| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
a Department of Surgery, Duke University, Durham, North Carolina
b Duke Clinical Research Institute, Durham, North Carolina
Accepted for publication March 26, 2010.
* Address correspondence to Dr Onaitis, Department of Surgery, Duke University, DUMC Box 3305, Durham, NC 27710 (Email: mark.onaitis{at}duke.edu).
Presented at the Forty-sixth Annual Meeting of The Society of Thoracic Surgeons, Fort Lauderdale, FL, Jan 25–27, 2010.
| GENERAL THORACIC SURGERY:
The Annals of Thoracic Surgery CME Program is located online at http://cme.ctsnetjournals.org. To take the CME activity related to this article, you must have either an STS member or an individual non-member subscription to the journal.
|
 |
Abstract |
|---|
 Top Abstract IntroductionPatients and MethodsResultsCommentDiscussionReferences |
|---|
Methods: The Society of Thoracic Surgeons (STS) database was queried for all lobectomy and pneumonectomy patients with a diagnosis of lung cancer. A multivariable logistic regression model was developed to predict the risk of atrial arrhythmia as a function of preoperative and perioperative factors. Generalized estimating equations methodology was used to account for correlation among observations from the same institution. Missing data were handled using the method of chained equations with 10 randomly imputed data sets.
Results: A total of 13,906 patients who underwent resection for lung cancer at participating institutions had complete information for postoperative atrial arrhythmia, of whom 1,755 (12.6%) experienced the outcome. Multivariable logistic analysis indentified increasing age, increasing extent of operation, male sex, nonblack race, and stage II or greater tumors as predictors of postoperative atrial fibrillation.
Conclusions: Analysis of the STS database has identified five variables that predict postoperative atrial fibrillation. This predictive model may be useful to develop strategies for risk stratification, prophylaxis, and treatment.
Atrial arrhythmia occurs after 10% to 20% of major noncardiac thoracic operations. Although often minimized because it is generally self-limited and controlled with rate control and amiodarone, it has been demonstrated to increase length of stay [1–6]. In addition, patients who experience atrial arrhythmia have increased perioperative mortality [4–8], although the arrhythmia may be a result of other serious complications and not a direct contributor to mortality.
Published predictors of postoperative atrial arrhythmia after major thoracic surgery include increasing age, increasing extent of operation, preoperative increased resting heart rate, history of congestive heart failure, history of peripheral vascular disease, male sex, and intraoperative transfusions [4–7, 9–12]. Small trials of prophylaxis have identified diltiazem, digitalis, and amiodarone as potentially effective drugs [13–15]. While the existing literature provides useful information about prediction and possible prophylaxis of atrial arrhythmia, most reports document single institution experiences and include multiple different indications for thoracic surgery.
The Society of Thoracic Surgeons (STS) created the General Thoracic Surgery Database (GTSD) in 1999. This voluntary database contains a multitude of preoperative, operative, and early postoperative variables and has grown to include more than 100 centers. Because atrial arrhythmia most often occurs in the first week after major thoracic surgery, and because the database is designed to carefully collect perioperative data, the STS GTSD is ideal for this type of study. We thus sought to query this large national database to identify risk factors for atrial arrhythmia after lobectomy or greater for primary lung cancer with an eye toward future trials of pharmacologic prophylaxis.
|
Patients and Methods |
|---|
|
TopAbstractIntroductionPatients and MethodsResultsCommentDiscussionReferences |
|---|
Patient populations
Between 2002 and 2008, 14,041 patients were entered into the database having undergone lobectomy or greater for an indication of primary lung cancer. Of these, 135 had missing data for the binary variable atrial arrhythmia requiring treatment after surgery and were excluded from the analysis. This definition of this field in the STS general thoracic data specifications (http://www.sts.org/documents/pdf/ndb/GeneralThoracicDataSpecsV2081.pdf) follows: indicate whether the patient had a new onset of atrial fibrillation/flutter (AF) requiring treatment; does not include recurrence of AF that had been present preoperatively. Also excluded from the analysis were patients for whom age or sex were missing.
