C22-19 Validating Clinician Value-added Modeling for Mechanically Ventilated Patients: Quantifying the Roles Physicians and Nurses Play in Disease Trajectory

Abstract Rationale Multidisciplinary team-based care, an evidence-based care delivery model, may improve outcomes for mechanically ventilated patients. However, components of team-based care that contribute to improved outcomes have yet to be robustly quantified. We aimed to validate value-added modeling, an econometric technique that quantifies independent contributions of individual team members toward overall multidisciplinary team performance, in the context of mechanically ventilated patients. Methods Among a retrospective cohort study of mechanically ventilated patients across 6 hospitals at an academic health system between 2018 and 2022, we used multiple linear regression to measure relationships between change in individual patient Laboratory Acute Physiology Score from mechanical ventilation initiation to ICU discharge and clinical assignment of individual physicians and nurses. After partitioning the cohort into a development and testing set, we evaluated models that regressed physician assignments, nursing assignments, both, or neither to determine optimal model fit and precision based on the adjusted coefficient of determination (R2) and the Akaike information criterion (AIC). We assessed correlation of nurse and physician value-add across partitions to establish reproducibility. Lastly, we evaluated the random distribution of clinician value using analysis of variance and the likelihood ratio test of nested linear models, hypothesizing that mechanically ventilated patients have equal access to clinicians based on their value-add. Results Among 17,495 unique patient encounters, 215 distinct physicians and 1719 nurses contributed to the care of 15,521 patients. Models including both physician and nursing variables demonstrated superior model fit based on adjusted R2 and AIC (0.19 and 3953236 respectively) when compared to adjusted models without either variable (0.12 and 3980745 respectively), models containing only physicians (0.16 and 3963114 respectively), and models containing only nurses (0.18 and 3959350 respectively). Clinician value-add was moderately correlated across development and testing partitions (0.31 for nurses and 0.4 for physicians). Clinician value-add was associated with improved disease severity scores, ranging from a 20- to 70-point decrease for physicians, and a 10- to 55-point decrease for nurses (Figure). Clinician value-add was nonrandomly distributed with respect to disease severity change based on analysis of variance (F-statistic 6.18, p 0.001), and nested model testing (Chi-square 7432.7, p 0.001). Conclusion Clinician value can be independently quantified by using value-added modeling, allowing future studies to measure relationships between clinician value and patient-centered outcomes. Additionally, clinician value-add is nonrandomly distributed; future work should evaluate for potential disparities related to clinician performance. This abstract is funded by: NHLBI