TABLE 3.

Comparisons of the predictive value of the HLA classifier versus disease severity and SRS.

Models	Univariable analysis
	OR (95% CI)	p
Use of hydrocortisone in the HLA low-risk subgroup	1.63 (0.21–12.71)	0.644
Use of hydrocortisone in the HLA high-risk subgroup	2.84 (1.07–7.57)	0.037*
Use of hydrocortisone in SRS 2	3.76 (1.41–10.04)	0.008*
Use of hydrocortisone in SRS 1	1.25 (0.47–3.36)	0.658
Use of hydrocortisone by APACHE II	0.94 (0.86–1.03)	0.210
Use of vasopressin in SRS 2	0.69 (0.18–2.62)	0.583
Use of vasopressin in SRS 1	1.50 (0.40–3.89)	0.403
Use of vasopressin in the HLA low-risk subgroup	2.91 (0.29–29.45)	0.366
Use of vasopressin in the HLA high-risk subgroup	1.26 (0.58–2.76)	0.559
Use of vasopressin by APACHE II	0.96 (0.87–1.06)	0.427

The logistic regression models integrating interactions between treatment allocation and SRS, HLA, classifier, or APACHE II, were built in the E-MTAB-7581, dataset. A total of six logistic regression models were built by using mortality as the response variable and respective predictors and interactions were: hydrocortisone & class, hydrocortisone and SRS, hydrocortisone and APACHE II, vasopressin and class, vasopressin and SRS, and vasopressin and APACHE II. A significant ( p <0.05) interaction indicated that the classification method was of predictive value because it identified that a subgroup of patients responded differently to treatment. SRS, classification was used as previously reported. Abbreviations: OR, odds ratio; CI, confidence intervals; MARS, the Molecular Diagnosis and Risk Stratification of Sepsis; DM, diabetes mellitus, APACHE II, acute physiology and chronic health evaluation; SRS, sepsis response signature.