Data definitions
Other postoperative events were those defined by the STS GTSD guidelines (http://www.ctsnet.org/file/ThoracicDCFV2_07_Nonannotated.pdf). Operative mortality was defined as death during the same hospitalization or within 30 days of the procedure.
Previously reported predictors of atrial arrhythmia as well as potentially interesting variables were included in the analysis. These included the continuous variables age, body mass index, height, weight, pulmonary function test (PFT) values, and smoking pack-years. Binary variables included sex, operative laterality, operative approach (thoracotomy/thoracoscopy), presence/history of hypertension, congestive heart failure, coronary artery disease, induction therapy, steroid use, peripheral vascular disease, cerebrovascular events, diabetes mellitus, and chronic renal insufficiency. Categorical variables included resection type, race, and clinical stage. Several of these variables contained significant proportions of missing data. These ranged from 0% for age and sex (as missing were excluded from the analysis) to 20% for the American Joint Committee on Cancer (AJCC) pathologic stage, 24% for forced expiratory volume of air in 1 second (FEV1), 36% for diffusion capacity of lung for carbon monoxide (DLCO), and 42% for AJCC clinical stage.
The predictor variables in the full model were chosen based on clinical relevance and consideration of sample size in each variable category. Those variables are age, body mass index, sex, operative laterality, operative approach (thoracotomy/thoracoscopy), presence/history of hypertension, congestive heart failure, coronary artery disease, clinical stage II and above, resection type (three groups: lobectomy as reference, bilobectomy, and standard/carinal/completion/extrapleural pneumonectomy/intrapericardial pneumonectomy), race (three groups: Caucasian as reference, black, other).
Statistical Analysis
Summary statistics, based on nonmissing values, are presented as frequency and percentage for categorical variables, or median and interquartile range for continuous variables. The distributions of risk factors and outcomes for those who did and did not experience atrial arrhythmia were compared using Mantel-Haenszel tests for categorical variables and Wilcoxon rank sum tests for continuous variables.
Given the extent of missing data in the risk model predictors, multiple imputation was carried out using the R (www.R-project.org) package Multivariate Imputation by Chained Equations (MICE). This package generates multiple imputations for incomplete multivariate data by Gibbs sampling [16]. Variables included in the imputation models were atrial arrhythmia, predictors in full model (as described previously), any postoperative complication, operative death, preoperative induction therapy, and pathologic staging. Ten complete imputed datasets were created. A proposed model was then fitted to each of completed data sets. The 10 sets of results were combined using the method proposed by Rubin [17].
Patients were randomly divided into training and validation datasets (60% training set, 40% validation set). The training set was used to estimate model coefficients. Data from the validation sample were used to assess model fit and performance evaluation. Logistic regression modeling was used to estimate risk of atrial arrhythmia as a function of preoperative characteristics (ie, predictor variables). The predictor coefficients were estimated using the generalized estimating equations method with exchangeable correlation structure to account for correlations within same participants. To reduce the number of variables, a reduced model was also developed including only variables with significant coefficient using criterion of p values = 0.1, and logistic regression was performed in the same fashion. Using validation datasets, c statistics were calculated and compared for full model and reduced model. The performance of the reduced model was then evaluated. Briefly, the predicted probability of atrial arrhythmia from the reduced model was computed in each of the 10 validation sets using coefficients estimated from the corresponding training set. Based on the rank of combined predicted risks, the patients in validation set were grouped into 10 ordinal risk score groups with approximately equal size. The actual proportion of atrial arrhythmia and the mean predicted risk were then calculated for each risk score group.
|
Results |
|---|
|
TopAbstractIntroductionPatients and MethodsResultsCommentDiscussionReferences |
|---|
|
|
|
|
|
|
|
Finally, the calibration of the model was assessed. For each patient in validation set, the predicted risk of reduced model was computed, based on coefficients obtained from training set. The patients in validation set were then grouped into 10 ordinal groups based on the ranks of their predicted probability of atrial arrhythmia. The actual proportion of atrial arrhythmia and the mean of the predicted risk were calculated for each group, and are presented in Figure 1. Across each decile, the actual number of patients experiencing atrial arrhythmia is similar to that predicted by the reduced model. Both the ninth and tenth deciles contain patients with actual and predicted rates of atrial arrhythmia at or above 20%.
