| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Ann Thorac Surg 2007;84:528-536
© 2007 The Society of Thoracic Surgeons
a Department of Cardiothoracic Surgery, The Cardiothoracic Centre-Liverpool, Liverpool, United Kingdom
b Department of Clinical Governance, The Cardiothoracic Centre-Liverpool, Liverpool, United Kingdom
Accepted for publication April 2, 2007.
* Address correspondence to Mr Grayson, Clinical Governance Department, The Cardiothoracic Centre-Liverpool, Thomas Drive, Liverpool, L14 3PE, United Kingdom (Email: tony.grayson{at}ctc.nhs.uk).
Presented at the Poster Session of the Forty-third Annual Meeting of The Society of Thoracic Surgeons, San Diego, CA, Jan 29–31, 2007.
 |
Abstract |
|---|
 Top Abstract IntroductionMaterial and MethodsResultsCommentAcknowledgmentsReferences |
|---|
Methods: This is a retrospective analysis of prospectively collected data on 12,662 consecutive patients undergoing adult cardiac surgery between April 1997 and March 2005. Data were randomly split into a development dataset (n = 6,000) and a validation dataset (n = 6,662). A multivariate logistic regression analysis was undertaken using a forward stepwise technique to identify independent risk factors for prolonged ventilation (defined as ventilation greater than 48 hours). 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 the model, respectively. Patients were split into low-, medium-, and high-risk groups based on their predicted probability of prolonged ventilation.
Results: Three hundred thirty-three patients had prolonged ventilation (5.5%). Independent variables, identified with prolonged ventilation, are shown with relevant coefficient values and p values as follows: (1) age 65 to 75 years, 0.7831, p < 0.001; (2) age 75 to 80 years, 1.5605, p < 0.001; (3) age greater than 80 years, 1.7115, p < 0.001; (4) forced expiratory volume less than 70% predicted, 0.3707, p = 0.013; (5) current smoker, 0.5315, p = 0.001; (6) serum creatinine 125 to 175 µmol/L, 0.6371, p < 0.001; (7) serum creatinine greater than175 µmol/L, 1.3817, p < 0.001; (8) peripheral vascular disease, 0.6212, p < 0.001; (9) ejection fraction less than 0.30, 0.7839, p < 0.001; (10) myocardial infraction less than 90 days, 0.7415, p < 0.001; (11) preoperative ventilation, 1.3540, p = 0.004; (12) prior cardiac surgery, 0.8946, p < 0.001; (13) urgent surgery, 0.4414, p = 0.004; (14) emergency surgery, 0.7421, p = 0.005; (15) mitral valve surgery, 0.7715, p < 0.001; (16) aortic surgery, 1.7043, p < 0.001; and (17) use of cardiopulmonary bypass, 0.4052, p = 0.025; intercept, –4.7666. The ROC curve for the predicted probability of prolonged ventilation was 0.79, indicating a good discrimination power. The prediction equation was well-calibrated, predicting well at all levels of risk. A simplified additive scoring system was also developed. In the validation dataset, 5.1% of patients had prolonged ventilation compared with 5.4% expected. The ROC curve for the validation dataset was 0.75.
Conclusions: We developed a contemporaneous multivariate prediction model for prolonged ventilation after cardiac surgery. This tool can be used in day-to-day practice to calculate patient-specific risk by the logistic equation or a simple scoring system with an equivalent predicted risk.
|
Introduction |
|---|
|
TopAbstractIntroductionMaterial and MethodsResultsCommentAcknowledgmentsReferences |
|---|
Patients who require prolonged ventilation will have increased intensive care unit lengths of stay. This in turn decreases hospital bed availability and is associated with increased resource utilization and health care costs. A randomized controlled trial by Cheng and colleagues [7] also suggested that prolonged ventilation can result in worse physiologic outcomes for patients postextubation as a result of atelectasis and intrapulmonary shunting.
As a consequence, several studies have examined risk factors for prolonged ventilation or assessed the ability of models such as the Intensive Care Unit Risk Stratification Score (ICURSS) to predict length of intubation [4–6, 8, 9]. These studies have added valuable knowledge to the cardiac surgical community and helped aid clinicians during a patient’s postoperative stay. However, very few studies have assessed preoperative risk factors only, thus making it difficult to have a preoperative assessment of the risk of prolonged ventilation for patient consent purposes and help planning of resource allocation. Both studies by Spivack and colleagues [1] and Legare and colleagues [2] have focused on preoperative risk factors and identified groups of patients at greater risk of prolonged ventilation. However, there is an absence of a multivariate prediction model that could provide a clinician with a percentage risk for each patient, similar to that estimated by the logistic version of the European System for Cardiac Operative Risk Evaluation (EuroSCORE) for predicting mortality [10]. We therefore aimed to develop and validate a multivariate prediction model for prolonged ventilation after cardiac surgery in a specialist cardiothoracic center to establish a contemporaneous tool for risk stratification and adjustment.
