Outcomes and predictors of prolonged ventilation in patients undergoing elective coronary surgery

Abstract

OBJECTIVES

Despite the seriousness of prolonged mechanical ventilation (PMV) as a postoperative complication, previously proposed risk prediction models were met with limited success. The purpose of this study was to identify perioperative variables associated with PMV in elective primary coronary bypass surgery. PMV was defined as the need for intubation and mechanical ventilation for >72 h, after completion of the operation.

METHODS

Between April 1997 and September 2010, 10 977 consecutive patients were retrospectively reviewed. A series of two multivariate logistic regression analyses were carried out to identify preoperative predictors of prolonged ventilation and the impact of operative variables.

RESULTS

PMV occurred in 215 (1.96%) patients; 119 (55.3%) of these underwent tracheostomy. At multivariate analysis, predictors included NYHA higher than class II (odds ratio [OR], 1.77; 95% confidence intervals [CI], 1.34–2.34), renal dialysis (OR, 5.5; 95% CI, 2.08–14.65), age at operation (OR, 1.04; 95% CI, 1.02–1.06), reduced FEV₁ (OR, 0.99; 95% CI, 0.98–0.99), body mass index >35 kg/m² (OR, 1.73; 95% CI, 1.14–2.63). On serial logistic regression analyses, operative variables added little to the discriminatory power of the model. Kaplan–Meier survival curves showed reduced survival among PMV patients (P < 0.001) with an improved survival in the tracheostomy subgroup.

CONCLUSIONS

PMV after coronary bypass is associated with a reduction in early and mid-term survival. Risk modelling for PMV remains problematic even when examining a more homogenous lower risk group.

INTRODUCTION

Despite advances in surgical techniques and anaesthetic management, prolonged mechanical ventilation (PMV) after coronary artery bypass grafting (CABG) continues to be a relatively common problem, with a reported incidence between 2.9 and 8.6% [1–5]. The significant reduction in early and mid-term survival associated with PMV has been demonstrated previously [2, 6, 7]. However, the heterogeneity of patient populations studied may have limited the success of previously suggested risk prediction models [5, 8–10].

As a trial to verify the validity of previously identified predictors in a more homogenous group of relatively lower risk patients, the aim of this work was to identify the incidence and predictors of PMV after elective primary isolated CABG and to assess the impact of this occurrence on those patients’ early and mid-term survival.

MATERIALS AND METHODS

Patient population and data

We retrospectively analysed a series of 11 016 consecutive patients who underwent elective primary isolated CABG at our institution between April 1997 and September 2010. Thirty-nine patients who did not survive until the defined cut-off point of PMV were excluded from the study. The remaining 10 977 patients composed the study population (control group: n = 10 762 98.4%; PMV group: n = 215 patients, 1.96%). The study was approved as an audit by the local ethics committee and patient consent was waived.

All data were prospectively collected in our cardiac surgery database during the patient's admission as a part of routine practice. The cardiac surgery database represents a mandatory data collection system regularly validated to national standards. Definitions and data collection methods, unless otherwise specified, have previously been published [11]. The different surgical techniques used in our patients, on-pump (n = 7226) and off-pump (n = 3751), have already been published [12]. In-hospital mortality was defined as death within the same hospital admission, regardless of cause. All patients transferred from the base hospital to another hospital were followed up to confirm their discharge. To establish follow-up mortality, patient records were linked to the demographics batch service (DBS), which records all deaths in the UK. Patients were matched to the DBS based on patient name, National Health Service number, date of birth, gender and postcode.

Prolonged mechanical ventilation

PMV was defined as the need for intubation and mechanical ventilation for >72 h, after completion of the operation. This includes both patients with early and persistent ventilatory dependency who were not extubated within the initial 72 h (n = 76, 35.4%) and those who had one or more unsuccessful extubation attempts (n = 139, 64.6%) eventually accumulating >72 h of endotracheal intubation and mechanical ventilation.

