Gershengorn 2016: Association between overnight extubations and outcomes in the intensive care unit.
This is a retrospective cohort study based a database of 97,844 patients in 165 ICUs between 2000-2009. Patients extubated during the day were compared to patients extubated at night (7PM-7AM). The primary endpoint was reintubation, with secondary endpoints including mortality and length of stay. Propensity matching was used to avoid confounding factors. Patients were divided into two groups, based on whether they were intubated for over or under 12 hours. The main results are shown here:
Failure of results to be confirmed by a second dataset
To confirm these results, the authors repeated this analysis using a second dataset including over one million ICU admissions between 2010-2013. This yielded the following:
The two datasets produce conflicting results:
- The primary outcome was reintubation rate. In the 2000-2009 dataset, reintubation correlated with nocturnal extubation among patients who were intubated >12 hours. In the 2010-2013 database there were no differences at all in reintubation rates.
- The most important secondary outcome was mortality. In the 2000-2009 database, mortality correlated with nocturnal extubation in both patient groups. However, in the 2010-2013 database, this correlation only existed among patients who had been intubated >12 hours.
These differences might reflect improvements in care between 2000 and 2010, which increased the safety of nocturnal extubation. Or they might be statistical flukes due to confounding.
Why did the authors focus on the 2000-2009 dataset?
The study focused on the 2000-2009 dataset, relegating the 2010-2013 dataset to the purgatory of the supplemental data section. Disagreements between these datasets were glossed over in the results section, which merely states that “Similarly, in the APACHE Outcomes cohort [2010-2013 dataset], mortality was significantly higher in patients with MV duration of at least 12 hours who underwent over-night extubation.” This ignores differences between the two datasets regarding reintubation rates (the primary outcome).
According to the Methods section, the 2010-2013 dataset was used only for secondary analysis because it was less complete (lacking information about tracheostomies and limitation of care). However, it doesn’t seem that limitation of care was clearly defined in the 2000-2009 database either (1). It is uncommon that patients with tracheostomies undergo disconnection from mechanical ventilation in the ICU, making it unlikely that this would affect the results. Thus, the two datasets may have similar utility. Since the 2010-2013 dataset is more current and larger, it may have deserved greater consideration.
Regardless of which dataset we examine, this analysis cannot prove causality.
Let's consider the correlation between nocturnal extubation and mortality among patients intubated for <12 hours in the 2000-2009 dataset (table below). Given lack of any differences in reintubation rate, it’s hard to believe that nocturnal extubation caused patients to die. The mortality difference likely reflects confounding.
Propensity matching is a statistical technique to account for confounding variables, similar to multivariable regression. It’s a nice technique, but it’s not magic. In order to work, all confounding variables must be included in the propensity matching model. However, many confounding variables weren't included. For example:
- Extubation delayed until after 7 PM due to difficulty awakening from sedation
- Palliative extubation may be more likely to be delayed until after 7 PM
For this complex situation, it is impossible to include all potential confounding variables. Therefore, the results from this analysis are essentially adjusted correlations, which cannot prove causality.
Besides, the study attempts to answer an unanswerable question.
For example, during my training I rotated through a small hospital where nobody was available to intubate at night. If a patient required intubation, an anesthesiologist would drive in from home. Nocturnal extubation in that hospital would not be wise.
Currently I work at a tertiary care center with intensivists and anesthesiologists in-house 24-7, collaborating with experienced respiratory therapists and critical care nurses. We routinely extubate patients around the clock, without any apparent problems.
Is nocturnal extubation good or bad? It depends. The question cannot be answered using big data. My answer is 42, and I’m sticking to it.
- Gershengorn et al. 2016 correlates nocturnal extubation with patient outcomes within two datasets.
- The study focuses on the dataset from 2000-2009 including 97,844 patients. Results from the dataset from 2010-2013 including over one million patients were mostly ignored.
- There are major differences in results obtained with the different datasets. In particular, the primary outcome (reintubation rate) was found to correlate with nocturnal extubation in a subset of patients in the 2000-2009 dataset, but not in either subset within the 2010-2013 dataset.
- Given the correlational nature of this study and disagreement between different datasets, no clear conclusion can be drawn from it.
- It’s doubtful that any universal answer exists to the question of whether nocturnal extubation is advisable. This may depend on the specific patient and amount of resources available.
- In order to perform a sensitivity analysis of the data from the 2000-2009 database excluding patientswho had undergone palliative extubation, “we restricted each subcohort to patients who survived (1) to hospital discharge and separately (2) at least 24 hours after extubation.” Thus, it doesn’t seem that palliative extubation was specifically delineated in this database. This is probably a significant confounding factor present in both datasets.