OBJECTIVES: 

Unplanned extubations are an infrequent but life-threatening adverse event in pediatric critical care. Due to the rarity of these events, previous studies have been small, limiting the generalizability of findings and the ability to detect associations. Our objectives were to describe unplanned extubations and explore predictors of unplanned extubation requiring reintubation in PICUs.

DESIGN: 

Retrospective observational study and multilevel regression model.

SETTING: 

PICUs participating in Virtual Pediatric Systems (LLC).

PATIENTS: 

Patients (≤ 18 yr) who had an unplanned extubation in PICU (2012–2020).

INTERVENTIONS: 

None.

MEASUREMENTS AND MAIN RESULTS: 

We developed and trained a multilevel least absolute shrinkage and selection operator (LASSO) logistic regression model in the 2012–2016 sample that accounted for between-PICU variations as a random effect to predict reintubation after unplanned extubation. The remaining sample (2017–2020) was used to externally validate the model. Predictors included age, weight, sex, primary diagnosis, admission type, and readmission status. Model calibration and discriminatory performance were evaluated using Hosmer-Lemeshow goodness-of-fit (HL-GOF) and area under the receiver operating characteristic curve (AUROC), respectively. Of the 5,703 patients included, 1,661 (29.1%) required reintubation. Variables associated with increased risk of reintubation were age (< 2 yr; odds ratio [OR], 1.5; 95% CI, 1.1–1.9) and diagnosis (respiratory; OR, 1.3; 95% CI, 1.1–1.6). Scheduled admission was associated with decreased risk of reintubation (OR, 0.7; 95% CI, 0.6–0.9). With LASSO (lambda = 0.011), remaining variables were age, weight, diagnosis, and scheduled admission. The predictors resulted in AUROC of 0.59 (95% CI, 0.57–0.61); HL-GOF showed the model was well calibrated (p = 0.88). The model performed similarly in external validation (AUROC, 0.58; 95% CI, 0.56–0.61).

CONCLUSIONS: 

Predictors associated with increased risk of reintubation included age and respiratory primary diagnosis. Including clinical factors (e.g., oxygen and ventilatory requirements at the time of unplanned extubation) in the model may increase predictive ability.