Decision to extubate

The criteria for extubation were the following: (a) that the patients were haemodynamically stable, (b) drainage <50 ml/h, (c) fully awake and able to move all limbs to command (in cases of stroke the decision to extubate was individualized depending on the extent of neurological deficit), (d) partial pressure of oxygen >12 kPa on 0.5% fraction of inspired oxygen, (e) tympanic temperature >36°C, (f) base deficit <3, and (g) respiratory rate >10 min⁻¹. Primary extubation was nurse led according to the protocol and any deviation from the norm was assessed by the anaesthetist and appropriate intervention made to address individual patient’s requirements. Patients who required PMV were assessed by the senior anaesthetist on a daily basis and a ventilation weaning protocol suggested. All senior anaesthetists at the hospital specialize in cardiothoracic critical care and have extensive airway management skills.

Statistical methods

Continuous variables are shown as median values with 25th and 75th percentiles. Categorical data are shown as percentages with patient numbers. Univariate comparisons were made with Wilcoxon rank-sum tests, χ² and Fisher's exact tests as appropriate. Deaths occurring over time were described using Kaplan–Meier survival curves.

A series of two multivariate logistic regression analyses were carried out on the data, each using the forward stepwise technique to identify risk factors associated with PMV. The first logistic regression analysis produced a model based on preoperative variables, the coefficient from this was used to create a risk score to which the operative variables were added, forming a second risk score.

Candidate variables with a P-value of <0.05 were entered into each model. The area under the receiver operating characteristic (ROC) curve and the Hosmer–Lemeshow goodness-of-fit statistic were calculated to assess the performance and calibration of each model, respectively. The relative contribution of each variable to the prediction of PMV was calculated. In all cases, a P-value of <0.05 was considered significant. All statistical analyses were performed with SAS for Windows Version 9.2 (SAS, Cary, NC, USA).

Results

Patients’ characteristics

The 10 977 patients included represent a typical group of elective primary CABG at a tertiary centre, and compare well with the UK referral patterns (http://www.scts.org/modules/resources/default.aspx?type=bluebook). The patients were mostly male (82%) with a median age of 65.2 (58.6–71.2) years and 24% were diabetics. The majority had triple vessel disease (78.4%) with 8.0% of the study population having an ejection fraction (EF) <30%. The median EuroSCORE was 2.5 (1.4–4.4).

Differences between the groups at preoperative, operative and postoperative levels are shown in Table 1. The median (Q1–Q3) ventilation time in the control group was 8 (5–12) h compared with 240 (119–456) h in the PMV group. Among the PMV group, 119 (55.3%) patients required a tracheostomy.

Table 1:

