Risk Stratification Using The Society of Thoracic Surgeons Program

Transcription

1 Risk Stratification Using The Society of Thoracic Surgeons Program Brack G. Hattler, MD, PhD, Carol Madia, PA, Carol Johnson, CRNP, John M. Armitage, MD, Robert L. Hardesty, MD, Robert L. Kormos, MD, Si M. Pham, MD, Dale N. Payne, MD, and Bartley P. Griffith, MD Division of Cardiothoracic Surgery, University of Pittsburgh Medical Center, Pittsburgh, Pennsylvania In an era of progressive cost containment and public scrutiny, the wisdom of aggressive surgical therapy for high-risk candidates has been questioned. At our center in the previous 24 months, 728 patients with coronary artery disease were entered into The Society of Thoracic Surgeons national database, and the hospital outcomes plus length of stay were analyzed. Patients were separated according to the predicted mortality based on the groupings in The Society of Thoracic Surgeons database: o to 5% (453 patients); 5% to 10% (126 patients); 10% to 20% (96 patients); 20% to 30% (17 patients); and 30% and greater (36 patients). There was a dose correlation with the predicted rates of mortality. Importantly, the preoperative risk stratification demonstrated a strong correlation with the significant morbidity and excessive length of stay in the highest-risk groups (predicted risk of 20% to 2':30%). The incidences of the most common complications in the group with the highest predicted risk (2':30%) were 28%, renal failure; 33%, ventilator dependence; and 17%, cardiac arrest. In addition, at short-term follow-up (6 to 8 months), a 24.3% mortality was identified in patients with a predicted mortality that exceeded 20%. These data quantify the risks and morbidities associated with the care of seriously ill patients with coronary artery disease and demonstrate the need for professional and public discussions focusing on the association of a high preoperative risk status and the consumption of resources. (Ann Thorae Surg ) A t a time of increasing scrutiny from government agencies, third-party payers, and consumer groups, accurate, valid, and risk-stratified data with which to assess various outcome criteria are critically needed by hospitals, academic centers, and practicing cardiothoracic surgeons. To be of practical use to the clinician, a computer-based risk assessment should provide a means for accurately predicting not only the operative mortality, but also the major morbidity, the length of hospital stay, and, ultimately, the cost that the institution will incur in caring for patients in various risk categories. To this end, The Society of Thoracic Surgeons (STS) National Cardiac Surgery Database operating in more than 600 hospitals across the country now contains the records for approximately 240,000 patients who underwent isolated coronary artery surgical procedures from 1980 through 1993 [1]. This extensive repository of cases in the STS database allows for accurate statistical modeling. A portion of these stored data has been used recently to formulate a risk equation based on a detailed and exhaustive analysis of risk factors in patients undergoing isolated coronary artery operations. This has provided a tool for predicting the probability of an operative death (all deaths occurring during hospitalization or after discharge from the hospital, but within 30 days of the procedure) in any individual patient [2]. Presented at the Poster Session of the Thirtieth Annual Meeting of The Society of Thoracic Surgeons, New Orleans, LA, Jan 31-Feb 2, Address reprint requests to Dr Hattler, University of Pittsburgh Medical Center, 200 Lothrop st. C-700 PUH, Pittsburgh, PA by The Society of Thoracic Surgeons Whereas variate analysis identifies and weighs risk factors predictive of operative mortality, the STS risk equation looks at combinations of preoperative risk factors as they apply to the outcomes of individual patients. The sample size for anyone institution or practice group is necessarily small relative to the STS database. The grouping of patients, however, into risk categories according to the STS model allows individual hospitals and surgeons to compare their results with a large, statistically valid patient population representative of practice patterns across the United States. Using the STS database and statistical analysis package, we have analyzed our own recent experience in patients who have undergone isolated coronary artery operations. In our subset of patients, we have been able to examine not only the ability of the risk model to predict operative mortality, but also its ability to correlate various categories of risk with major morbidity and length of hospital stay. Material and Methods A total of 728 patients undergoing isolated coronary artery bypass grafting at our institution over a 2-year interval ending in October 1993 form the basis for this study. Patient variables as defined and published by the STS database were used to complete standard database forms for each patient entered into the study [3]. Two trained nurse practitioners working solely within the division of cardiothoracic surgery abstracted charts at the time of the patients' discharge. Once they were familiar with the /94/$7.00

