Use of a Bayesian statistical model for risk assessment in coronary artery surgery

Superlative performance in cardiovascular surgery

Achieving excellence in surgery is an ongoing endeavor, gained through experience, observation, and practice. It is difficult to evaluate enterprise excellence, but attempts include the ratings of the Society of Thoracic Surgeons. The surgery team at Baylor University Medical Center has achieved three-star ratings for 9 of the past 10 evaluations for coronary artery bypass. This accomplishment is a result of many factors, including teamwork, multidisciplinary conferences, application of the latest evidence, continuous efforts at quality improvement, and effective governance. Some aspects of the latter include individual excellence, enjoying the work, being bold, having psychological safety, and employing meritocracy. Discernment of contemporary issues, a clear vision of the common good, and virtuous service to all must be attained while preserving the highest level of patient-centered service to patients and the institution.

Aprotinin does not increase the risk of renal failure in cardiac surgery patients

BACKGROUND

Aprotinin is frequently used in high-risk cardiac surgery patients to decrease bleeding complications and transfusions of packed red blood cells (PRBC). Transfusions of PRBC are known to directly increase the risk of new onset postoperative renal failure (ARF) in cardiac surgery patients. A recent highly publicized report implicated aprotinin as an independent causal factor for postoperative renal failure, but ignored the potential confounding affect of numerical PRBC data on ARF. We sought to investigate that claim with an analysis that included all perioperative risk factors for renal failure, including PRBC transfusion data.

METHODS AND RESULTS

Prospectively collected patient data from 12 centers contributing to the Merged Cardiac Registry, an international multicenter cardiac surgery database, operated on between January 2000 and February 2006 were retrospectively analyzed. A previously published risk model for ARF incorporating 12 variables was used to calculate a baseline ARF risk score for each patient in whom those variables were available (n=15,174). After adding transfused PRBC data 11,198 patients remained for risk-adjusted assessment of ARF in relation to aprotinin use. Risk-adjusted multivariable analyses were carried out with, and without, consideration of transfused PRBC. Aprotinin was used in 24.6% (2757/11,198). The overall incidence of ARF was 1.6% (180/11,198) and was higher in the aprotinin subset (2.6%, 72/2757 versus 1.3%, 108/8441; P<0.001). The incidence of ARF directly and significantly increased with increasing transfusions of PRBC (P<0.001). Risk-adjusted analysis without transfused PRBC in the model suggests that aprotinin significantly impacts ARF (P=0.008; OR=1.5). However, further risk adjustment with the addition of the highly significant transfused PRBC variable (P<0.0001; OR=1.23/transfused PRBC) to the model attenuates the purported independent affect of aprotinin (P=0.231) on ARF.

CONCLUSIONS

The increase in renal failure seen in patients who were administered aprotinin was directly related to increased number of transfusions in that high-risk patient population. Aprotinin use does not independently increase the risk of renal failure in cardiac surgery patients.

Aborted off-pump coronary artery bypass patients have much worse outcomes than on-pump or successful off-pump patients

BACKGROUND

Off-pump coronary artery bypass graft (CABG) surgery is purported to reduce perioperative mortality and morbidity compared with on-pump coronary bypass graft surgery. However, the outcomes of patients for whom an off-pump strategy must be changed to an on-pump procedure during surgery have not been extensively studied.

METHODS AND RESULTS

The Merged Cardiac Registry (Health Data Research, Inc) contains 70 514 isolated CABG performed from January 1998 to March 2004 in 40 facilities. Among them, 62 634 patients begun and completed on-pump bypass (CPB); 7880 patients begun off-pump, of which 7424 (94.2%) completed off-pump coronary artery bypass (OPCAB), whereas 456 (5.8%) were converted to on-pump (CONVERT). CONVERT patients were more severely ill. The observed mortality of CONVERT, CPB, and OPCAB was 9.9%, 3.0%, and 1.6%, respectively, and the observed-to-predicted ratio was 2.77, 1.20, and 0.74, respectively. CONVERT also had more morbidity than either OPCAB or CPB. Finally, a risk model was created to identify patients who might be at risk for conversion from off-pump to on-pump CABG.

CONCLUSIONS

Patients who are intended for an off-pump strategy and then require conversion to on-pump have significantly higher operative mortality and morbidity than either completed OPCAB or CPB patients. In addition, the operative mortality and morbidity are far in excess of that predicted preoperatively. Based on these results, strong consideration should be given for a planned strategy of CPB for those patients with preoperative hemodynamic instability requiring a salvage CABG operation, left ventricular hypertrophy, or previous CABG.

Computer-aided disease prediction system: development of application software with SAS component language

AIMS

The intricacy of predictive models associated with prognosis and risk classification of disease often discourages medical personnel who are interested in this field. The aim of this study was therefore to develop a computer-aided disease prediction model underpinning a step-by-step statistics-guided approach including five components: (1) data management; (2) exploratory analysis; (3) type of predictive model; (4) model verification; (5) interactive mode of disease prediction using SAS 8.02 Windows 2000 as a platform.

