Surgical site infections (SSIs) rank among the most common healthcare-associated infections (HAIs), leading to increased patient morbidity, extended hospital stays, and greater healthcare costs. For effective infection prevention and control (IPC) strategies, the World Health Organization identifies the surveillance of HAIs, such as SSIs, as crucial. Existing surveillance of SSIs is mainly performed by IPC professionals through laborious chart reviews and manual evaluations, which are not only time-consuming but also display low inter-rater reliability.
With the advent of electronic health records (EHR), there’s increasing enthusiasm to transition from manual to automated and semi-automated SSI surveillance. Machine learning (ML) models and rule-based algorithms offer promising avenues for this transition. This article dwells on the development and assessment of machine learning and rule-based models for semi-automated detection of deep and organ/space SSIs among a study pool of 3931 surgical patients.
These models aim to minimize undetected SSI cases while significantly reducing the manual workload of IPC professionals. In their performance, the best-striving ML models (Naïve Bayes and dense neural network) projected sensitivity up to 0.90 and workload reduction over 90%. In contrast, the rule-based model exhibited higher sensitivity but lower workload reduction.
The primary advantage of these semi-automated strategies is their potential to release IPC professionals’ time, thus facilitating more preventive actions rather than retrospective case identification. However, there’s a critical trade-off to be considered between accuracy and operational efficiency during the selection of surveillance methodologies for implementation.
This comprehensive examination of ML and rule-based models underscores how both can be integrated rather than solely relying on the latter due to its interpretability and ease of implementation. Therefore, ML models should supplement wider strategies that incorporate rule-based logic, expert validation, and data-driven decision-making to optimize automated surveillance systems.