COMPLEX DETERMINANTS OF THE LEAST COSTLY WAY TO DESIGN A STUDY TO DETECT DIFFERENCES IN QALYS WITH A GIVEN POWER

Purpose: The costs to design a randomized controlled trial (RCT) rise with increased sample size and study duration. These costs can be minimized by accounting for complex determinants that affect the power calculation for an RCT designed to detect differences in quality adjusted life years (QALYs).  This study's purpose is to illustrate comprehensively how study costs are affected by the choice of data in such an RCT.

Methods: A conceptual framework was developed to construct a constrained cost minimization function to analyze how data to be collected in an RCT designed to detect differences in QALYs and the sample size affect overall costs of the study.

Results: The dotted rectangle in the conceptual model highlights the common considerations in a sample size calculations: a given power conditioned on an expected standard deviation and calculated Type I error.  The standard deviation of QALYs is affected by the number of observations during a year, the recall period of each observation, and the correlation of a study subject's observations over time (which, in turn, may be related to the recall period of the instrument).  The amount of data to be collected per study subject also affects potential study subjects' willingness to participate and the attrition rate.  The cost function includes the costs of recruiting and retaining study subjects and the standard protocol for enrolled participants.  This is constrained by the need to achieve a given power. The sample size, number of interviews, and recall period of the health utility instrument affect both the cost and power of the study.  The marginal cost of recruitment and retention depends on the frequency of interviews and the recall period for the health utility instrument.

Conclusions: Selecting the sample size based only on the power achieved through the calculated type I error and standard deviations of QALYs will not minimize cost to conduct an RCT. Not accounting for the effects of the frequency of interviews and recall period on the marginal cost of study subjects' recruitment and retention results in a failure to design the least costly study possible to detect a difference in QALYs with a given power.