| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Ann Thorac Surg 2007;84:829-835
© 2007 The Society of Thoracic Surgeons
a Division of Cardiac Surgery, Johns Hopkins Hospital, Baltimore, Maryland
b Department of Surgery, Johns Hopkins Hospital, Baltimore, Maryland
Accepted for publication April 23, 2007.
* Address correspondence to Dr Yuh, Division of Cardiac Surgery, Johns Hopkins Hospital, 600 N Wolfe St, Blalock 618, Baltimore, MD 21287-4618 (Email: dyuh{at}csurg.jhmi.jhu.edu).
Presented at the Forty-third Annual Meeting of The Society of Thoracic Surgeons, San Diego, CA, Jan 29–31, 2007.
 |
Abstract |
|---|
 Top Abstract IntroductionMaterial and MethodsResultsCommentDiscussionAcknowledgmentsReferences |
|---|
Methods: Logistic regression analysis of the California hospital discharge database during a 5-year period was performed to identify the most prevalent preoperative International Classification of Disease, 9th Revision Clinical Modification (ICD-9-CM) diagnoses associated with SNF admission after primary CABG in patients aged 65 years or older. Each diagnosis was weighted according to odds ratios to develop an index that predicts the likelihood of discharge to a SNF. The index was validated using our institutional database.
Results: A total of 26,040 patients (mean age, 74.2 years; 67.2% men) fit our criteria. They had an in-hospital mortality rate of 3.09% and a 17.3% SNF discharge rate. Our index was a summation of nine selected preoperative ICD-9-CM diagnoses, which were assigned a value of 1 point (osteoarthritis, congestive heart failure, atrial fibrillation, myocardial infarction, anemia, obesity) or 2 points (female, chronic obstructive pulmonary disease, renal failure). Validation analysis produced a C statistic and pseudo r 2 value of 0.6435 and 0.0408, respectively. Cut-point analysis suggests that patients with scores of 3 or higher can be considered "high-risk."
Conclusions: We describe a simple index to identify older patients at low-risk and high-risk for SNF admission after CABG. Such tools may be useful in counseling older patients considering CABG.
|
Introduction |
|---|
|
TopAbstractIntroductionMaterial and MethodsResultsCommentDiscussionAcknowledgmentsReferences |
|---|
In a survey of 226 older individuals with illnesses potentially treatable with interventional therapies, Fried and colleagues [10] concluded that functional and cognitive outcomes of a given therapy weighed more heavily than mortality rates in the patient’s decision to undergo treatment. These priorities underscore the importance of accurate preoperative counseling for prospective older CABG patients and their families. The purpose of this study was (1) to identify which preoperative risk factors were associated with admission to a SNF and (2) to develop a predictive index from these data to help clinicians counsel older patients considering CABG.
|
Material and Methods |
|---|
|
TopAbstractIntroductionMaterial and MethodsResultsCommentDiscussionAcknowledgmentsReferences |
|---|
Inclusion criteria were patients aged 65 years or older who had International Classification of Diseases, 9th Revision Clinical Modification (ICD-9-CM) procedure code 36.1 (bypass anastomosis for heart revascularization) in the first (primary coding) position. To focus on isolated CABG cases, patients with procedure code 35.xx (all types of valve repairs) were excluded. To focus on patients who were functionally independent on admission, patients who were admitted from a SNF, residential care facility, or other hospital setting were excluded.
The primary outcome variable was disposition to a SNF, coded as skilled nursing or intermediate care, either within the same hospital, to different hospital, or to a residential care facility.
