A comparative analysis of predictive models of morbidity in intensive care unit after cardiac surgery – Part II: an illustrative example (original) (raw)
Related papers
A comparative study of four intensive care outcome prediction models in cardiac surgery patients
Journal of Cardiothoracic Surgery, 2011
Background: Outcome prediction scoring systems are increasingly used in intensive care medicine, but most were not developed for use in cardiac surgery patients. We compared the performance of four intensive care outcome prediction scoring systems (Acute Physiology and Chronic Health Evaluation II [APACHE II], Simplified Acute Physiology Score II [SAPS II], Sequential Organ Failure Assessment [SOFA], and Cardiac Surgery Score [CASUS]) in patients after open heart surgery. Methods: We prospectively included all consecutive adult patients who underwent open heart surgery and were admitted to the intensive care unit (ICU) between January 1 st 2007 and December 31 st 2008. Scores were calculated daily from ICU admission until discharge. The outcome measure was ICU mortality. The performance of the four scores was assessed by calibration and discrimination statistics. Derived variables (Mean-and Max-scores) were also evaluated. Results: During the study period, 2801 patients (29.6% female) were included. Mean age was 66.9 ± 10.7 years and the ICU mortality rate was 5.2%. Calibration tests for SOFA and CASUS were reliable throughout (p-value not < 0.05), but there were significant differences between predicted and observed outcome for SAPS II (days 1, 2, 3 and 5) and APACHE II (days 2 and 3). CASUS, and its mean-and maximum-derivatives, discriminated better between survivors and non-survivors than the other scores throughout the study (area under curve ≥ 0.90). In order of best discrimination, CASUS was followed by SOFA, then SAPS II, and finally APACHE II. SAPS II and APACHE II derivatives had discrimination results that were superior to those of the SOFA derivatives. Conclusions: CASUS and SOFA are reliable ICU mortality risk stratification models for cardiac surgery patients. SAPS II and APACHE II did not perform well in terms of calibration and discrimination statistics.
PLOS ONE, 2015
Background Advanced predictive analytical techniques are being increasingly applied to clinical risk assessment. This study compared a neural network model to several other models in predicting the length of stay (LOS) in the cardiac surgical intensive care unit (ICU) based on pre-incision patient characteristics. Methods Thirty six variables collected from 185 cardiac surgical patients were analyzed for contribution to ICU LOS. The Automatic Linear Modeling (ALM) module of IBM-SPSS software identified 8 factors with statistically significant associations with ICU LOS; these factors were also analyzed with the Artificial Neural Network (ANN) module of the same software. The weighted contributions of each factor ("trained" data) were then applied to data for a "new" patient to predict ICU LOS for that individual.
Clinical Cardiology, 1994
The traditional approach to developing models predictive of cardiac events has been to pertorm logistic regression (LR) analysis on a variety of potential predictors. An altemative is to use an artificial intelligence system called a neural network (NN) which simulates biological intelligence. To evaluate the potential applicability of the latter method, we compared the ability of LR and NN techniques to predict cardiac events after noncardiac surgery. A total of 200 patients (hairing p u p) underwent cardiac risk assessment before major noncardiac surgery using 17 clinical parameters and 7 quantitative indices based on dipyndamole-thallium imaging. There were 21 postoperative myocardial infarctions andor cardiac deaths. Data from the training group were used to develop two predictive models: one based on backward stepwise LR multivariate statistical analysis and the other one using a neural network. Both models were then validated on a second group of 160 consecutive patients also referred for preoperative risk stratification (validation group). The NN consisted of 14 input, 29 hidden, and I output neurons and used a back-propagation algorithm (leaming rate 0.2, training tolerance 0.5, sigmoid transfer function). The sensitivity, specificity, positive and negative predictive accuracies for the prediction of postoperative events in the vali-The views presented in this study are those of the authors and do not speak for the United States Government or its Department of Defense.