2 Ann Thorac Surg HATILER ET AL 1349 system records, it took an average of 15 minutes for them to review each chart. Checks by physicians confirmed the accuracy of the data for the initial study patients. Chart reviews are now conducted solely by these trained personnel. This information was stored in the STS database, but was also available locally for inhouse use where data were extracted and subjected to statistical analyses. The statistical techniques applicable to the STS database have recently been described in detail by Edwards and associates [2]. Preoperative risk factors were examined by univariate analysis using a JI technique with two-by-two contingency tables and one degree of freedom. Multivariate analysis was carried out using step-wise, logistic regression to determine those variables having an independent impact on patient survival. Calculated p values less than 0.05 were considered statistically significant. The risk model developed for the STS database was used to examine the effect of multiple risk factors as they pertain to individual patient survival. This model uses a modification of the bayesian algorithm and has been validated using a training set/test set approach. Patients can thus be grouped according to their predicted probability of operative death. After grouping into risk categories, the observed mortality in our patient population was compared with the preoperative predicted mortality. Using these same groupings, morbidity and length of hospital stay were readily extracted from the STS database. Short-term follow-up in our patient population was accomplished either through clinic visits or telephone contacts with the patients or their referring physicians. A questionnaire was administered to determine the New York Heart Association class and the patient's subjective assessment of his or her own condition. Results The clinical profiles for our patient sample using the risk factors from the STS database are shown in Table 1 and compared to information recently published from the STS database that covers the years 1980 through 1990 [2]. Our data represent a more recent time interval (1992 to 1993), and reflect a higher percentage of morbid risk factors in our patient population than that for the STS database. The mortality statistics as derived from the STS database for our sample of 728 patients are depicted in Table 2. The predicted mortality rate is high (6.94%) and reflects, in part, the activity of a high-risk cardiac surgery service that functions as a separate entity within our institution. The overall observed mean mortality (3.98%) for these patients, however, is significantly lower than the predicted mortality (p < 0.005). The predicted versus the observed mortalities for the various risk-stratified groups as derived from the risk model developed for the STS database are depicted in Figure 1. The upper and lower curves represent the calculated range for the risk interval. The data do indicate that the mortality associated with higher-risk patients does not exceed the predicted norm. Not included in the data depicted in Figure 1 are 13 patients with a predicted risk exceeding 50% and an observed mortality rate of 30.8%. Table 1. Summary of Study Patient Profiles as Compared With STS National Database Patient Profiles" STS Database" (% with (% with risk factor; risk factor; Risk Factor n = 728) n = 78,927) Age <70 years >70 years Female Male White Tobacco abuse Family history of coronary artery disease Renal failure Diabetes Hypertension Morbid obesity Hypercholesterolemia Pulmonary hypertension Cerebrovascular disease Cardiomegaly Chronic obstructive pulmonary disease Immunosuppressive treatment Percutaneous transluminal coronary angioplasty <6 hours before CABG Myocardial infarction < days before CABG Myocardial infarction > days before CABG Cardiogenic shock Arrhythmia Unstable angina Intravenous nitrates Intravenous inotropic agents New York Heart Association class I II III IV First operation Reoperation Elective operation Nonelective operation Single-vessel disease Double-vessel disease Triple-vessel disease Preoperative intraaortic balloon pump Ejection fraction < > a Data on these 728 patients were gathered in 1992 and the STS database were gathered from 1980 through 1990 [2]. CABG = coronary artery bypass grafting; Surgeons. b Data on STS = The Society of Thoracic