The analysis began with a tabulation of all diagnosis codes versus the outcome of interest, after eliminating any diagnosis code for a patient that was not present on admission. We then focused only on diagnosis codes that had a frequency exceeding 5% in this patient population; this elimination allowed us to focus on conditions that would occur relatively frequently in order for our tool to be clinically useful. We then grouped some of the codes together that represented clinically similar conditions; specifically, the various myocardial infarction (MI) codes (410.11, acute anterior wall MI; 410.41, acute inferior wall MI; and 410.71, subendocardial infarction), the lipid disorder codes (272.0, pure hypercholesterolemia; and 272.4, hyperlipidemia not elsewhere classified), and tobacco use codes (305.1, tobacco use disorder; and V15.82, history of tobacco use).
The presence and absence of each of these conditions were then evaluated by multiple logistic regression modeling to identify possible associations with our outcome of interest. We did not include patient age in the regression analysis because we found age to be significantly correlated with all of the comorbidities that we examined. Including patient age and the identified comorbidities in the same analysis would violate the independence principle for multivariate analysis.
The multiple logistic regressions allowed us to then develop a scoring index based on previously described methodology [11] by taking the odds ratios for each of the variable and using them as weighting factors in an additive formula. For example, if condition X had an odds ratio of 2.5 and condition Y had an odds ratio of 3.2, then the scoring index would be the sum of 2.5 points if condition X was present (zero if condition X was not present), plus another 3.2 points if condition Y was present (zero if condition Y was not present).
We then attempted to simplify the weighting factors, with a goal to arrive at a scale that would be easy to use clinically while preserving its diagnostic properties. The diagnostic properties of different versions of our index were evaluated by two statistical measures throughout this process: their discriminative power (how well it was able to differentiate between a positive and a negative outcome), and their goodness-of-fit to the data (how well did the prediction from the index fit the data). Discrimination was measured by the C statistics from receiver-operating characteristics (ROC) curve analysis, which ranges from 0.5 to 1.0 and reflects the ability of the index to differentiate positive events (in this case, being discharged to a SNF) and negative events (being discharged home). Goodness-of-fit to data was measured by pseudo r 2, which ranges from 0 (none of the variance is explained) to 1 (all variance is explained) [12].
With an exemption from our Institutional Review Board, the index was then validated using our internal institutional data from 1999 through 2005 by calculating the scores based on our index for patients meeting the inclusion criteria as initially used and then calculating the discrimination and goodness-of-fit properties of our new index again as it was applied to patients in this validation data set. Our institution is in Maryland, hence there was no overlap between the development data set from California and our institutional validation data set.
|
Results |
|---|
|
TopAbstractIntroductionMaterial and MethodsResultsCommentDiscussionAcknowledgmentsReferences |
|---|
Among all of the different ICD-9CM diagnosis codes recorded in these patients, 25 codes had prevalence rates of 5% or more among patients discharged to a SNF, and 22 such codes were identified among patients who were not discharged to SNF (Table 1). These codes were then evaluated in the multiple logistic regression analysis. Ten codes associated with SNF admission emerged from this process; specifically, female gender, osteoarthrosis, CHF, atrial fibrillation, MI, anemia, obesity, occlusive carotid disease, chronic obstructive pulmonary disease (COPD), and renal and ureteral diseases (Table 2).
|
|
|
|
|
|
|
Comment |
|---|
|
TopAbstractIntroductionMaterial and MethodsResultsCommentDiscussionAcknowledgmentsReferences |
|---|
65) requiring discharge to a SNF after a CABG. This index includes nine items, including six items with point values of 1 (osteoarthrosis, CHF, atrial fibrillation, MI, anemia, and obesity) and three items with point values of 2 (female gender, COPD, and renal and ureteral diseases). With this system, patients with overall scores of 3 or less are considered low risk and are 60% less likely than high-risk patients (with scores of 4 or more) to be discharged to a SNF. Previous studies have evaluated SNF utilization among elderly CABG patients. In a single institution study of 128 CABG patients by Garza and colleagues [8], the discharge patterns were similar to our data, with 20% of patients requiring admission to a SNF before returning home. Nallamothu and colleagues [7] reported a much higher SNF admission rate of 50% in a series of 1217 patients older than age of 80 who underwent CABG in the state of Michigan. In a second study by the same author, however, the rate of discharge to extended care facilities was 24.9% among the national EXPLORE database, a comparative health care outcomes benchmarking database from Solucient LLC (Evanston, IL) [9]. These discrepancies likely represent differences in SNF usage across different geographic regions with different payment structures. Our findings support the notion of high SNF use among elderly CABG patients, as suggested by these prior publications. With 17.3% of elderly patients discharged to a SNF, the need for models to predict postoperative disposition is obvious. This index is a tool that can be used in this regard.
