POS0825 Development of a New And Simplified Formula to Predict Mortality in Patients with Anca-Associated Vasculitis Admitted to the Intensive Care Unit

Abstract

Background: ANCA-associated vasculitis (AAV) can present with a wide range of symptoms, including acute kidney injury (AKI) requiring renal replacement therapy or diffuse alveolar hemorrhage (1). These two manifestations often require admission to an intensive care unit (ICU) and are associated with increased mortality. To predict ICU mortality, the Simplified Acute Physiology Score version 2 (SAPS2) is often used but has not been formally tested in AAV patients (2). In addition, it is cumbersome to assess. Objectives: To develop a novel, simplified formula to predict ICU mortality in an AAV ICU population from an academic tertiary care center. Methods: We retrospectively recorded clinical and laboratory parameters in patients admitted to our ICU from 2000-2018. We performed risk factor analysis using univariate and multivariate logistic regression. In the multivariate case we applied the least absolute shrinkage and selection operator (lasso) method for variable selection. We considered average marginal effects and partial dependence plots in order to describe the influence of various independent variables on the probability of death more specifically. We evaluated our new score by comparing the corresponding area under the curve (AUC) to the AUC corresponding to the established SAPS2 score. Results: We analyzed 58 patients with AAV (39 granulomatosis with polyangiitis, 19 with microscopic polyangiitis) with a mean age of 74±14 (GPA) and 73±12 (MPA). 19/39 (48.7%) of GPA and 9/19 (47.4%) were female. Reasons for admission included disease manifestations or infectious complications from treatment (e. g. pneumonia, urinary tract infection). In total, 13/58 (22.4%) patients died throughout the study (10 GPA, 3 MPA patients). Using a cut-off threshold of 40 for SAPS2, sensitivity and specificity for mortality were 0.92 and 0.60, respectively. Confidence interval for the AUC was [0.68,0.95]. In the fitted multivariate logistic regression model, lasso was applied for variable selection. The identified variables included: disease duration, pH, procalcitonin, hemoglobin, leukocytes on admission, coronary heart disease, and pneumonia on admission. The estimated mortality is given by the formula ƒ (β0 + β1χ1 + …+ β7χ7), where ƒ( u )=1/(1+exp(− u )). Table 1 shows the estimated mortality for various values of the new score. Table 1. Example scores predicting mortality using the novel formula. Score Predicted mortality -2.2 0.1 -1.1 0.25 0 0.5 1.1 0.75 2.2 0.9 Testing if the AUC corresponding to the new model is significantly larger than the one corresponding to the SAPS2 score as independent variable resulted in p-value of 0.037. To identify possible overfitting, a 5-fold cross validation was performed. This resulted in a CI for the AUC of [0.64,0.96], suggesting that the new score allows for simpler prediction of mortality. Conclusion: We developed a novel formula corresponding to a score which is able to simpler predict mortality in patients with AAV admitted to the ICU. We will test our formula in the available ICU database MIMIC III, which comprises a large dataset of ICU patients. References: [1]Kitching AR, Anders H-J, Basu N, Brouwer E, Gordon J, Jayne DR, et al. ANCA-associated vasculitis. Nature Reviews Disease Primers. 2020 Aug 27;6(1):1–27. [2]Le Gall JR, Lemeshow S, Saulnier F. A new Simplified Acute Physiology Score (SAPS II) based on a European/North American multicenter study. JAMA. 1993 Dec 22;270(24):2957–63. Disclosure of Interests: PETER KORSTEN Speakers bureau: Chugai, Boehringer-Ingelheim, Sanofi Aventis, Abbvie, GSK, Novartis, Consultant of: Lilly, Gilead, Abbvie, Boehringer-Ingelheim, GSK, Novartis, Grant/research support from: GSK, Fabian Kück: None declared, Karaine Tejiozem Donfack: None declared, Radovan Vasko: None declared, Andreas Lena: None declared, Björn Tampe: None declared