Ms Stephanie J. Curtis1, Associate Professor Leon J Worth1, Associate Professor Rhonda L. Stuart2, Associate Professor Caroline Marshall3, Professor Paul D.R. Johnson4, Dr Lucy O. Attwood4, Dr Andie S. Lee5, Professor Allen C. Cheng1, Dr Andrew J. Stewardson1
1Department of Infectious Diseases, Alfred Health and Monash University, Melbourne, Australia,
2Department of Infectious Diseases, Monash Medical Centre , Clayton, Australia,
3Infection Prevention and Surveillance Service, Royal Melbourne Hospital, Melbourne, Australia,
4Department of Infectious Diseases, Austin Health, Melbourne, Australia,
5Department of Infectious Diseases, Royal Prince Alfred Hospital , Camperdown, Australia
The Australian Commission on Safety and Quality in Healthcare developed a list of sixteen potentially preventable Hospital-Acquired Complications (HACs) and an algorithm using International Classification of Disease (ICD) codes to detect them. We evaluated the HAC algorithm’s performance for diagnosing hospital-onset bloodstream infections (HO-BSI) compared to a reference surveillance definition.
We extracted administrative records for all acute patient episodes from July 2016 to June 2017 at four principal referral hospitals. We applied the BSI HAC algorithm to each episode, then randomly selected 50 patients that flagged positive and negative at each site. Reviewers blinded to HAC status applied the reference surveillance definition for HO-BSI. We computed the positive predictive value (PPV) and negative predictive value (NPV) for the BSI HAC.
Our cohort included 297,338 patient episodes; median (IQR) age was 56 (35–72), 48% were female, and 48.1% were elective admissions. According to the HAC algorithm, 1,936 (0.7%) had a HO-BSI. Among the 400 manually reviewed episodes, the PPV for the HAC algorithm was 0.25 (95% CI, 0.20–0.31) and the NPV was 1.00 (95% CI, 0.98–1.00). The codes, ‘Sepsis, unspecified’ and ‘bacterial sepsis of newborn, unspecified’ were both relatively common triggers for BSI HACs (37% and 18%, respectively) and had poor PPV (0.05 and 0.03, respectively). Removal of these ICD codes from the algorithm increased its PPV to 0.55 (95% CI 0.44–0.65).
The HAC algorithm has sub-optimal PPV for HO-BSI. This performance can be improved by removing the ‘unspecified’ ICD codes.
Stephanie is a researcher in the epidemiology of antimicrobial resistance and healthcare-associated infections. She has a background in international relations and community development, including a Master of Public Health (Health Economics & Economic Evaluation), Bachelor of Arts (International Studies & Spanish and Latin American Studies) and Diploma of Project Management. In 2018 she travelled across Australia to conduct a national point prevalence survey on HAIs, and is preparing to expand this work into the Pacific region. Her work also looks at the performance of algorithms for detecting HAIs, treatment and outcomes of urinary tract infections in the community and hospital presentations for injection-related infections.