|
Material and Methods |
|---|
|
TopAbstractIntroductionMaterial and MethodsResultsCommentAcknowledgmentsReferences |
|---|
Patient characteristics collected are listed in Table 1 and were defined in accordance with the Society of Cardiothoracic Surgeons of Great Britain and Ireland minimum dataset (Fifth National Adult Cardiac Database Report – The Society of Cardiothoracic Surgeons of Great Britain and Ireland [www.ctsnet.org/file/5thBlueBook2003.pdf)]. Current smokers were defined as patients who had smoked a cigarette less than two months prior to surgery. Peripheral vascular disease (PVD) was defined as history or evidence of aneurysm or occlusive PVD, including femoral or carotid bruit, aortic aneurysm, and reduced or absent peripheral pulses. Urgent surgery was defined as patients "who have not been scheduled for routine admission and require urgent surgery during the current admission for medical reasons." Such patients are unable to be sent home prior to receiving surgery. Emergency surgery was defined as patients who require treatment within 24 hours of the admission or even cardiopulmonary resuscitation on route to theatre. The outcome for the study was prolonged ventilation, defined as ventilation after cardiac surgery that had a duration greater than 48 hours.
|
Statistical Analysis
Continuous data are shown as median values with 25th and 75th percentiles. Categoric variables are shown as a percentage and comparisons were made with 
2 tests as appropriate. Standard statistical tests were used to calculate odds ratios and 95% confidence intervals. Data were randomly split using a simple random sample method without replacement (’proc surveyselect’ function within SAS) into a development dataset (n = 6,000) and a validation dataset (n = 6,662). A multivariate logistic regression analysis was undertaken on the development dataset, using the forward stepwise technique, to identify independent risk factors for prolonged mechanical ventilation [12]. Candidate variables with a p value less than 0.1 were entered into the model. The area under the receiver operating characteristic (ROC) curve [13] and the Hosmer-Lemeshow goodness-of-fit statistic [12] were calculated to assess the performance and calibration of the model, respectively. The relative contribution of each variable to the prediction of prolonged ventilation was calculated. Patients were split into low-, medium-, and high-risk groups based on their predicted probability of prolonged ventilation. After ordering patients into order of risk, low-risk was considered the bottom 50% of the cohort and high-risk was considered the top 10%.
The statistical model was then tested on the validation dataset. Observed and expected rates of prolonged ventilation were compared. The ROC curve and the Hosmer-Lemeshow goodness-of-fit statistic were calculated to assess the performance and calibration of the model. The logistic EuroSCORE [10] was also calculated for each patient to assess its ability to predict the risk of prolonged ventilation compared with our own local model.
A simplified clinical risk assessment tool was developed from the multivariate risk prediction model and was scored by rounding the adjusted odds ratio for each variable to the nearest 0.5. These weights were then summed. The relationship between this clinical risk score and the probability calculated from the risk prediction model was read from a graph. This clinical risk assessment tool therefore approximates the risk that would have been calculated from the risk prediction model. In all cases a p value less than 0.05 was considered significant. All statistical analyses were performed with SAS for Windows Version 8.2 (SAS, Cary, NC).
|
Results |
|---|
|
TopAbstractIntroductionMaterial and MethodsResultsCommentAcknowledgmentsReferences |
|---|
|
|
Independent Risk Factors for Prolonged Ventilation in Development Dataset
The independent risk factors for prolonged ventilation, along with coefficients, odds ratios, confidence limits, and p values, are shown in Table 3. The area under the ROC curve for the multivariate prediction model was 0.79. The predicted risks of individual patients were rank-ordered and divided into ten groups. Within each group of estimated risk, the number of prolonged ventilation predicted was compared with the number of observed prolonged ventilation. The Hosmer-Lemeshow goodness-of-fit statistic across groups of risk was not statistically significant (Fig 2;
p = 0.29), indicating little departure from a perfect fit.
|
|
Low-, medium-, and high-risk groups were created based on the patients’ expected or predicted risk of prolonged ventilation. Low-risk patients had a 3% or less risk, medium-risk had between 3% and 10% risk, while high-risk patients had a greater than 10% risk of prolonged ventilation. The observed incidence of prolonged ventilation depending on whether a patient was considered low-, medium-, or high-risk is shown in Figure 3. In the high-risk group 157 patients had prolonged ventilation, which accounts for almost half of the 330 observed prolonged ventilations (47.6%).