|
|
Comment |
|---|
|
TopAbstractIntroductionPatients and MethodsResultsCommentDiscussionReferences |
|---|
Strengths of this study include the sample size, inclusion of patients in high- and low-volume centers, a large number of both thoracoscopy and thoracotomy patients, the inclusion of only patients with a diagnosis of primary lung cancer, and the simple nature of the reduced model. Obvious limitations of the trial include its retrospective nature, the necessity to impute significant amounts of data, and possible differences in measurement of atrial arrhythmia and other variables across the various centers in the database. As with any retrospective study, particularly one in which centers voluntarily enter information, the possibility exists that the incidence of atrial arrhythmia will be underreported. However, the incidence in our study of 12.6% is well within the recently published range [6, 10].
The largest previously reported analysis of atrial fibrillation after major thoracic surgery using a prospectively maintained database at a single center identified male sex, increasing age, history of congestive heart failure, history of arrhythmias, history of peripheral vascular disease, resection of mediastinal tumor/thymus, increasing extent of resection, and intraoperative transfusion as predictive of atrial arrhythmia [6]. Although patients with a previous history of atrial arrhythmia and nonlung cancer patients were excluded from the present study, our model corroborates many of these predictors. Possible explanations for congestive heart failure, history of peripheral vascular disease, and transfusion not predicting atrial arrhythmia in our study include increased sample size, different measurement of predictors across centers, and possible interactions with other variables in the models.
The predictive ability of increasing extent of lung resection and clinical disease above stage 1 (at least N1 disease) as well as the lack of predictive ability of operative approach (thoracoscopy versus thoracotomy) combine to implicate extent of hilar dissection (rather than extent of parenchymal resection and degree of postoperative pain/sympathetic surge) as a probable causal factor for atrial arrhythmia development. This fits with purported origin of paroxysmal atrial fibrillation in the pulmonary veins [18]. However, this is unlikely to explain all of the cases, and competing autonomic inputs [19], atrial size (which is larger in males) [20, 21], and inflammation [22], all of which are unmeasured in our database, have been reported to contribute.
Our reduced model has a moderate ability to predict atrial arrhythmia (c statistic 0.66 for the reduced model). This area under the receiver operating characteristic curve is similar to that seen in a recent single center prediction rule for atrial fibrillation [10]. However, the predictive factors in this model (444 patients in training set and 412 patients in validation set) included only male sex, age, and preoperative resting heart rate. Our model corroborates the first two of these, and the last is unmeasured in the STS database. Extent of pulmonary resection, nonblack race, and clinical stage are also strongly predictive in our model. That may again be due to the increased power owing to our large sample size and different variable definitions.
As mentioned above, the need to impute missing data for several variables is a weakness of this study and will be a weakness in all studies using the STS database. As the database matures, it is hoped that missing data will become less of an issue. One of the final predictor variables, clinical stage, contained more than 40% of missing data. However, the imputation of this field may be less problematic given pathologic stage information. Also, as noted above, when applying the reduced model to only patients with complete information (n = 6,843), the same predictors are obtained.
Finally, we believe that this large model that represents the risk factors of atrial fibrillation across centers throughout the United States sets the stage for possible trials of pharmacologic prophylaxis. The top two deciles of atrial fibrillation risk include patients who have a greater than 20% incidence of predicted and actual atrial fibrillation. This prediction rule that contains factors that are known preoperatively could easily allow randomization of patients in these deciles to prophylaxis with diltiazem or placebo. We propose to include this simple postoperative arrhythmia calculator on the STS website for this purpose as well as for clinicians to better understand this risk.
|
Discussion |
|---|
|
TopAbstractIntroductionPatients and MethodsResultsCommentDiscussionReferences |
|---|
The relationship of postoperative A-fib and lung resection has long been a topic of investigation in general thoracic surgery. For decades, colleagues have documented the incidence, especially after pneumonectomy, where it can occur in as many as a quarter or a third of patients. Similarly, the search for effective and simple prophylaxis has been difficult, even in the pneumonectomy patient who we know is at the highest risk. There is simply little convincing multi-institutional evidence that pharmacologic prophylaxis can mitigate this risk. Clearly, this area is a fruitful arena for collaborative, thoughtful, and prospective investigations.
