In our NMA for short-term mortality, there was significant difference in the comparison results of tracheostomy performed within 4 days after intubation and 13 days or later (RR 0.65 [95% CI 0.46–0.93]). In NMA for adverse events, there were no significant differences in all comparisons. In addition, ranking analysis of short-term mortality and adverse events were as follows: within 4 days (P-score 0.96) > 5–12 days (P-score 0.32) > 13 days or later (P-score 0.22) and within 4 days (P-score 0.95) > 13 days or later (P-score 0.40) > 5–12 days (P-score 0.15). In summary, compared with tracheostomy performed 13 days or later after intubation, tracheostomy within 4 days might lower short-term mortality. However, the upper limit of the 95% CI was close to 1, and it is difficult to conclude that there is a benefit.
In our NMA, in the comparison between tracheostomy within 4 days after intubation vs. 13 days or later, the significant decrease in mortality in the group of within 4 days might have been influenced by the rehabilitation and decreased dosage of sedative drug due to the decreased physical restraint and improved activity. The RCTs included in our NMA did not report results on the implementation of rehabilitation, but two RCTs did report on the use of sedative drugs [8, 13]. Rumbak et al. compared tracheostomy within 2 days after intubation with 14-16 days after in 120 patients admitted to the ICU. The duration of sedation (mean ± standard deviation [SD]) was significantly shorter in the early tracheostomy group (3.2 ± 0.4 vs. 14.1 ± 2.9 days; p < 0.001) [8]. Similarly, Bösel et al. compared tracheostomy within 3 days after intubation and 7–14 days after for 60 patients with severe stroke, and the rate of sedative drug use and ICU stay was significantly shorter in the early tracheostomy group (62% vs. 42%; p = 0.02) [13]. However, previous studies suggest that muscle strength showed improvement and ventilator-free days are shortened by enhanced rehabilitation [20], and that prognoses improved through the decreased use of sedative drugs [21, 22]. As the pathophysiological benefits of enhanced rehabilitation and decreased use of sedative drugs are apparent, the benefit attributed to early tracheostomy is arguable. It should also be noted that the results of our NMA suggest that early tracheostomy might improve the prognosis in patients who require tracheostomy. All RCTs included patients who were predicted to require tracheostomy; therefore, the results of our NMA are not applicable to patients who are not predicted to require tracheostomy. If patients who are predicted to require tracheostomy are identified earlier and more accurately, the possibility of performing an early tracheostomy in the optimal population may result in improved patient prognoses. Therefore, for future studies, a model for accurately predicting patients who need tracheostomy might be beneficial.
Two RCTs included in our NMA reported different results regarding short-term mortality when comparing tracheostomy within 4 days after intubation vs. 13 days or later [9, 12]. The reason for the difference in results might be that the patients included in these RCTs were less severely ill, and the timing of tracheostomy might have had a lesser effect on mortality. Zheng et al. compared tracheostomy on 3 days after intubation vs. 15 days for 119 patients admitted to the ICU, and 28-day mortality was similar in both groups (13.8% vs. 9.8%; p = 0.55) [12]. The mean value of Acute Physiology and Chronic Health Evaluation II scores in both groups was 19.5 with a predicted mortality rate of 10%–20%, which is not severe. Therefore, due to the low severity of the disease, it is possible that there was no difference in 28-day mortality [23]. Alternatively, Blot et al. compared tracheostomy within 4 days vs. 14 days after intubation for 123 patients admitted to the ICU [9]. The median value of Simplified Acute Physiology Score II for both groups was 50, and the predicted mortality rate was 40%–50%, but there was no difference in 28-day mortality (20% vs. 24%; p = 0.62) [9, 24]. However, in this study, 60 patients (98%) in the early tracheostomy group underwent a tracheostomy compared with only 16 patients (26%) in the late tracheostomy group. Although the study does not provide any justification for this, we presume that patients in the late tracheostomy group who did not undergo tracheostomy may have been extubated. The severity of patient conditions included in this study was high at the time of inclusion; however, over time, the severity of respiratory failure and their prognosis may have improved. Therefore, the 28-day mortality in the late tracheostomy group might have been underestimated, and there would have been no difference in 28-day mortality in this study without bias.
The results of our NMA might be useful for selecting an appropriate timing of tracheostomy for critically ill patients because tracheostomy performed within 4 days from intubation might reduce short-term mortality in a greater proportion of patients (27–209 patients per 1000) than tracheostomies 13 days or later after intubation. This means that, in patients who are predicted to be on prolonged intubation for more than 13 days and who are tracheostomy candidates, tracheostomy performed within 4 days might reduce short-term mortality. Empirically, patients with severe traumatic brain injury (TBI) or severe respiratory failure are predicted to be on prolonged intubation for more than 13 days. Therefore, it might be important and beneficial to consider an early tracheostomy for these patients, as they are critically ill and predicted to be on prolonged intubation.
There are some limitations in our NMA. First, different grouping for tracheostomy timing might yield different results. In our NMA, we divided the patients into three groups to ensure clinical validity and inclusion of as many studies as possible. This means that, for example, RCTs comparing tracheostomy within 5 days after intubation vs. 5 days after were excluded because they did not fit into our grouping. If the groups were subdivided for smaller time spans or different study methods and if different RCTs were included, our NMA may have yielded different results. Second, the results of our NMA might not be wholly accurate because the number of patients who underwent tracheostomies in the late tracheostomy group in each of the RCTs differed significantly. For example, in a study by Blot et al., only 26% of patients in the late tracheostomy group underwent tracheostomy [9]. In contrast, in the studies by Koch et al. and Bösel et al., 100% of patients in the late tracheostomy group underwent tracheostomy [11, 13]. If extubation was a common reason for not performing tracheostomy, patients with good prognoses were analysed as the late tracheostomy group and the mortality rate may have been underestimated. Therefore, in our NMA, the short-term mortality of the tracheostomy 5–12 days after intubation group and 13 days or later group might have been underestimated, resulting in skewed or inaccurate results. Third, due to the conceptual heterogeneity of the integrated studies, the results of our NMA might not be as reliable as they seem. In other words, if the NMA was limited to a specific clinical cause for ICU admission (i.e.,, disease), different results might have been obtained. For example, in patients with severe TBI who are expected to be on prolonged intubation, early tracheostomy will allow for more enhanced rehabilitation earlier than usual. Thus, early tracheostomy might lead to a decrease in VAP and an improvement in ADLs, thereby leading to better prognoses. Fourth, the number of patients included in our NMA were insufficient for detecting statistically significant differences. An example would be 160 patients included for the comparison of tracheostomy within 4 days after intubation vs. 5–12 days and short-term mortality in our NMA; thus, 160 patients may not have been enough to yield generalisable results. If more knowledge about this subject is accumulated in the future and more studies or patients are included in future NMAs, different results might be obtained. Finally, the results of our NMA were influenced by the effect modifier; patients included in the RCTs in our NMA differed in the causes of respiratory failure and the severity of illness in each study, which might have been effect modifiers. In addition, there are other factors that might have been effect modifiers, such as the use or no use of sedative drugs, organ failure, and the dose of sedative administered. Therefore, the analysis of our NMA might have been influenced by these effect modifiers.