Predictors of extubation success for premature infants

We aimed to determine whether clinical variables predict the success of extubation for premature infants.

Using variables preceding 320 extubations of infants ≤30 weeks or ≤1250 g at birth, we built predictive models for success at 1-, 3-, and 7-days using machine learning algorithms. We also determined whether lung ultrasound (LUS) scores (n = 15) were associated with success, or predicted success, of extubation.

Of 84 factors considered, nine were associated with success at 1 day, seven at 3 days, and six at 7 days. The accuracies of the predictive models were 78-84%. Median LUS scores were significantly lower preceding successful extubations (at 3 days) but not correlated with the findings of the predictive models.

We devised robust models for predicting extubation success based on clinical antecedents. Further study is needed to determine whether LUS can further improve prediction of extubation readiness.