METHODS

The application of this system was illustrated by using data from the Swedish Two-County Trial on breast cancer screening. The effects of tumour size, node status, and histological grade on breast cancer death using logistic regression model or survival models were predicted. A total of 20 questions were designed to exemplify the usefulness of each component. We also evaluated the system using a controlled randomized trial. Times to finish the above 20 questions were used as endpoint to evaluate the performance of the current system. User satisfaction with the current system such as easy to use, the efficiency of risk prediction, and the reduction of barrier to predictive model was also evaluated.

RESULTS

The intervention group not only performed more efficiently than the control group but also satisfied with this application software.

CONCLUSIONS

The MD-DP-SOS system characterized by menu-driven style, comprehensiveness, accuracy and adequacy assessment, and interactive mode of disease prediction is helpful for medical personnel who are involved in disease prediction.

Cardiac surgery risk modeling for mortality: a review of current practice and suggestions for improvement

Risk models play a vital role in monitoring health care performances. Despite extensive research and widespread use of risk models in cardiac surgery, there are methodologic problems. We reviewed the methodology used for risk models for short-term mortality. The findings suggest that many risk models are developed in an ad hoc manner. Important aspects such as selection of risk factors, handling of missing values, and size of the data used for model development are not dealt with adequately. Methodologic details presented in publications are often sparse and unclear. Model development and validation processes are not always linked to the clinical aim of the model, which may affect their clinical validity. We make some suggestions in this review for improvement in methodology and reporting.

Risk stratification for adverse outcome in cardiac surgery

Risk-adjusted outcome prediction is mainly important in two separate fields. The first is quality monitoring: measuring actual versus predicted mortality in an institution allows assessment of the clinical surgical and anaesthesia performance while adjusting for the risk profile of the patients. Without risk stratification, surgeons and hospitals treating high-risk patients will appear to have worse results than others. This may prejudice referral patterns, affect the allocation of resources and even discourage the treatment of high-risk patients. The second field is that of informed consent and clinical decision-making. Risk-adjusted predicted mortality should form an important part of patient and surgeon decisions on whether or not to proceed with surgery. Clearly, no 'perfect' model can be produced as some aspects of mortality will always be related to risk factors not included in the model (e.g. the quality of the distal coronary artery vessels in coronary artery surgery) or due to chance happenings not related to preoperative patient characteristics (such as surgical error). An individual patient will either survive or die after cardiac surgery. Clearly, no scoring system will predict the specific outcome for every patient. However, risk stratification will inform patients and clinicians of the likely risk of death for a group of patients with a similar risk profile undergoing the proposed operation. This information is useful and should form part of the basis on which the patient and surgeon decide whether to proceed.

A Bayesian hierarchical approach to comparative audit for carotid surgery

OBJECTIVES

the aim of this study was to illustrate how a Bayesian hierarchical modelling approach can aid the reliable comparison of outcome rates between surgeons.

DESIGN

retrospective analysis of prospective and retrospective data.

MATERIALS

binary outcome data (death/stroke within 30 days), together with information on 15 possible risk factors specific for CEA were available on 836 CEAs performed by four vascular surgeons from 1992-99. The median patient age was 68 (range 38-86) years and 60% were men.

METHODS

the model was developed using the WinBUGS software. After adjusting for patient-level risk factors, a cross-validatory approach was adopted to identify "divergent" performance. A ranking exercise was also carried out.

RESULTS

the overall observed 30-day stroke/death rate was 3.9% (33/836). The model found diabetes, stroke and heart disease to be significant risk factors. There was no significant difference between the predicted and observed outcome rates for any surgeon (Bayesian p -value>0.05). Each surgeon had a median rank of 3 with associated 95% CI 1.0-5.0, despite the variability of observed stroke/death rate from 2.9-4.4%. After risk adjustment, there was very little residual between-surgeon variability in outcome rate.

CONCLUSIONS

Bayesian hierarchical models can help to accurately quantify the uncertainty associated with surgeons' performance and rank.

Surgical performance measurement

The need for effective surgical performance measurement has gained an increasingly high profile in recent years, particularly since events at Bristol Royal Infirmary, where apparent poor performance has prompted the UK Department of Health to instigate a major Public Inquiry. This paper describes issues that concern the measuring and monitoring of surgical performance, and methods that have been devised for judging a good surgeon from the less competent. The authors are a collaborative team composed of specialists in Cardiothoracic surgery and Operational Research analysts with experience of monitoring performance in cardiac surgery. This paper describes concrete examples from that knowledge base.