|
|
Simplified Model
A simplified clinical risk assessment tool derived from the logistic regression equation, described at the bottom of Table 3, is shown in Figure 5. Using the logistic equation a 77-year-old current smoker with peripheral vascular disease, and undergoing coronary artery bypass surgery with cardiopulmonary bypass, would have a 16.1% risk of prolonged ventilation. However, using the simplified model in Figure 5, the same patient would have a prolonged ventilation score of 10, which approximates to around an 18% risk of prolonged ventilation. Any patients with a prolonged ventilation score of 18 or greater would have an approximate risk of prolonged ventilation of greater than 80%.
|
|
Comment |
|---|
|
TopAbstractIntroductionMaterial and MethodsResultsCommentAcknowledgmentsReferences |
|---|
The incidence of prolonged ventilation in our series is similar, if not lower, to other reports at 5.5% in the development dataset and 5.1% in the validation dataset (5.3% overall). Kern and colleagues [4] used the same definition of prolonged ventilation, as greater than 48 hours, in 687 cardiac surgery patients; however, prolonged ventilation occurred in 9%. Although, it should be noted that in such a small sample it is difficult and unfair to compare the rate of prolonged ventilation with our series. Other studies have used greater than 48 hours as prolonged ventilation after coronary artery bypass graft surgery with rates varying between 6% and 8% [1, 3, 5].
Several studies have examined risk factors for prolonged ventilation or assessed the usefulness of existing intensive care scores, albeit in small sample sizes. Estenssoro and colleagues [9] recently examined predictors of prolonged ventilation in medical and surgical patients admitted to intensive care and found shock on intensive care to be the only risk factor after multivariate analysis. Factors such as the Simplified Physiology Score (SAPS) II, the Acute Physiology and Chronic Health Evaluation (APACHE) II, and the Therapeutic Intervention Scoring System (TISS) were not identified as risk factors after multivariate analysis, although they were different between prolonged and nonprolonged ventilation patients on univariate analysis [9]. Kern and colleagues [4] concluded that data collected postoperatively using established scoring systems (TISS and SAPS), as well as documented events of high clinic impact for risk assessment and quality control, were reliable predictors of prolonged ventilation. However, the ICURSS model was found not to be useful for the prediction of length of ventilation by Serrano and colleagues [5], who examined 569 patients undergoing coronary artery bypass grafting with a prolonged ventilation rate at 48 hours of 7.6%.
Other studies have focused on preoperative factors in an attempt to identify patients at risk of prolonged ventilation prior to undergoing surgery. Spivack and colleagues [1] found that the combination of a reduced left ventricular ejection fraction and the presence of selected preexisting comorbidities (congestive heart failure, angina, current smoking, and diabetes) served as modest risk factors for prolonged ventilation. Legare and colleagues from Canada [2] identified unstable angina, ejection fraction less than 0.50, respiratory disease, renal failure, female gender, and age greater than 70 years to be predictors of prolonged ventilation. They also identified intraoperative factors such as stroke, reoperation for bleeding, and perioperative myocardial infarction to be strong predictors. Bezanson and colleagues [14] found older age, urgent or emergent preoperative clinical status, and prior coronary artery bypass surgery to be predictors of prolonged ventilation. An analysis of the Society of Thoracic Surgeons database by Shroyer and colleagues [3] found a vast number of preoperative risk factors for prolonged ventilation ranging from patient demographics to comorbidities, heart disease severity, and acuity. They developed a risk prediction model with this data; however, they published only the odds ratio value for each factor and did not include the logistic regression equation. Also, their analysis was only in coronary artery bypass surgery patients, therefore making it difficult to assess the potential resources needed in intensive care for patients undergoing other cardiac surgery.
These studies have found preoperative risk factors similar to our own study including increasing age, current smokers, respiratory disease, renal dysfunction, peripheral vascular disease, poor ejection fraction, myocardial infarction, prior surgery, and urgent or emergent surgery. Our study took advantage of having data available such as forced expiratory volume in the first second percentage predicted and serum creatinine levels. We were able to identify preoperative ventilation support as a predictor of prolonged ventilation after cardiac surgery and also found that patients undergoing mitral valve surgery or surgery on the aorta and use of cardiopulmonary bypass to be at increased risk. We included the type of procedure and use of cardiopulmonary bypass, although operative factors, as we felt that these types of data would already be known in the vast majority of cases prior to surgery.
We have previously identified cardiopulmonary bypass as an important factor in determining the length of ventilation [15]. Other studies have suggested that patients avoiding the use of cardiopulmonary bypass when undergoing coronary artery surgery benefit from shorter ventilation times [16, 17].
