Table 2. Characteristics of Derivation or Validation Models for Clinical Prediction of Foreign Body Aspiration.

Source	No. of candidate predictors (EPV)a	Predictors in final model	Model method, presentation	Model performance measures
Heyer et al,32 2006; derivation	20 (6.1)	Witnessed choking crisis Focal hyperinflation on CXR WBC count, >10 000 ×103/μL	Multivariable logistic regression Clinical decision tool based on scoring system Recommend bronchoscopy with >2 risk factors	Cumulative proportions of FBA by number of risk factors: None: 16% 1: 40%-60% 2: 80%-100% 3: 100%
Janahi et al,33 2017; derivation	26 (3.5)	Witnessed choking (1)b Noisy breathing/stridor/dysphonia (1) Wheeze (2) Abnormal CXR findings (1) Unilateral reduced air entry (2)	Multivariable logistic regression Clinical algorithm with scoring system Consider bronchoscopy with score ≥2	Score ≥2: Sensitivity, 89.1%; PPV, 41.3% Specificity, 45.0%; NPV, 90.4% LR+ 1.6; LR− 0.24 C statistic, 0.756
Haller et al,31 2018; derivation	16 (1.4)	Choking Cyanosis Apnea Decreased breath sounds Atelectasis Mediastinal shift Air trapping	Multivariable logistic regression Clinical algorithm with scoring system Recommend bronchoscopy with sudden choking + ≥1 factor	Score ≥3c: Sensitivity, 0.59; PPV, 0.91 Specificity, 0.97; NPV, 0.83 C statistic, 0.94
Kadmon et al,34 2008; derivation	24 (3.3)	All history and physical predictors included in logistic regression model	Multivariable logistic regression Clinical predictive algorithm with score calculator	Score >0.05 (admit for observation): sensitivity, 100%; specificity 35% Score >0.3 (consider flexible bronchoscopy): sensitivity, 94.5%; specificity, 70% Score = 1 (consider rigid bronchoscopy): sensitivity, 77%; specificity, 81% C statistic, 0.884 Hosmer-Lemsehow test, P = .68
		Radiographic variables included: Unilateral findings on CXR (atelectasis or infiltrate) Bilateral findings on CXR Unilateral hyperinflation Obstructive emphysema Suspicious tracheal radiography	Consider flexible bronchoscopy with score >0.3
Stafler et al,36 2020; validation	21 (0.6)	All history, physical examination, and radiographic predictors included in logistic regression model	Multivariable logistic regression Clinical predictive algorithm with score calculator Recommend diagnostic flexible bronchoscopy with score >0.6	Score >0.6 (high risk, diagnostic bronchoscopy): Sensitivity, 100% Specificity, 41% C statistic, 0.74
Özyüksel et al,35 2020; derivation	15 (25)	Physical examination: Normal findings (0) Wheezing, stridor (1) Decreased breath sounds on 1 side (2) Cyanosis-respiratory insufficiency (2) Radiological findings: Normal CXR (0) Hyperinflation on 1 side (1) Shift in mediastinum (2) Foreign body in chest CT (3) Opaque foreign body on CXR (3)	Multivariable logistic regression, clinical scoring system Recommend bronchoscopy with score ≥2	Score >2: Sensitivity, 77.9% Specificity, 74.8% PPV, 77.0% NPV, 75.7% C statistic, 0.816
Zaupa et al,37 2009; derivation	8 (3.1)	All variables selected, presence of 1 variable within category resulted in positive binary decision outcome (positive clinical and/or radiographic findings)	Binary decision tree clinical algorithm Recommend bronchoscopy with positive clinical and/or pathological radiography findings	Sensitivity, 80% Specificity, 95% PPV, 0.85 NPV, 0.93 C statistic, 0.95c

Abbreviations: CT, computed tomography; CXR, chest radiography; EPV, events per variable; FBA, foreign body aspiration; LR, likelihood ratio; NPV, negative predictive value; PPV, positive predictive value.

^aNumber of outcome events divided by number of candidate predictor variables.

^bWeighted risk scores.

^cMeasure values calculated.