ATHOS-3 trial

The ATHOS-3 trial has been previously described (clinicaltrials.gov identifier NCT 02338843) [8]. Briefly, adults with persistent vasodilatory shock after ≥ 25 mL/kg of volume resuscitation requiring high-dose vasopressors (i.e., norepinephrine-equivalent dose [NED] > 0.2 μg/kg/min) were randomly assigned 1:1 to receive synthetic human angiotensin-II (La Jolla Pharmaceutical Co.) or saline placebo plus standard vasopressors. Randomization was stratified by MAP at screening and Acute Physiology and Chronic Health Evaluation II (APACHE-II) score.

Objectives

The present study reflects a post-hoc analysis of ATHOS-3 that aimed to answer the following questions:

1.

Does prior exposure to ARB or ACEi therapy alter the efficacy of angiotensin-II treatment for hypotension in vasodilatory shock?

2.

Does prior exposure to ARB or ACEi therapy, alter the systemic RAS biomarker profile in vasodilatory shock and the RAS response to angiotensin-II treatment?

Exposures

The co-primary exposures of interest were exposure to ARB or ACEi therapy within the 7 days prior to randomization, which were assessed as potential effect modifiers for angiotensin-II treatment. Exposure was determined through a combination of history taken from patients or their representative and review of the electronic/patient medical record by study personnel.

As an exploratory analysis, the degree of ARB exposure was further assessed as a continuous variable by determining the equipotent dose levels in losartan equivalents (Additional file 1: Table-S1), using previously established conversion factors [10].

In the ATHOS-3 trial, study drug infusion was started at 20 ng/kg/min and titrated during hr0-hr3 to achieve MAP ≥ 75 mmHg while keeping other vasopressor doses constant. Thereafter, study drug and other vasopressors were titrated at treating clinicians’ discretion to maintain MAP between 65 and 75 mmHg. At hr48, study drug infusion was discontinued according to a protocol-specified tapering process but could be continued for up to 7 days per clinician discretion.

Outcomes

The primary outcome was MAP at hr1. This was selected because we reasoned the immediate MAP change best reflected the hemodynamic efficacy of angiotensin-II and because using a continuous variable (rather than the binary primary endpoint of the ATHOS-3 trial) would reduce the sample size needed to statistically interrogate an interaction effect. Additionally, we compared the NED and the study-drug titration level during hr0-3 (i.e., during the titration period where background vasopressors were held constant), and during hr4-hr48 (i.e., the period where background vasopressors were not held constant). To investigate the association of prior ACEi and ARB exposure with the biomolecular RAS profile, we compared the baseline levels of renin, angiotensin-I, angiotensin-II, and the angiotensin-I/angiotensin-II ratio, and the change in renin at hr3.

Patient-centered outcomes (e.g., 28-day mortality) were not analyzed due to insufficient sample size.

Statistical analysis

Continuous variables are reported as mean (SD) or median (interquartile range), as appropriate, and categorical variables as frequency (percent). Analyses were performed in SAS (SAS Institute, Cary, NC, USA.)

In the primary analysis, MAP and NED at hr1-hr3 were modeled in a linear regression that included terms for treatment and hr0 value of the dependent variable in accordance with best statistical practice for evaluating continuous outcomes in clinical trials [11, 12]. These models were performed when stratified by exposure status (ARB, ACEi, no ACEi/ARB), and including an interaction-term for whether exposure modified the effect of treatment.

For longitudinal modeling of NED and study drug dose, a mixed-effects repeated measures model was used with a random intercept for individual patients and unstructured covariance. The model included fixed-effects for time, treatment, ACEi and ARB exposure, as well as interaction effects between these terms to assess heterogeneity of treatment effect. Models were separately constructed for the active titration period (hr0-3) and the subsequent intervention period (hr4-48). These models have the advantage of being robust to missing data bias, even when data are not missing at random [13]. To account for right-skewed data in NED, data were log-transformed for modeling.

Differences in baseline levels of RAS biomarkers were compared with generalized linear models with post-hoc adjustment for multiple comparisons. Again, to account for right-skewed data, biomarker levels were log-transformed for analysis. To facilitate intuitive interpretation, we report back-transformed differences and least-squares means (i.e., the geometric mean [gmean]).

We used two sets of multivariable models to account for potential confounding in ACEi/ARB exposure status. The primary analysis models used a parsimonious approach, adjusted for age, sex, baseline APACHE-II score (as a measure of overall illness severity), and baseline NED (as a measure of circulatory failure severity, specifically). These covariates were selected prior to analysis as they were felt to reflect the most important sources of potential confounding. In sensitivity analyses, we also assessed the main outcomes using models that adjusted for the previous covariates as well as baseline MAP, history of chronic kidney disease, chronic hypertension, and factors that had previously been associated with response to angiotensin-II [8, 14,15,16]: baseline albumin, ARDS at baseline, and whether patients were on renal replacement therapy (RRT) at baseline. We used complete case-analysis as the first set of models had no missing data and the second set of models had only one covariate with missing data (albumin), which was missing in less than 5.0% of patients in all three exposure groups. We considered the use of inverse-probability of treatment weighting as an additional method of covariate adjustment, but the sample size was too small to facilitate effective balancing of the weighted population. As such we limit the presented analysis to traditional covariate adjustment.

For the exploratory analysis of ARB dose exposure, the analysis dataset was limited to ARB-exposed and ACEi/ARB-unexposed patients. The unexposed patients were considered to have a losartan-equivalent dose of 0 mg/day.