Automated Surveillance of Surgical Site Infections in a VA Hospital

Background: Surgical site infections (SSIs) account for approximately 17% of hospital-acquired infections. These infections result in an increase in emergency room visits, outpatient visits, radiology services, home health aide services, and readmissions adding an estimated $1 billion-10 billion in indirect and direct medical costs each year. The CDC and the Surgical Infection Society recommend routine surveillance as a method for decreasing the rates of these infections. By monitoring SSI rates, areas of improvement can be identified and interventions can be made to reduce the incidence of SSIs in the hospital. Reductions of up to 35% have been documented with the implementation of SSI surveillance programs. Current methods of surveillance in the VA are only partially automated and are labor intensive. Automated methods of surveillance using electronic medical records have been proposed to decrease the resources involved in SSI monitoring. The VA is well-suited for this with their extensive medical records database and relatively closed system of patients. Purpose: To construct an automated SSI surveillance system using electronic patient medical record data and validate this system by comparing its performance to the current surveillance method used at the Durham VA hospital. Methods: In this project, we modified the methods previously described by Richard Platt to create an automated SSI surveillance system at the VA hospital in Durham, North Carolina. We used ICD-9 codes, vital signs, microbiology data, consult orders, and pharmacy records sensitive and specific for SSIs to identify patients with potential infections. Logistic regression was used to create predictive models for SSIs of different severity. This system was validated by comparing its performance to that of the current manual record review performed by the infection control department in the hospital on patients who underwent surgery at the Durham VA hospital from May 1st, 2002 to April 30th, 2004. All surgical-site infections met the criteria set forth by the National Nosocomial Infections Surveillance (NNIS) report. The system was evaluated using the framework set forth by the CDC Working Group for public health surveillance systems Results: SSIs occurred in 195 of 7340 surgeries conducted in the study period (2.7% attack rate). Of these, 91 were superficial SSIs, 45 were deep SSIs, and 59 were organ/space SSIs. Logistic regression models using data found to be strongly correlated with SSI diagnoses had a sensitivity and specificity of 90.9% and 61.2% for all types of SSIs, 89.2% and 74.2% for severe SSIs (deep and organ/space) and 89.5% and 74.0% for organ/space SSIs, respectively. Conclusions: This study demonstrates that an automated SSI surveillance system with reasonable sensitivity and specificity can be created by using data from electronic medical records. Such a system can drastically reduce the amount of labor necessary for SSI monitoring and increase the speed these complications are detected. The information technology used at the Durham VA hospital is similar to that used in other VA hospitals, so this system can be exported to other hospitals throughout the country.Master of Public Healt