Dr Onaitis and colleagues from Duke are to be congratulated for leveraging the general thoracic database to begin to dissect this problem. They have confirmed the implied linkage of postoperative A-fib and adverse outcome, showing a greater than threefold increase in 30-day mortality in patients having perioperative A-fib. The simplified risk model offers clinicians the benefit to use it to stratify risk and better counsel their patients, but unfortunately their prognostic factors did not provide an opportunity to modify that risk.
Perhaps the most beneficial outcome of the work is to identify the groups at higher risk as the best candidates for pharmacologic prophylaxis in hopes of defining a therapeutic signal of efficacy in the noise encountered in other studies where all preoperative patients were given prophylaxix. I have a couple of questions for the authors.
Why did you choose to limit your analysis to only lobectomy and pneumonectomy? Given the more frequent detection of smaller primary lesions and a renewed interest in sublobar resection, inclusion of segmentectomy and wedge resection in your models would add timely information. At what level of missing variables do the imputation methods used lose their validity? I would remind the audience that 24% of patients had no FEV1 recorded and 36%, more than a third, had no DLCO. And why are thoracotomy and VATS included as individual versus dichotomous (either/or) variables in your model? Did some patients have both, thereby underestimating the benefit of VATS?
And finally, Dr Harpole and ourselves, in previous work, showed that the peak incidence of A-fib after pneumonectomy was 3 to 4 days postoperatively and that the complication developed in 84% of patients within 7 days. With variable standards of postoperative monitoring in our patients, how can we assure that all or even most of the cases are detected? Were any of the cases in that STS database reported after discharge? And with ever-increasing efforts on our part to decrease length of stay, how can we be sure that we adequately are monitoring our patients?
I would like to thank the Society for the opportunity to discuss this paper.
DR ONAITIS: Thank you for those comments and questions, Dr DeCamp. I will answer those questions in order.
We thought about including all sorts of resection, especially given renewed interest in sublobar resection. However, many of these patients go home in the first two days after surgery. We would probably miss arrhythmia cases in this group.
In terms of missing data, that is a huge problem in this database, and I think Dr Kozower highlighted it yesterday. We need to do a better job as thoracic surgeons of making sure that our database is complete, because we did have to impute a lot of data. That is one of the reasons why we didn't include FEV1 or DLCO in the analysis. We didn't put PFTs in our multivariate model just because there were so many missing data points.
We did include both VATS and thoracotomy as separate variables just for the reason you said. There is really no way to know if someone had a VATS converted to thoracotomy, whether that was planned or an emergent operation, and so we included each of them as their own operation. And you could argue we could have done it the other way, but at the present time in the database there is no way to characterize this in a binary fashion.
And finally, we do recognize your and Dr Harpole's previous contributions to this literature, and I think monitoring is a big issue. At our facility, everyone is on telemetry through their whole hospitalization; however, I am sure at other places that is not done. The postoperative data in our case are entered by a nurse practitioner, who keeps tracks of these patients postoperatively, but there is no way to know how many occurred after discharge. My sense is that is a small number and that the AF episodes that are in the database occurred during the postoperative hospitalization. That is a weakness of this kind of study, but we will just have to live with it.
DR FRANK C. DETTERBECK (New Haven, CT): I have always thought that A-fib is not A-fib in this setting. There are some patients who have just a brief run of A-fib that is asymptomatic and really doesn't matter. There are other patients where it is a bigger issue and they stay in the hospital longer. And then there are some patients who have pneumonia, sepsis, ARDS, and later on in their course as a premorbid event, A-fib develops. I wonder if you could tease some of that out and if there is a reasonable way to separate out different groups that perhaps we should be looking at differently?