Validation of a preoperative risk index as a predictor of perioperative morbidity and hospital costs in coronary artery bypass graft surgery

OBJECTIVE

To validate a previously developed model (CABDEAL) for predicting postoperative morbidity for coronary artery bypass graft (CABG) surgery patients using the New York State Statewide Planning and Research Cooperative System (SPARCS) database and to examine the effects of preoperative risk factors, postoperative complications, and death on costs of care for CABG surgery.

DESIGN

Retrospective database review.

SETTING

Governmental agency database of cardiac surgery.

PARTICIPANTS

CABG surgery patients (n = 15,388).

INTERVENTIONS

A previously developed preoperative risk model (CABDEAL) was applied to all patients. Predicted length of hospital stay and costs were compared with actual length of stay and costs, using a charge-to-cost conversion formula.

MEASUREMENTS AND MAIN RESULTS

The CABDEAL model was moderately predictive of outcomes. The specificity was 64%, the sensitivity was 73.8%, and the receiver operating characteristic curve area was 0.728. Morbidity in the form of postoperative complications was recorded in 24.5% (3,770 patients), and the mortality rate was 3.4% (527 patients). The mean (+/- SD) total hospital cost was 28,408 US dollars +/-28,982, and the median cost was 21,644 US dollars. Based on the linear regression model, an equation was developed for predicting total costs: Cost (in US dollars) = 22,952 + (3,277. [CABDEAL score]).

CONCLUSION

The previously developed CABDEAL model was predictive of increased morbidity in the SPARCS database. Total hospital costs increased nearly linearly with increasing CABDEAL score. These results encourage the development of models for preoperative estimation of costs related to perioperative morbidity.

Cardiac risk stratification for postmyocardial infarction dental patients

Traditional dental management guidelines of myocardial infarction survivors mandate a 6-month waiting period before elective treatment can be considered. Technological advances in cardiac disease diagnosis, management, and revascularization treatment may make this older mandatory 6-month waiting period obsolete. The purposes of this literature review are to provide an overview of the historical development of cardiac risk stratification and discuss current developments and guidelines in cardiac risk assessment. We hope that this review and update will stimulate the development of updated dental guidelines for treating the cardiac patient.

Centrifugal pumping during routine open heart surgery improves clinical outcome

Carrying out a 1,000 patient prospective, randomized study comparing a roller pump and the BioMedicus centrifugal pump (CP), hematological parameters, blood loss, renal function, postoperative complications, and lethality data were evaluated. Using a validated preoperative risk stratification method (Cardiac RiskMaster), patients were divided into different risk categories for statistical analysis. This study verified an improved outcome with the use of a CP in routine cardiac surgery, demonstrated by blood handling, blood loss, renal function, and nephrological complication data. There was also a significant reduction in neurological complications. There was no significant difference in postoperative lethality, but high risk patients demonstrated outcomes comparable to those being defined for medium risk patients. Routine cardiac surgical patients as well as multimorbid patients benefit from the use of a CP. Preoperative risk stratification is a valid tool to demonstrate how the employment of new technologies can provide for an improved outcome without increasing overall costs at the same time.

Predicting in-hospital deaths from coronary artery bypass graft surgery. Do different severity measures give different predictions?

OBJECTIVES

Severity-adjusted death rates for coronary artery bypass graft (CABG) surgery by provider are published throughout the country. Whether five severity measures rated severity differently for identical patients was examined in this study.

METHODS

Two severity measures rate patients using clinical data taken from the first two hospital days (MedisGroups, physiology scores); three use diagnoses and other information coded on standard, computerized hospital discharge abstracts (Disease Staging, Patient Management Categories, all patient refined diagnosis related groups). The database contained 7,764 coronary artery bypass graft patients from 38 hospitals with 3.2% in-hospital deaths. Logistic regression was performed to predict deaths from age, age squared, sex, and severity scores, and c statistics from these regressions were used to indicate model discrimination. Odds ratios of death predicted by different severity measures were compared.

RESULTS

Code-based measures had better c statistics than clinical measures: all patient refined diagnosis related groups, c = 0.83 (95% C.I. 0.81, 0.86) versus MedisGroups, c = 0.73 (95% C.I. 0.70, 0.76). Code-based measures predicted very different odds of dying than clinical measures for more than 30% of patients. Diagnosis codes indicting postoperative, life-threatening conditions may contribute to the superior predictive power of code-based measures.

CONCLUSIONS

Clinical and code-based severity measures predicted different odds of dying for many coronary artery bypass graft patients. Although code-based measures had better statistical performance, this may reflect their reliance on diagnosis codes for life-threatening conditions occurring late in the hospitalization, possibly as complications of care. This compromises their utility for drawing inferences about quality of care based on severity-adjusted coronary artery bypass graft death rates.

Coronary artery bypass risk prediction using neural networks

BACKGROUND

Neural networks are nonparametric, robust, pattern recognition techniques that can be used to model complex relationships.

