Lars Anderson

Association Between Tracheal Intubation During In-Hospital Cardiac Arrest and Survival

Anderson, Lars W. JAMA 2017;317(5):494-506. Doi:10.1001

Clinical Question

  • In adult patients with in-hospital cardiac arrest, is tracheal intubation associated with survival to hospital discharge?

Design

  • Retrospective, observational matched cohort study
  • Data from Get With the Guidelines-Resuscitation (GWTG-R) A US-based multi-centre prospective registry of in-hospital cardiac arrest
  • Time-dependent propensity score matching used to compare intervention vs. control group
  • 1:1 risk matching on the propensity score using nearest neighbour-matching algorithm
  • Primary outcome was survival to discharge
  • Secondary outcome of Return of Spontaneous Circulation (ROSC), defined as no further need for chest compressions (including bypass)
  • Secondary outcome of good functional outcome defined as a cerebral perfusion performance score of 1 or 2 (consistent with Utstein guidelines)
  • Data abstractors not blinded to exposure status but unaware of study hypothesis
  • Sample size calculation not performed

Setting

  • 668 Hospitals in the U.S.
  • Data collected from January 1st 2000 – December 31st 2014

Population

  • Inclusion critieria:
    • Adults >18 years old
    • In-hospital cardiac arrests for which they received chest compressions
  • Exclusion criteria:
    • Invasive airway in place at time of arrest
    • Visitor or employee of the hospital
    • DNACPR in place
    • Patients with missing data on tracheal intubation or co-variates (excluded from the main analysis, they were then included after imputation of missing values in a pre-planned sensitivity analysis)
  • 143,810 met inclusion criteria
    • 35,731 excluded from main analysis due to missing data (included in sensitivity analysis)
    • 108,079 included in main analysis
      • 71,615 intubated within 15 minutes
      • 3,964 intubated after 15 minutes
      • 32,500 not intubated
  • Comparing baseline characteristics of patients intubated within 1st 15 minutes vs. not intubated/intubated >15 minutes
    • Age (median): 70 vs 68
    • Illness category
      • Cardiac: 41% vs. 48%
    • Pre-existing condition
      • Myocardial infarction: 17% vs. 16%
      • Diabetes: 34% vs. 30%
      • Renal insufficiency: 35% vs. 32%
    • In-place at time of cardiac arrest
      • Non-invasive assisted ventilation: 4% vs. 22%
      • Intra-arterial catheter: 3% vs. 6%
      • Electrocardiogram: 71% vs. 82%
      • Pulse oximeter: 53% vs. 68%
      • Vasoactive drugs: 12% vs. 21%
    • Location of cardiac arrest
      • Intensive care unit: 30% vs. 48%
      • Floor without telemetry: 30% vs. 17%
    • Witnessed cardiac arrest: 75% vs. 84%
    • 1st documented pulseless rhythm
      • Asystole: 38% vs. 32%
      • PEA: 47% vs. 44%
      • VF: 10% vs. 15%
      • Pulseless VT: 5% vs. 9%
    • Adrenaline administration: 95% vs. 76%
    • Time to defibrillation, median (IQR), min: 2 (0-4), vs. 1 (0-2)
    • evenly matched between groups
  • In time-dependent propensity score all above baseline characteristics were evenly matched. Comparing the intubated vs. non-intubated group
    • Intubated group: time to intubation 4 minutes (IQR 2-6min)
    • In non-intubated group: 68.2% intubated. For these patients time to intubation 8min (IQR 5-12min)

Intervention

  • Patients successfully intubated within first 15 minutes
    • Defined as intubation of a tracheal or tracheostomy tube during the cardiac arrest
    • Unsuccessful intubations were not recorded
    • Time to tracheal intubation was the interval in whole minutes from loss of pulses to tracheal tube insertion

Control

  • Patients not intubated within first 15 minutes
    • Included patients who were subsequently intubated
    • In time dependent propensity score matching, patients that were being intubated at any given minute (from 0-15min, intubation group) were matched with patients at risk of being intubated within the same minute i.e. still receiving resuscitation, no intubation group). This meant that a large number of patients in this group (68.2%) were actually intubated at a later stage but the majority were still intubated within 15 minutes

