Optimal and Efficient Repeated Measurements (Crossover) Designs

Co-Principal Investigators: Howard Kushner, Ph.D., Morris Meisner, Ph.D., Eugene Laska, Ph.D.

PROJECT GOALS

When the aim of a randomized trial is the comparison of two or more interventions, it is best accomplished by using optimal or efficient experimental design. Such designs provide accurate measurements of the interventions’ effects. Repeated measurements (crossover) designs have advantages when they can be employed because each experimental subject is his own control, and also because of their cost advantages. To specify an optimal or efficient repeated measurements design before the experiment begins, however, one must know the correlation structure – which is generally not known to the practitioner – of a given subject’s observations. A design that is optimal or efficient under one correlation structure may, unfortunately, be quite inefficient under a different correlation structure. In this period, the project is concerned with the response-adaptive approach to the specification of efficient crossover designs. In this approach, a design is specified as the accrued information in previous responses is analyzed. A related aim concerned the statistical testing for treatment effects in response-adaptive designs. During this period our specific aim was to publish our response-adaptive allocation methods in a statistical journal concerned with experimental design and inference.

RESEARCH ACTIVITIES AND RESULTS

A paper was published that gives a method for adaptively allocating new subjects to treatment sequences in repeated measurements designs. Simulations indicated that the efficiency of the resulting designs increases with N (the number of subjects), avoiding the low efficiency and poor power that can result from an unfortunate design choice. An LR test of no treatment differences is presented.

POLICY IMPLICATIONS

Adaptive repeated measurement designs can avoid the possibly seriously inefficient designs that are specified using erroneous values for a subject’s correlation structure

PLANS

In the previous period, the efficiency of our adaptive designs was examined via statistical simulations. In the current period, we will be concerned with proofs that the adaptive designs will actually tend to optimal designs as N increases.

Publication:

Kushner HB (2003). Allocation rules for adaptive repeated measurements designs. Journal of Statistical Planning and Inference 11: 293-313
(Available on-line in PDF format)

Entered: July 17, 2003