Determinants of duration of ICU stay after coronary artery bypass graft surgery

Transcription

1 British Journal of Anaesthesia 1996; 77: Determinants of duration of ICU stay after coronary artery bypass graft surgery A. MICHALOPOULOS, G. TZELEPIS, G. PAVLIDES, J. KRIARAS, U. DAFNI AND S. GEROULANOS Summary Prediction of duration of a patient s stay in the ICU after cardiac surgery is difficult. In 652 consecutive adult patients undergoing elective coronary artery bypass graft (CABG) surgery, we analysed prospectively preoperative and immediate postoperative variables thought to influence duration of stay in the ICU. With univariate analysis, we found that age, preoperative left ventricular ejection fraction, bypass time, aortic cross-clamp time, blood transfusions and the number of inotropic agents administered in the immediate postoperative period (for at least 6 h) were significant correlates of duration of stay in the ICU. However, logistic regression analysis showed that the number of inotropes was the most important determinant of stay in the ICU, with an overall prediction accuracy of 94.8 %. The main cause of prolonged stay in the ICU (more than 2 days) was low cardiac output syndrome. We conclude that analysis of perioperative variables enhanced our ability to accurately predict duration of stay in the ICU in cardiac surgery patients. The number of inotropic agents administered during the first 6 h after operation was the most important determinant of duration of stay in the ICU. (Br. J. Anaesth. 1996; 77: ) Key words Surgery, cardiovascular. Intensive care. Heart, coronary artery bypass. Risk. Despite recent efforts to bypass intensive care unit (ICU) stay [1], cardiac surgery patients are invariably monitored in the ICU for a period of time that varies from 1 to several days. Prolonged stay in the ICU not only increases the overall costs of cardiac surgery, but it may also limit the number of operations performed. Therefore, the ability to accurately predict the duration of stay in the ICU and patient outcome is important. The scoring systems used to predict a patient s stay in a general ICU have not been found to be useful in the postoperative cardiac ICU [2, 3]. Over the past few years, attempts have been made to develop predictive models of stay in the ICU with special applicability to cardiac surgery patients. However, the majority of reported models are complex, impractical for intercentre use and of modest predictive ability [4 6]. This may result partly from the fact that all studies analysed preoperative variables and ignored variables related to the operation itself. In this study, we have assessed determinants of length of stay in the ICU in postoperative coronary artery bypass graft (CABG) surgery patients by analysing both preoperative and immediate postoperative variables. This approach has enabled us to develop a simple and more accurate predictive index of stay in the ICU. Patients and methods From January 1994 to December 1994, we studied prospectively 652 consecutive patients undergoing elective CABG surgery in our hospital. We recorded patients variables, New York Heart Association (NYHA) class and preoperative left ventricular ejection fraction assessed by angiography (grade 1 50 %; grade %; grade %; and grade 4 20 %). We also recorded the type of cardioplegia used, number of grafts, aortic cross-clamp time and bypass time. In addition, we recorded the number of inotropic drugs administered during the first 6 h in the ICU. Collection of these data in no way interfered with or affected the care of patients being studied; therefore, it was deemed that patient consent was not required. Approval for this study was obtained from the Ethics and Research Committee for our institution. Inotropic drugs were used to treat low cardiac output syndrome, defined as a cardiac index 2 litre min 1 m 2, despite optimal filling pressures (wedge pressure 12 mm Hg). Inotropes were included only if they were given in the ICU for at least 6 h, immediately after surgery. In this situation, dobutamine, adrenaline or dopamine, up to maximum doses (adrenaline 2 g kg 1 min 1, dobutamine 20 g kg 1 min 1 or dopamine 20 g kg 1 min 1 ) were A. Michalopoulos*, md, G. Tzelepis, md, G. Pavlides, md, J. Kriaras, md, U. Dafni, md, S. Geroulanos, md, Surgical Intensive Care Unit, Onassis Cardiac Surgery Centre, Athens, Greece. Accepted for publication: March 13, * Address for correspondence: Surgical Intensive Care Unit, Onassis Cardiac Surgery Centre, 356 Sygrou Ave, Athens, Greece.