3 1350 HATILER ET AL Ann Thorac Surg Table 2. The Society of Thoracic Surgeons Risk Stratification Analysis Variable Prediction of operative mortality Sample size for study (No.) Observed mortality in sample (No.) Observed mean mortality Predicted mean mortality Range of predicted mortality in sample Minimum Maximum Standard deviation of predicted mortality Standard error of predicted mortality >I analysis of results predicted versus observed mortality Corrected >I statistic p Value for >I statistic Mortality Groups (Bars = Observed Mortality) Value % 6.94% 0.17% 58.56% <0.005 The risk factors listed in Table 3 were found to be statistically significant in predicting mortality in our patient sample. In addition to cardiogenic shock, the need for intravenously administered nitrates preoperatively, emergency procedures, and a preoperative intraaortic balloon pump for managing hemodynamic instability, as well as the patient being in New York Heart Association class IV, were all found to be highly significant risk factors (mortality risk present versus mortality risk absent; p < 0.005). The risk ratio is derived by dividing the mortality rate with the risk present by the mortality rate with the risk absent. As an example, a patient requiring inotropic agents preoperatively had a 5.1 times greater likelihood of dying after coronary artery bypass grafting at our institution than did a patient not requiring inotropic agents preoperatively. The predicted-risk intervals, as determined preoperatively, and the observed morbidity and length of stay (procedure to discharge) are given in Table 4. The utility of the STS preoperative risk categories is evident as they pertain to postoperative complications. The average number of complications per patient increased in a linear 60% 50% "C 40% Gl. ū 30% "C Gl... II.. 20% % Mortality 10% 2.21% 0% 0-5% 5-10% 10-20% 20-30% 30-50% Fig 1. Predicted mortality for The Society of Thoracic Surgeons model and actual observed mortality for patients in various preoperative risk categories. Not in the data depicted are 13 patients with a predicted risk exceeding 50% and an observed mortality rate of 30.8%. Table 3. Univariate Analysis of Preoperative Risk Variables Found to be Statistically Significant for Entire Population Group (n = 728) Mortality Mortality Risk p Risk Variable (risk present) (risk absent) Ratio Value" Morbid obesity 3/18 (16.67%) 26/710 (3.66%) 4.6 <0.050 Time from 2/8 (25.00%) 27/718 (3.76%) 6.6 <0.050 failed PTCA to CABG <6 hours Prior 12/136 (8.82%) 8/309 (2.59%) 3.4 <0.010 myocardial infarction <21 days Cardiogenic 7/25 (28.00%) 22/703 (3.13%) 8.9 <0.005 shock Intravenous 13/152 (8.55%) 16/576 (2.78%) 3.1 <0.005 nitrates preoperatively Inotropic 3/ %) 26/712 (3.65%) 5.1 <0.025 agents preoperatively New York 17/230 (7.39%) 12/490 (2.45%) 3.0 <0.005 Heart Association class IV Nonelective 18/ %) 11/548 (2.01%) 5.0 <0.005 procedure Preoperative 11/ %) 18/659 (2.73%) 5.8 <0.005 intraaortic balloon pump Ejection 8/ %) 21/659 (3.19%) 3.6 <0.05 fraction <0.30 a it? analysis of results (mortality risk present versus mortality risk absent). PTCA = percutaneous trans CABG = coronaryarterybypass grafting; luminal coronary angioplasty. fashion as the predicted risk became progressively greater. Patients in the greater than 30% predicted-risk category sustained an average of three complications per patient, with two-thirds of patients experiencing complications. Short-term follow-up was obtained in 94% of the 149 patients categorized preoperatively into high-risk (>10%) groups (Table 5). As an example, of the 36 patients with a preoperative risk assessment of greater than 30%, there were seven hospital deaths (19.4%). Of the remaining 27 patients (2 were lost to follow-up), 8 (29.6%) were dead at a mean follow-up of 6.4 months. Of the 19 patients available for follow-up, 18 (94.7%) showed an improvement in their New York Heart Association class of at least two grades. A lesser number (78.9%) felt better subjectively. Comment Using the statistical techniques available through the STS database, we have analyzed our 2-year experience with isolated coronary artery bypass grafting. As relatively recent subscribers to this system, we were interested in determining the validity of these results as they relate to our practice. The comparison of predicted and observed

