| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
|
Interact CardioVasc Thorac Surg 2009;9:463-466. doi:10.1510/icvts.2008.201178 © 2009 European Association of Cardio-Thoracic Surgery
External validation of the modified Thoracoscore in a new thoracic surgery program: prediction of in-hospital mortalityDepartment of Cardiothoracic Surgery, ‘Attikon’ Hospital, Sofokleous 36, Voula 16673, Athens, Greece Received 23 December 2008; received in revised form 24 May 2009; accepted 28 May 2009
*Corresponding author. Tel.: +302105832150.
Informed consent prior to any surgical intervention should include in-hospital survival estimation after the procedure performed. The recently developed Thoracoscore predicts well the postoperative mortality possibility. The purpose of our study was to test the modified Thoracoscore performance in our new thoracic program. One hundred and fifty-five consecutive patients underwent thoracic surgery procedure within two years. The procedures performed were: 62 lung resections, 10 open tumor biopsies, 21 neck and mediastinal procedures, 33 chest wall and pleural procedures, 8 tracheal procedures, 3 esophageal procedures, 13 minor cardiac procedures, and 5 chest trauma cases. The modified Thoracoscore was calculated based on the following variables: age, gender, priority of the procedure, malignancy, type of procedure, Zubrod score, ASA class, and number of co-morbidities. The observed mortality was 5.2% (eight deaths) while the predicted one based on the modified Thoracoscore was 4.9%. The scoring system we used had excellent discriminatory ability with a C statistic (0.95, 95% CIs 0.91–0.99). The Hosmer–Lemeshow goodness-of-fit was not statistically significant (P=0.82), indicating acceptable calibration of the model for the present series. The modified Thoracoscore's ability to predict postoperative survival in the whole context of thoracic surgery performs well in our program. Application of any risk scoring system requires external validation and provides comparison of the actual outcomes with other programs.
Key Words: In-hospital mortality; Thoracic surgery; Risk prediction
Thoracoscore is the first scoring system predicting in-hospital outcome after general thoracic procedures [1]. It was derived from 15,183 patients who underwent thoracic surgery in 59 institutions. The modified Thoracoscore has been shown to predict accurately both short- and mid-term outcomes in a northern American thoracic database [2]. The purpose of this study is to validate the modified Thoracoscore in a recently developed University Thoracic Surgery Program.
2.1. Patient population and data One hundred and fifty-five consecutive patients were operated between June 2006 and September 2008 at the Hospital of Athens ‘Attikon’ affiliated with the University of Athens Medical School by one consultant thoracic surgeon. Patients data were collected prospectively during admission as part of routine clinical practice and were entered into a database modeled by the Society for Thoracic Surgery General Thoracic Database; age, gender, priority of the procedure, disease category, associated co-morbidities, ASA score, Zubrod score, type of procedure performed and in-hospital outcome were recorded. Seventy-three patients had malignancy and 44 cases were performed on an urgent or emergent basis. Patient characteristics are presented in Table 1.
2.2. Data analysis and statistical methods All discrete variables were recorded as percentages and continuous variables as mean value±1 S.D. Difference for discrete variables between two groups was analyzed with the  2-test and for continuous variables with the one-way ANOVA test.
The modified Thoracoscore was calculated as follows based on our previous publication [2]: modified Thoracoscore=–6.975+[–0.108 if patients age >55 and <65 or 1.057 if patients age was Performance of the modified Thoracoscore was tested with assessment of the discriminatory ability and the calibration of the model. A C statistic (or the area under the receiver operating characteristic curve) was used to assess the discriminatory ability [3]. The area under the receiver operating curve was calculated as an index (C statistic) for how well the modified Thoracoscore could discriminate patients who lived and those who died during their hospitalization after thoracic surgery. The discriminative power of the model is thought to be excellent if the area under the receiver operating characteristic curve is >0.80, very good if >0.75, and good if >0.70. The calibration of the model was assessed by the Hosmer–Lemeshow goodness-of-fit statistic [4]. For the Hosmer–Lemeshow statistic, the predicted risks of individual patients were rank-ordered and divided into quartiles of roughly equal size, based on their predicted probability. Within each quartile of estimated risk, the number of predicted deaths was accumulated against the number of observed deaths; a P>0.05 indicates acceptable calibration of the model. All analyses were performed with SPSS 11.0 (SPSS, Inc, Chicago, IL, USA) and P-values were two-tailed.
One hundred and fifty-five patients underwent an equal number of procedures. Eighty-nine patients were male (57.4%) and 44 (28.4%) cases were performed on an urgent or emergent basis. The distribution of procedure categories were as follows: 62 lung resections, 10 open tumor biopsies, 21 neck and mediastinal procedures (including mediastinoscopies), 33 chest wall and pleural procedures, 8 tracheal procedures, 3 esophageal procedures, 13 minor cardiac procedures without use of cardiopulmonary bypass, and 5 chest trauma cases requiring hemothorax drainage. The actual in-hospital mortality was 5.2% (eight cases) (Table 2). From the eight patients who died, two had undergone a pneumonectomy (one developed a broncho-pleural fistula and another pneumonia), two had an open lung biopsy (were critically ill patients in the intensive care unit), another two patients had a tracheo-esophageal fistula repair (from prolonged intubation), one patient had an open tumor biopsy, and finally another patient died from sepsis after sternal debridement following coronary artery bypass surgery. The mean predicted in-hospital mortality based on the modified Thoracoscore was 4.9% (range, 0.1–92.1%). The mean modified Thoracoscore of the patients who died was 37.4±35.2%. The mean modified Thoracoscore of the patients who survived after their procedure was 3.2±8.8%; the difference was statistically significant (P<0.001).
