Study objective: To identify variables associated with weaning outcome in long-term ventilator-dependent patients. Using those variables, to construct models to predict weaning success and to test the accuracy of those models.
Design: Retrospective medical record review.
Setting: Regional weaning center (RWC).
Patients: An initial group of 421 and a subsequent group of 170 consecutive patients referred for attempted weaning after 6 weeks of mechanical ventilation.
Measurements and results: Data obtained on admission to our facility were analyzed for correlation with weaning outcome. In the initial patient group, selected variables which correlated with weaning success were alveolar-arterial oxygen pressure difference (P[A-a]O2), BUN, BUN/creatinine ratio (each with p < or = 0.001), and female gender (p = 0.04). We used these variables in logistic regression models to predict weaning success in this population. We then tested the models in the 170-patient validation group using both standard and receiver operating characteristic (ROC) curve analysis. The ROC analysis indicated 59% accuracy using P(A-a)O2 alone and 68% accuracy using all previously mentioned variables. We used data from all 565 patients with known outcome and omitted BUN/creatinine ratio to fashion a simple scoring system to predict weaning success with 70% accuracy using P(A-a)O2, BUN, and Gender--the A+B+G score.
Conclusion: In patients suffering prolonged mechanical ventilation, models incorporating simple measurements allowed construction of a score to predict weaning success at our RWC.