ABSTRACT: The prognosis of steroid-refractory acute graft-versus-host disease (aGVHD) is poor, and predictors of response and survival are unclear. In an exploratory analysis of 203 steroid-refractory aGVHD patients with prospectively collected GVHD data who received antithymocyte globulin, etanercept, or mycophenolate mofetil (MMF) as second-line treatment, we determined the predictors of day 28 response, 2-year overall survival, and 2-year nonrelapse mortality (NRM). To minimize the risk of finding false-positive results, we used least absolute shrinkage and selection operator regression, aggressively eliminating variables that are unlikely to be associated with outcome. Day 28 response to second-line therapy was 38% (complete response, 23%), with a 2-year overall survival of 25% and a 2-year NRM of 62%. Factors associated with response were GVHD prophylaxis, organ involvement, and initial aGVHD to steroid-refractory aGVHD interval. Specifically, compared with cyclosporine/MMF as GVHD prophylaxis, the odds ratio (OR) for calcineurin inhibitor/methotrexate was .8 and for cyclosporine/prednisone .6. The OR for aGVHD to steroid-refractory aGVHD interval ? 14 versus <14 days was 1.3. The ORs for skin only involvement and gut or liver only involvement when compared with multiorgan involvement were 1.4 and 1.2, respectively. The only variable associated with worse survival was age, with a hazard ratio (HR) per decade of 1.04 for overall mortality. Similarly, age was the only variable associated with NRM (HR per decade, 1.02). When compared with complete response, no response at day 28 increased the risk of death (HR, 2.4; 95% confidence interval, 1.5 to 3.7). In conclusion, by means of an underused statistical technique in the field of transplantation, we identified predictors of response and survival in steroid-refractory aGVHD. Our results highlight the importance of developing novel treatment strategies because current treatments yield poor outcomes.