J. R. Bergquist1,2, C. B. Storlie2, K. L. Mathis1, J. C. Boughey3, D. A. Etzioni4, E. B. Habermann2, R. R. Cima1 1Mayo Clinic,Division Of Colon And Rectal Surgery,Rochester, MN, USA 2Mayo Clinic,Robert D And Patricia E Kern Center For The Science Of Health Care Delivery,Rochester, MN, USA 3Mayo Clinic,Department Of Surgery,Rochester, MN, USA 4Mayo Clinic In Arizona,Colon And Rectal Surgery,Phoenix, AZ, USA
Introduction: Key drivers of colorectal surgical-site-infection (C-SSI) occurrence are institution-specific, and early identification of patients who will develop C-SSI requiring readmission remains challenging. We developed an analytic tool which would utilize institution-specific data for C-SSI screening and treatment during index hospitalization.
Methods: Elective colorectal resections from institutional ACS-NSQIP datasets (2006-2014) at 2 locations were included. A Bayesian-Probit regression model with multiple-imputation (BPMI) via Dirichlet process handled missing data. The baseline for comparison was a multivariate logistic regression model (GLM) with indicator variables for missing data (e.g., adding a “missing” level to factors) and stepwise variable selection. Out-of-sample performance was evaluated with Receiver Operating Characteristic (ROC) and Net Reclassification Improvement (NRI) analysis of 10-fold cross-validated samples. Primary endpoint was C-SSI requiring hospital readmission.
Results: Among 2376 resections, deep/organ space C-SSI rate was 4.6% (N=108: Figure-patients 3,4). Among patients developing C-SSI, N=65(60.1%) were discharged prior to clinical diagnosis (Figure-patient 3). The tool identified N=15(23.1%) of these patients prior to discharge (3 requiring re-operation), with 10% false alarm rate. Among patients clinically diagnosed with C-SSI prior to discharge (patient 4), the tool identified C-SSI 4.5 (mean) days prior to clinical identification. Tool performance generated ROC=0.77 and NRI=21.7%, demonstrating high predictive accuracy. When applied to independent validation data (N=478 cases, N=20 SSI), the tool identified during hospitalization 40% of patients discharged then readmitted with C-SSI (ROC=0.75; NRI=8.4%).
Conclusion: Identification of C-SSI prior to clinical presentation can facilitate early intervention, potentially reducing morbidity, re-admission, and re-operation. Our tool correctly identified a substantial proportion of patients who were discharged and readmitted with C-SSI in two independent datasets. This institutionally-generic analytic tool can improve outcomes and reduce costs associated with readmission and late C-SSI identification.
