Application of a Pediatric Tracheostomy-Specific Risk Tier System Using Administrative Data.

IMPORTANCE

Children undergoing tracheostomy have heterogeneous clinical profiles, yet no scalable, tracheostomy-specific system exists to stratify risk using administrative data. General severity measures such as All Patient Refined Diagnosis Related Group (APR-DRG) provide limited case mix differentiation for this population.

OBJECTIVE

To evaluate whether a literature-derived, tracheostomy-specific risk-tier system improves stratification of pediatric index hospitalization outcomes compared with APR-DRG severity classification.

DESIGN, SETTING, AND

PARTICIPANTS

In this retrospective cohort study using the Pediatric Health Information System database, participants were children younger than 18 years undergoing index tracheostomy placement at 44 tertiary children's hospitals from January 1, 2016, through December 31, 2024, in the US.

MAIN OUTCOMES AND MEASURES: The primary outcomes were prolonged length of stay (greater than 90 days) and in-hospital mortality. Risk stratification was assessed by separation of outcomes across proposed tiers (standard, moderate, or critical risk) and supported by discrimination metrics (C statistics).

RESULTS

Among 14 275 patients (6027 female [42.2%]; 8237 male [57.7%]; median [IQR] age, 0 [0-7] years), 3431 (24%) were in the critical-risk tier, 9363 (66%) in the moderate-risk tier, and 1481 (10%) in the standard-risk tier. The median (IQR) length of stay increased from 34 (16-69) days (standard-risk tier) to 105 (54-192) days (moderate-risk tier) to 173 (104-263) days (critical-risk tier). Mortality increased from 70 (4.7%) in the standard-risk tier to 710 (7.6%) in the moderate-risk tier to 554 (16.1%) in the critical-risk tier. The tier system C statistic was 0.665 (95% CI, 0.658-0.672) for prolonged hospitalization and 0.608 (95% CI, 0.594-0.622) for mortality, compared with 0.521 (95% CI, 0.519-0.524) and 0.509 (95% CI, 0.507-0.511) for APR-DRG. Intraclass correlation coefficients were 0.065 (95% CI, -0.214 to 0.335) for prolonged stay and 0.005 (95% CI, -0.001 to 0.011) for mortality. The intraclass coefficient is bounded at 0. The negative lower limit reflects the Wald approximation for a variance component estimated near zero, not a genuinely negative correlation, consistent with negligible between-hospital variance.

CONCLUSIONS AND RELEVANCE: In this cohort study, a tracheostomy-specific risk-tier system using administrative data provides meaningful case mix stratification for pediatric tracheostomy hospitalizations and outperforms general severity classification. This approach may support risk-adjusted benchmarking and multicenter comparisons, though modest discrimination limits individual-level prognostic use and warrants further validation.