While there is no known cure for Huntington's disease (HD), there are early-phase clinical trials aimed at altering disease progression patterns. There is, however, no obvious single outcome for these trials to evaluate treatment efficacy. Currently used outcomes are, while reasonable, not optimal in any sense. In this paper we derive a method for constructing a composite variable via a linear combination of clinical measures. Our composite variable optimizes the signal-to-noise ratio (SNR) within the context of a longitudinal study design. We also demonstrate how to induce sparsity using a soft-approximation of an L 1 penalty on the coefficients of the composite variable. We applied our method to data from the TRACK-HD study, a longitudinal study aimed at establishing good outcome measures for HD, and found that compared to the existing composite measurement our composite variable provides a larger SNR and allows clinical trials with smaller sample sizes to achieve equivalent power.
Keywords: longitudinal data; mixed effects models; optimization; signal-to-noise ratio; translational science.