Strategies for prevention of prolonged intensive care unit stay following cardiac surgery by identifying the determinants

Retrospective data analysis on 1592 consecutive patients admitted to intensive care following cardiac surgery (2011-2014). Dichotomous and categorical data were compared using Chi-square or Fisher's Exact tests. P- value of < 0.05 was significant. Univariate and Multivariate Regression identified predictors of prolonged intensive care stay. A score model for prolonged intensive care stay was developed as a logistic probability unit (z=logit (p)= log e (p/1-p); The area under the receiver curve (AUC) generated. The best cut-off point of the scoring model was identified, the likelihood ratio of a positive test result calculated. Logistic regression showed predictors of prolonged intensive care unit stay as ; NYHA class 3-4 (OR,1.5; p = 0.0029), FEV1 (OR, 0.76; p = 0.0026), emergency operation (OR,8.75; p = 0.0022), age (OR,1.02; p = 0.00007), LVEF<50% (OR,2.21; p = 0.00001), creatinine (OR,1.01; p = 0.000001), bypass time (OR, 1.01; p = 0.000001). Intensive care unit stay score was determined by logistic probability (AUC=0.76, 95% CI, 0.73;0.78, p = 0.00001) suggesting that a cut-off score of 35 predicts prolonged intensive care stay with a sensitivity of 0.66, specificity of 0.72 and accuracy of 0.70. The likelihood ratio of a positive test was 2.34.