Table 2.

GLM scenario comparison.

Scenario 1	Poisson Regression		Lasso Poisson Regression
	AIC	BIC	AIC	BIC
Scenario 1: main effects, no interaction terms	39,548	39,593	39,559	39,612
Scenario 2: main effects, top significant two-way interaction identified by CTree analysis	39,531	39,592	39,559	39,627
Scenario 3: main effects, all significant two-way interactions identified by CTree analysis	39,525	39,601	39,563	39,608
Scenario 4: main effects, all significant two- and three-way interactions identified by CTree analysis	39,520	39,618	39,544	39,604

Abbreviations: AIC, Akaike information criterion; BIC, Bayesian information criterion; CTree, conditional tree; GLM, generalized linear model. ¹ Refer to text for complete description of each scenario. The final GLM model that we applied is described in Equation (3): Y = Exp(0.0916 + β₁Race + β₂MaxDailyPM2.5Exposure_StudyMax_cat + β₃(Prednisone * Race) + β₄ObesityDx) (3), where Y = predicted annual number of ED or inpatient visits for respiratory issues; β₁ = estimated model parameters for Race; β₂ = estimated model parameters for MaximumDailyPM2.5Exposure_StudyMax; β₃ = estimated model parameters for interaction between Prednisone and Race; and β₄ = estimated model parameters for ObesityDx. Bold font was used to highlight the metrics that were used to select the final model.