Patients’ demographics

Variable	Control (n = 10762)	PMV (n = 215)	P-value
Preoperative characteristics
Age at operation (years)	65.2 (58.5–71.1)	68.6 (63.2–73.7)	<0.001
Female	18.2 (1963)	22.8 (49)	0.09
Body mass index
<20	1.1 (114)	1.4 (3)	0.50
>35	7.7 (832)	13.0 (28)	0.004
% FEV1	84.8 (70.3–96.7)	78.7 (70.0–91.4)	<0.001
Ejection fraction
30–50%	30.3 (3255)	36.3 (78)	0.06
<30%	7.9 (850)	14.0 (30)	0.001
Triple vessel disease	78.2 (8413)	90.7 (195)	<0.001
Left main stem lesion	17.7 (1907)	14.4 (31)	0.21
Diabetes	20.7 (2228)	30.7 (66)	<0.001
Dialysis	0.4 (42)	2.3 (5)	<0.001
Respiratory disease	29.3 (3155)	43.7 (94)	<0.001
Peripheral vascular disease	13.3 (1434)	27.9 (60)	<0.001
Cerebrovascular disease	8.2 (877)	14.0 (30)	0.002
NYHA class > 2	27.7 (2984)	47.0 (101)	<0.001
Previous Q-wave MI	46.6 (5010)	53.5 (115)	0.04
Atrial fibrillation	3.1 (332)	7.0 (15)	0.001
Current smoker	58.1 (6248)	67.9 (146)	0.004
Hypertension	60.1 (6472)	73.5 (158)	<0.001
IABP	0.2 (19)	1.4 (3)	0.009
Logistic EuroSCORE	2.4 (1.4–4.3)	4.9 (2.5–7.9)	<0.001
Intraoperative characteristics
Date of operation (years since 1997)	7 (4–10)	7 (4–9)	0.12
Off-pump procedure	34.3 (3692)	27.4 (59)	0.04
IABP	0.9 (97)	8.4 (18)	<0.001
CPB time	105 (87–126)	120 (104–147)	<0.001
AXC time	63 (50–79)	75 (61–93)	<0.001
Postoperative complications
MI	1.6 (172)	19.5 (42)	<0.001
Stroke	0.9 (94)	21.9 (47)	<0.001
IABP	0.3 (32)	13.5 (29)	<0.001
Acute renal failure	2.8 (296)	39.1 (84)	<0.001
Re-operation for bleeding	2.5 (272)	8.4 (18)	<0.001
Surgical wound infection	2.6 (282)	17.2 (37)	<0.001
Inotrope delivery	28.0 (3016)	78.1 (168)	<0.001
Atrial fibrillation	4.5 (480)	38.6 (83)	<0.001
Tracheostomy	0.5 (48)	55.4 (119)	<0.001
Re-intubation	1.7 (185)	64.7 (139)	<0.001
In-hospital mortality	0.5 (57)	28.4 (61)	<0.001

AF: atrial fibrillation; AXC: aortic cross-clamping; CPB: cardiopulmonary bypass; FEV1: forced expiratory volume 1st second; IABP: intra-aortic balloon pump; MI: myocardial infarction; NYHA: New York Heart Association.

Predictors of prolonged mechanical ventilation

Univariate analysis showed statistically significant differences for all preoperative variables except female sex, low BMI, moderate impairment of systolic function and left main coronary disease. Apart from the date of the operation, all the included operative variables were significantly associated with PMV (Table 1).

Table 2 shows predictors of PMV after serial logistic regression analyses as the above description. The Hosmer–Lemeshow goodness-of-fit test was not statistically significant throughout the two stages of the analysis (P = 0.23 and 0.10, respectively), indicating good calibration of the model. The area under the ROC curve for the multivariate prediction model was 0.72 in the first logistic regression analysis based on preoperative risk factors only, and it remained relatively constant (0.73) after the introduction of the operative variables of interest, indicating an acceptable predictive power.

Table 2:

Multivariate predictors of prolonged ventilation

Variable	Odds ratio	95% CI	P-value
Preoperative predictors
NYHA class >2	1.77	1.34–2.34	<0.001
Peripheral vascular disease	1.96	1.44–2.67	<0.001
Age at operation (years)	1.04	1.02–1.06	<0.001
Dialysis	5.53	2.08–14.67	<0.001
Triple vessel disease	2.4	1.5–3.82	<0.001
FEV1 %	0.99	0.98–0.99	<0.001
BMI > 35	1.73	1.14–2.63	0.003
Intra-aortic balloon pump	5.07	1.41–18.14	0.005
Hypertension	1.48	1.08–2.02	0.01
C =  0.72; HL test = 0.23
Intraoperative predictors
Preoperative score	1.24	1.20–1.29	<0.001
CPB time (per 30 min intervals)	1.26	1.16–1.36	<0.001
Date of operation (years since 1997)	0.97	0.94–1.00	0.004
C = 0.73; HL test = 0.10
Logistic EuroSCORE	1.07	1.05–1.09	<0.001
C = 0.69; HL test < 0.001

CPB: cardiopulmonary bypass.

Table 3 compares the two subgroups of PMV patients depending on whether or not a tracheostomy was resorted to. Patients requiring a tracheostomy included more current smokers, more patients with respiratory disease, with a lower preoperative FEV₁ and higher NYHA class. There was no difference in post-operative complications between both subgroups except for a higher incidence of re-operation for bleeding and a significantly better survival in the tracheostomy group (19.3 versus 39.6%).

