1574 - Multivariate Regression Analysis to Determine Independent Predictors of Treatment Outcomes in the RESTORE-IMI 2 Trial

Background: In the RESTORE-IMI 2 trial, imipenem/cilastatin/relebactam (IMI/REL) was non-inferior to PIP/TAZ for treating hospital-acquired/ventilator-associated bacterial pneumonia (HABP/VABP) in the primary endpoint of Day 28 all-cause mortality (D28 ACM) and the key secondary endpoint of clinical response (CR) at early follow-up (EFU; 7-14 d after end of therapy). We performed a multivariate regression analysis to determine independent predictors of treatment outcomes in this trial.

Methods: Randomized, controlled, double-blind, phase 3, non-inferiority trial comparing IMI/REL 500 mg/250 mg vs PIP/TAZ 4 g/500 mg, every 6 h for 7-14 d, in adult patients (pts) with HABP/VABP. Stepwise-selection logistic regression modeling was used to determine independent predictors of D28 ACM and favorable CR at EFU, in the MITT population (randomized pts with ≥1 dose of study drug, except pts with only gram-positive cocci at baseline). Baseline variables (n=19) were pre-selected as candidates for inclusion (Table 1), based on clinical relevance. Variables were added to the model if significant (p < 0.05) and removed if their significance was reduced (p > 0.1) by addition of other variables.

Results: Baseline variables that met criteria for significant independent predictors of D28 ACM and CR at EFU in the final selected regression model are in Fig 1 and Fig 2, respectively. As expected, APACHE II score, renal impairment, elderly age, and mechanical ventilation were significant predictors for both outcomes. Bacteremia and P. aeruginosa as a causative pathogen were predictors of unfavorable CR, but not of D28 ACM. Geographic region and the hospital service unit a patient was admitted to were found to be significant predictors, likely explained by their collinearity with other variables. Treatment allocation (IMI/REL vs PIP/TAZ) was not a significant predictor for ACM or CR; this was not unexpected, since the trial showed non-inferiority of the two HABP/VABP therapies. No interactions between the significant predictors and treatment arm were observed.

Conclusion: This analysis validated known predictors for mortality and clinical outcomes in pts with HABP/VABP and supports the main study results by showing no interactions between predictors and treatment arm.Table 1. Candidate baseline variables pre-selected for inclusion
Figure 1. Independent predictors of greater Day 28 all-cause mortality (MITT population; N=531)
Figure 2. Independent predictors of favorable clinical response at EFU (MITT population; N=531)