O0018: External Validation of the Pediatric Submersion Score

Background: Pediatric drowning victims may be hospitalized due to concern for delayed respiratory complications. Our aim was to externally validate a previously developed, 5-factor risk score that identifies children at low risk for drowning injury who can be safely discharged from the ED after a period of observation. A secondary aim was to explore whether the score could be improved using external data.

Methods: This was a retrospective, cross-sectional study of children (0-18 years) who presented to 32 pediatric emergency departments from 2010-2017. We reviewed demographics, comorbidities, and prehospital and ED course from EMS, medical and fatality records. Primary outcome was safe discharge parameters at 8 hours post-drowning: normal mentation, oxygenation and vital signs. The same five independent predictors at ED presentation (normal mentation, respiratory rate, absence of dyspnea, systolic hypotension and airway support) that identified low risk drowning victims were validated. Receiver operating characteristic (ROC) curves with area under the curve (AUC) were generated to test sensitivity and specificity. To improve the diagnostic ability of the risk score, we attempted to identify potentially new factors by applying 3 regression models: binary logistic (forward-step), LASSO and Recursive Partitioning and Regression Trees modeling (with pruning) to the external dataset. Any factor with an overlap in ≥ 2 models was considered a potential score factor. Using the validation set from the original study, we tested for discrimination (ROC curve), calibration and (re)classification measures with safe discharge parameters at 8 hours post-drowning as the outcome. Diagnostic testing was used to assess the effectiveness of the score.

Results: The external validation cohort consisted of 4,147 patients (Median age: 3 years). When all five factors were included in the regression model, absence of dyspnea (aOR=3.1; 95%CI:2.4-3.9), normal respiratory rate (aOR=1.7;95%CI:1.3-2.1) and absence of airway support (aOR=75.3;95% CI:17.5-323.1) remained significant; while normal mentation (aOR=1.3;95%CI:0.9-2.1) and absence of systolic hypotension (aOR=0.4;95%CI:0.1-1.2) were not significant. Using the recommended cut-off score of ≥4, diagnostic tests suggested a safe discharge at 8 hours: Sensitivity:82%(95%CI:79%-85%); Specificity:59%(95%CI:56%-62%); PPV:58%(95%CI:55%-61%); NPV:83%(95%CI:80%-86%); AUC:0.77(95%CI:0.74–0.79).
Based on regression analysis of the external dataset, the risk score can be updated to include 7 parameters obtained at ED arrival: normal mentation, normal respiratory rate, absent dyspnea, no airway support, absent hypoxia, age < 6 years old and normal heart rate. Using the validation set from the original study (N = 89), diagnostic testing of the updated risk score revealed that a score of ≥ 5 (out of 7) predicts safe discharge at 8 hours with a sensitivity:91%(77-98); specificity:61%(44-77); PPV:70%(54–82) and NPV:88%(69–98) and AUC:0.86(0.77–0.94).

Conclusion: We externally validated the clinical risk score for drowning in a nationwide cohort. The diagnostic ability of the risk score can be improved but it needs to be prospectively validated.