2 Length of ICU stay after CABG 209 Table 1 Patients data (mean (range) or number (%)) (n 652) n (%) Sex (M/F) 592/ /9.2 Age (yr) 62.1 (31 81) Smokers Left ventricular ejection fraction Grade 1 ( 50 %) Grade 2 (35 50 %) Grade 3 (20 34 %) Grade 4 ( 20 %) NYHA classification system Class I Class II Class III Class IV usually used initially. If there was need for additional inotropic support (persistence of condition) within this critical period, milrinone or noradrenaline (1 g kg 1 min 1 or 0.3 g kg 1 min 1, respectively) were then added [7]. Dopamine at renal doses was excluded. The amount of blood transfused in the operating room and during the first 6 h after operation was also recorded. Blood transfusion was given when the patient s packed cell volume (PCV) was 27 %. Duration of mechanical ventilation, all postoperative complications, duration of stay in the ICU and patient outcome were recorded for all patients. For this analysis, prolonged stay in the ICU was defined as greater than 2 days. No patient was excluded. Patient outcome was evaluated on the day of ICU discharge. Data were analysed using BMDP statistical software (Los Angeles, CA, USA) [8]. Continuous variables were analysed using the two-sample t test. The chi-square test was used for discrete variables. To assess the ability of independent variables to predict a dependent variable, the odds ratio was calculated [9]. Furthermore, the data were subjected to logistic regression analysis where ICU duration constituted the binary dependent variable (2 vs 2 days). Sex and cardioplegia (crystalloid vs blood) were categorical predictors, and ejection fraction, aortic cross-clamp time, bypass time, number of grafts, units of blood transfused and number of inotropic drugs administered were interval-scaled predictors. The fit of the model was assessed by the Hosmer Lemeshow goodness-of-fit statistic [10]. Data are reported as the mean (SD) and were considered significant when P Results Patient characteristics and preoperative data are shown in table 1. For all patients, mean duration of stay in the ICU was 2.3 (SD 1.6) days (range days) and duration of mechanical ventilation 0.8 (0.3) days. Of all patients, 585 patients (89.7 %) were discharged in less than 2 days (group A), whereas 67 patients (10.3 %) remained in the ICU for more than 2 days (group B). Patient characteristics and differences in perioperative variables for both groups are presented in table 2. The two groups did not differ in sex distribution, cardioplegia, preoperative haemoglobin (PCV) or number of grafts. Ejection fraction was a significant predictor of prolonged stay in the ICU (P 0.001). Use of inotropes was also associated with prolonged stay in the ICU. For example, use of one inotrope was associated with an odds ratio of 2, and use of two inotropes with an odds ratio of 104 (P ). Aortic cross-clamp time in excess of 90 min was associated with a statistically significant and sharp increase in odds ratio for prolonged stay in the ICU (P 0.001). Bypass time longer than 120 min increased the risk for prolonged stay in the ICU, corresponding to an odds ratio of 4.3 (P 0.001). There was a non-monotonic relation of risk for prolonged stay in the ICU and the amount of blood transfused; patients who received more than 6 u. of blood were at a substantially increased risk for prolonged stay in the ICU (odds ratio 13.5; P 0.001). Table 3 shows the results of a forward stepwise selection of variables into the logistic model. Five variables were entered based on their statistical significance (entering criterion P 0.15). A backward selection procedure led to the same set of five variables in the model (removing criterion P 0.20). The great majority of improvement in this model fit Table 2 Patient characteristics and perioperative variables in patients with short ( 2 days) (group A) or long ( 2 days) (group B) ICU stay (mean (SD or range) or number (%)) Group A Group B n (%) n (%) P No. of patients Sex (M/F) 531/ /9.2 61/6 91.0/ Age (yr) 61.7 (31 81) 66.1 (47 79) Ejection fraction (%) 0.46 (0.1) 0.34 (0.07) Aortic cross-clamp time (min) 54 (17) 70 (19) Bypass time (min) 94 (28) 131 (44) Type of cardioplegia Blood Crystalloid Bloods units transfused 4.2 (1.7) 7.5 (2.5) Inotropes Yes No Grafts

3 210 British Journal of Anaesthesia Table 3 Summary of stepwise selection results Step df Log likelihood Improvement chi-square (df 1) P Goodness of fit (Hosmer Lemeshow) Chi-square (df ) P Inotropes , (8) Blood units , (8) Ejection fraction , (8) Age , (8) Cardioplegia , (8) Table 4 Number of patients (%) in ICU groups as a function of number of inotropes used Inotropes days 418 (98.3) 147 (96.7) 19 (31.1) 1 (7.1) 2 days 7 (1.7) 5 (3.3) 42 (68.9) 13 (92.9) Total 425 (100) 152 (100) 61 (100) 14 (100) Table 5 Number of patients (%) in ICU groups as a function of inotrope groups used Inotropes 0 or 1 2 Total 2 days 564 (96.4) 21 (3.6) 585 (100) 2 days 13 (19.4) 54 (80.6) 67 (100) was associated with the number of inotropes used, the number of blood units and preoperative ejection fraction. Table 4 shows the number and percentage