The Mount Sinai Acute GVHD International Consortium (MAGIC) algorithm probability (MAP), derived from 2 serum biomarkers, measures damage to crypts in the gastrointestinal tract during graft-versus-host disease (GVHD). We hypothesized that changes in MAP after treatment could validate it as a response biomarker. We prospectively collected serum samples and clinical stages of acute GVHD from 615 patients receiving hematopoietic cell transplantation in 20 centers at initiation of first-line systemic treatment and 4 weeks later. We computed MAPs and clinical responses and compared their abilities to predict 6-month nonrelapse mortality (NRM) in the validation cohort (n = 367). After 4 weeks of treatment, MAPs predicted NRM better than the change in clinical symptoms in all patients and identified 2 groups with significantly different NRM in both clinical responders (40% vs 12%, P < .0001) and nonresponders (65% vs 25%, P < .0001). MAPs successfully reclassified patients for NRM risk within every clinical grade of acute GVHD after 4 weeks of treatment. At the beginning of treatment, patients with a low MAP that rose above the threshold of 0.290 after 4 weeks of treatment had a significant increase in NRM, whereas patients with a high MAP at onset that fell below that threshold after treatment had a striking decrease in NRM that translated into clear differences in overall survival. We conclude that a MAP measured before and after treatment of acute GVHD is a response biomarker that predicts long-term outcomes more accurately than change in clinical symptoms. MAPs have the potential to guide therapy for acute GVHD and may function as a useful end point in clinical trials.