METHODS

The applicability of multilayer perceptron neural networks (MLP) to coronary artery bypass grafting risk prediction was assessed using The Society of Thoracic Surgeons database of 80,606 patients who underwent coronary artery bypass grafting in 1993. The results of traditional logistic regression and Bayesian analysis were compared with single-layer (no hidden layer), two-layer (one hidden layer), and three-layer (two hidden layer) MLP neural networks. These networks were trained using stochastic gradient descent with early stopping. All prediction models used the same variables and were evaluated by training on 40,480 patients and cross-validation testing on a separate group of 40,126 patients. Techniques were also developed to calculate effective odds ratios for MLP networks and to generate confidence intervals for MLP risk predictions using an auxiliary "confidence MLP."

RESULTS

Receiver operating characteristic curve areas for predicting mortality were approximately 76% for all classifiers, including neural networks. Calibration (accuracy of posterior probability prediction) was slightly better with a two-member committee classifier that averaged the outputs of a MLP network and a logistic regression model. Unlike the individual methods, the committee classifier did not overestimate or underestimate risk for high-risk patients.

CONCLUSIONS

A committee classifier combining the best neural network and logistic regression provided the best model calibration, but the receiver operating characteristic curve area was only 76% irrespective of which predictive model was used.

Use of a probabilistic neural network to estimate the risk of mortality after cardiac surgery

OBJECTIVE

To develop a probabilistic neural network (PNN) to estimate mortality risk following cardiac surgery.

DESIGN AND SETTING

The PNN model was created using an institutional database obtained as part of routine quality assurance activity. Patient records (from 1991 to 1993) were randomly divided into training (n = 732) and validation (n = 380) sets. The model uses seven variables, each obtainable during routine clinical patient care. After completion of the initial validation phase, newer data (1994) became available and were used as an independent source of validation (n = 365).

PATIENTS

1,477 consecutive cardiac surgery patients operated on in a teaching hospital during a four-year period (1991-94).

RESULTS

The overall accuracy of the neural network was 91.5% in the training set; it was 92.3% in the validation set. The model was well calibrated (p = 0.21 for the Hosmer-Lemeshow goodness-of-fit test) and discriminated well (areas under the ROC curves were 0.72 and 0.81 for the training and validation sets). The trained network also performed well on the 1994 data (ROC = 0.74, p = 0.19 for the Hosmer-Lemeshow test), albeit with a slight decrement in overall accuracy (88.2%).

CONCLUSION

A neural network may be implemented to estimate mortality risk following cardiac surgery. Implementation is relatively rapid, and it is an alternative to standard statistical approaches.

Preoperative prediction of postoperative morbidity in coronary artery bypass grafting

BACKGROUND

The risk factors of patients selected for coronary artery bypass grafting have increased in recent years because of the aging population. Prediction of postoperative complications is essential for optimal use of the available resources. The aim of this study was to develop a scoring method for prediction of postoperative morbidity of individual patients undergoing bypass grafting.

METHODS

Data from 386 consecutive patients who underwent coronary artery bypass grafting in a single center were retrospectively collected. The relationship between the preoperative risk factors and the postoperative morbidity was analyzed by the Bayesian approach. Three risk indices (15-factor and seven-factor computed and seven-factor manual models) were developed for the prediction of morbidity. The criterion for morbidity was a prolonged hospital stay postoperatively (> 12 days) because of adverse events.

RESULTS

The best predictive preoperative factors for increased morbidity were emergency operation, diabetes, rhythm other than sinus rhythm on the electrocardiogram or recent myocardial infarction, low ejection fraction (< 0.49), age greater than 70 years, decreased renal function, chronic pulmonary disease, cerebrovascular disease, and obesity. The sensitivity of the scoring methods ranged from 51% to 72% and the specificity, from 77% to 86%.

CONCLUSIONS

The results show that individual patients can be stratified according to postoperative risk for complications on the basis of preoperative information that is available for most patients.

Risk stratification in cardiac surgery: is it worthwhile?

Outcome analysis of many surgical procedures has become increasingly important to surgeons, institutions, and the public. Because there may be wide differences in case mix, outcomes must be evaluated in light of the patient's preoperative status. All relevant preoperative conditions must be identified and weighted, so that when risk factor scores are combined in some fashion, they will provide a single preoperative risk estimate for the individual patient, representing the likelihood of dying as a consequence of the operation. Comparing the mean risk adjusted score of a group of patients undergoing the same procedure with the observed mortality rate for the same group yields an index of the quality of care, provided all preoperative risk scores are calculated with reference to the same benchmark. We question the logic and wisdom of surgical outcome analysis because of the infinitely complex nature of biological and pathological processes, as well as the practical problems of reliable data collection. The assumption of true scientific accuracy may be illusory. Even though risk adjusted outcome analysis has merit in studying trends in therapy, it should be regarded with caution when used as a tool for evaluating quality of care. If publicized at all, the results should not be represented as "hard" scientific fact.

