Web site designed and maintained by
Chris Cairns
© SICS EBM Group 2004
|
|
|
|
Timing of tracheostomy: a meta-analysis.
Corresponding author: J Duncan Young: duncan.young@nda.ox.ac.uk Three-part Clinical Question: Does early tracheostomy in adult intensive care patients alter the incidence of death, pneumonia; and lengths of ventilation and critical care stays? Search Terms: tracheostomy AND respiration, artificial AND randomised controlled trial. The Review: Data Sources: Cochrane Library, Medline, Embase, CINAHL, National Research Register, NHS Clinical Trials Register, MRC Database, BHF Database, NHS R+D health technology assessment program. Study Selection: Randomised or quasi-randomised trials. Patients were adults who required artificial ventilation. Early tracheostomy was compared to either late tracheostomy or prolonged trans-laryngeal intubation. Early tracheostomy meant at less than seven days from start of ventilation. Data Extraction: Data on mortality from 4 studies; pneumonia from 5; length of ventilation from 4; length of critical care stay from 2. The studies were multiple independent reviews of individual reports. They were tested for heterogeneity. The Evidence: 1. Mortality. Information from 4 studies (332 patients). The relative risk with the random effects meta-analysis is 0.79 for death following early tracheostomy (95% CI 0.45 – 1.39). 2. Pneumonia. Information from 5 studies (406 patients). The relative risk with the random effects meta-analysis is 0.90 for pneumonia following early tracheostomy (95% CI 0.66 – 1.21). 3. Duration of ventilation. Information from 4 studies (332 patients). The weighted mean difference was 8.5 days less in the early tracheostomy group (95% CI 1.7-15.3 days; p=0.03) 4. Duration of critical care stay. Information from two studies (226 patients). The weighted mean difference was 15.3 days less in the early tracheostomy group (95% CI 6.1 to 24.6 days; p=0.001) Comments and EBM questions: 1) Do the methods allow accurate testing of the hypothesis?
In principle the methodology seems sound. Only English language articles were included. The total patient numbers were small (406) from 5 trials; there was not always agreement between the trials. One conclusion was based on only two studies. One paper only compared early tracheostomy to no tracheostomy. One paper probably should not have been included at all, although this perhaps would not change the take home message. There are several errors that suggest inadequate scrutiny of the composite studies.
There are specific problems with each paper as detailed in their respective CATs. A summary of some of the broader issues is presented in the table.
The specific problems with the meta-analysis of each of the outcomes are:
1. Mortality. There was inconsistency between the results with an I2 of 57.8%.
2.Pneumonia. There was considerable inconsistency between the results (again discussed) with an I2 of 86.5% (statistical heterogeneity). The definitions used for pneumonia varied widely between the five studies, and the authors did discuss this. The incidences of pneumonia varied, perhaps relating to the definition used to make the diagnosis (methodological heterogeneity). The weighting method used in the meta-analysis considers both the number of patients and the incidence of the outcome studied. Therefore, the study by Rumbak et al, which showed the greatest difference and a low incidence in the early tracheostomy group, got the least weight (5.3%) in the meta-analysis. The converse applied to the study of Saffle et al. This study had the smallest number of patients. Nearly all of the patients got pneumonia, so this study was weighted the greatest (29%). Additionally, there are likely to be differences between the studies in the approach to prevention of pneumonia (clinical heterogeneity). Only Rumbak et al discussed their approach. These might include the use of head up tilt – 3 of these studies looked at either trauma or head injuries; antibiotic protocols and use of interchangeable inner cannulae. There may be a detection bias as it appears that the diagnosis of pneumonia was not independent of the study, except in that of Dunham & LaMonica. This study considered “respiratory sepsis”, of which pneumonia was part. It probably was inappropriate to include this paper in the meta-analysis as we do not know what the incidence of pneumonia was at all. The authors of the meta-analysis grouped all “respiratory sepsis” as pneumonia.
3. Ventilated days. There was considerable inconsistency between the results with an I2 of 81.3%. This is compounded by the possibility of a sedation effect: were patients with a trans-laryngeal tube over-sedated, or is the advantage of tracheostomy that sedation can be reduced: i.e. are the studies largely pragmatic? The other issue is that at least in head injury patients (Bouderka et al), there is the need for airway protection and in the absence of a strict weaning protocol these patients may remain ventilated.
4. ICU stay. There was considerable inconsistency between the results with an I2 of 86.9%. There were only two studies contributing to this. Rumbak et al used a weaning floor, separate from the ICU. Not only does this make the ICU stay a very soft outcome measure, it is also clearly prone to both bias and logistical issues. This result is so limited in it usefulness that perhaps it should not have been subject to analysis.
Table. Comparison of the five contributory papers:
* all patients received the allocated treatment: there is no need for intention to treat analysis.
2) Do the statistical tests allow differentiation of statistically significant results? Yes. The use of random effect model is appropriate since the outcome may vary. There was considerable statistical heterogeneity between the studies. This suggests that differing results were less likely to have occurred by chance.
3) Are conclusions valid in light of the results? That the incidence of death does not differ is a valid conclusion from the methods used and the contributing studies. There are problems with the conclusions on pneumonia as discussed in section 1. There are difficulties in interpreting the time in critical care and this is discussed in section 1. Is the conclusion that the time of mechanical ventilation differs by 8.5 days physiologically plausible?
4) Did results get omitted, and why? No. However, only five randomised or quasi-randomised trials were identified and not all of these reported each outcome. Two studies reported hospital stay as an outcome; this may be more robust than critical care stay but was not subject to meta-analysis.
Yes, although this was a mention of the recently started Trachman study. This, and other studies aim to be sufficiently powered in a heterogeneous intensive care population to see if early compared to late tracheostomy alters outcome.
Yes. They suggest that a tracheostomy is placed early in the course of artificial ventilation. The meta-analysis suggests that this may reduce duration of ventilation. As discussed by the authors, predicting who would need a tracheostomy may be difficult and this was also an issue in the contributing studies.
Yes, many intensive care patients may need prolonged ventilation and be considered for tracheostomy.
1-: meta-analysis with inconsistent results in a mixed population.
9) What grade of recommendation can I make on this result alone? C
10) What grade of recommendation can I make when this study is considered along with other available evidence? This is the only recent meta-analysis. It may be more valid, from a strength of evidence perspective, to consider the results of Rumbak et al in isolation and proceed to early tracheostomy in sick medical patients likely to need prolonged ventilation.
11) Should I change my practice because of these results? This meta-analysis does not provide sufficient evidence to change practice. The results of the large multi-centre studies may provide more information.
Not as a result of this meta-analysis. Appraised by: Richard Price, SpR, Dept
Anaesthetics: 30 Shelley Court, Gartnavel Hospital, Glasgow, G12 0YN. 09 May
2006. Kill or Update By: May 2011. ©SICS EBM 2006
|