4 Ann Thorae Surg HATTLER ET AL 1351 Table 4. Univariate Analysis of Morbidity and Length of Hospital Stay Comparing Low-Risk and High-Risk Patient Groups The Society of Thoracic Surgeons Predicted-Risk Intervals 0%-5% 5%-10% 10%-20% 20%-30% >30% Complication (n = 453) (n = 126) (n = 96) (n = 17) (n = 36) Reoperative bleeding (%) a Perioperative myocardial infarction (%) Infection (%) Mediastinal b Septicemia b Stroke (%) Permanent a b Transient Ventilator >5 days (%) a 1O.42 b b b Renal failure (no dialysis) (%) b b b Dialysis required (%) a b b Heart block-permanent (%) Cardiac arrest (%) a a b Anticoagulant complication (%) b Tamponade (%) b Gastrointestinal complication (%) b 1O.42 b b b Multisystem failure (%) a b Inhospital mortality (%) a b Average No. complications per patient a b Patients having complications (%) " b Length of stay-procedure to discharge ± ± b ± b ± c ± b (days ± SD) a p < versus 0-5% group. SD = standard deviation. b P < versus 0-5% group. c p < versus 0-5% group. mortality rates revealed that patients in various risk categories at our institution are being cared for in a manner that meets standards established by the STS database. The small numbers of patients in groups at greater risk makes Table 5. Short-Term Follow-Up in High-Risk Patients Patients (No.) Mean follow-up (mo. ± SD) Lost to follow-up (No. of patients) Follow-up mortality Patient's assessment Better Worse Same Improvement in New York Heart Association class SD = standard deviation. The Society of Thoracic Surgeons Predicted-Risk Intervals 10%-20% (n = 89) 9.9 ± /84 (6.0%) 58/79 (73.4%) 13/7906.1%) 8/7900.1%) 71/79 (89.9%) 20%-30% (n = 15) 5.9 ± 4.4 2/1404.3%) 9/12 (75.0%) 1/12 (8.3%) 2/1206.7%) 11/12 (91.7%) >30% (n = 29) 6.4 ± /27 (29.6%) 15/19 (78.9%) 3/1905.8%) 1/19 (5.3%) 18/19 (94.7%) any conclusion relative to predicted versus observed mortality less reliable (see Fig 1). However, if those deaths in high-risk (>10%) patients that occur during short-term follow-up are included (see Table 5), the observed mortality in each risk category then lies within the range of the predicted mortality. This suggests that, in high-risk patients with numerous associated risk factors, the hazard function as defined by Kirklin and Barratt-Boyes [4] extends beyond the traditional time interval for an operative death to include the early months after discharge from the hospital. Other models besides that of the STS have been used to predict operative risk [5-8]. These have recently been reviewed [2]. In general, most institutional models consist of smaller sample sizes, and their ability to accurately predict outcomes has been found to provide useful results of interest largely to local users. Interest in the STS model, however, has been prompted by the need for a system that goes beyond any local institutional needs to a more flexible and complete database made up of large numbers of patients and against which individual outcomes can be measured over time and as patient characteristics may change [2, 9, 10]. Even early users of the STS database can derive useful data by arranging relatively small numbers of patients into risk category groups. To the extent that the STS database represents a broad national experience, it forms a basis for tracking trends in surgical outcome within various risk categories. Our analyses of patients' preoperative risk through STS modeling have provided not only postoperative correlations

5 1352 HATTLER ET AL Ann Thorac Surg 1994;58:] with mortality data, but also correlations with morbidity and length of hospital stay. In addition, at short-term follow-up, a significant mortality (24.3%) was identified in the first 6 to 8 months after operation in patients with a greater than 20% predicted mortality. These data help, therefore, to quantify the risks, morbidity, and consumption of resources associated with operations in seriously ill patients with coronary artery disease. They should not be used, however, as the only variable in the decision-making process concerning the management of individual patients at increased risk, as the majority of patients in these categories had shown improvement in their New York Heart Association classification on short-term follow-up. There are no data to allow adequate comparison of the natural history of patients in these same high-risk categories. Even death rates for patients on our waiting list for cardiac transplantation with ejection fractions similar to those in the high-risk patients were unsuitable for comparison because they lacked comparable comorbid conditions. The Health Care Financing Administration continues to collect and publish data on mortality statistics for all hospitals performing coronary artery bypass grafting on Medicare patients. Many states now follow a practice similar to that adopted by New York and Pennsylvania in publishing not only hospital, but also physician-specific, mortality rates [11]. Objections have been raised to both the Health Care Financing Administration and various local state systems, criticizing them as being "insufficiently sensitive to make valid inferences about the quality or effectiveness" of various outcome criteria [12, 13]. The model used by the Pennsylvania Health Care Cost Containment Council (Medis), a model with which we are most familiar, has been scrutinized for its ability to provide