This preview has intentionally blurred sections. Sign up to view the full version.View Full Document
Unformatted text preview: Some Current Research Topics in Clinical Trials Recent issues of statistical journals show that clinical trials remain a very active and fruitful area for statistical research. All the examples here are taken from recent issues of Biometrics . Some other journals that often have papers in this area are JASA , Stat in Med , and Biometrika . Example 1: Phase I trials to determine schedule of cytotoxic agent We have focused on Phase I trials to determine maximum-tolerated dosage (MTD). In these trials, the outcome is a binary response (unacceptable toxicity) and the study design an increasing series of possible dosages. Braun et al. considered the setting where physicians administer an agent repeatedly and monitor long-term toxicity. Clinical setting for the example from their paper: • Patients with leukemia who receive bone marrow transplant (BMT) from a donor. • The desired response: T and natural killer cells coordinate an immune response that kills residual leukemia cells, or graft-vs.-leukemia (GVL). • Bad side effect: these same cells also evoke an immune response against patient’s own skin, liver, and gastro-intestinal tract, graft-vs.-host-disease (GVHD). • A recombinant human growth factor, KGF) has been shown to reduce chemo or ra- diation therapy induced damage to GI tract, in treatment for colon cancer, and may have this effect in leukemia patients who get MTD. • But KGF has its own toxicities that can cause damage to pancreas. • KGF can be administered safely on 3 consecutive days, then must have some days off. How many times can you do this? Statistical model for a new approach to getting maximum tolerated schedule in this setting: • Define the outcome for patient i to be T i , the time to toxicity. • We allow this to be censored by end of study. • The times (on the patient’s cloc k) when the patient gets the drug are s ij . • The hazard of toxicity is assumed to be a sum of a sequence of hazards, each associated with one administration. • They assume a parametric hazard contribution, triangle-shaped, that starts at 0 before agent is given, rises to a peak of θ 2 at time θ 1 after administration, then declines linearly, returning to 0 at time θ 3 . • The goal is to ensure that you pick a dosage schedule that keeps the probability of a response before τ days after initiation of treatment to no more than a fixed value p τ . Estimation approach: They establish a prior distribution for θ that is uninformative, more or less, based on physician comments. They estimated parameters via MCMC. Their results are somewhat sensitive to the prior distribution tuning parameters. Note that conduct of the trial is sequential, with continuous monitoring of patients. Results of simulation helped to tune their procedure under various scenarios....
View Full Document
This note was uploaded on 12/30/2010 for the course BST 252 taught by Professor Tsodikov during the Winter '06 term at UC Davis.
- Winter '06