Prediction of mortality in intensive care unit cardiac surgical patients☆☆☆
European Journal of Cardio-Thoracic Surgery, 2010
Objectives: The purpose of this study was to develop a specific postoperative score in intensive care unit (ICU) cardiac surgical patients for the assessment of organ dysfunction and survival. To prove the reliability of the new scoring system, we compared its performance to existing ICU scores. Methods: This prospective study consisted of all consecutive adult patients admitted after cardiac surgery to our ICU over a period of 5.5 years. Variables were evaluated using the patients of the first year who stayed in ICU for at least 24 h. The reproducibility was then tested in two validation sets using all patients. Performance was assessed with the Hosmer-Lemeshow (HL) goodness-of-fit test and receiver operating characteristic (ROC) curves and compared with the Acute Physiology and Chronic Health Evaluation (APACHE II) and Multiple Organ Dysfunction Score (MODS). The outcome measure was defined as 30-day mortality. Results: A total of 6007 patients were admitted to the ICU after cardiac surgery. Mean HL values for the new score were 5.8 (APACHE II, 11.3; MODS, 9.7) for the construction set, 7.2 (APACHE II, 8.0; MODS, 4.5) for the validation set I and 4.9 for the validation set II. The mean area under the ROC curve was 0.91 (APACHE II, 0.86; MODS, 0.84) for the new score in the construction set, 0.88 (APACHE II, 0.84; MODS, 0.84) in the validation set I and 0.92 in the validation set II. Conclusions: Most of general ICU scoring systems use extensive data collection and focus on the first day of ICU stay. Despite this fact, general scores do not perform well in the prediction of outcome in cardiac surgical patients. Our new 10-variable risk index performs very well, with calibration and discrimination very high, better than general severity systems, and it is an appropriate tool for daily risk stratification in ICU cardiac surgery patients. Thus, it may serve as an expert system for diagnosing organ failure and predicting mortality in ICU cardiac surgical patients. #
A non-linear time series based artificial intelligence model to predict outcome in cardiac surgery
Health and Technology, 2022
Background Adverse lifestyles have led to increased cardiac complications, further accelerating the burden of cardiac surgeries in tertiary care hospitals. For optimum management of cardiac surgical patients in the hospital, it is essential to have an accurate idea regarding the patients' expected ICU stay and hospital stay. Additionally, forecasting patients' survival outcome is also essential for ICU management. Objectives This study aims to develop artificial intelligence models based on non-linear time-series data of blood pressure and heart rate to predict the ICU stay, hospital stay, and survival outcome of cardiac surgical patients. Methods The intraoperative heart rate and blood pressure data of 6064 patients undergoing cardiac surgeries at a single tertiary care hospital were recorded every minute. After data cleaning, the data was split into 781 patients in the train data set and 296 patients in the test data set. Feature engineering and balancing of data were performed on the train data set. Various classification models for survival outcome and regression models for ICU stay and hospital stay were trained using the balanced train data set. These models were tested on the test data set, and the prediction results were evaluated on the following performance metrics: area under the curve (AUC), accuracy, F1-score, RMSE, and R2-score. Results The Gaussian Naive Bayes + Logistic Regression (GNB + LR) model is the best for survival analysis, having the highest AUC of 0.72, Accuracy of 83%, and an F1-score of 0.86. The Gradient boosting (GB) model is the best model for the analysis of hospital stay, offering the highest R2-score (0.023). The XGBoost regressor is the best model for ICU stay analysis, offering the highest R2-score (0.125). Conclusion Artificial intelligence models based upon the intraoperative time series data were developed to analyse outcomes in cardiac surgery with high accuracy. These models can be used in cardiac surgeries to predict the ICU stay, hospital stay, and overall survival of the patients for better ICU management at the hospital.
BMC Medical Informatics and Decision Making
Background A disease severity classification system is widely used to predict the survival of patients admitted to the intensive care unit with different diagnoses. In the present study, conventional severity classification systems were compared with artificial intelligence predictive models (Artificial Neural Network and Decision Tree) in terms of the prediction of the survival rate of the patients admitted to the intensive care unit. Methods This retrospective cohort study was performed on the data of the patients admitted to the ICU of Ghaemshahr’s Razi Teaching Care Center from March 20th, 2017, to September 22nd, 2019. The required data for calculating conventional severity classification models (SOFA, SAPS II, APACHE II, and APACHE IV) were collected from the patients’ medical records. Subsequently, the score of each model was calculated. Artificial intelligence predictive models (Artificial Neural Network and Decision Tree) were developed in the next step. Lastly, the perform...
