiac Arrest Hospital Prognosis (CAHP) score,7 the most relevant of several previously published prediction tools.8 The CAHP score also addressed prognosis assessed at the time of hospital admission for the outcome of death or poor neurological recovery, and the outcome was at time of intensive care unit discharge. The CAHP risk score also performed well, with a c-index in the derivation dataset of 0.93, and with 99% of patients in the high-risk group having unfavourable outcome (although this dropped to 96% in one of the two validation datasets).8 Strengths of the current study are the longer term follow-up and the use of two independent datasets for validation, also from centres performing primary percutaneous coronary intervention (PCI), in Slovenia and in London. The derivation population had a high rate of ST elevation myocardial infarction (56%) and were treated with primary PCI (47%), as well as a high incidence of shockable rhythm (70%). Forty percent of the population survived with good neurological function, showing that this was a lower than usual risk population. This is important, as most out-of-hospital cardiac arrest patients do not resemble this population, thus limiting the generalizability of the MIRACLE2 score. Half of the patients fell in the high-risk category (MIRACLE2 score ≥5), and 92% of those died or had poor neurological outcome at 6 months. In one validation dataset, that figure dropped to 90%, such that 10% of those deemed ‘high risk’ were alive with good neurological function at 6 months.
The measure of performance of a model depends on its intended use. The C-statistic or c-index is the standard tool to test how well the model discriminates between those who do and do not have the outcome. A c-statistic of 0.90 is excellent, meaning 9 of 10 times the model correctly identifies the higher risk patient given randomly selected pairs of patients with and without the event. The calibration, which reflects how the actual risk correlates with the estimated risk, was also good in the validation sets. However, for the purpose of this model, the most important measure is the positive predictive value, or how frequently did a patient die when the model predicted high risk of death. Because one would generally want to continue to provide life support unless poor outcome was nearly certain, this has to be extremely high to be clinically helpful, at least for most patients. Is a 92% chance of death high enough to forgo potentially lifesaving care, such as primary PCI? If it is a younger previously healthy patient, should it be 98% or 99%? This will vary based on family preference and on clinical judgement, which will depend on other factors. Having the MIRACLE2 score will not change acute care for most patients with cardiac arrest, since the ability to identify those who have an extremely high chance of poor outcome is limited. International guidelines recommend taking patients with cardiac arrest and ST elevation, including those with cardiogenic shock, immediately for primary PCI, unless they have extenuating circumstances that make survival with desirable quality of life extremely unlikely. Having a more accurate estimate of longer term outcome to partner with families to make the best decisions will be helpful for selected patients, such as those with relatively poor medical or functional status at baseline. It is important to note, however, that patients with very poor baseline health status were excluded from the population from which MIRACLE2 was derived.
Some important factors may not have been included in this score. For example, even though not awake [Glasgow Coma Score (GCS) score 15 were excluded] at admission, was the patient beginning to respond, which would identify a patient highly likely to recover regardless of the MIRACLE2 score? Bystander CPR and bystander AED (automated external defibrillator) placement, now used commonly in some northern European countries,9 may be important predictors of better outcome. While a simple score may be helpful, other factors must also be considered to make the best decisions.
A situation where better prognostic tools are even more needed is at the assessment several days later, for the patient who remains comatose, to decide whether to withdraw life support care. A study from the Ruscitations Outcome Consortion10 found that a third of patients had withdrawal of life support within 72 h after admission because of perceived poor neurological prognosis, and it was estimated that up to 65% of those patients might have recovered if care had been continued. Thus, continuing supportive care until one can determine very low likelihood of recovery is the goal. Some believe that a major benefit of targeted temperature management is simply that it delays the decision to withdraw life support, providing a longer opportunity for recovery. Important prognostic factors for outcome for those still comatose at 4–5 days include clinical (age, prior status, comorbidities), imaging findings, evoked potentials, electroencaphalography (EEG) findings, and biomarkers. The available studies that have systematically addressed the dynamic status of risk assessment and the multiple potentially important factors are generally limited by small numbers and lack of validation. High-quality, large studies are urgently needed for this risk assessment purpose.11
For now, the MIRACLE2 score is an effective tool for assessing longer term outcome for patients with cardiac arrest presenting to hospitals providing primary PCI for acute myocardial infarction. However, it should not be used as the sole factor to decide who should have aggressive care withheld. It can be included as one of several parameters to identify the unusual patient who may have such a low chance of good recovery at presentation that limiting care may be appropriate. The greatest need now is to develop tools to predict futile care for those with persistent coma several days after arrest.
The opinions expressed in this article are not necessarily those of the Editors of the European Heart Journal or of the European Society of Cardiology.
Footnotes
† doi:10.1093/eurheartj/ehaa570.
Acknowledgements
We acknowledge Karen Pieper, MS, for her statistical review of and input to our Editorial.
Conflict of interest: none declared.
References
