An epistatic interaction between pre-natal smoke exposure and socioeconomic status has a significant impact on bronchodilator drug response in African American youth with asthma

Table 3 Age Adjusted Higher-Order Interaction Models Associated with BDR identified by ViSEN

A. Full Dataset
				ViSEN Analysis		Logistic Regression
				Age Adjusted		Logistic Regression
Variable 1		Variable 2	Variable 3	IG	p-value	OR	p-value
Experience of Discrimination		Age	Socioeconomic Status	2.39%	0.003	0.029	0.008
Experience of Discrimination		African Ancestry	Socioeconomic Status	1.71%	0.012	0.026	0.040
Experience of Discrimination		NO₂ Air Pollution	Socioeconomic Status	1.80%	0.017	0.043	0.013
Sex		Pre-natal Smoke Exposure	Socioeconomic Status	1.82%	0.041	1.525	0.817
Experience of Discrimination		Pre-natal Smoke Exposure	NO₂ Air Pollution	1.66%	0.043	–	–
B. Sex-Stratified Subsets
				ViSEN Analysis		Logistic Regression
				Age Adjusted		Logistic Regression
Subset	Variable 1	Variable 2	Variable 3	IG	p-value	OR	p-value
Female	Experience of Discrimination	Pre-natal Smoke Exposure	African Ancestry	2.73%	0.018	1.000	1.000
	Pre-natal Smoke Exposure	Socioeconomic Status	NO₂ Air Pollution	4.92%	0.030	0.091	0.413
	Experience of Discrimination	Pre-natal Smoke Exposure	Age	4.01%	0.039	–	–
Male	Experience of Discrimination	African Ancestry	Socioeconomic Status	5.11%	0.001^a	0.004	0.005
	Experience of Discrimination	NO₂ Air Pollution	Socioeconomic Status	3.95%	0.008	0.012	0.015
	Experience of Discrimination	Age	Socioeconomic Status	3.90%	0.012	0.012	0.017
	Experience of Discrimination	African Ancestry	NO₂ Air Pollution	2.44%	0.038	–	–

Information Gain (IG) and unadjusted permutation p-value results for select interaction models associated (permutation unadj. p < 0.05) with BDR identified by the age adjusted ViSEN and logistic regression analyses. Positive IG values indicate synergistic interactions, negative IG values indicate redundant models. The Bonferroni method was used to correct for multiple testing. Bonferroni familywise error rate (FWER) thresholds of 0.05 and 0.1 were used to determine significantly associated models (permutation unadj. p ≤ 0.0009) and suggestively associated (permutation unadj. p ≤ 0.002), respectively. Models significantly associated with BDR after correction for multiple testing are highlighted in BOLD. Models suggestively associated with BDR after correction for multiple testing are indicated with ^a. Logistic regression analysis was performed in the same LCC age adjusted dataset as VISEN analyses for accurate comparison of results; to maintain consistency with ViSEN analyses regression models were adjusted for the marginal effects of each independent variable included in the specified interaction model. When interaction bin size was < 5, Firth’s Bias-Reduced logistic regression OR and p-values are presented. ---: model could not be accurately assessed by regression modeling due to deviation from model assumptions of no collinearity and no complete or quasi-complete separation; OR: Odd’s Ratio

ISSN: 1756-0381