2DkcTXceO

T.L. Lai 371(Rush et al., 2004) that compares seven treatment options in patients who didnot attain a satisfactory response with citalopram, an inhibitor antidepressant.After receiving citalopram, participants without sufficient symptomatic benefitwere eligible for randomization among these options. A novel feature ofthe study design is that it ascertains before randomization the set of optionsthat the patient-clinician dyad consider to be equally reasonable, given thepatient’s preferences, and his or her state after a trial of citalopram. Thisset of options characterizes the patient’s Equipoise Stratum (ES). A total of1429 patients were randomized under this scheme. The largest ES were the“Medication Switch Only” group, allowing randomization among the threemedications (40%) and the “Medication Augmentation Only,” allowing randomizationbetween two options (29%). The “Any Augmentation” (10%) and“Any Switch” (7%) were the next largest, and only 5% of patients were randomizedamong options that contrasted a switch and augment condition. Inretrospect, it became clear that patients (and their clinicians) were roughlydivided into two groups, those who obtained partial benefit from citalopramand therefore were interested in augmentation, and those who obtained nobenefit and were interested only in switching. Thus, the ES design allowedthe study to self-design in assigning patients to the parts of the experimentthat were relevant to current practice and to patient preferences.A third approach is to design point-of-care (POC) clinical trials which canbe regarded as experiments embedded into clinical care; see Fiore et al. (2011)on a VA-sponsored trial that compares the effectiveness of two insulin dosingregimens for hospitalized diabetic patients. In POC trials, subjects are randomizedat the health care encounter, clinician equipoise defines the referencepopulation, and baseline and/or outcome data are captured through electronicmedical records. By using outcome-adaptive randomization, POC trials integrateexperimentation into implementation and learn sequentially the superiortreatment(s). This is similar to the classical multi-arm bandit problem (Lai,2001) except that POC uses adaptive randomization to implement it in aclincial setting with clinician equipoise and patient’s informed consent. Groupsequential generalized likelihood ratio tests with efficient outcome-adaptiverandomization to multiple arms have recently been developed and can beused for POC trials (Lai and Liao, 2012; Shih and Lavori, 2013).33.3 Innovative clinical trial designs in translationalmedicineBesides CER studies, Lai and Lavori (2011) describe novel design methodsfor clinical studies of personalized treatments and targeted cancer therapies intranslational medicine. “From bench to bedside” — in which “bench” refers tolaboratory experiments to study new biochemical principles and novel molec-