of patients in each ICU stay-group for the four levels of inotropes observed. Using inotropes as a binary predictor, with values of 0 and 1 indicating stay in the ICU of 2 days or less, and 2 or more indicating more than 2 days, the overall prediction accuracy was 95.1 % (620 of 652) based on 96.6 % for predicting short stays (565 of 585) and 82.1 % for long stays (55 of 67) (table 5). These rates are close to the five predictor model of table 3 and also to a five predictor model where the binary inotropes variable itself was substituted for the interval variable. Thus the binary inotropes variable is an accurate and simple predictor of stay in the ICU. The causes resulting in prolonged stay in the ICU were low cardiac output syndrome in 35 (52.2 %), respiratory failure in eight (11.9 %), neuro-psychiatric abnormalities in five (7.5 %) and combi- nation of the above in 19 (28.3 %) patients. For both groups, mean duration of mechanical ventilation, ICU and hospital stay, and mortality rate are presented in table 6. The overall ICU mortality rate was 0.8 %. Causes for a longer stay in the ICU for the group receiving less than two inotropes (n 13) were hypoxaemia (n 6), atrial fibrillation (n 3), their combination (n 3) and mental disturbances (n 1). Discussion This study included both preoperative and immediate postoperative variables in a logistic regression model and developed an index to predict duration of stay in the ICU after cardiac surgery. We found that patients who had a low ejection fraction, had received a large number of units of blood or more than one inotropic agent remained in the ICU for a longer period than usual. Sex, bypass time, aortic cross-clamp time and number of grafts were not significant predictors. Among all variables examined, the number of inotropic drugs administered in the immediate postoperative period was the most reliable predictor of stay in the ICU, with an overall predictive accuracy of 94.8 %. This is supported by our finding that the most frequent cause of prolonged stay in the ICU was low cardiac output syndrome; however, it should be noted that the number of inotropes rather than inotrope use was the significant factor. The major limitation of this study was that this index was derived after the patient s stay in the ICU for 6 h. In this sense, it cannot be used as a preoperative prognostic indicator of duration of stay in the ICU. In our study, there was a delay of a few hours for predicting the duration of stay in the ICU, but we believe that there is still considerable value in predicting the length of stay in the intensive care unit in the early postoperative period. The accuracy Table 6 Duration of mechanical ventilatory support, ICU and hospital stay, and mortality rate for both groups (mean (SD)) Group A Group B (ICU stay 2 days) (ICU stay 2 days) P Mech. ventilation (days) 0.5 (0.05) 3.8 (1.4) ICU stay (days) 1.3 (0.3) 7 (4.5) Hospital stay (days) 7.1 (0.8) 14.6 (3.9) ICU mortality 0.3 % 4.5 %

4 Length of ICU stay after CABG 211 of our prediction was much higher compared with similar, previous studies which used only preoperative variables. The determinants of this study can be considered similar to the APACHE II scoring system, which predicts patient outcome at the time of admission to the ICU, but has a poor performance in cardiac surgery patients [11]. Over the past 5 yr, several studies have developed predictive indices of stay in the ICU after cardiac surgery. The majority of these studies have included preoperative variables and, in general, have not taken into account events that affect patient outcome during the operative or immediate postoperative period. A great disparity in type and number of independent variables analysed has been reported. Tuman and co-workers [4] found 11 variables to be significantly associated with stay in the ICU, including emergency surgery, age, preoperative renal dysfunction, previous myocardial infarction, cerebrovascular disease, type of surgery, congestive heart failure and left ventricular dysfunction. Although the duration of stay in the ICU in this study correlated with a clinical risk score derived from these variables, no accurate predictive index of stay in the ICU based on clinical score was provided. Furthermore, it cannot be determined which risk factor is the most significant predictor of length of stay in the ICU. In a similar study of preoperative variables, Tu and co-workers [12] suggested that length of stay in the ICU after cardiac surgery could be predicted by a multivariate predictive index. Five variables were found to be statistically significant independent predictors for a prolonged stay in the ICU (age, female sex, left ventricular function, urgency of surgery and type of surgery). A predictive index was developed for length of stay in the ICU with a risk score of Patients with a low risk score remained in the ICU for a few days, while patients with a high risk score had a prolonged stay in the ICU. However, this model has a poor predictive ability; 31.7 % of patients with a low risk score (0 3) stayed more than 2 days stay