Table 3:

Demographics of PMV with and without tracheostomy

Variable	PMV, no tracheostomy (n = 96)	PMV + tracheostomy (n = 119)	P-value
Preoperative characteristics
Age at operation (years)	68.3 (62.9–73.5)	68.7 (63.2–74.5)	0.42
Female	28.1 (27)	18.5 (22)	0.09
Body mass index
<20	3.1 (3)	0 (0)	0.09
>35	12.5 (12)	13.5 (16)	0.84
% FEV1	79.8 (70.0–93.3)	77.5 (70.0–86.7)	0.02
Ejection fraction (%)
30–50	33.3 (32)	38.7 (46)	0.42
<30	15.6 (15)	12.6 (15)	0.53
Triple vessel disease	90.6 (87)	90.8 (108)	0.97
Left main stem lesion	11.5 (11)	16.8 (20)	0.27
Diabetes	30.2 (29)	31.1 (37)	0.89
Dialysis	2.1 (2)	2.5 (3)	>0.99
Respiratory disease	33.3 (32)	52.1 (62)	0.006
Peripheral vascular disease	27.1 (26)	28.6 (34)	0.81
Cerebrovascular disease	13.5 (13)	14.3 (17)	0.88
NYHA class >2	38.5 (37)	53.8 (64)	0.03
Previous Q-wave MI	44.8 (43)	60.5 (72)	0.02
Atrial fibrillation	5.2 (5)	8.4 (10)	0.36
Current smoker	60.4 (58)	74.0 (88)	0.03
Hypertension	70.8 (68)	75.6 (90)	0.43
IABP	0 (0)	2.5 (3)	0.26
Intraoperative characteristics
Date of operation (years since 1997)	6 (3.5–9)	7 (5–9)	0.01
Off-pump procedure	24.0 (23)	30.3 (36)	0.30
IABP	7.3 (7)	8.4 (10)	0.76
CPB time	119 (102–147)	120 (108–147.5)	0.54
AXC time	71 (58–92)	76 (62–95)	0.28
Postoperative complications
MI	24.0 (23)	16.0 (19)	0.14
Stroke	18.8 (18)	24.4 (29)	0.32
IABP	12.5 (12)	14.3 (17)	0.70
Acute renal failure	39.6 (38)	38.7 (46)	0.89
Re-operation for bleeding	4.2 (4)	11.8 (14)	0.045
Surgical wound infection	11.5 (11)	21.9 (26)	0.04
Inotrope delivery	76.0 (73)	79.8 (95)	0.50
Atrial fibrillation	37.5 (36)	39.5 (47)	0.77
In-hospital mortality	39.6 (38)	19.3 (23)	0.001

IABP: intra-aortic balloon pump; CPB: cardiopulmonary bypass.

Early and mid-term survival

Overall in-hospital mortality was 1.07% (n = 118). Mortality was significantly higher in the PMV group (28.4%, n = 61) when compared with the control group (0.5%, n = 57). There was no significant difference in early mortality in both subgroups of PMV patients.

The marked reduction in survival among patients requiring PMV persisted at mid-term. At 60 months, the mortality rate in the PMV group was 48.4% (n = 104) compared with 8% (n = 862) in the control group (Fig. 1). Survival was significantly better in the tracheostomy subgroup (Fig. 2).

Figure 1:

Mid-term survival by PMV group.

Figure 2:

Mid-term survival by tracheostomy subgroup.