1988: use of a Bayesian statistical model for risk assessment in coronary artery surgery. Updated in 1995

A computerized statistical model based on the theorem of Bayes was developed to predict mortality after coronary artery bypass grafting. From January, 1984, to April, 1987, at our hospital, 700 patients underwent isolated coronary artery bypass grafting. The presence or absence of 20 risk factors was determined for each patient. The first 300 patients formed the initial database of the Bayesian predictive model, and the remaining 400 patients were prospectively evaluated in four groups of 100 each. Each group was prospectively evaluated and then incorporated into the database to update the model. There was good agreement between predicted and observed results. Bayesian theory is particularly suited to this task because it (1) accommodates multiple risk factors, (2) is tailored to one's specific practice, (3) determines individual, rather than group, prognosis, and (4) can be updated with time to compensate for a changing patient population. These flexible attributes are especially valuable in light of recent changes in the coronary artery bypass graft patient profile.

Application of technetium-99m hexamethylpropylene amine oxime single-photon emission tomography to neurologic prognosis in patients undergoing urgent carotid surgery

In this study we aimed to work out a quantitative prognostic index for preoperative assessment of brain technetium-99m hexamethylpropylene amine oxime (HMPAO) single-photon emission tomography (SPET) in patients referred for urgent carotid endarterectomy due to acute obstructive disease of the internal carotid artery (ICA) and neurological deficit. To this end we compared data from preoperative SPET studies with the postinterventional changes in neurological status in 20 patients (17 males, three females; mean age 53 years, SD 4 years) with acute ischaemic cerebral disorders induced by obstruction of the ICA. Carotid obstruction was diagnosed by ultrasound B-mode study. All patients underwent urgent carotid endarterectomy from the ICA. Patients were divided into two groups in accordance with the results of postoperative follow-up: group A comprised patients with significant (more than 3 points) postoperative improvement in neurological condition as quantified by the Canadian Neurological Scale (11 patients); group B consisted of patients with minimal improvement or deterioration (nine, three of whom died). All patients were studied preoperatively by 99mTc-HMPAO SPET. The volume of nonperfused tissue (VS, cm3) was quantified using the Mountz technique. Hypoperfused volume (Vhypoperf, cm3) in the affected hemisphere was calculated as the total volume of voxels with 99mTc-HMPAO uptake < 90% of the contralateral symmetric voxels. Discriminant prognostic function was calculated by discriminant analysis as: PF = 0.072 x VS + 29.46x(VS/Vhypoperf). Patients with preoperative PF values < 8.20 demonstrated postoperative improvement in neurological status, while the group with PF > 8.90 comprised patients who demonstrated minimal improvement or deterioration. PF values in the range 8.20-8.90 carried an indefinite prognosis. We conclude that the preoperative 99mTc-HMPAO SPET can be used for the selection of patients in whom improvement in neurological status may be expected after urgent surgical correction of acute extracranial obstruction of the ICA.

Determinants of the length of stay in intensive care and in hospital after coronary artery surgery

BACKGROUND

Patients who have coronary artery surgery normally occupy intensive care beds for less than 24 hours. Longer stays may result in under use of cardiac surgical capacity. One approach to optimise surgical throughput is prospectively to identify fast track patients--that is, those who occupy an intensive care bed for less than 24 hours. A prospective audit of patients was performed to identify fast track patients by simple clinical criteria. Total length of hospital stay was also assessed in an attempt to predict which patients were likely to have a short postoperative stay, defined as < or = 7 days.

METHODS

Baseline demographic details, cardiovascular risk factors, angiographic and operative details were recorded for 431 consecutive patients who underwent coronary surgery at a regional centre over a nine month period. Outcome measures were the duration of the stay in the intensive care unit in hours and total duration of the postoperative stay in hospital in days. In addition, two groups of patients who were thought to be fast track were identified prospectively. Fast track 1 patients were identified by criteria selected by cardiovascular physicians. These were age less than 60 years, stable angina, good left ventricular function (ejection fraction > 50%), good renal function (serum creatinine < 120 mumol/l), and no obesity, diabetes, or other serious disease. Fast track 2 patients were identified by criteria defined by cardiovascular surgeons. These were male sex, age less than 65 years, good left ventricular function and no peripheral vascular disease, diabetes, or other serious disease. The efficacy of both sets of criteria in predicting outcome was tested.

RESULTS

344 (79.8%) patients were fast track. Significant factors for the prediction of fast track patients by univariate analysis (with positive predictive accuracy and sensitivity) were left ventricular ejection fraction > 50% (83%, 80%), left ventricular end diastolic pressure < 13 mm Hg (90%, 59%), creatinine less than 120 mumol/l (83%, 87%), and one or two vessel coronary disease (89%, 34%). Of the patients categorised as fast track 1 89% proved to be fast track (sensitivity 24%), however, the fast track 2 characteristics were not significant. Age, sex, obesity, diabetes, hypertension, a history of obstructive pulmonary disease and unstable angina were not predictive of the duration of intensive care stay. Multivariate analysis indicated that only left ventricular end diastolic pressure and the number of diseased coronary arteries predicted fast track patients. These criteria separated patients into three groups. Those who were good risk had one or two vessel disease and left ventricular end diastolic pressure < 13 mm Hg. They comprised 19% of the total and 93% of them were fast track. Those who were intermediate risk had either three vessel disease or left ventricular end diastolic pressure > 13 mm Hg but not both. They comprised 49% of the total and 85% of them were fast track. Those who were poor risk had both three vessel disease and left ventricular end diastolic pressure > 13 mm Hg. They comprised 32% of the total and 62% of them were fast track. The 106 (24%) patients who spent < or = 7 days in hospital after surgery were significantly younger (mean (SD) 55(8) v 58(8) years; P < 0.001) with a lower incidence of previous myocardial infarction (positive predictive accuracy 30%, sensitivity 53%), were less likely to have a history of obstructive pulmonary disease (25%, 98%), and more likely to have one or two vessel coronary disease (33%, 41%). They were more likely to have an internal mammary artery as a bypass conduit (27%, 89%) and more likely to need fewer than three distal anastomoses of the vein graft (29%, 63%). By multivariate analysis only age was significantly predictive of hospital stay. Total hospital stay could not be satisfactorily modelled on the basis of the criteria tested here. Sex, obesity, diabetes, hypertension, unstable angina, renal function, and left ventricular function were not associated with hospital stay. CONCLUSIONS-Most patients who had coronary artery surgery spent less than or equal to 24 hours in intensive care, but most spent > 7 days in hospital. The chance of a patient spending less than or equal to 24 hours in intensive care could be predicted by the number of coronary arteries diseased and the left ventricular end diastolic pressure. Poor risks patients (32%) had only a 62% chance of an intensive care unit stay of less than or equal to 24 hours. A policy of scheduling no more than one such patient for surgery per day would be simple to institute and would maximise the use of surgical capacity.

Criteria by which to evaluate risk-adjusted outcomes programs in cardiac surgery

Risk adjustment models for hospitalized patients are most advanced for the assessment of the clinical outcome of cardiac procedures, and for coronary artery bypass grafting in particular. The goal of being able to use outcomes as a credible indicator of quality of care has stimulated the development of several programs that use reliable, valid patient data collected during the surgical episode to adjust outcomes for the severity of illness. Several criteria that are useful in the assessment of risk adjustment methods for outcome and quality-of-care investigations are discussed in detail and five of these programs are compared. The programs have more similarities than differences and identify many of the same patient characteristics predictive of a higher likelihood of mortality in the period immediately after operation. Whether persistent differences in mortality after risk adjustment across institutions or individual surgeons, or both, may ultimately be attributed to the process and structure of care needs further study and investigation. Similar methods should be applied to other outcomes of importance to patients, their families, and their physicians, such as surgically related morbidity, functional status, quality of life, costs, and patient-reported perceptions of the nontechnical aspects of their care.

Assessment of predictive models for binary outcomes: an empirical approach using operative death from cardiac surgery

Predictive models in medical research have gained popularity among physicians as an important tool in medical decision making. Eight methodological strategies for creating predictive models are compared in a large, complex data base consisting of preoperative risk and operative outcome data on 12,712 patients undergoing coronary artery bypass grafting and entered into the Department of Veterans Affairs Cardiac Surgery Risk Assessment Program between April 1987 and March 1990. The models under consideration were developed to predict operative death (any death within 30 days following the surgical procedure or later if the result of a perioperative complication). The two strategies with the best predictive power among the eight examined were stepwise logistic regression alone and data reduction by cluster analysis combined with clinical judgement followed by a logistic regression model. The additive model based on unadjusted relative risks, the model based on Bayes' Theorem, and the logistic model using all candidate variables were good alternatives. Whether or not we imputed values did not have a significant impact on the predictive power of the models.

Bayesian-logit model for risk assessment in coronary artery bypass grafting

Predictive models for the assessment of operative risk using patient risk factors have gained popularity in the medical community as an important tool for the adjustment of surgical outcome. The Bayes' theorem model is among the various models used to predict mortality among patients undergoing coronary artery bypass grafting procedures. Comparative studies of the various classic statistical techniques, such as logistic regression, cluster of variables followed by a logistic regression, a subjectively created sickness score, classification trees model, and the Bayes' theorem model, have shown that the Bayes' model is among those with the highest predictive power. In this study, the Bayes' theorem model is reformulated as a logistic equation and extended to include qualitative and quantitative risk factors. We show that the resulting model, the Bayesian-logit model, is a mixture of logistic regression and linear discriminant analysis. This new model can be created easily without complex computer programs. Using 12,712 patients undergoing coronary artery bypass grafting procedures at the Department of Veterans Affairs Continuous Improvement in Cardiac Surgery Study between April 1987 and March 1990, the predictive power of the Bayesian-logit model is compared with the Bayes' theorem model, logistic regression, and discriminant analysis. The ability of the Bayesian-logit model to discriminate between operative deaths and operative survivors is comparable with that of logistic regression and discriminant analysis.

Continuous assessment and improvement in quality of care. A model from the Department of Veterans Affairs Cardiac Surgery

OBJECTIVE

The authors organized the Department of Veterans Affairs (VA) Continuous Improvement in Cardiac Surgery Study (CICSS) to provide risk-adjusted outcome data for the continuous assessment and improvement of quality of care for all patients undergoing cardiac surgery in the VA.

BACKGROUND

The use of risk-adjusted outcomes to monitor quality of health care has the potential advantage over consensus-derived standards of being free of preconceived biases about how health care should be provided. Monitoring outcomes of all health care episodes, as opposed to review of selected cases (e.g., adverse outcomes), has the advantages of greater statistical power, the opportunity to compare processes of care between good and bad outcomes, and the positive psychology of treating all providers equally. These two concepts, together with a pre-existing peer committee (the VA Cardiac Surgery Consultants Committee) to review, interpret, and act on the risk-adjusted outcome data, form the primary design considerations for CICSS.

METHODS

Patient-level risk and outcome (operative mortality and morbidity) data are collected prospectively on each of the approximately 7000 patients undergoing cardiac surgery in the VA each year. These outcomes, adjusted for patient risk using logistic regression, are provided every 6 months to each cardiac surgery program and to a national peer review committee for internal and external quality assessment and improvement.

RESULTS

For the most recent 12-month period with complete data collection, observed-to-expected (O/E) ratios ranged from 0.2 to 2.2, with eight centers falling outside of the 90% confidence limits for an O/E ratio equaling 1.0. The O/E ratio for all centers has fallen by 14% over the 4.5-year period of this program (p = 0.06).

CONCLUSIONS

A large-scale, low-cost program of continuous quality improvement using risk-adjusted outcome is feasible. This program has been associated with a decrease in risk-adjusted operative mortality.

The Society of Thoracic Surgeons National Database status report

This report describes the development of the first known national surgical database designed for the practicing community cardiothoracic surgeon. Acceptance by members of The Society of Thoracic Surgeons has been gratifying. The number of patients on the system has grown from 116,109 at the end of 1991 to an anticipated 350,000 to 450,000 by the end of 1993. At the time of this report, 842 surgeons were participating, and more than 1,200 will be on the system by the end of 1993. A risk stratification system has been incorporated into the software, which predicts each patient's risk based on the individual surgeon's past experience. Trend analyses demonstrate a substantial increase in the number of patients at increased risk for perioperative death for coronary artery bypass operations over the past 5 years, while observed mortality has remained relatively constant. Programs are available for adult and congenital heart disease, lung cancer, and esophageal cancer, and modules for mediastinal tumors, pleural disorders, and benign pulmonary disease will soon be added. We anticipate that growth will continue as the need for practice profile data increases because of reimbursement issues.

A simple classification of the risk in cardiac surgery: the first decade

Since 1980, the operative risk in all our cardiac surgical patients has been assessed before surgery. In light of reports of changes in cardiac surgical populations, we reexamined our practice and risk classification. The purpose of this study was to compare the surgery performed, the characteristics of the patients operated upon and the hospital mortality in our institution in two epochs ten years apart. In 1989-90, the 2029 consecutive cardiac surgical patients who had the same operations as the 500 patients of a 1980 study in our institution were prospectively stratified using our risk classification based on the number of risk factors (RFs) present: normal-risk patient = no RF, increased risk = 1 RF, high risk > or = 2 RFs. These two cohorts of patients were compared. From 1980 to 1990, the proportion of high-risk patients tripled whereas the proportion of normal-risk patients diminished by one third and the proportion of increased risk remained unchanged. The incidence of the following RFs increased: poor left ventricular function, advanced age, emergency surgery, reoperation and other systemic disorders. In coronary artery surgery patients, the incidence of unstable angina/recent myocardial infarction and of obesity also increased. In noncoronary artery surgery patients, the incidence of heart failure increased while obesity remained unchanged. The difference in hospital mortality among the three risk classes was significant within both study periods. The mortality in each risk class and total mortality did not change between 1980 and 1990.(ABSTRACT TRUNCATED AT 250 WORDS)

Morbidity and duration of ICU stay after cardiac surgery. A model for preoperative risk assessment

Although risk factors for mortality after cardiac surgery have been identified, there is no widely applicable method for readily determining risk of postoperative morbidity based on preoperative severity of illness. The goal of this study was to develop a model for stratifying the risk of serious morbidity after adult cardiac surgery using readily available and objective clinical data. After univariate analysis of risk factors in 3,156 operations, 11 variables were identified as important predictors by logistic regression (LR) analysis and used to construct an additive model to calculate the probability of serious morbidity. Reliable correlation was found between a simplified additive model for clinical use and the LR model. The clinical and logistic models were then tested prospectively in 394 patients and demonstrated a pattern of increasing morbidity with ascending scores similar to that predicted by the reference group. Increasing clinical risk score was also associated with a greater frequency of individual complications as well as prolongation of ICU stay. This study demonstrates that it is feasible to design a simple method to stratify the risk of serious morbidity after adult cardiac surgery. With further prospective multicenter refinement and testing, such a model is likely to be useful for adjusting severity of illness when reporting outcome statistics as well as planning resource utilization.

Multivariate prediction of in-hospital mortality associated with coronary artery bypass graft surgery. Northern New England Cardiovascular Disease Study Group

BACKGROUND

A prospective regional study was conducted to identify factors associated with in-hospital mortality among patients undergoing isolated coronary artery bypass graft surgery (CABG). A prediction rule was developed and validated based on the data collected.

METHODS AND RESULTS

Data from 3,055 patients were collected from five clinical centers between July 1, 1987, and April 15, 1989. Logistic regression analysis was used to predict the risk of in-hospital mortality. A prediction rule was developed on a training set of data and validated on an independent test set. The metric used to assess the performance of the prediction rule was the area under the relative operating characteristic (ROC) curve. Variables used to construct the regression model of in-hospital mortality included age, sex, body surface area, presence of comorbid disease, history of CABG, left ventricular end-diastolic pressure, ejection fraction score, and priority of surgery. The model significantly predicted the occurrence of in-hospital mortality. The area under the ROC curve obtained from the training set of data was 0.74 (perfect, 1.0). The prediction rule performed well when used on a test set of data (area, 0.76). The correlation between observed and expected numbers of deaths was 0.99.

CONCLUSIONS

The prediction rule described in this report was developed using regional data, uses only eight variables, has good performance characteristics, and is easily available to clinicians with access to a microcomputer or programmable calculator. This validated multivariate prediction rule would be useful both to calculate the risk of mortality for an individual patient and to contrast observed and expected mortality rates for an institution or a particular clinician.

Risk assessment in urgent/emergent coronary artery surgery

A statistical model has been developed to allow for prediction of individual patient prognosis following urgent/emergent coronary artery bypass grafting (CABG). None of the models previously described for use in coronary artery surgery has been tested with a prospective patient series that confirms the true predictive capacity of the model. Ideally, the predictive ability of such models should be validated with prospective trials. To examine the feasibility of statistical modeling in this clinical context, a computerized model based on the theorem of Bayes was developed to predict operative mortality for urgent coronary artery surgery. The presence or absence of 20 risk factors was determined for each of 405 consecutive patients undergoing urgent coronary artery surgery from January 1984 to January 1989. The first 100 patients were used to develop a database for the model, which was then used to prospectively evaluate the remaining 305 patients. There was good agreement between predicted and observed results. Models of this kind are particularly advantageous because of the ability to (1) accommodate multiple risk factors, (2) become tailored to a specific practice, and (3) determine individual rather than group prognosis. Validation with a prospective trial confirms the practical utility of this approach. This model has reliably predicted the risk associated with urgent coronary artery surgery and may provide important clinical information for the management of patients being evaluated for urgent revascularization.

A quality assurance model of operative mortality in coronary artery surgery

Quality assurance in coronary artery bypass grafting (CABG) surgery requires a comparison of operative mortality against an accepted standard of care. Raw mortality statistics are unacceptable in this context, and risk factor analysis is essential. However, this principle has not been adequately demonstrated in previous reports. Our goal in this study was to develop a risk model of accepted CABG mortality and illustrate its proper use in coronary artery surgery. The model was derived from a Bayesian analysis of 6,630 patients undergoing CABG in the Coronary Artery Surgery Study (CASS) registry. Age, sex, ventricular function, previous myocardial infarction, extent of coronary artery disease, unstable angina, and surgical priority were used by the model to sort patients into risk categories. From January 1984 through December 1987, 840 patients underwent isolated CABG at our hospital. With raw mortality data, the 3.9% (33/840) mortality of our patients was significantly different from the 2.3% (153/6,630) CASS mortality (p less than 0.001). When our patients were entered into the CASS model for risk stratification, however, our CABG mortality conformed to the CASS experience. These results illustrate the fallacy of using raw mortality statistics for interinstitutional comparisons. This type of risk model is a fundamental element of CABG quality assurance.