Circulation, 2010
Background-Several models have been developed to predict prolonged stay in the intensive care unit (ICU) after cardiac surgery. However, no extensive quantitative validation of these models has yet been conducted. This study sought to identify and validate existing prediction models for prolonged ICU length of stay after cardiac surgery. Methods and Results-After a systematic review of the literature, the identified models were applied on a large registry database comprising 11 395 cardiac surgical interventions. The probabilities of prolonged ICU length of stay based on the models were compared with the actual outcome to assess the discrimination and calibration performance of the models. Literature review identified 20 models, of which 14 could be included. Of the 6 models for the general cardiac surgery population, the Parsonnet model showed the best discrimination (area under the receiver operating characteristic curveϭ0.75 [95% confidence interval, 0.73 to 0.76]), followed by the European system for cardiac operative risk evaluation (EuroSCORE) (0.71 [0.70 to 0.72]) and a model by Huijskes and colleagues (0.71 [0.70 to 0.73]). Most of the models showed good calibration. Conclusions-In this validation of prediction models for prolonged ICU length of stay, 2 widely implemented models (Parsonnet, EuroSCORE), although originally designed for prediction of mortality, were superior in identifying patients with prolonged ICU length of stay. (Circulation. 2010;122:682-689.) Key Words: cardiovascular diseases Ⅲ complications Ⅲ epidemiology Ⅲ risk factors Ⅲ surgery I n the past decades, mortality during or shortly after cardiac surgery has decreased. 1 However, morbidity has increased, 2 mainly because cardiac surgery is increasingly utilized in older and more vulnerable patients. This often results in more complications after surgery and potential reduction in quality of life. 3-5 One method of assessing complications occurring directly after cardiac surgery is a prolonged stay in the intensive care unit (ICU). 6-9 Prolonged ICU stay also leads to incremental use of resources. In practice, prediction models are being used for efficient use of ICU resources. Patients with a low risk of complications are being scheduled for surgery before patients with a high risk. 5-13 Various prediction models have been developed to preoperatively identify patients with an increased risk for postoperative complications and prolonged ICU stay. 12-28 Interestingly, all of these prediction models were derived from samples including different patients, as reflected by the different distributions of patient and outcome characteristics. Hence, which model should be preferred in which situation is still unclear. Recently, in a qualitative review, Messaoudi and colleagues 14 reviewed 13 of these prediction models by comparing their published prognostic values for predicting ICU stay. They found that the 13 different prediction models indeed used different definitions of prolonged ICU stay and different definitions of predictors. Clinical Perspective on p 689 Even though it is widely accepted that no prediction model should be applied in practice before being formally validated on its predictive accuracy in new patients, 29-31 no study has previously performed a formal, quantitative (external) validation of these prediction models in an independent patient population. Therefore, we first conducted a systematic review to identify all existing prediction models for prolonged ICU length of stay (PICULOS) after cardiac surgery. Subsequently, we validated the performance of the identified models in a large independent cohort of cardiac surgery patients. Methods Systematic Literature Review
American Heart Journal, 2000
can identify patterns of variables that predict an outcome. They provide a nonlinear approach to data analysis 6-8 and have been successfully used to model clinical data with results comparable to traditional modeling techniques. 9,10 Examples of artificial neural networks include applications in cardiovascular medicine, 11 the prediction of perioperative cardiac risk in vascular patients, 12 intensive care unit length of stay, 13,14 weaning from respiratory support, 15 prognosis in heart failure, 16 cardiac complications after noncardiac surgery, 17 and the diagnosis of myocardial infarction. 18 Critics have argued that limitations of artificial neural networks include the subjective judgement in selecting which initial input variables to present to the artificial neural network for analysis. 19 It has been suggested that an artificial neural network, if elaborate enough, can find a relation between completely unrelated input variables. 20-23 To limit the emphasis on incidental associations between subjectively determined input Clinical outcomes after percutaneous transluminal coronary angioplasty (PTCA) are influenced by individual patient comorbidities and procedural variables. The recognition of variable patterns that correlate with an adverse outcome is valuable for risk assessment and stratification. Multivariate models have recently been developed to identify risk factors for adverse outcomes after PTCA and to allow the comparison of the results of different operators and institutions. 1-5 Artificial neural networks are computer programs that