Machine Learning: Prediction of Central Line Associated Blood Stream Infections (CLABSI)

Session Description: Central Line Blood Stream Infections (CLABSI) are a major source of hospital-acquired infections (HAI) and can increase patient mortality by 10-30%, adding $30,000 to $50,000 per patient. Annually in the United States, CLABSI results in an estimated 250,000-500,000 infections and over 28,000 deaths. Machine learning (ML) algorithms have successfully identified variables that predict CLABSI using a risk rating score. These variables and algorithms can help in early identification of high-risk patients, enhancing patient safety, reducing Length of Stay (LOS), and lowering healthcare costs.

With healthcare resources becoming increasingly scarce due to financial challenges from the aftermath of the COVID-19 pandemic and staffing constraints, an efficient tool is needed to reduce HAI. Automation tools can expand the capacity of Infection Prevention (IP) personnel, allowing them to focus on other interventions and support the field amid turnover and fewer trained IPs entering the workforce.

We explored a risk rating tool using previously researched patient variables to extract data from Electronic Medical Records. This tool distinguishes high-risk from low-risk patients, impacting individualized care plans and directing resources to higher-risk patients. The ML model can predict which patients with a central line in place for 48 hours will likely develop a positive blood culture. Predictions are made daily, forecasting a possible CLABSI 48 to 72 hours in the future. This rolling time window was chosen to predict future possible CLABSI not yet clinically evident. In testing, the model had an Area Under the Curve (AUC) of 0.96, precision of 0.91, and recall of 0.94.

We are conducting a responsible Artificial Intelligence (AI) analysis of our model before live deployment. Upon a CLABSI alert, the Infection Preventionist would review potential mitigating strategies like central line removal, adherence to best practice bundle elements, and antibiotic lock therapy (ALT) with the care team.