Monday, 2 February 2015

TRACMAN - Early versus late tracheostomy for prolonged ventilation in ICU

Key Question

Does the insertion of a tracheostomy within 4 days of ICU admission in mechanically ventilated patients, who are predicted to require a further 7 days or more of ventilation, improve 30 day all-cause mortality?

Paper Link

Young D et al for the TracMan collaborators
Journal of the American Medical Association (2013); 309:2121-2129

At a glance

  • Randomised - A qualified yes, as explained below
  • Controlled - Yes - early tracheostomy compared with late
  • Placebo used - No
  • Blinded - No
  • Population / Sample size calculation - Study failed to recruit 60% of initial planned numbers, resulting in under powering to detect clinically meaningful outcomes
  • Groups same at randomisation - Yes
  • Groups treated the same after randomisation - Yes
  • Intention to treat analysis - Yes
  • Follow-up - 10 of 908 patients were lost to followup
  • Significant primary endpoint - No difference found in all cause 30 day mortality
  • Harms analysed - Partly - some clinically relevant outcomes were not recorded, including long term complications


Tracheostomy is a commonly performed clinical procedure in the ICU.  Since the advent of percutaneous dilational tracheostomy at the bedside in the mid 1980s, the technique has gained popularity, with over 15,000 procedures performed annually in the United Kingdom alone.  Advocates for the procedure argue that as a tracheostomy is more comfortable for the patient than an oral endotracheal tube, its insertion is associated with significantly less sedation with consequent improvements in ventilation times, ICU and hospital length of stay, costs and ventilator-associated complications such as pneumonia.  Furthermore, early insertion of a tracheostomy may be associated with greater benefit than late.

Timing of tracheostomy has thus been somewhat controversial with very wide range of practice.  In a large survey of UK-based units, the median day of insertion was day 10 of ICU admission, though up to 13% of tracheostomies were placed prior to day 3.  Further complicating the discussion was the lack of well validated prediction rules for anticipated duration of mechanical ventilation, with numerous studies demonstrating poor discrimination compared with a clinician's best guess.

In the early 2000s, the Intensive Care Society of the United Kingdom identified the issue as a priority for research.  A meta-analysis of available data was conducted, suggesting that early tracheostomy led to reduced ICU length-of-stay and reduced ventilation days, and possibly a reduction in mortality.

The Study

Based on this data, the TracMan collaborators set about studying the hypothesis that early tracheostomy would reduce sedation use, and thereby ventilation duration and associated complications.

The study initially aimed to recruit just over 1600 patients, with the intent of detecting a 6.3% mortality reduction, from 30 to 22.7% (based on the results of a previous study).  Sadly, they were unable to maintain recruitment and the study was eventually ceased at 900 patients, the investigators blaming "recruitment fatigue" and lack of ongoing funds.

The patients were "randomised" to two groups.  In truth, this was not a true randomisation process, rather a process known as Prospective Minimisation.  This statistically valid alternative to randomisation is a method used when investigators seek to balance out combinations of known risk factors within a sample.  For example, if you wanted to ensure that the number of elderly smoking patients with diabetes was equal in both groups, not just that the numbers of elderly patients were equal, the number of smokers were equal and the number of diabetics were equal, you would need a very large sample size to balance out all these variables using a pure randomisation method, where patients have a 50:50 chance of going into either arm.  

In Prospective Minimisation, patients are allocated a likelihood of randomisation to one or other arm of the study based on the allocation of previous patients with similar sets of conditions.  In the example above, lets say the last 5 elderly smoking diabetic patients had been allocated to arm A. The next one would be randomised with an increased chance (e.g. 80%) of going into arm B, rather than 50:50.  Its still partly chance -  a bit like a toss of a coin, but one side of the coin weighs more than the other.  In this way, the balance of the two groups tends to be improved.