An important property of a diagnostic index is to discriminate between two events or two ultimate outcomes. The discriminative property of this tool, then, would be its fixed property, even though the exactly probability of an event predicted by the index would differ between patient populations with different characteristics. This is similar in concept to the issue that the positive predictive value (PPV) and negative predictive value (NPV) of a test may vary depending on prevalence of disease in the population, but differing prevalence rates would not affect the sensitivity and specificity properties of the test instrument. This is demonstrated by our study here, which developed the original index on a national data set that has much lower prevalence of the outcome of interest. Yet, when the index is applied to our own institutional data, a tertiary-care academic medical center with a much sicker patient population who would have higher risk of being discharged to a SNF, the index maintains its diagnostic properties in terms of discrimination and goodness-of-fit to data, even though the probabilities of outcome predicted at different point levels were much higher than they were from the development data set.
An interesting finding from our index is the observation that female gender carries significant risk of being discharged to a SNF. Previous publications have highlighted gender differences after CABG, demonstrating a more difficult recovery period in women that was not attributed to preoperative fitness or illness severity but rather to differences in ambulation dysfunction, physical symptoms, perceptions of physical health, symptoms of depression, and home management dysfunction [13, 14]. For nonsurgical patients, however, a study from Minnesota [15] revealed that women with angina or MI were significantly less likely to use cardiac rehabilitation than men. Another study of both surgical and nonsurgical patients revealed that low-income women (annual income of <$20,000) were less likely to be referred to cardiac rehabilitation after a percutaneous coronary intervention, CABG, or an MI that did not require revascularization [16]. A large French study of patients admitted to the hospital after acute coronary syndrome also revealed a gender difference, with 82% of men going to cardiac rehabilitation compared with only 68% of women [17].
Various social explanations for the gender difference in admission to a cardiac rehabilitation unit have not yet been explored. Previous research in the impact of social support on postsurgical outcomes has been limited to white men [17, 18]. Future studies are therefore needed to identify the precise relationship between gender and use of a SNF after CABG to better account for this disparity.
Our study has important limitations, many of which relate to using a large outcomes database. It is conceivable that errors could have been introduced due to under-reporting on the postoperative disposition of some patients. Such potential errors were most likely randomly distributed across the patients in our analysis, however, and therefore less apt to affect our findings. Furthermore, the use of administrative databases did not permit the analysis of some unrecorded clinical variables that may have been significant.
An alternative study design would have consisted of a comprehensive study on a relatively smaller series of patients (probably single-institutional); however, such a study would have sample size limitations in making the inferences we were seeking. Our objective was to develop a scoring system consisting of relevant, readily acquired preoperative clinical characteristics as the basis for a prospective validation study. We recognize that it is quite likely that, in the course of such a study, additional important preoperative characteristics will be identified.
Our study is also limited by the inability to distinguish between internal versus external rehabilitation facilities. We believe that patient characteristics affecting discharge to internal versus external facilities are largely dependent on social or practical issues, such as insurance coverage and closeness to home, and would have little impact on our findings. Nevertheless, because this could not be examined directly in our data sets, this is an important limitation to acknowledge.
We also recognize that our findings may not be applicable to hospitals in other geographic regions, although we sought to address this by validating our indexing system with our own institutional data by using data from a different state on the opposite continental coast.
