The Medical Information Mart for Intensive Care database (MIMIC-III) is a single-centre database from Beth Israel Deaconess Medical Center, a large, tertiary care teaching hospital affiliated with Harvard Medical School. This database includes data from over 50,000 critical care admissions from 2001through 2012.Reference Johnson, Pollard, Shen, Lehman, Feng and Ghassemi11 MIMIC-III includes detailed patient information with data ranging from imaging reports with radiologist interpretation to various interventions and laboratory results with time stamps. The Institutional Review Boards of Beth Israel Deaconess Medical Center and the Massachusetts Institute of Technology approved use of these data, which are freely accessible. Because MIMIC-III is a de-identified database, approval from the Rutgers New Jersey Medical School Institutional Review Board was not required.

MIMIC-III was queried for all patients that received a tracheostomy during their stay within the ICU via the World Health Organization's International Classification of Disease, 9th revision procedure codes. The procedure codes included were 31.1, 31.2, 31.21, 31.29 and 31.74. The chart date and time for the tracheostomy International Classification of Disease codes were identified for each patient and defined as the start time. Patients that had a start time prior to their ICU admission or after downgrade were excluded. The time from ICU admission to tracheostomy was calculated for each patient.

Patient demographic information included sex, ethnicity and age. In compliance with the Health Insurance Portability and Accountability Act, MIMIC-III recodes all patient ages over 89 years to greater than 300 years. These patients were identified and had their ages changed to 90 years. Additional variables included type of admission (elective, urgent or emergency), mortality during hospital stay and various comorbid conditions reported within the database. Included comorbidities were congestive heart failure, cardiac arrythmias, valvular disease, peripheral vascular disease, hypertension, uncomplicated diabetes mellitus, complicated diabetes mellitus, hypothyroidism, renal failure, liver disease, metastatic cancer, rheumatoid arthritis, coagulopathy, obesity and fluid electrolyte imbalance.

Patients were also categorized as having a prolonged length of stay in the ICU if they had a stay greater than the 75th percentile. Patients were grouped into early and late tracheostomy groups. Patients were defined to have received an early tracheostomy if their time to tracheostomy was less than or equal to the 25th percentile. Univariate and multivariate analyses were conducted to find associations between patient characteristics, time to tracheostomy, ICU length of stay and mortality. Multivariate logistic regression analyses adjusted for all demographic information and comorbidities significantly associated with the outcome of interest on univariate analyses. The threshold for statistical significance was set at p < 0.05. All tests were performed using SPSS version 24 (IBM, Armonk, NY, USA).

A total of 1566 patients met inclusion criteria. Table 1 details information of patients between the early and late tracheostomy group. Overall, patients had an average age of 61.85 years. Patients were most frequently male (n = 924, 59.0 per cent), of white race (n = 1095, 69.9 per cent), and admitted emergently (n = 1362, 87.0 per cent); 11.9 per cent of the cohort died during their hospital stay. Patients had an average 12.22 days from ICU admission to tracheostomy. The 25th percentile cutoff for early versus late time to tracheostomy groups was 6.82 days. The average time to tracheostomy for the early group was 3.28 days compared to 15.20 for the late group (p < 0.001). A total of 391 patients (25 per cent) met criteria for ICU prolonged length of stay (75th percentile = 30.21 days), of which 94.1 per cent were patients that had a late tracheostomy. Patients in the late tracheostomy group had significantly longer ICU lengths of stay (27.32 days vs 12.55 days, p < 0.001).

Table 1. Demographic information of patients in early vs late tracheostomy groups

Late tracheostomy group patients had a higher average age (63.18 vs 57.88 years, p < 0.001). Significant comorbidity differences existed between the early and late tracheostomy groups. Patients receiving a delayed tracheostomy had higher rates of congestive heart failure (30.2 per cent vs 17.6 per cent, p < 0.001), cardiac arrhythmias (36.3 per cent vs 23.8 per cent, p < 0.001), valvular disease (36.3 per cent vs 23.8 per cent, p < 0.001), renal failure (12.7 per cent vs 7.9 per cent, p = 0.011), a coagulopathy (18.8 per cent vs 7.4 per cent, p < 0.001) or a fluid electrolyte disorder (38.3 per cent vs 28.9 per cent, p = 0.001). Table 2 reflects a multivariate regression analysis to identify factors associated with early tracheostomy within this patient population. Analyses demonstrated that patients with metastatic cancer (odds ratio: 2.590, p = 0.001) and rheumatoid arthritis (odds ratio: 2.789, p = 0.006) were more likely to have an early tracheostomy. In contrast, patients with congestive heart failure (odds ratio: 0.673, p = 0.020), cardiac arrhythmia (odds ratio: 0.730, p = 0.048), coagulopathy (odds ratio: 0.334, p < 0.002), or a fluid electrolyte disorder (odds ratio: 0.696, p = 0.009) were less likely to receive an early tracheostomy.

