Using Machine Learning To Define the Impact of Beta-Lactam Early and Cumulative Target Attainment on Outcomes in Intensive Care Unit Patients with Hospital-Acquired and Ventilator-Associated Pneumonia

Hospital-acquired pneumonia (HAP) and ventilator-associated pneumonia (VAP) are the most common intensive care unit (ICU) infections. We aimed to evaluate the association of early and cumulative beta-lactam pharmacokinetic/ pharmacodynamic (PK/PD) parameters with therapy outcomes in pneumonia. Adult ICU patients who received cefepime, meropenem, or piperacillin-tazobactam for HAP or VAP and had its concentration measured were included. Beta-lactam exposure was generated for every patient for the entire duration of therapy, and the time free concentration remained above the MIC (fT(>MIC)) and the time free concentration remained above four multiples of the MIC (fT(>4xMIC)) were calculated for time frames of 0 to 24 h, 0 to 10 days, and day 0 to end of therapy. Regression analyses and machine learning were performed to evaluate the impact of PK/PD on therapy outcomes. A total of 735 patients and 840 HAP/VAP episodes (47% HAP) were included. The mean age was 56 years, and the mean weight was 80 kg. Sequential organ failure assessment (SOFA), hemodialysis, age, and weight were significantly associated with the clinical outcomes and kept in the final model. In the full cohort including all pneumonia episodes, PK/PD parameters at different time windows were associated with a favorable composite outcome, clinical cure, and mechanical ventilation (MV)-free days. In patients who had positive cultures and reported MICs, almost all PK/PD parameters were significant predictors of therapy outcomes. In the machine learning analysis, PK/PD parameters ranked high and were the primary overall predictors of clinical cure. Early target attainment and cumulative target attainment have a great impact on pneumonia outcomes. Beta-lactam exposure should be optimized early and maintained through therapy duration.