valid information on risk-adjusted mortality rates to consumers in Pennsylvania. The model has been found to lack sensitivity in its ability to portray comorbid conditions and to assess accurately the severity of illness for each patient entered into the state's database [12]. More specifically, risk factors that are recognized as important (ie, ejection fraction, left main occlusive disease, urgent failed percutaneous transluminal coronary angioplasty, acute cardiac catheterization emergencies, requirement for preoperative intraaortic balloon pump, and preoperative need for intravenously administered inotropic agents) are not recognized by this model [11]. Of even greater concern is the fact that each patient's risk-adjusted mortality and morbidity statistics are partly determined by a derived admission severity grouping. This grouping is arrived at by using a system that assigns a score to each entry from a list of key clinical findings. For each patient's hospital chart, the key clinical findings are searched for during a review of the admission record. After the key clinical findings list is compared with the patient's chart, missing data are recorded as normal data. The admission severity grouping is not revised if the patient, days later and before operation, suffers a myocardial infarction and cardiogenic shock develops. The patient's status on admission to the operating room can, consequently, be ignored. We have found that 12% of the patients at our institution with an admission severity grouping of 1 (0.2% to 1.1% risk) have an STS database-predicted mortality that exceeds 10%. The system is therefore particularly onerous to institutions that attract high-risk patients in unstable conditions [14]. The STS along with the American Medical Association have clearly stated their support for accountability and the release to the public of risk-adjusted hospital and physician-specific data on coronary artery bypass grafting when the data are accurate, carefully gathered, and complete, and when the statistical analyses used are demonstrated to be valid by peer review [12, 15]. In this era of close scrutiny, it is imperative that cardiothoracic surgeons be empowered in their dealings with various agencies by a database that approaches a national average against which individual performance can be measured. The extensive national database established by the STS represents a beginning in arriving at this goal. It will be important for many other groups to confirm its utility. We thank Dr Fred H. Edwards, Dr Richard E. Clark, and Mr Marc Schwartz for their assistance and helpful suggestions, and also Rhonda J. White for her technical assistance. References 1. Data analyses of The Society of Thoracic Surgeons National Cardiac Surgery Database. The third year. Minneapolis: Summit Medical Systems, January Edwards FH, Clark RE, Schwartz M. Coronary artery bypass grafting: The Society of Thoracic Surgeons national database experience. Ann Thorac Surg 1994;57: The Society of Thoracic Surgeons National Cardiac Surgery Database manual for data managers. Minneapolis: Summit Medical Systems, August Kirklin JW, Barratt-Boyes BG, eds. The generation of knowledge from information, data, and analyses. In: Cardiac Surgery, 2nd ed. New York: Churchill Livingstone, 1993: Junod FL, Harlan BI, Payne I, et al. Preoperative risk assessment in cardiac surgery: comparison of predicted and observed results. Ann Thorac Surg 1987;43: Parsonnet V, Dean 0, Bernstein AD. A method of uniform stratification of risk for evaluating the results of surgery in acquired adult heart disease. Circulation 1989;79: Hannan EL, Kilburn H, O'Donnell JF, Lukacik G, Shields EP. Adult open heart surgery in New York state. JAMA 1990;264: Higgins TL, Estafanous FG, Loop FD. Stratification of morbidity and mortality outcome by pre-operative risk factors in coronary artery bypass patients. JAMA 1992;267: Edwards FH, Graeber GM. The theorem of Bayes as a clinical research tool. Surg Gynecol Obstet 1987;165: Edwards FH, Albus RA, Zajtchuk R, Graeber GM, Barry M. A quality assurance model of operative mortality in coronary artery surgery. Ann Thorac Surg 1989;47: The Pennsylvania Health Care Cost Containment Council. Coronary artery bypass graft surgery: a technical report. Harrisburg, PA: The Pennsylvania Health Care Cost Containment Council, Report of the Ad Hoc Committee on physician-specific mortality rates for cardiac surgery. Ann Thorac Surg 1993;56: Green I, Wintfeld N, Sharkey P, Passman LJ. The importance of severity of illness in assessing hospital mortality. JAMA 1990;263: Greenfield S, Aronou HU, Elashoff RM, Watanabe D. Flaws in mortality data. The hazards of ignoring comorbid disease. JAMA 1988;260: American Medical Association Policy Compendium. Chicago: American Medical Association, 1993:131-2.