BACKGROUNDClinical trials using American College of Chest Physicians/Society of Critical Care Medicine Consensus sepsis definitions as entry criteria fail to reduce septic mortality. We hypothesized that the systemic mediator-associated response test (SMART) methodology could match sepsis therapies biologically to individual patients by relating baseline data statistically to outcomes and treatment effects. This article reports the SMART analyses of four failed sepsis investigations.METHODSDatabases from the E5 antiendotoxin antibody, North American Sepsis Trial (NORASEPT) and NORASEPT II anti-tumor necrosis factor antibody (TNFMAb), interleukin (IL)-1ra, and platelet-activation factor acetylhydrolase (PAF-AH) sepsis clinical trials were evaluated with SMART using multivariate logistic regression. From baseline data, within each study, mortality prediction models were built separately for the placebo and active drug populations. Subjects among whom each drug's effects were greatest were then identified by excluding from efficacy analysis subjects predicted by SMART to survive on placebo or to expire on active drug. Finally, prerandomization data from patients in each study were entered into SMART models, and placebo or active drug treatment effects were evaluated for parent populations and SMART cohorts.RESULTSE5-consensus mortality: 27.4% placebo, 26.2% E5; SMART mortality: 17.1% placebo, 8.0% E5 (p < 0.01). NORASEPT-consensus mortality; 33.4% placebo, 29.5% TNFMAb; SMART mortality: 47.2% placebo, 34.7% TNFMAb (p = 0.03). IL-1ra-consensus mortality: 33.9% placebo, 29.8% IL-1ra; SMART mortality: 55.6% placebo, 34.9% IL-1ra (p < 0.001). PAF-AH-consensus mortality: 22.4% placebo, 23.9% PAF-AH; SMART mortality: 17.7% placebo, 28.9% PAF-AH (p = 0.039).CONCLUSIONSUsing prerandomization clinical trial data, SMART identifies septic patients whose host-inflammatory responses can benefit from specific drugs. SMART also predicts ineffective drugs and patients whom they might harm.