Discussion

Defining the predictors of postoperative PMV is a difficult task. Previous studies included heterogeneous groups of patients undergoing different cardiac or cardiovascular surgical procedures [2, 3, 5, 7, 8, 10]. Given this heterogeneity of patient groups, and the variability in methodology and definition of PMV, caution should be exercised while interpreting the results of various studies. In the current study, we excluded combined, non-elective and redo cases mainly to add to the homogeneity of the study population. Excluding such higher risk patients, may explain the relatively lower incidence of PMV (1.96%) in our experience, when compared with other studies [1–5]. Because postoperative extubation is rather a clinical decision that retains a level of variability despite trials to adhere to rigid protocols, like many other authors [2, 7, 13] we opted for a definition of PMV using a higher cut-off point (72 h) to provide ample time for the resolution of minor clinical problems. A lower cut-off point would have included a much larger and more heterogeneous group of patients.

In the initial logistic regression analysis including only pre-operative factors, advanced age and increased body mass index (BMI > 35 kg/m²) were the only demographic variables that were found to be independent predictors of PMV. Advanced age is a marker of reduced physiological reserve and multiple co-morbidities, and has been almost consistently demonstrated to be a predictor of PMV [1, 3–5, 7–10, 13, 14]. Although it would be reasonable to assume that marked obesity would increase the complexity of the operation and the difficulty with post-operative pulmonary rehabilitation, the impact of obesity on the length of postoperative ventilation remains unclear. In parallel with some studies [7, 15, 16], and in contrast with others [14], we found BMI > 35 kg/m² to be associated with a higher incidence of PMV. The conflicting evidence about obesity as a risk factor may simply reflect differences in the cut-off point used to define it or the variable measurement of comorbidities in different studies [17]. Among other demographic variables examined, female sex was not found to be an independent predictor of PMV in agreement with some authors [3, 9] and in contrast to others [5, 13, 14].

Among preoperative variables, renal failure requiring dialysis was found to have the highest odds ratio (5.5; 95% CI, 2.1–14.7) for PMV. Apart from predisposing to volume overload and pulmonary oedema, renal failure is also associated with a delayed referral, more advanced coronary disease and a higher atherosclerotic burden, which predisposes to other complications such as stroke that may also result in PMV. This association was previously documented in many studies concerned with cardiac operations [3, 5, 8, 13, 14]. Similarly, three other indicators of increased atherosclerotic burden entailing the potential for more operative time and risk were found to be independent predictors of PMV. These were peripheral vascular disease, hypertension and triple vessel disease. This echoes the findings of previous studies, which established peripheral vascular disease [3, 8, 14], and to a lesser extent hypertension [14], as independent risk factors for PMV.

Although most previous studies found low ejection fraction (EF) to be a risk factor for PMV [3–5, 8, 13], this association was not found in the present study. Instead, an advanced NYHA class, another potential indicator of poor ventricular function, was found to be an independent predictor of PMV. Another related variable found to predict PMV was the need for an intra-aortic balloon pump (IABP). Although only used in a total of 22 patients, IABP was found to be a significant predictor of PMV (OR, 5.0; 95% CI, 1.4–18.1). Based on previously published work from our institution [18], earlier IABP insertion in high-risk CABG patients was standard practice in a large part of the study timeframe. Although our database does not detail the exact indication for IABP insertion, it would be reasonable to assume that, in the setting of elective CABG, this was largely related to severe ventricular impairment rather than refractory angina.

