Abstract
Background: APACHE II, APACHE III and SOFA are widely used intensive care severity scores, but their comparative discrimination for mortality in acute respiratory distress syndrome (ARDS) is not well established in Indian cohorts. Methods: In this prospective observational study in the medical intensive care unit of a tertiary care hospital, 100 patients fulfilling the Berlin definition of ARDS were enrolled. All three scores were calculated within 24 hours of admission and related to Berlin severity and outcome by logistic regression and receiver operating characteristic analysis. Results: All three scores rose across severity strata and were higher in the 27 non-survivors than in the 73 survivors (p<0.001 throughout): APACHE II 35.8 versus 21.9, APACHE III 87.85 versus 53.14 and SOFA 9.0 versus 6.6. Each discriminated mortality better than chance — APACHE II area under the curve 0.895 (95% confidence interval 0.812– 0.978), APACHE III 0.797 (0.689–0.905), SOFA 0.765 (0.651–0.879) — although the intervals overlap and no paired comparison of the curves was performed. At its optimum cut-off of ≥27, APACHE II gave 85.2% sensitivity and 78.1% specificity, with negative predictive value 93.4% but a positive predictive value of only 59.0%. On logistic regression, APACHE II (Wald 12.418), APACHE III (Wald 7.854) and mode of ventilatory support (Wald 10.29) were retained and age, sex, primary diagnosis and SOFA were not; with 27 events and 14 model degrees of freedom this model is overparameterised. Conclusion: APACHE II had the highest point estimate of discrimination, but overlapping confidence intervals do not establish superiority over APACHE III or SOFA.
Keywords: ARDS, APACHE II, APACHE III, SOFA, ROC Curve, Mortality Prediction, Severity Scoring