Outcome

  • Primary outcome: Survival to hospital discharge – significantly lower in intubation group, p<0.001
    • Comparing no intubation vs. intubation
      • Unadjusted analysis
        • 33.2% vs. 17%, risk ratio (RR) 0.58 (95% C.I. 0.57-0.59)
        • 12,116/36,464 vs. 12,140/71,615
      • Propensity Score-Matched Analysis
        • 19.4% vs. 16.3%, RR 0.84 (95% C.I. 0.81-0.87)
        • 8,407/43,314 vs. 7,052/43,314
  • Secondary outcomes:
    • Return of spontaneous circulation – significantly lower in intubation group, p<0.001
      • Unadjusted analysis (data missing for 7 patients)
        • 69% vs. 59.2%, RR 0.75 (95% C.I. 0.73-0.76)
      • Propensity Score-Matched Analysis (data missing for 7 patients)
        • 59.3% vs. 57.8%, RR 0.97 (95% C.I. 0.96-0.99)
        • 25,685/43,310 vs. 25,022/43,311
    • Favourable functional outcome at discharge – significantly lower in intubation group, p<0.001
      • Unadjusted analysis (data missing for 4,631 patients, 4.3%)
        • 25.7% vs. 11.2%, RR 0.55 (95% C.I. 0.54-0.56)
      • Propensity Score-Matched Analysis (data missing for 3,027 patients, 3.5%)
        • 13.6% vs. 10.6%, RR 0.78 (95% C.I. 0.75-0.81)
        • 5,672/41,733 vs. 4,439/41,868
  • Sensitivity analysis (included patients with missing data)
    • Reported similar results for primary and secondary outcomes
  • Sub-group analysis for propensity matched – intubation vs. no intubation group
    • Survival to hospital discharge based on initial rhythm
      • Intubation was associated with a significantly greater negative impact in patients with shockable vs. non-shockable rhythms
        • Shockable (n=13,321)
          • 26.8% vs. 39.2% (RR 0.68, 95% C.I. 0.65-0.72)
        • Non-shockable (n=73,307)
          • 14.4% vs. 15.8% (RR 0.91, 95%C.I. 0.88-0.94)

Authors’ Conclusions

  • In adult patients with in-hospital cardiac arrest, early endotracheal intubation (<15minutes) compared with no intubation is associated with decreased survival to hospital discharge.

Strengths

  • Large multi-centre study
  • The data set is large and generalisable to our population
  • The data set is from a prospective registry
  • The use of time-based propensity attempts to minimise effects of confounding variables

Weaknesses

  • Unsuccessful intubations not recorded or time taken to intubate which is likely to be a large confounding factor in outcome.
  • Patients who are intubated may have greater severity of illness in the first place, and efforts to adjust for severity of illness may fail to fully account for residual indication bias
  • Intubation may facilitate higher oxygen concentration delivery, which has been associated with harm, therefore it’s not necessarily the intubation but the oxygen management that could be related to worse outcomes.
  • Practices in resuscitation have changed over the 15 years of data collection
  • Control patients are individuals who were not intubated during the same minute that an exposed case was intubated. However, these controls may become exposed cases in subsequent minutes. In the propensity matched groups, a number of patients that were intubated within 15 minutes were included in the no intubation group. These patients could have also been included in the intubation group.
  • Baseline difference – significantly more patients with VF/VT and witnessed arrest in no intubation group; and significantly lower use of adrenaline
  • Due to the difficulty of collecting accurate data during a cardiac arrest there is the  possibility of timings being misclassified

The Bottom Line

  • In patients that have an in-hospital cardiac arrest the use of intubation was associated with an increased mortality. These results were significant in both the unadjusted results and the propensity score-matched analysis. Due to weaknesses in the study design and concerns over possible missed confounding variables, further studies are required. This study has helped to demonstrate equipoise that should allow a RCT to be performed. In the mean time, this paper will certainly make me think twice before I reach for the laryngoscope. It also highlights to me the importance of minimal interruptions to effective chest compressions and as always to make decisions on management of patients in a case-by-case fashion.