The discriminatory ability of the modified model was excellent as measured by the C statistic (0.95, 95% CIs 0.91–0.99) (Fig. 1). The Hosmer–Lemeshow goodness-of-fit was not statistically significant (P=0.82), indicating acceptable calibration of the model for the given small number of in-hospital deaths in the present study.
We further analyzed the performance of the modified Thoracoscore algorithm in predicting combined mobidity and mortality. We included 22 major postoperative complications (six respiratory failures with a ventilation support of >24 h, four sepsis, one wound infection, two bleedings requiring transfusion, two atelectasis requiring bronchoscopy, two atrial arrhythmias, one dilatation of esophagus prior to discharge, one recurrent nerve paresis, two prolonged air leaks and one reoperation) along with the eight in-hospital deaths. By analyzing these 30 combined events we found that the group with the adverse events had statistically significant higher modified Thoracoscore (17.0±28.4% vs. 3.3±9.0%, P=0.003). However, the discriminatory ability of the model as measured by the C statistic was not good (0.69, 95% CIs 0.66–0.72), although the Hosmer–Lemeshow test was not statistically significant (P=0.752).
In this study, we examined the ability of the modified Thoracoscore to predict the outcome after general thoracic operations. The observed in-hospital mortality (5.2%) is very close to the one predicted by the scoring system we used (4.9%). In addition, the discriminatory ability and the calibration are satisfactory indicating good validation of the external scoring system in our new thoracic surgery program. Similarly, the observed adverse events in this study, including major postoperative complications and in-hospital deaths (19.4%), are close to the predicted ones from modified Thoracoscore (17.0%). However, in this case the algorithm showed inability in terms of the discriminatory ability as measured by C statistic to predict combined outcomes. A possible explanation for this is that the algorithm has been initially developed for the prediction of in-hospital mortality. It may also have the ability to predict accurately some of the major postoperative complications, but this has yet to be determined in large general thoracic databases with large numbers of major complications in each category. Predicting postoperative outcome must be part of patients' preoperative assessment and informed consent. The identification of independent predictors for in-hospital mortality entails multivariate logistic analysis based on a larger patient population than the one included in the present analysis. We, therefore, decided to apply an external predicting score from our previous experience [2]. The external model must be validated at the initial application and periodically thereafter as the patients' database increases. The Thoracoscore is the first scoring system predicting in-hospital outcome after general thoracic procedure which was derived from 15,183 patients who underwent thoracic surgery in 59 institutions [1]. This includes patient's age, gender, priority of the procedure, ASA class, Zubrod score, number of co-morbidities, presence of malignancy, dyspnea score, and type of procedure (pneumonectomy). We decided to use the modified Thoracoscore simply because we had prospectively collected data during admission for all variables except the dyspnea score; this is not included in the database modeled by the Society for Thoracic Surgery General Thoracic Database. From the eight independent variables included in our model pneumonectomy, Zubrod score, and ASA class have the highest impact on short-term outcome. Pneumonectomy carries a significant perioperative mortality risk; the largest pneumonectomy series from the Mayo Clinic reported 11% deaths within a 30-day period [5]. Zubrod score reflects patient's ambulatory status; bedridden patients are prone to developing pulmonary postoperative complications and, therefore, have a higher risk for bad outcome. Ferguson and Durkin showed that performance status is an independent predictor for pulmonary complications after esophagectomy for cancer [6]. Prause et al. [7] found that ASA score was a good predictor for perioperative mortality in the whole context of thoracic surgery. Additionally, Berrisford et al. [8] and Harpole et al. [9] identified similar independent variables for in-hospital outcome after lung resection in the European Thoracic Surgery Database Project and the National Veterans Affairs Surgical Quality Improvement Program, respectively. Performance status and ASA score can vary based on observer's subjective assessment; this limitation can be minimized if experienced medical staff is involved in patient's preoperative evaluation. There are few limitations in this analysis. First, the number of cases is not large. When the database increases the performance of the model applied may change; therefore, testing the scoring system from time to time is advisable. When in-hospital deaths exceed a minimum number, multivariate logistic regression analysis may be used to identify internal independent variables for postoperative death. Second, variables such as dyspnea score and postoperative complications have not been included in the model. It is known from the cardiac surgery literature that complications increase the likelihood of death [10]; it is, therefore, quite possible that the same may hold true for thoracic surgery patients. The omission of the above parameters, however, does not appear to affect the model's performance based on the measured C statistic (0.95, 95% CIs 0.91–0.99). In addition, the model's prediction ability of the combined morbidity and mortality was not assessed. Finally, prediction of mid- and long-term outcome is not provided in this study; long-term survival data are not available at this point, since our program is only two years old. In conclusion, we demonstrated that the modified Thoracoscore has excellent postoperative outcome predictability in the whole context of thoracic surgery procedures in our program. This scoring system is easily applied at the patient's bedside and can be part of history and physical examination. In addition, it is a useful tool to monitor early results in new surgical programs comparing the actual mortality with what the expected should be based on northern American and western European standards. Incorporation of specific postoperative complications in the model may make it more robust.
|
||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| 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 |
Top
Abstract
1. Introduction
65]+[0.402 if patients sex was male]+[1.909 if American Society of Anesthesiologists class 
2 or 0.761 if number of co-morbidities was 