Table 2. Multivariate logistic regression on factors associated with early tracheostomy

Patients with a delayed tracheostomy had a higher rate of death (12.9 per cent vs 9.0 per cent, p = 0.039). Table 3 reports factors associated with mortality. This analysis demonstrated that increased time to tracheostomy was associated with an increased risk for death (odds ratio: 1.029, p = 0.009). Similarly, increased time to tracheostomy was associated with having an ICU prolonged length of stay (odds ratio: 1.246, p < 0.001) (Table 4). Examining the linear relationship between time to tracheostomy and ICU length of stay, we found that a 1-day increase in time to tracheostomy was associated with a 1.27-day increase in ICU length of stay (95 per cent confidence interval = 1.203–1.333, p < 0.001) (Table 5).

Table 3. Multivariate regression for factors associated with mortality

Table 4. Multivariate regression for factors associated with prolonged ICU length of stay

Table 5. Linear regression analysis for association between time to tracheostomy and ICU length of stay

Our study elucidates the effect of tracheostomy time on outcomes for patients admitted into a single-centre ICU. Understanding optimal times for tracheostomy has clinical significance.Reference Terragni, Antonelli, Fumagalli, Faggiano, Berardino and Pallavicini10,Reference Scales and Kahn12 Early tracheostomy has been hypothesized to offer significant benefits because early tracheostomy patients may benefit from shorter duration of mechanical ventilation and decreased exposure to sedating medications.Reference Scales and Kahn12 Unfortunately, prospective trials on tracheostomy are difficult to conduct, as explained by Scales and Kahn, due to difficulty in patient enrollment.Reference Scales and Kahn12 Consequently, several studies suffer from limited sample size which affects their ability to reach significant associations for certain outcomes.Reference Scales and Kahn12 The MIMIC-III database offers a robust sample to potentially detect these relationships, if they exist. Our study suggests an association of early tracheostomy with both shorter ICU length of stay and lower mortality.

The cutoff for early tracheostomy in our study was 6.82 days. However, there is heterogeneity in the classification of early versus late tracheostomy within the literature. For example, different studies have used demarcation points of 4 daysReference Young, Harrison, Cuthbertson and Rowan13–Reference Hosokawa, Nishimura, Egi and Vincent157 daysReference Okada, Watanuki, Masato, Sugiyama, Futamura and Matsuyama16–Reference Rodriguez, Steinberg, Luchetti, Gibbons, Taheri and Flint19, and 10 daysReference Villwock and Jones4,Reference Brook, Sherman, Malen and Kollef6,Reference Andriolo, Andriolo, Saconato, Atallah and Valente9,Reference Devarajan, Vydyanathan, Xu, Murthy, McCurry and Sessler20–Reference Scales, Thiruchelvam, Kiss and Redelmeier22 to identify early tracheostomy. Our estimation therefore aligns with previous studies. Furthermore, we decided to proceed with an approximately 7-day cutoff given the findings of Liu et al.'s systematic review on early versus late tracheostomy.Reference Liu, Livingstone, Dixon and Dort8 Specifically, Liu et al.'s data suggested that early tracheostomy (less than 7 days) was associated with a decrease in ICU length of stay.Reference Liu, Livingstone, Dixon and Dort8 In an earlier systematic review from 2005, Griffiths et al. also concluded that early tracheostomy may reduce duration of ICU length of stay.Reference Griffiths, Barber, Morgan and Young23 Similarly, in a meta-analysis specifically on trauma patients, Cai et al. found that early tracheostomy was associated with a significantly lower ICU length of stay.Reference Cai, Hu, Liu, Bai, Xie and Chen24

