Development and prospective validation of a model for predicting weaning in chronic ventilator dependent patients

Abstract

Background: Approximately ten percent of patients placed on mechanical ventilation during acute illness will require long-term ventilator support. Unfortunately, despite rehabilitation, some will never be liberated from the ventilator. A method of predicting weaning outcomes for these patients could help conserve resources and minimize frustrating failed weaning attempts for this population. The objective of this investigation was to identify predictors of weaning outcome for patients admitted to a chronic ventilator unit (CVU).

Methods: This was a retrospective analysis with prospective validation. The study setting was a 25 bed CVU within a rehabilitation hospital. The training group consisted of 43 patients referred to our facility for weaning after > 3 weeks of mechanical ventilation. A multivariate model to predict weaning outcome was constructed in this group and applied to a prospective group of 31 patients followed during an 18-month period.

Results: A modified Glasgow Coma Scale (GCS) and the presence of sustained spontaneous respirations (SSR), defined as the presence of 2 breaths recorded above the ventilator settings on four occasions, were highly predictive of weaning success within six months of CVU admission. Patients with a modified GCS > or = 8 were 6.5 times more likely to wean than those with a modified GCS < 8 (95% confidence interval 1.6-26.3) and those with SSR were 25.5 times more likely to wean than those without SSR (95% confidence interval 4.3-51.9).

Conclusions: In our population of CVU patients, simple parameters that were available on admission and did not directly reflect cardiopulmonary function were useful predictors of weaning outcome.

Figure 1

Predicted and Observed Values of Weaning Success in the Validation Cohort. Positive predictive values were calculated from the development cohort and applied to the validation cohort. Predicted and observed proportions of patients successfully weaned are shown.

Figure 2

Predicted and Observed Values of Weaning Failure in the Validation Cohort. Negative predictive values were calculated from the development cohort and applied to the validation cohort. Predicted and observed proportions of patients failing to wean are shown.