Interestingly, the literature remains divided regarding the impact of chronic obstructive pulmonary disease (COPD) on the duration of postoperative ventilation. Although some authors established COPD as an independent predictor of PMV [5, 13], others failed to find such an association. Spivack et al.  [4] reported the lack of predictive ability that a clinical diagnosis of COPD has for time receiving ventilation after elective CABG. Similarly, Filsoufi et al. [3, 14] found no association between COPD and PMV in patients undergoing valvular surgery [19]. Other authors found only a modest increase in the risk of prolonged ventilation in COPD patients. These controversial findings may be related to the wide spectrum of pulmonary disease severity embraced by COPD. In this regard, preoperative spirometric data can be a helpful tool to quantify disease severity in the studied patients. Fuster et al. [20] found a higher incidence of PMV only in patients with moderate and severe COPD as stratified according to spirometric criteria. Recently, Ad et al. [21] showed that the severity of chronic lung disease was significantly underreported when spirometric data were neglected, with a resultant adverse change in the predicted risk model for mortality and morbidity including PMV. Previous work from our institution established an FEV₁ <70% predicted, rather than a clinical diagnosis of COPD, as an independent predictor of PMV in patients undergoing cardiac operations [8]. Likewise, in our study, we found a significant association between decreasing preoperative FEV₁ values and increasing risk of PMV. Another important finding was shown when comparing the two subgroups of PMV patients, where the only significant difference between patients with PMV without the need for tracheostomy (n = 96, 44.7%) and those who required a tracheostomy (n = 119, 55.3%) was a significantly higher prevalence of chronic respiratory disease, current smoking status, higher NYHA class, along with a significantly lower preoperative FEV₁ values in the tracheostomy subgroup (Table 3). Survival in the tracheostomy subgroup was significantly better. On examining the characteristics of both subgroups, this survival advantage seems to be related to a higher proportion of salvageable patients who required PMV solely due to perioperative limited respiratory reserve in the tracheostomy group, rather than being related to tracheostomy itself. A recent randomized trial confirmed that early tracheostomy does not impact survival post-cardiac surgery [22]. Further studies are needed to decide whether any preoperative measures may decrease the risk of PMV in these higher risk respiratory patients. Despite the evidence supporting the value of preoperative optimization of respiratory function in these patients [23], such practices continue to be adopted only sporadically.

In the second regression model, and in parallel with other studies [3, 5, 19], prolonged cardiopulmonary bypass (CPB) time was found to be a significant predictor of PMV. Similar to other authors [7], we also noted a decline in the likelihood of PMV over the study period, which may be related to improvements in operative and postoperative care.

Although the preoperative model performed reasonably (C-statistic: 0.72), with minimal improvement after adding operative variables (C-statistic: 0.73), the discriminatory power of such models remains inadequate for the use in the clinical settings. This is partly due to the previously demonstrated complex interactions of preoperative, operative and postoperative events that make risk prediction models for PMV insensitive [5, 7, 10]. Although in some instances PMV may be related to specific pre-operative variables, in most cases it is more of a systemic problem that is precipitated, or even precipitates different post-operative complications. The latter already has a variable contribution from antecedent preoperative or operative characteristics [3, 7, 10].

Not surprisingly, in-hospital mortality was remarkably higher in the PMV group when compared with the control group (28.4 vs. 0.5%). This falls within the previously reported 18.5–33% early mortality risk in patients requiring PMV [2, 5–7]. However, once again such comparisons should be done cautiously given the variation in PMV definitions and the differences in the patient populations studied. The significant difference in survival persisted at mid-term (Fig. 1), denoting the poor risk profile of these patients, with various comorbidities affecting their outcome even after surviving the initial postoperative period.

There are some limitations to the current study. It is a single-centre observational study, conducted in a specialized referral centre performing a relatively large number of operations per year, and therefore the results may not be completely generalizable to other centres with a lower volume of CABG cases. The retrospective nature along with the relatively long time period this study covers may carry along variable changes in clinical practices and processes that might have affected the outcomes. Finally, like most of the previous studies, the exact clinical circumstances surrounding the occurrence of PMV could not be detailed.

In conclusion, in patients undergoing CABG, advanced age, renal dialysis, peripheral vascular disease, hypertension, advanced NYHA stage, elevated BMI, reduced FEV₁ and prolonged CPB times are associated with PMV. Although our risk model showed reasonable predictive power, the complexity of PMV as a clinical problem makes such models inadequate for the clinical use even in the lower risk elective patients. Given the associated significant reduction in early and mid-term survival, further studies are needed to assess the impact of preoperative optimization of the identified medical conditions, to modify the risk of PMV and to elucidate the potential role of tracheostomy in improving the outcomes. These factors should also be borne in mind, while counselling patients judged at risk of PMV, prior to CABG surgery.

Conflict of interest: none declared.