This technique is usually only used when the sample size is small or these permutations are particularly important.  Its not clear from the paper why they would use it in a study aiming to recruit 1600 people.  For more information on this check out the following reference

In the end, the study enrolled 908 patients within 4 days of admission, from 72 adult intensive care units including 2 cardiothoracic units, 13 tertiary units and 59 community hospitals, over a 4 year period from 2004.  70% of patients had primary pulmonary failure and less than 10% had primary neurological failure as the indication for ventilation.  The mean APACHE 2 score was 20.

The patients selected were mechanically ventilated patients, for whom the clinician felt that ventilation would be expected to be required for at least 7 more days.  No guidelines were provided for this prediction, and the authors note that there is currently no accepted prediction tool available.  Patients with severe neurological or spinal disease, those with anatomical abnormalities and those with primary ENT reasons for tracheostomy were not part of the study.  This is very important when considering the outcomes of the paper - in this group, clinicians were unsure of the likely time-course of the disease

Patients were randomised to either early or tracheostomy.  The early group underwent tracheostomy as early as practical, with 85% receiving it before day 4 of their admission.  7% never received a tracheostomy, due to instability, early death or more rapid recovery than expected.  In the late group, tracheostomy was intended to be delayed until day 10 or later, and only conducted if the clinician still believed it would be necessary.  Interestingly, 53% of the late group never received a tracheostomy - in two thirds of these patients (one third of the overall group), this was because the patient had improved more than the clinician expected.

The results

The primary outcome, all cause 30 day mortality, was not significantly different.

The major premise of the trial is that tracheostomy likely assists patients by reducing the requirement for sedation.  The TracMan study measured this by recording the number of days free of sedative drugs.  In the first 30 days after randomisation, the survivors of the early group had 3 fewer days of sedation compared with the late group; no difference was seen in the non-survivors, which is not unexpected.  This suggests that the intervention arm successfully achieved separation from the control arm, though the authors had believed this would be more impressive.

The early group was ventilated for 1.7 days less than the late group, though the p value of 0.06 narrowly missed significance.

No other secondary outcome measure was statistically significant, including length of ICU stay, length of hospital stay, 1 year or 2 year mortality, or antibiotic usage (a surrogate for infective complications).

Complications were recorded, but only those at the time of insertion.  Major bleeding was the most common complication in all tracheostomies, occurring in 3% of cases.  No long term complications such as tracheal stenosis, laryngeal injury or fistula formation were reported.  No direct measure of patient comfort was recorded.

The rub

So where does this leave us?

One major limitation of the trial is the inability to recruit the numbers they'd hoped for.  The study power was based on a mortality benefit of 7.5%, seen only in one paper included in the original meta-analysis.  To be able to detect this absolute risk reduction, 1600 patients were required, TracMan achieving a little over half of this.  Consequently, the study is underpowered to detect any mortality reduction less than 8.3% (which of course would still be important).

A key finding of the trial is the inability of clinicians to accurately predict which patients would ultimately require a tracheostomy in the late group.  This meant that potentially 1 third of patients with the eligibility criteria could be spared an invasive, potentially dangerous procedure by simply waiting, with the almost-statistically-significant reduction in ventilation time of 1.7 days the only consequence.  

Notably, this is a sick group.  Over 50% of patients in the study were dead at 24 months, a sobering statistic to say the least.

Overall, this study provides some further insights into the use of tracheostomy in prolonged ventilation.  The results of the trial are consistent with other studies in the area, including those published during the study by Blot and Terrangi, and highlight the difficulties still faced by clinicians when deciding on timing.  The results are likely to be generalisable across similar inclusion criteria.  Its worthy of note that patients in whom tracheostomy is obviously necessary - severe TBI, severe Guillain-Barre Syndrome etc - were NOT included in this trial.

Of course, the use of day 10 is entirely arbitrary.  The study results do not provide us with the ideal timing of tracheostomy, so its likely the debate around this topic will continue for some time to come.  In short, until better predictors of prolonged ventilation come along, early tracheostomy cannot be recommended.

Additional Resources

  • Duncan Young, lead investigator of TRACMAN, discusses his paper

Reviews and Editorials