Predictive Model for Length of Stay in Intensive Care Unit Based on Preoperative and Perioperative Variables. Four-Year Analysis of Mitral Valve Surgery in One Single Center

Abstract

Submit Manuscript | http://medcraveonline.com IntroductionThe cost of intensive care unit (ICU) stay is very high for patients after cardiac surgery in all settings. The costs of cardiac valve surgery depends mainly on the expenses of the operation and length of hospital (LOSH) and ICU (LOSI) stay.As ICUs are the most expensive part during hospital admission, it is important to assess which are the risk factors contributing to a prolonged stay in the ICU. Several risk models such as the Cleveland model, EuroSCORE, CORRADscore, and Parsonnet model have been developed originally to predict postoperative morbidity and mortality of patients undergoing cardiac surgery [1][2][3]. High preoperative risk scores have been associated with longer LOSH and increased total costs in coronary artery bypass surgery [4]. Those scores have been successfully validated also for prediction of prolonged LOSI. Parsonnet model and EuroSCORE were superior in identifying patients with prolonged LOSI [5].Predicting LOSI after mitral valve surgery is challenging due to the wide spectrum of comorbidities these patients sustain, and to the cardiovascular procedures and potential complications related to the surgical procedures. In addition, patients with a AbstractStudy objective: Assessing length of stay in Intensive Care Units (LOSI) after mitral valve surgery is challenging due to wide spectrum of comorbidities and cardiovascular procedures. The study aim was to elaborate a predictive model in order to identify risk factors and to predict the expected resources consumption for better monitoring for both patients and caregivers.Design: Retrospective study of prospectively collected data in the period 2009-2013. Setting: University teaching hospital.Patients: We studied all consecutive adult patients (≥ 18 years) who underwent elective mitral valve surgery. We enrolled 212 patients. Measurements:The outcome variable was LOSI. Comorbidities, postoperative morbidity and mortality were collected during the perioperative period. Multivariate analysis was used and data were studied as an additive model to assess the associations between the variables and correlations between them. A generalized additive model was fitted with least squared means estimated by the additive model due the non linearity of Euroscore I. Histograms of the residuals were examined visually to assess the fit of the models above. The model was selected using the statistical software R. Results:Median LOSI was 4 days (Interquantile range 2 to 4 days). The model for LOSI includes: infection post surgery (p<0.05), cardiogenic shock (p<0.05), Euroscore I (p<0.05), valvular replacement (p<0.05), double valvular disease (p<0.05) and associated tricuspid valve repair (p<0.05). The adjusted r squared for this model was 0.74, which implies a high predictive value for LOSI. In this model the ponderate weights for the main variables were 36.3% for infection post surgery, 15.2% for aortic clamp time, 6.9% for cardiogenic shock and 5.4% for the Euroscore I. Conclusion:Longer LOSI is st...