Our study supports these findings. Patients with early tracheostomy had significantly shorter ICU length of stay (27.3 days vs 12.6 days, p < 0.001). Our results also suggest that delaying tracheostomy by one day is associated with a 1.27-day increase in total ICU length of stay. Of course, there are several factors that influence the clinical decision to proceed with a tracheostomy such as severity of disease and anticipated ICU course. The confluence of these factors may result in delayed tracheostomy being considered the ideal treatment option for select patients. However, it is important to be aware of certain benefits of tracheostomy which may result in a shortened ICU stay. Specifically, tracheostomy facilitates better oral and airway care and results in reduced airway resistance.Reference Mahafza, Batarseh, Bsoul, Massad, Qudaisat and Al-Layla25,Reference Durbin26 With a lowered work of breathing, patients can benefit from shorter mechanical ventilation periods, thereby reducing their rate of complications such as airway injuries.Reference Mahafza, Batarseh, Bsoul, Massad, Qudaisat and Al-Layla25 Shortening ICU length of stay is an important consideration in the current climate with limited ICU bed availability. Prolonged ICU length of stay is a significant financial burden on the healthcare system.Reference Villwock and Jones4,Reference Halpern, Goldman, Tan and Pastores27 Previous studies have demonstrated that early tracheostomy can lead to significant cost savings.Reference Villwock and Jones4,Reference Herritt, Chaudhuri, Thavorn, Kubelik and Kyeremanteng28–Reference Koch, Hecker, Hecker, Brenck, Preuß and Schmelzer30 In a systematic review on this topic, Herritt et al. reported that early tracheostomy had an average cost saving of $4316, indicating that early tracheostomy may be a financially prudent decision if patient outcomes are not jeopardized.Reference Herritt, Chaudhuri, Thavorn, Kubelik and Kyeremanteng28

Identifying patients requiring extended ventilatory support is a significant challenge and an important factor in determining tracheostomy timing. Physicians have a limited ability to accurately gauge the required time for mechanical ventilation, and many clinical tools to aid decision making have low predictive value.Reference Young, Harrison, Cuthbertson and Rowan13,Reference Sanabria, Gómez, Vega, Domínguez and Osorio31 As such, several studies rely on the clinical acumen of physicians to accurately project mechanical ventilation time.Reference Liu, Livingstone, Dixon and Dort8 Our study sheds light on certain clinical risk factors that may be influencing physician decisions to perform a tracheostomy. In this study, early tracheostomy was associated with metastatic cancer and rheumatoid arthritis. Metastatic cancer has been shown to be associated with early tracheostomy, however, we present this relationship with rheumatoid arthritis.Reference Villwock and Jones4 In a study utilizing the Nationwide Inpatient Sample (https://hcup-us.ahrq.gov/nisoverview.jsp), Villwock et al. reported predictors of late tracheostomy which included fluid/electrolyte disorders.Reference Villwock and Jones4 Our study aligns with their findings and reports additional predictors. Specifically, patients with cardiac abnormalities, coagulopathy, and a fluid/electrolyte disorder were significantly more likely to have a delayed tracheostomy. This may be due to patients being unsuitable for surgical intervention early in their ICU course. As such, physicians may have delayed tracheostomy until patients were surgically cleared. Alternatively, these patients may have had more severe disease and been given a more dire prognosis on admission. As such, tracheostomy may not have been considered appropriate at admission, given the anticipated clinical course, and only reconsidered after the patient survived for more than 7 days. This phenomenon is an important consideration when interpreting this study's results.

Patients with early tracheostomy had a decreased incidence of mortality (12.9 per cent vs 9.0 per cent). Previous studies have reached conflicting conclusions on this relationship. Koch et al. noted that early tracheostomy did not decrease mortality in critically ill patients.Reference Koch, Hecker, Hecker, Brenck, Preuß and Schmelzer30 However, they used early and late tracheostomy time definitions as 4 days and 6 days, respectively, which may be too short of a duration to detect a mortality difference.Reference Koch, Hecker, Hecker, Brenck, Preuß and Schmelzer30 Ben-Avi et al. found that early tracheostomy, defined as less than 14 days, was associated with reduced mortality in cardiac surgery patients.Reference Ben-Avi, Ben-Nun, Levin, Simansky, Zeitlin and Sternik32 Also in cardiovascular surgery patients, Okada et al. reported decreased morbidity and mortality in early tracheostomy patients, which was defined as less than 7 days.Reference Okada, Watanuki, Masato, Sugiyama, Futamura and Matsuyama16 Tong et al., using a 7-day cutoff, found that early tracheostomy patients did not have reduced mortality.Reference Tong, Kleinberger, Paolino and Altman17 These studies reflect the lack of consensus on the relationship between early tracheostomy and mortality. This finding is likely partly influenced by differences in disease severity upon presentation. Similarly, heterogeneity amongst study populations and parameters across different analyses also likely contributed to the conflicting literature.

Our study did find an association between mortality and early tracheostomy after accounting for potential confounding comorbid conditions. While this analysis could not judge disease severity, which is especially difficult given its partially subjective nature, our findings do attempt to account for differences between patient groups. In this context, our study suggests that patients with early tracheostomy did experience a mortality benefit.