A word of caution with respect to our designated cut-point score of 3 defining high-risk patients is warranted. Such categoric assignments can be useful for the practical application of our system, but as can be seen from Table 4, no cut-point is optimal. Higher or lower cut-point scores can be used, depending on whether there is more concern for sensitivity (ie, identifying patients who will be discharged to a SNF) versus specificity (ie, identifying patients who will not go to a SNF) in a given patient population. Higher cut-point scores decrease sensitivity but improve specificity of our system.
By introducing a mechanism for identifying patients who are likely to require admission to a SNF, we hope to improve patient selection for CABG among the elderly population. The index’s simplicity and validation through applying it to patients within our own institution make it a useful tool in the preoperative evaluation of these patients.
|
Discussion |
|---|
|
TopAbstractIntroductionMaterial and MethodsResultsCommentDiscussionAcknowledgmentsReferences |
|---|
DR CHANG: This was based on making cutoffs at different score levels, and then examining the different two-tiered scales that are created for their discrimination, goodness-of-fit to data, and tradeoffs in sensitivity and specificity.
DR DOTY: The other comment or question I had is I understand that your variables came from a database, and some of these clearly, as you know, aren’t sort of yes–no things. Congestive heart failure and myocardial infarction and anemia, that is a diagnosis in the database, but how low of anemia, or how much failure, or how long was their infarct? So, do you have any way of measuring that and working that into your model?
DR CHANG: That is an important limitation of our index: it does not include clinical data and all the details that would go along with them. Like you say, in an administrative database, it is a yes-or-no diagnosis. We have no way of knowing the severity, or the length of degradation of the condition. We would be able to answer that question if we had more data probably from a prospective trial if we were to collect this prospectively.
DR JOHN D. PUSKAS (Atlanta, GA): Have you begun to use your index? Do patients like it? Does it change your practice?
DR CHANG: We have not. We believe it is important to put the tool, with its development and validation process, through peer-review process first before we begin a prospective trial.
DR THOMAS R. CALHOUN (Houston, TX): This is interesting and probably very pertinent in this day and age of increased use of extended and skilled nursing facilities. One question. In this day and age of early discharge, even in older patients and higher risk patients, did you have any way to retrieve early readmissions to the same or different hospitals, and how much might this have skewed your data?
DR CHANG: That is an interesting question. I think with the exception of the Medicare database, most of these large databases are de-identified to the point that you don’t know if the same patient comes back a second time. So, no, we do not know about readmission problems, and that is obviously, again, a limitation of the study.
DR GLENN J. R. WHITMAN (Philadelphia, PA): Although I had the feeling from your initial slide that showed the New England Journal article that quality of life after a procedure was something that was relevant to patients in terms of their decision-making, I think that many patients would be comforted to know that should they need a skilled nursing facility after open heart surgery it would be available to them, and I wouldn’t think it would be a decision-making issue for them. Am I misunderstanding that? Would, in fact, an elderly patient turn down surgery if he were told he might be going to a skilled nursing facility after the operation?
DR CHANG: Yes, that was the feeling of our group and that was the premise of our study, as well as the conclusion of that NEJM article, that discharge status is an important consideration for patients. We believed it is important for us to provide this information and the data to assist patients. Patients have their own unique set of judgments and values, and this would just be data that we can provide to fit into their value system to see whether or not being discharged in a non-independent status is important to them, and if so, would that prevent them from going through with surgery? That was the premise of our study. Ultimately, I think the interpretation and final decision still is left up to the individual patient; some patients may think it is acceptable to be discharged to a skilled nursing facility, but some may not.
|
Acknowledgments |
|---|
|
TopAbstractIntroductionMaterial and MethodsResultsCommentDiscussionAcknowledgmentsReferences |
|---|
|
References |
|---|
|
TopAbstractIntroductionMaterial and MethodsResultsCommentDiscussionAcknowledgmentsReferences |
|---|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| 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 |
