59: PREDICTION MODEL TO IDENTIFY SEPSIS PATIENTS FOR MINIMALLY INVASIVE MANAGEMENT OUTSIDE THE ICU

Abstract

Introduction: Patients with sepsis who need intensive care unit (ICU) admission for short-term monitoring (< 48 hr) with low-dose vasopressors could be candidates for minimally invasive sepsis (MIS) approach outside the ICU. We aimed to describe such patients and develop a prediction model to find candidates for MIS approach. Methods: In a secondary analysis of institutional data, we identified adults (≥18 yr) with septic shock admitted to the Mayo Clinic medical ICU from emergency. MIS approach candidates were with < 48 hr of ICU stay, without advanced respiratory support or renal replacement therapy and alive at ICU & hospital discharge. Comparison group consisted of septic shock patients with an ICU stay of >48 hr who didn’t need advanced respiratory support before ICU admission. Wilcoxon rank, Chi Square, and Fisher’s exact tests were used to compare baseline characteristics. Logistic regression and prediction profiler were used to select predictive variables and assign cut-offs. Using points assigned to variables based on OR and clinical rational, a nominal logistic regression model was created. Model discrimination & calibration were assessed by receiver operating characteristic (ROC) curves and Hosmer-Lemeshow (H-L) goodness-of-fit test. Results: Out of 1795 MICU admissions, 109 patients (6 %) met MIS approach criteria and 97 in comparison group. MIS group was younger with lower respiratory rate, temperature, lactate, creatinine, BUN, and WBCs (p < 0.05 for all). Predictive variables (age >65 yr, O2 flow >4 L/mint, temperature < 37℃, creatinine >1.6 mg/dL, lactate >3 mmol/L, WBCs >15x109, heart rate >100 beats/mint, and respiration rate >25 breaths/ mint) were translated into an 8-point score. Model discrimination yielded ROC: AUC of 80%. Model fit was confirmed by agreement between observed and predicted probability and was well calibrated (H-L p = 0.94). MIS score cut-off of 3 resulted in a predictive model OR of 0.15 (CI: 0.08–0.28) and a negative predictive value of 91% [(CI: 88.69-92.92) (using prevalence of MIS patients ~20%)] Conclusions: Study implies that a significant minority of sepsis patients can be managed outside the ICU. Once validated in an independent, prospective sample our prediction model can be used to identify candidates for minimally invasive management outside of ICU