Ventilator-acquired pneumonia is a significant cause of in hospital mortality and is a manifest risk of prolonged mechanical ventilation.Reference Adly, Youssef, El-Begermy and Younis1,Reference Okada, Watanuki, Masato, Sugiyama, Futamura and Matsuyama16,Reference Nieszkowska, Combes, Luyt, Ksibi, Trouillet and Gibert33–Reference Rumbak, Newton, Truncale, Schwartz, Adams and Hazard36 As such, one of the purported advantages of early tracheostomy is decreased risk for pneumonia acquisition. Villwock et al. noted that early tracheostomy was associated with a 1.5 per cent decrease in ventilator-acquired pneumonia incidence.Reference Villwock and Jones4 This may result from tracheostomy reducing airway resistance and the resulting decrease in tracheobronchial bacterial colonization.Reference Adly, Youssef, El-Begermy and Younis1,Reference Villwock and Jones4,Reference Tong, Kleinberger, Paolino and Altman17,Reference Koch, Hecker, Hecker, Brenck, Preuß and Schmelzer30 Several studies have noted this relationship between early tracheostomy and decreased ventilator-acquired pneumonia incidence; however, there has been no clear link to mortality. In a systematic review of ventilator-acquired pneumonia and mortality by Melsen et al., ventilator-acquired pneumonia was found to be significantly associated with increased risk of death.Reference Melsen, Rovers and Bonten37 However, they did note high levels of heterogeneity among the outcomes of the various included observational studies.Reference Melsen, Rovers and Bonten37 When Melsen et al. sub-selected for studies solely concerning trauma or acute respiratory distress syndrome, there was no mortality attributable to ventilator-acquired pneumonia.Reference Melsen, Rovers and Bonten37

These findings indicate specific subgroup analysis is needed to clarify the nature of the relationship between ICU mortality and ventilator-acquired pneumonia. Our study includes patients with a variety of indications, likely reflecting a similar heterogeneity with our patient cohort. Given database limitations, our study was unable to monitor pneumonia incidence, however our results demonstrated that patients in the late tracheostomy group did have higher rates of death. Associations between ventilator-acquired pneumonia and mortality for a potentially significant portion of our patient cohort may be responsible for our findings.

Surgical interventions, such as tracheostomy, inherently present risk for patient morbidity. As such, physicians are cautious about subjecting patients to additional, potentially unnecessary, procedures. Early tracheostomy does carry complication risks that our study was unable to evaluate. For example, a potential complication for early tracheostomy is laryngotracheal stenosis. Studies have reported laryngotracheal stenosis incidence rates of 0.0–20.8 per cent.Reference Curry and Rowan38,Reference Goldenberg, Ari, Golz, Danino, Netzer and Joachims39 Rumbak et al. suggested a potential increase in tracheal stenosis amongst early tracheostomy patients, but their data were not significant.Reference Rumbak, Newton, Truncale, Schwartz, Adams and Hazard36 Similarly, in a systematic review, Curry et al. concluded that patients were at higher risk for laryngotracheal stenosis if undergoing conversion of endotracheal intubation to tracheostomy within 7 days.Reference Curry and Rowan38 However, Liu et al. did not find a significant association between laryngotracheal injury and early tracheostomy, but noted concerns about sample size.Reference Liu, Livingstone, Dixon and Dort8 As such, further study on the relationship between early tracheostomy and complications is needed.

• Optimal timing of tracheostomy is controversial, with conflicting conclusions on the benefits of early tracheostomy versus prolonged endotracheal intubation

• Patients with early tracheostomy had significantly shorter intensive care unit length of stay and lower mortality than patients with late tracheostomy

• Significant predictors for patients receiving early versus late tracheostomy were metastatic cancer and rheumatoid arthritis

• Our data suggest that early tracheostomy should be given strong consideration in appropriately selected patients

Our study has several limitations such as our inability to identify patient indication for ICU admission. Relationships between tracheostomy timing and many patient outcomes have been found to have no statistical significance when sub-selecting for specific patient groups.Reference Melsen, Rovers and Bonten37 However, understanding general themes may inform areas of future investigation because statistically non-significant relationships may have clinical significance. Furthermore, our study was unable to compare tracheostomy complication rates between the early and late tracheostomy groups. This is an important piece of information as it could help elucidate driving factors for the observed differences in outcomes for our cohort. We were also unable to assess disease severity at admission, which likely drove the clinical decision-making process on when tracheostomy could be considered or performed. Unfortunately, this limitation is common to several studies in the literature given the difficulty in accurately predicting ventilation needs amongst patients.Reference Young, Harrison, Cuthbertson and Rowan13,Reference Sanabria, Gómez, Vega, Domínguez and Osorio31 Our study, however, does attempt to account for patient characteristics at admission via the inclusion of patient comorbidities in our multivariate analysis.