in the ICU, whereas only 69.4 % of patients with a high risk score ( 8) had a prolonged ( 2 days) stay in the ICU. The authors acknowledged that length of stay in the ICU can never be predicted with certainty. One year later, Tu, Jaglal and Naylor [13] reported that duration of stay in the ICU after cardiac surgery could be predicted using a six-variable risk index which included age, sex, preoperative ejection fraction, type of surgery, urgency of surgery and reoperation. Although the authors considered prolonged stay in the ICU as longer than 6 days, the overall predictive ability of their index was poor (67 %). The fundamental difference between our study and those described above is that we took into account not only preoperative variables, but also significant variables related to both the operative and immediate postoperative period, such as number of inotropes administered and blood transfusions. We found that duration of stay in the ICU depended predominantly on a single critical variable, number of inotropes administered during the immediate postoperative period. This is not surprising because inotropic support reflects poor cardiac function. It follows that the group requiring intensive inotropic support would include patients with intraoperative complications (e.g. myocardial ischaemia or infarction), inadequate revascularization, low cardiac output syndrome related to systemic inflammatory response syndrome (SIRS), stunned myocardium or inadequate myocardial protection during bypass. Apparently, none of these conditions can be predicted accurately by any preoperative variable. On the other hand, in some high-risk cases (e.g. low ejection fraction) there may be great improvement in left ventricular function after operation as a result of successful revascularization. This is supported by the study of Zaroff and colleagues who found that, although low preoperative ejection fraction is a known predictor of poor immediate postoperative outcome after cardiac surgery, not all patients with low preoperative ejection fraction required inotropic support [14]. In both cases, the models based on preoperative variables are likely to fail in predicting stay in the ICU. Our findings are in agreement with those reported by Weintraub and co-workers [15], who showed that the correlates of prolonged hospital stay could not be predicted by preoperative variables alone. The authors emphasized that serious complications were responsible for prolonging the length of stay after coronary artery bypass grafting surgery. Acknowledgements We thank Dr Mark Cohen for his assistance with the statistical analysis. References 1. Chong JL, Pillai R, Fisher A, Grebenik C, Sinclair M, Westaby S. Cardiac surgery: moving away from intensive care. British Heart Journal 1992; 68: Hadorn D. Who really needs to be in the intensive care unit: Using clinical guidelines to define healthcare needs. Critical Care Medicine 1994; 22: Green J, Wintfeld N, Sharkey P, Passman LJ. The importance of severity of illness in assessing hospital mortality. Journal of the American Medical Association 1990; 263: Tuman K, McCarthy R, March R, Najafi H, Ivankovich A. Morbidity and duration of ICU stay after cardiac surgery. A model for preoperative risk assessment. Chest 1992; 102: Geraci JM, Rosen AK, Ash AS, McNiff KJ, Moskowitz MA. Predicting the occurrence of adverse events after coronary artery bypass surgery. Annals of Internal Medicine 1993; 118: Tu JV, Guerriere MR. Use of a neural network as a predictive instrument for length of stay in the intensive care unit following cardiac surgery. Proceedings of the Annual Conference on Computer Applied Medical Care 1992; 5: Trujillo M, Arai K, Bellorin-Font E. Practical guide for drug administration by intravenous infusion in intensive care units. Critical Care Medicine 1994; 22: Dixon WJ. BMDP Statistical Software Manual. Berkeley, CA: University of California Press, Dawson-Saunders B, Trapp R. Basic and Clinical Biostatistics. A Lange Medical Book, 2nd Edn. New York: Prentice-Hall International Inc., Hosmer DW, Lemeshow S. Applied Logistic Regression. New York: John Wiley and Sons, Turner JS, Mudaliar YM, Chang RWS, Morgan CJ. Acute

5 212 British Journal of Anaesthesia physiology and chronic health evaluation (APACHE II) scoring in a cardiothoracic intensive care unit. Critical Care Medicine 1991; 19: Tu JV, Mazer CD, Levinton C, Armstrong PW, Naylor CD. A predictive index for length of stay in the intensive care unit following cardiac surgery. Canadian Medical Association Journal 1994; 151: Tu JV, Jaglal SB, Naylor CD. Multicenter validation of a risk index for mortality, intensive care unit stay, and overall hospital length of stay after cardiac surgery. Circulation 1995; 91: Zaroff J, Aronson S, Lee BK, Feinstein SB, Walker R, Wiencek JG. The relationship between immediate outcome after cardiac surgery, homogeneous cardioplegia delivery and ejection fraction. Chest 1994; 106: Weintraub W, Jones E, Craver J, Guyton R, Cohen C. Determinants of prolonged length of hospital stay after coronary bypass surgery. Circulation 1989; 80: