| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Ann Thorac Surg 2002;74:464-473
© 2002 The Society of Thoracic Surgeons
a The Outcomes Research and Assessment Group, The Duke Clinical Research Institute, Durham, North Carolina, USA
b Louisiana State University Health Sciences Center, New Orleans, Louisiana, USA
c Departments of Surgery and Medicine, University of Colorado Health Sciences Center, Denver, Colorado, USA
d University of Florida Health Science Center, Jacksonville, Florida, USA
Accepted for publication April 16, 2002.
* Address reprint requests to Dr Peterson, Associate Professor of Medicine, Duke University Medical Center, Box 3236, Durham, NC, USA27710
e-mail: peter016{at}mc.duke.edu
 |
Abstract |
|---|
 Top Abstract IntroductionMaterial and methodsResultsCommentReferences |
|---|
Methods. We analyzed 496,797 isolated CABG procedures performed between January 1997 to January 2001 at 587 US hospitals participating in the Society of Thoracic Surgeon’s National Cardiac Database. Logistic and linear regression were used to identify independent preoperative factors affecting a patient’s likelihood for early discharge (PLOS 
5 day), prolonged stay (>14 days), and overall PLOS. Hierarchical models were used to determine the degree to which hospital factors influenced PLOS beyond patient factors.
Results. Overall, 53% of CABG patients were discharged within 5 days of CABG, whereas 5% required prolonged (>14 days) stays. More than 25 preoperative patient factors were independently associated with a patients’ likelihood for early discharge and prolonged stay (model C index 0.70 and 0.75, respectively). After adjusting for patient factors, however, there remained wide unexplained variability among hospitals in PLOS and limited correlation between these PLOS metrics and hospitals’ risk-adjusted mortality results (Spearman correlation coefficient -0.15 and 0.35).
Conclusions. Our study provides a method for institutions to receive meaningful risk-adjusted bypass PLOS information. Given the marked variability among hospitals in CABG PLOS, institutions should consider benchmarking metrics of efficiency, as well as patient outcomes.
|
Introduction |
|---|
|
TopAbstractIntroductionMaterial and methodsResultsCommentReferences |
|---|
When benchmarking either clinical outcomes or resource metrics among providers, it is important to consider the clinical characteristics of the patients receiving the procedure [1, 2]. For example, institutions whose surgical population consisted of a high percentage of elderly patients and those with multiple comorbid conditions, might be expected to have fewer patients discharged early and more with prolonged stays. Multiple statistical models exist that account for differences in patient case-mix and risk for bypass mortality outcomes. In contrast, few applications of risk adjustment techniques have been published for benchmarking institutional performance regarding PLOS.
Using the Society of Thoracic Surgeons National Cardiac Database, we conducted contemporary analyses of bypass surgery PLOS among a national sample of US hospitals. Our goals were to first display the distribution of length of stay among patients undergoing isolated CABG. Second, we investigated the degree to which PLOS is affected by certain preoperative patient clinical factors, developing and validating models that estimated patients likelihood for "early discharge" (within five days), "prolonged postoperative stay" (>14 days), and the continuous measure of PLOS. Third, we compared the degree to which hospitals varied in PLOS after adjusting for patient case-mix. Finally, we examined the correlation among hospitals’ risk-adjusted provider resource and mortality metrics.
|
Material and methods |
|---|
|
TopAbstractIntroductionMaterial and methodsResultsCommentReferences |
|---|
Patient population
Our analysis population consists of patients who underwent isolated CABG surgery from January 1, 1997 to December 31, 1999 at centers contributing data to the STS database. We excluded those undergoing valvular surgery (repair or replacement) or other combined surgical procedures (e.g., ventricular septal repair or carotid endarterectomy). We also excluded those with missing age, gender, date of surgery, or discharge data (<2% of the total). The final analysis database contains clinical information on 496,797 patients from 587 hospitals. These hospitals come from 48 US states (excluding Vermont and Maine) and represent approximately 60% of all US centers performing bypass. This large sample was then randomly divided at the patient level into a model development set (80%, n = 397,993) and a validation set (20%, n = 98,804).
Statistical analysis
We defined PLOS as the number of days from the date of surgery to the date of hospital discharge. The distribution of PLOS for all patients is displayed in Figure 1.
Patients who survived are indicated by filled boxes, whereas the deaths are denoted by empty boxes. Although PLOS is a continuous function, we also defined two clinically meaningful binary end-points: "Early Discharge" was defined as discharged or death in five days or less following CABG, a metric representing care patterns for the routine bypass patient. "Prolonged Stay" was defined as patients discharged or dying more than 14 days after CABG, representing a metric of complicated postoperative cases. Baseline demographics are displayed by these subgroups in Table 1.
We also displayed the frequency of PLOS as a function of important patient subsets and among those with procedural complications in Tables 2 and 3.
|
|
|
|
Three separate multivariable models were developed. These included two logistic regression models, estimating each patient’s likelihood for Early Discharge (5 days vs not) and also estimating each patient’s likelihood of having a Prolonged Stay (>14 days vs not). Linear regression was used to examine patients’ log transformed PLOS as a continuous function. In each of these three models, we considered all variables with significant univariable relationship to PLOS (p < 0.05). We then used multivariable regression (stepwise) to develop and refine our final models. Model entry and exit criteria were both set at p less than 0.05. In our modeling, we also determined whether the relationship between base line continuous predictors of PLOS (eg, age, body surface area, and ejection fraction) were linear or required transformation for best fit. Finally, we tested the incremental value of clinically plausible interaction terms. To the extent possible, we used similar risk factors and transformation techniques among the three models.
These models were first developed in the overall CABG analysis development dataset, (including those who died during hospitalization). The models were then rerun on those patients surviving to hospital discharge. As the risk factors remained the same and the coefficients of the risk factors changed only marginally for three variables (status, shock, and prior surgery) in the survivor-only models and all-patient models, only the results of the latter analysis are presented. Independent clinical predictors of Early and Prolonged PLOS were displayed as Odds Ratios with 95% confidence intervals. For the continuous variable PLOS model, because of the log transform, the estimated parameter for a risk represents a multiplicative effect. The "multiplier" factor is provided, representing the degree to which a clinical factor increases or decreases PLOS relative to those patients without this risk factor present.
A number of standard criteria were used to examine model performance. The c-index, or area under the receiver operator curve (ROC), reflects a model’s ability to correctly discriminate patients who will have an event (i.e., either Early Discharge or Prolonged Stay) from those who did not. Model C index generally ranges from 0.5 (no better than chance) to 1.0 (perfect discrimination). Model calibration reflects the degree to which predicted outcomes from the model match those actually observed among various patient risk groups. Specifically, event probabilities were calculated for each patient, based on the model. Patients were then separated into ten equal sized risk strata. Within each stratum, the average predicted likelihood for Early Discharge (or Prolonged Stay) was compared with that actually observed. For the continuous PLOS model, we assessed the proportion of patient variance in PLOS that was explained by preoperative patient factors in the model (R2). All model performance characteristics were assessed both in the development population and the validation population.
To estimate and test the influence of hospital, we used a hierarchical (mixed effects) logistic regression model, where hospital effects were considered random while patient risk factors were considered fixed. By treating hospitals as random, we assume that the providers are a sample from a larger population of providers. In this analysis, we are assuming that the hospital effects follow a normal distribution on the logit scale. Once the hospital effect estimates were obtained, they were exponentiated to obtain estimated odds of Early Discharge or Prolonged Stay. Estimated odds of operative mortality were similarly obtained using the 2000 STS mortality risk model (http://www.sts.org).
To compare the relative influence of hospital—versus patient—characteristics on the three PLOS metrics, we estimated the ratio of the reduction in variation using patient risk factors only to that obtained from the mixed model that included hospital effects and then subtracted this value from one. Minus twice the maximized log likelihood was used to determine reduction in variation. Finally, we compared the correlation between hospitals’ CABG PLOS metrics and their in-hospital risk-adjusted mortality outcomes. This analysis was performed among 548 of the hospitals performing 50 or more CABG procedures per year (representing 99% of our overall sample). Specifically, our patient outcome metric was CABG operative mortality, defined as in-hospital or 30-day mortality (whichever was longest). Correlation among hospital effects on Early Discharge, Prolonged Stay, and Mortality were compared both graphically and using Spearman’s correlation coefficient. SAS version 8.2 (SAS Institute, Cary, NC) was used for all analyses.
|
Results |
|---|
|
TopAbstractIntroductionMaterial and methodsResultsCommentReferences |
|---|
Effect of clinical factors on PLOS
Table 1 provides clinical characteristics for the 397,993 patients included in the model development population, stratified according to whether patients’ PLOS was (5 days), 6 to 14 days, or more than 14 days. The overall mean age of our population was 64.9 years, 29.1% were female and, 10.1% were nonwhite. Relative to those with longer PLOS, patients discharged early (5 days) tended to be younger, male, had less acute presentations, and lacked comorbid illness.
Table 2 provides PLOS distributions for selected important CABG subgroups. Patients aged more than 75 years had significantly longer mean PLOS than did younger patients. They were half as likely to be discharged within five days of surgery and three times more likely to have prolonged hospital stays more than 14 days compared with those aged less than 65 years. Surgical procedure performed under emergent or salvage conditions had longer mean stays and a much higher percentage of patients with Prolonged Stays relative to elective cases. As a composite measure, PLOS was significantly associated with increasing preoperative mortality risk groups. For example, low risk patients (predicted mortality < 1%) had a mean PLOS of 5.3 days and more than 70% of low risk patients were discharged in 5 days or less. On the opposite end of the spectrum, those with high preoperative mortality risk ( > 3 to 99%) had average longer PLOS (mean 8.9 days) with 10.5% staying 14 days or more.
Table 3 provides PLOS distributions for patients with postoperative complications. Patients with deep sternal wound infection had the longest mean and median PLOS, with more than half staying more than 14 days. This was followed by prolonged ventilation, stroke, and renal failure which all increased mean PLOS by 10 days or more relative to uncomplicated CABG cases. Of the total 26,008 CABG patients with prolonged PLOS more than 14 days, 72% had at least one major morbid or mortal complication [death (14%), deep sternal wound infection (6%), stroke (12%), renal failure (25%), prolonged ventilation (50%), or reoperation (31%)] and 94% were coded as having one or more major or minor complication.
Table 4
provides the multivariable predictors of patients’ likelihood for Early Discharge (5 days), Prolonged Stays ( > 14 days), and the continuous variable, PLOS. The first two columns provide the multivarible odds ratio (OR) and 95% confidence interval for the OR. In a similar manner, the final column provides the independent impact of specific risk factors to prolong PLOS relative to those without this condition. For example, after accounting for other clinical factors, those with chronic lung disease are nearly 33% less likely to be discharged early, 70% more likely to have a Prolonged Stay, and stay 1.11 times longer than those without lung disease.
|
|
The influence of the within center, or "provider-effect," was quantified in our multivariable analysis. While the addition of a hospital indicator (i.e., where the procedure was performed) to the models had limited impact on the relationship between patient level risk factors and PLOS (i.e., their adjusted OR’s remained constant), it did increase the models’ predictive power. Adding the hospital indicator increased the Early Discharge model’s c-index from 0.70 to 0.75 and the Prolonged Stay model’s c-index from 0.74 to 0.77. In fact, knowing the hospital where the surgery was performed was the single most important factor in predicting all the PLOS metrics. This single provider-level factor accounted for approximately 40% of the explained variation in Early Discharge, 27% of the explained variation in Prolonged Stay, and 36% of the variation in the continuous variable, PLOS.
Correlation of hospital risk-adjusted PLOS and risk-adjusted outcome
Comparison of hospitals’ CABG resource use should ideally be done in conjunction with their procedural outcomes. The correlations between hospitals’ PLOS metrics with risk-adjusted mortality are displayed in Figures 3 and 4.
In these plots, a "provider" effect more than 1.0 indicates institutions whose patients had a higher than average odds for a given outcome (eg, Early Discharge) after accounting for other preoperative patient risk factors. Conversely, a provider effect of less than 1.0 indicates centers that were less likely than average to have the outcome after accounting for other clinical factors. Hospitals that were more likely to have patients with Prolonged Stays tended to have higher risk-adjusted operative mortality, Spearman correlation 0.35 (Fig 3). In contrast, there was almost no association between hospitals’ likelihood for Early Discharge and the institution’s risk-adjusted operative mortality, Spearman correlation -0.15 (Fig 4).
|
|
|
Comment |
|---|
|
TopAbstractIntroductionMaterial and methodsResultsCommentReferences |
|---|
The Society of Thoracic Surgeons National Cardiac Database provides an ideal setting to examine patterns of bypass surgery resource use. It represents a majority of US bypass centers including academic, private, public, military, and Veterans Affairs institutions. In this contemporary study of more than 500 hospitals, we found that discharges within 5 days of surgery has become the norm for US CABG patients. Even among the elderly, nearly one half of CABG patients aged 65 to 70 and one third third of those aged more than 75 years were discharged within 5 days postoperatively (Table 3). At the other end of the spectrum, our study found that nearly all patients with prolonged stays (PLOS >14 days) had suffered at least one postoperative complication.
Prolonged PLOS, therefore, represents a composite indicator of adverse events following CABG. In this regard, prolonged PLOS provides complementary data to that gained from other patient outcomes assessments such as procedural complication rates (ie, stroke, renal insufficiency, infection rates). Prolonged PLOS is objective and easy to measure, yet it is nonspecific. Provider comparisons of complication rates are specific and actionable, yet may be subjectively defined and dependent on an institution’s diligence in detection, making cross-institutional comparisons challenging.
Prior studies
Our study identified the major preoperative patient characteristics that affected one’s likelihood for Early Discharge, Prolonged Stay, as well as their overall estimated PLOS (Table 4). The major factors affecting resource use include: patient age, gender, prior surgery, disease severity, and comorbid illness were similar to those identified by prior investigators [2–4, 12–20]. For example, in an early single-institution study (1981 to 1989) found five major preoperative determinants of PLOS (ie, age, gender, angina class, ejection fraction, and emergency surgery) [4]. Another multicenter study of PLOS in 1992 to 1993 identified eight factors related to PLOS (ie, age, gender, prior surgery, lung disease, stroke, mitral valve disease, renal function, and preoperative intraaortic balloon pump) [12]. Similar risk factors also predicted those patients with prolonged hospital stays after CABG in two other institution studies [19, 20].
Our current analysis adds to this prior research in a number of ways. First, mean hospital stays after CABG have declined by up to 4 days since these earlier studies were published. Thus, our results are more reflective of contemporary patterns of care. Second, our larger sample of hospitals and patient cohort permitted the identification of a broader array of clinical risk factors (more than 25 clinical factors). This size and inclusiveness of our models was reflected in better model predictive performance than had been reported in prior models. For example, the ability to correctly discriminate those with prolonged stay from those without was higher in our study than published previously (C-index 0.75 vs 0.68, respectively) [20]. Similarly, the overall degree of variability explained by our PLOS model was also higher than that found in prior studies (R2 of 0.14 vs 0.04) [12].
Hospital variability in PLOS
Among this cross-sectional analysis of national bypass surgery, our study found significant hospital variability in patterns of PLOS. The mean PLOS varied by up to 10 days among centers. Much of this variability was due to differences in handling of the uncomplicated cases which is evident from the wide range of Early Discharge rates across centers (IQR 43% to 62%). Although most institutions appeared to have embraced strategies to achieve early discharge for the routine case, some hospitals still keep the majority of their CABG patients more than 5 days after surgery. This variability in achieving Early Discharge is not fully accounted for by patients’ preoperative risk factors. In fact, the strongest predictor of whether a patient would be discharged in 5 days or less was the hospital where the surgery was performed, accounting for 40% of the overall explained variation in Early Discharge. Although there was less absolute differences in the percentage of patients with Prolonged Stays, the relative variability among hospitals persisting after adjusting for patient case-mix remained high. The hospital at which the surgery was performed accounted for 27% of the total explainable variation in Prolonged Stay, again the strongest predictor.
It is also noteworthy that we found limited correlation between hospital performance with regards to PLOS and risk-adjusted operative mortality (Figures 3 and 4). For example, among centers with very low mortality risk (<1.0), there was wide diversity in their ability to achieve discharge within 5 days or less (Fig 3). This point emphasizes the important need for institutions to benchmark both resource and outcomes metrics to better understand how to deliver both efficient and high quality of care.
Application of risk models
Our study provides separate multivariable risk models for estimating patients’ likelihood for Early Discharge, Prolonged Stay, and their total number of postoperative stay (in days) following CABG. This set of models has multiple potential applications. When used in patient aggregates (eg, hospital-level), these models can permit meaningful comparison of "risk-adjusted" PLOS metrics among peers. Specifically, these models can limit the potential confounding of provider comparisons due to differences in the type of patients treated. Although our study found that the influence of preoperative clinical factors on PLOS was modest, removal of these biases will provide more accurate provider comparison and increase clinician acceptance.
The provision of risk-adjusted PLOS comparison data can be an important step to encourage caregivers to consider whether their postoperative care processes are optimal. The Institute of Medicine report defines quality of care as including the concepts of timeliness and efficiency of care in addition to those processes that lead to better patient outcomes [21]. By this same definition, PLOS results need to be considered in context with patient outcomes and downstream consequences. Achieving shorter PLOS at the expense of increased operative mortality or rates of rehospitalization following hospital discharge would not be considered ideal [22].
Use of these models to predict individual patient hospital stays also comes with certain caveats. Caregivers may be benefited to have a means of estimating patients’ likelihood for long postoperative stays to assure appropriate staff and bed availability. However, the ability to accurately predict PLOS for an individual patient (without accounting for hospital-specific effects), is slight. Preoperative factors were not the major forces determining PLOS requirements when compared against hospital- and surgeon-specific factors and chance events. Likewise, the extent of variability among providers in PLOS, estimates based on national STS results may not compare well with those seen for a given patient cared for at a specific institution. Finally, the decision whether or not to perform surgery should rest mainly on the procedure’s potential risks versus its long-term benefits, and not on its resource requirements.
Study limitations
We would like to acknowledge limitations for our study. First, we chose definitions for Early Discharge and Prolonged Stay that reflect current practice norms, but note that these cut-points are to some degree are arbitrary. Second, we chose to examine only PLOS, rather than total length of stay for CABG. This was due to the high degree of local practice factors (ie, catheterization facilities, patient travel factors, surgery capacity, etc) affecting the preoperative hospital phase. Similarly, our national database does not have detailed hospital cost information. While length of in-patient stay is a major predictor of total hospital costs, this relationship is dependent on the type of care provided (eg, ICU vs monitored bed vs general ward) and can vary markedly among centers and geographic region [2]. Finally, and perhaps most importantly, our study was not designed to understand or prescribe the "ideal length of stay" for patients. Specifically, we did not examine the consequences of shorter- or longer-length of stay on patient outcomes including the need for readmission following discharge, patient functional outcomes, or effects on patient and family satisfaction with care. This point was underscored by the fact that we found only slight correlation between risk-adjusted hospital resource metrics and patient risk-adjusted mortality outcomes.
Conclusions
The current study provides contemporary national benchmark information on hospital length of stay following isolated bypass surgery. It also adds to our armamentarium of risk-adjustment tools for bypass surgery. They allow for clinicians to compare how their postoperative length of stay outcomes compare with those from other institutions. This information can be useful both for comparing their relative efficiency in achieving rapid discharge in routine CABG cases, as well as for signaling potential quality issues if prolonged PLOS rates are higher than expected.
|
References |
|---|
|
TopAbstractIntroductionMaterial and methodsResultsCommentReferences |
|---|
This article has been cited by other articles:
|
|
 |
|
  L. Dos, A. Dadashev, D. Tanous, I. J. Ferreira-Gonzalez, K. Haberer, S. C. Siu, G. S. Van Arsdell, E. N. Oechslin, W. G. Williams, and C. K. Silversides Pulmonary valve replacement in repaired tetralogy of Fallot: Determinants of early postoperative adverse outcomes J. Thorac. Cardiovasc. Surg., September 1, 2009; 138(3): 553 - 559. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 A. A. Kramer and J. E. Zimmerman Predicting Outcomes for Cardiac Surgery Patients After Intensive Care Unit Admission Seminars in Cardiothoracic and Vascular Anesthesia, September 1, 2008; 12(3): 175 - 183. [Abstract] [PDF]
|
|
|
|
|
|
E. Ott, C. D. Mazer, I. C. Tudor, L. Shore-Lesserson, S. A. Snyder-Ramos, B. A. Finegan, P. Mohnle, C. B. Hantler, B. W. Bottiger, R. D. Latimer, et al. Coronary artery bypass graft surgery care globalization: The impact of national care on fatal and nonfatal outcome J. Thorac. Cardiovasc. Surg., May 1, 2007; 133(5): 1242 - 1251. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 L. Abrahamyan, A. Demirchyan, M. E Thompson, and H. Hovaguimian Determinants of Morbidity and Intensive Care Unit Stay after Coronary Surgery Asian Cardiovasc Thorac Ann, April 1, 2006; 14(2): 114 - 118. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
Determinants of operative mortality in valvular heart surgery. J. Thorac. Cardiovasc. Surg., March 1, 2006; 131(3): 547 - 557.
|
|
|
|
 |
|
 D. M. Shahian, E. H. Blackstone, F. H. Edwards, F. L. Grover, G. L. Grunkemeier, D. C. Naftel, S. A.M. Nashef, W. C. Nugent, and E. D. Peterson Cardiac Surgery Risk Models: A Position Article Ann. Thorac. Surg., November 1, 2004; 78(5): 1868 - 1877. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
C. Weissman Pulmonary Complications After Cardiac Surgery Seminars in Cardiothoracic and Vascular Anesthesia, September 1, 2004; 8(3): 185 - 211. [Abstract] [PDF]
|
|
|
|
 |
|
 T. B. Ferguson Jr, E. D. Peterson, L. P. Coombs, M. C. Eiken, M. L. Carey, F. L. Grover, and E. R. DeLong Use of Continuous Quality Improvement to Increase Use of Process Measures in Patients Undergoing Coronary Artery Bypass Graft Surgery: A Randomized Controlled Trial JAMA, July 2, 2003; 290(1): 49 - 56. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
R. M. Mentzer Jr, R. D. Lasley, A. Jessel, and M. Karmazyn Intracellular sodium hydrogen exchange inhibition and clinical myocardial protection Ann. Thorac. Surg., February 1, 2003; 75(2): S700 - 708. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
| ANN THORAC SURG | ASIAN CARDIOVASC THORAC ANN | EUR J CARDIOTHORAC SURG |
| J THORAC CARDIOVASC SURG | ICVTS | ALL CTSNet JOURNALS |
