Surveillance of healthcare-associated infections (HAIs) in Australia is disparate, resource intensive, unsustainable, and provides limited information. Traditional HAI surveillance is time intensive and agreement levels between clinicians have been shown to be variable.

To compare two methods: a semi-automated algorithm, and coding data, against traditional surgical site infection (SSI) surveillance methods.

This retrospective multi-centre cohort study included all patients undergoing a hip (HPRO) or knee (KPRO) prosthesis and coronary artery bypass graft (CABG) surgery during a two-year period at two large metropolitan hospitals. Routine SSI data were obtained via the infection prevention and control (IPC) team, a previously developed algorithm was applied to all patient records, and the ICD-10-AM data were searched for those categorized as having an SSI.

Overall, 1447, 1416, and 1026 patients who underwent HPRO, KPRO, and CABG, respectively, were included. The highest sensitivity values were generated by the algorithm: HPRO deep or organ-space (D/O) 0.87 (95% confidence interval: 0.66–0.96), CABG 0.86 (0.64–0.96), and HPRO all SSI 0.77 (0.57–89); the lowest sensitivity was Code CABG D/O 0.03 (0.00–0.21). The highest PPV values were generated by the algorithm: HPRO D/O 0.97 (0.77–0.99), CABG D/O 0.97 (0.76–0.99), and the Code HPRO D/O 0.9 (0.66–0.99). Both the algorithm and coding data resulted in a substantial reduction in the number of medical records required to review.

The application of algorithms to enhance SSI surveillance demonstrates high accuracy in identifying patient records that require review by IPC teams to determine the presence of an SSI. Coding data alone should not be used to identify SSIs.