Adaptive decision making in a lymphocyte infusion trial

We describe an adaptive Bayesian design for a clinical trial of an experimental treatment for patients with hematologic malignancies who initially received an allogeneic bone marrow transplant but subsequently suffered a disease recurrence. Treatment consists of up to two courses of targeted immunotherapy followed by allogeneic donor lymphocyte infusion. The immunotherapy is a necessary precursor to the lymphocyte infusion, but it may cause severe liver toxicity and is certain to cause a low white blood cell count and low platelets. The primary scientific goal is to determine the infusion time that has the highest probability of treatment success; defined as the event that the patient does not suffer severe toxicity and is alive with recovered white blood cell count 50 days from the; start of therapy. The method is based on a, parametric model accounting for toxicity, time to white blood cell recovery; and survival time. The design includes an algorithm for between-patient immunotherapy dose de-escalation based on the toxicity data and an adaptive randomization among five possible infusion times according to their most recent posterior success probabilities. A simulation study shows that, the design reliably selects the best infusion time while randomizing greater proportions of patients to superior infusion times.