Evaluation of APACHE-IV Predictive Scoring in Surgical Abdominal Sepsis: A Retrospective Cohort Study PC16-PC18
Dr. Tiffany Chan,
The University of British Columbia Faculty of Medicine
Department of Surgery, Division of General Surgery
3129A, 950 West 10th Avenuem Vancouver, B.C. V5Z 1M9
Introduction: Evaluation of the effectiveness of care and clinical outcomes in critically ill patients is dependent on predictive scoring models that calculate measures of disease severity and an associated likelihood of mortality. The APACHE scoring system is a logistic regression model incorporating physiologic and laboratory parameters. APACHE-IV is the most updated scoring system for ICU mortality prediction. However, APACHE scores may not accurately predict mortality in patients who require surgery for abdominal sepsis, whose trajectory is modulated by source control procedures.
Aim: To evaluate the accuracy of APACHE-IV mortality prediction in a cohort of ICU patients with surgical abdominal sepsis (SABS) requiring emergent laparotomy for source control.
Materials and Methods: The study was conducted in a combined medical and surgical intensive care unit in a large urban Canadian tertiary care hospital. Retrospective review of 211 consecutive adult ICU admissions that fulfilled the 2012 ACCP/SCCM criteria for severe sepsis/septic shock due to abdominal source was performed. APACHE-IV score and predicted mortality rate (PMR) were calculated and evaluated using area under the ROC curve (AUROC).
Results: Overall in-hospital mortality was 28.4%. There was overestimation of PMR by the APACHE-IV model in the overall cohort with an absolute difference of 16.6% (relative difference 36.9%). APACHE-IV crudely distinguished between survivors and non-survivors, with a PMR of 40% vs. 59% (p<0.001). AUROC of the APACHE-IV score was 0.67, 95% CI (0.58, 0.76) while the AUROC for the PMR was 0.72, 95% CI (0.64, 0.80), indicating poor performance in this cohort.
Conclusion: APACHE-IV has poor discrimination in SABS. Future research should explore disease-specific prediction models.