External Links

Metadata

Summary author: David Harvie
Summary date: 12/04/2017
Peer-review editor: @davidslessor

6 comments

  • Pingback: In-Hospital Cardiac Arrest: The First 15 Minutes - R.E.B.E.L. EM - Emergency Medicine Blog

  • Pierce geoghegan

    Propensity score matching (PSM) is a tool which makes it possible in principle to make robust assertions about cause and effect based on observational data. PSM makes some important and fundamental assumptions:
    1. Each patient could really have received either of the treatments being compared. The idea is that we compare patients who had equal probabilities of receiving a treatment, some of whom “by chance” were allocated to treatment and others not (we are supposed to have accounted for all other major determinants of treatment allocation in calculating the propensity score, therefore chance is the major determinant of treatment allocation that remains.) This is a fundamental assumption and if it’s not true PSM is an inappropriate analysis.
    2. We are comparing patients with equal treatment propensity (probability of treatment allocation). Note that this is a very very difficult thing to establish in practice. This is because to know any individual’s probability of treatment allocation you have to have a very robust causal model of treatment allocation (a good model of all the factors which might increase or decrease the probability of treatment allocation). But treatment allocation is an extremely complex phenomenon and mapping out it’s causal determinants is a hard problem. Doing this would involve a series of observational studies and experiments in it’s own right (which are usually not done). For some inexplicable reason, many researchers simply see which of a finite list of incidentally measured variables is associated with treatment propensity and use these instead of mapping out the causal determinants.

    Back to the paper at hand by Anderson et al. Consider assumption 1: Could all patients really have received the treatments being compared? The answer is that we do not know. We are comparing ventilation with or without intubation. But was it possible to ventilate all the patients without intubation? If not clinicians were forced to intubate (the decision did not occur by chance) – this violates assumption 1 and the analysis fails.

    Now consider assumption 2. What kind of factors determine the probability that intubation will occur in a given minute of cardiac arrest? The authors do not explicitly outline their causal model of treatment allocation as regards intubation vs not in a given minute of cardiac arrest. This is a major flaw that undermines the analysis. In fact, treatment propensity was based on factors that happened to be measured in an observational dataset not designed for this purpose, rather than factors likely to be causal determinants. As a clinician I would suggest some critical factors determining treatment propensity here might be ease of ventilation without intubation (likely a major determinant), operator experience, presence or absence of airway soiling, and the etiology of the arrest. Are these factors used in calculating the propensity score? Essentially no.

    In summary this analysis is fundamentally flawed because 1) It’s not clear that all patients could have been allocated to both treatments being compared (a fundamental assumption of PSM) and 2) The propensity scores for individual patients are probably not reflective of actual treatment propensity because they are not calculated using all (or even most) of the factors which causally determine whether clinicians will intubate during cardiac arrest, but rather were estimated based on other factors which probably are not causal determinants of treatment allocation.

    e.g. Suppose a patient vomits, aspirates and suffers a hypoxic cardiac arrest. The resuscitating team note gross airway contamination and suction and intubation occurs early during the resuscitation. ROSC occurs but ultimately the patient dies. Did early intubation play a causal role in this man’s death? Would he have done better if intubation had been delayed? All sensible clinicians know the answer to this question and it should illustrate why drawing conclusions from this flawed analysis will only lead clinicians astray.

    Conclusion: It is possible that patients may be harmed by the inappropriate deferral of intubation based on a very flawed analysis of this large dataset and overinterpretation of this flawed analysis. Practice personalised medicine. If you need to tube during a cardiac arrest, tube. If you don’t or can’t or it’s not going to help, don’t.

  • Just a few thoughts and please prove me wrong!

    Incidents: 46% on a regular floor, 25% even not on telemetry, 22% non witnessed (16% non-intubation vs. 25% intubation).
    Maybe we should discuss early warning scores before we abolish intubation based on this study?!?

    ‘Baseline difference – significantly more patients with VF/VT and witnessed arrest in no intubation group; and significantly lower use of adrenaline’
    Might lead to the hypothesis: The non-intubation group included those patients in need of just a short sequence of CPR vs. the intubation group with prolonged CPR.
    Nobody can tell or did I overlooked time to ROSC/duration of CPR?

    Even a little more provocative: The cohort in this study would have benefitted from intubation.
    Sure, not pre-ROSC but maybe 1 hour before coding.

    • Duncan Chambler

      Simon,
      Thanks for your comment and sorry I’m only just replying now. I agree with your points. In some bigger hospital institutions, an in hospital cardiac arrest is probably no better than a community cardiac arrest, as the patient may not be monitored and the arrest may not be detected until the next set of observations are due! Scary thought.
      I think this is a provocative study but it just raises more questions than it answers. I can’t see an RCT ever being conducted – do we ever have equipoise about such cases??

  • Pingback: SGEM#197: Die Trying – Intubation of In-Hospital Cardiac Arrests | The Skeptics Guide to Emergency Medicine

  • Pingback: BAOIA Presentation 2018 – The Bottom Line

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.