Looking at the risk factors identified in the prolonged ventilation risk model many of them are also included in the EuroSCORE, which although a prediction model for mortality has also been shown to predict major morbidity and length of stay [18]. It could therefore lead one to wonder whether the EuroSCORE would perform just as well at predicting prolonged ventilation. Our results show, however, that although the ROC curve is reasonable at 0.71, the EuroSCORE overpredicts prolonged ventilation in low-risk cases and underpredicts in the high-risk cases. The reason for this inaccuracy is easily explained by the fact the EuroSCORE was designed to predict mortality and not prolonged ventilation and, although the two outcomes will have a degree of association, not all patients who have prolonged ventilations will die. The risk model contained within this article is designed specifically for the prediction of prolonged ventilation and therefore predicts more accurately the risk of this outcome for patients undergoing cardiac surgery.
Even though the use of a simplified model converting additive points into approximate risk (as shown in Fig 5) is very convenient and user friendly, the use of the logistic equation is more reliable and accurate and is the recommended method of using our risk prediction tool. With the use of hand-held computers and modern day technology the logistic equation can be keyed into any electronic spreadsheet, therefore making additive models redundant. However, we have included a simplified model if one prefers to use it as a rough guide during clinical consultation, giving a fairly close estimate of the predicted risk to the patient during the consenting process.
We have also identified low-, medium-, and high-risk categories for prolonged ventilation, which may prove useful by enabling clinicians to spread out patients at high-risk throughout a given week. This in turn would help intensive care planning. Without the aid of a prediction model for prolonged ventilation several patients with a preoperative risk of greater than 10% (as predicted by the logistic regression equation) could end up being scheduled at the beginning of the week, leading to beds in intensive care being occupied for more than two days; also the likely possibility of having to cancel other patients surgical procedures during the week. However, if using the model patients identified as high-risk could be spread out accordingly among low-risk and medium-risk patients. Patients coming to surgery as urgent or even emergency cases with a high-risk of prolonged ventilation would of course have to be treated as soon as possible, but the impact of these patients on the intensive care unit would be minimized, compared with a situation where maybe three high-risk patients have already been treated electively on the same day.
There are limitations with this model that need to be considered. The model includes only the preoperative variables available on our dataset; there may be other variables that could potentially affect outcomes but which are not routinely collected. The model does not take into account intraoperative variables such as length of surgery or cardiopulmonary bypass time or perioperative myocardial infarction, which could affect the duration of postoperative ventilation. However, as the purpose of this model is to guide clinicians prior to surgery, such factors would be impossible to predict. Another limitation is the fact that, although the data used to develop the model have been subjected to local validation and have the confidence of clinicians, the performance of the model in predicting prolonged ventilation in adult cardiac surgery patients outside of our own institution is still to be tested. A process of external validation is necessary to check the validity of the model across other geographic areas. A further limitation is the fact that a definition of prolonged ventilation of greater than 48 hours could potentially cover a wide variation in resource use with patients ventilated for months. Thus, their care would be much more resource-intensive than those who come off the ventilator after two to three days and our model cannot distinguish between these types of patients. However, we would expect such extremes to be minimal with only 66 patients ventilated for more than one month within our 12,662 cases (0.5%).
In conclusion, we have developed a contemporaneous multivariate prediction model for prolonged ventilation after adult cardiac surgery. This tool can be used in day-to-day practice to calculate patient-specific risk by the logistic equation or a simple scoring system with an equivalent predicted risk. This tool may also be useful for planning of resource allocation, enabling clinicians to spread out patients at high risk of prolonged ventilation throughout a given week.
|
Acknowledgments |
|---|
|
TopAbstractIntroductionMaterial and MethodsResultsCommentAcknowledgmentsReferences |
|---|
|
References |
|---|
|
TopAbstractIntroductionMaterial and MethodsResultsCommentAcknowledgmentsReferences |
|---|
This article has been cited by other articles:
|
|
 |
|
  K. Lachapelle Invited Commentary Ann. Thorac. Surg., January 1, 2011; 91(1): 15 - 15. [Full Text] [PDF]
|
|
|
|
 |
|
 J.-L. Trouillet, A. Combes, E. Vaissier, C.-E. Luyt, A. Ouattara, A. Pavie, and J. Chastre Prolonged mechanical ventilation after cardiac surgery: Outcome and predictors J. Thorac. Cardiovasc. Surg., October 1, 2009; 138(4): 948 - 953. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
L. Dos, A. Dadashev, D. Tanous, I. J. Ferreira-Gonzalez, K. Haberer, S. C. Siu, G. S. Van Arsdell, E. N. Oechslin, W. G. Williams, and C. K. Silversides Pulmonary valve replacement in repaired tetralogy of Fallot: Determinants of early postoperative adverse outcomes J. Thorac. Cardiovasc. Surg., September 1, 2009; 138(3): 553 - 559. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
| ANN THORAC SURG | ASIAN CARDIOVASC THORAC ANN | EUR J CARDIOTHORAC SURG |
| J THORAC CARDIOVASC SURG | ICVTS | ALL CTSNet JOURNALS |