DR ONAITIS: Those are great questions. At present there is no way in the database to tell when the patient had A-fib or whether it was a result of a more serious complication. We have no way of knowing the circumstances of each patient's arrhythmia. However, per the STS database guidelines, the arrhythmia needed to require treatment in order to be "counted." Because of these issues, I was a little bit careful during the talk not to say that atrial fibrillation caused these other things that are associated with it. There is no way to say that these things are causative or that one results from the other, and those are definitely good things to study in the future.
|
References |
|---|
|
TopAbstractIntroductionPatients and MethodsResultsCommentDiscussionReferences |
|---|
This article has been cited by other articles:
|
|
 |
|
  S. Shiono, M. Abiko, and T. Sato Postoperative complications in elderly patients after lung cancer surgery Interact CardioVasc Thorac Surg, June 1, 2013; 16(6): 819 - 823. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
P. L. Filosso, E. Ruffini, A. Sandri, P. O. Lausi, R. Giobbe, and A. Oliaro Efficacy and safety of human fibrinogen-thrombin patch (TachoSil(R)) in the treatment of postoperative air leakage in patients submitted to redo surgery for lung malignancies: a randomized trial Interact CardioVasc Thorac Surg, May 1, 2013; 16(5): 661 - 666. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 D. J. LaPar, G. J. Stukenborg, C. L. Lau, D. R. Jones, and B. D. Kozower Differences in reported esophageal cancer resection outcomes between national clinical and administrative databases J. Thorac. Cardiovasc. Surg., November 1, 2012; 144(5): 1152 - 1159. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 V. P. Rao, E. Addae-Boateng, A. Barua, A. E. Martin-Ucar, and J. P. Duffy Age and neo-adjuvant chemotherapy increase the risk of atrial fibrillation following oesophagectomy Eur J Cardiothorac Surg, September 1, 2012; 42(3): 438 - 443. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 D.-h. Wu, M.-y. Xu, T. Mao, H. Cao, D.-j. Wu, and Y.-f. Shen Risk Factors for Intraoperative Atrial Fibrillation: A Retrospective Analysis of 10,563 Lung Operations in a Single Center Ann. Thorac. Surg., July 1, 2012; 94(1): 193 - 197. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
D. J. LaPar, C. M. Bhamidipati, C. L. Lau, D. R. Jones, and B. D. Kozower The Society of Thoracic Surgeons General Thoracic Surgery Database: Establishing Generalizability to National Lung Cancer Resection Outcomes Ann. Thorac. Surg., July 1, 2012; 94(1): 216 - 221. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
M. Anile, V. Telha, D. Diso, T. De Giacomo, S. Sciomer, E. A. Rendina, G. F. Coloni, and F. Venuta Left atrial size predicts the onset of atrial fibrillation after major pulmonary resections Eur J Cardiothorac Surg, May 1, 2012; 41(5): 1094 - 1097. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 S. Suksompong, S. Thamtanavit, B. von Bormann, and P. Thongcharoen Thoracic surgery mortality and morbidity in a university hospital Asian Cardiovascular and Thoracic Annals, April 1, 2012; 20(2): 182 - 187. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
T. Raman, N. Roistacher, J. Liu, H. Zhang, W. Shi, H. T. Thaler, and D. Amar Preoperative left atrial dysfunction and risk of postoperative atrial fibrillation complicating thoracic surgery J. Thorac. Cardiovasc. Surg., February 1, 2012; 143(2): 482 - 487. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 S. Sabate, A. Mases, N. Guilera, J. Canet, J. Castillo, C. Orrego, A. Sabate, G. Fita, F. Parramon, P. Paniagua, et al. Incidence and predictors of major perioperative adverse cardiac and cerebrovascular events in non-cardiac surgery Br. J. Anaesth., December 1, 2011; 107(6): 879 - 890. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
Z. Makhija, M. S. Allen, D. A. Wigle, K. R. Shen, S. D. Cassivi, F. C. Nichols III, and C. Deschamps Routine Anticoagulation Is Not Indicated for Postoperative General Thoracic Surgical Patients With New-Onset Atrial Fibrillation Ann. Thorac. Surg., August 1, 2011; 92(2): 421 - 427. [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 |
