Assessment of a Precision Medicine Analysis of a Behavioral Counseling Strategy to Improve Adherence to Diabetes Self-management Among Youth

Abstract

Key Points Question Are there subgroups of participants in the Flexible Lifestyles Empowering Change (FLEX) trial for whom the intervention is estimated to be optimal, for whom usual care is estimated to be optimal, and for whom control conditions and intervention are estimated to be equivalent? Findings In this post hoc analysis of the FLEX randomized clinical trial of 258 adolescents with type 1 diabetes, an individualized treatment rule showed that a large proportion of participants had equivalent predicted outcomes under intervention vs usual care settings, regardless of randomization. Meaning The precision medicine approach is a conceptually and analytically novel method for post hoc subgroup analysis of randomized clinical trials, capturing data-driven response subgroups without relying on predefined categories.

Estimation and evaluation of individualized treatment rules following multiple imputation

“…$$ After estimating the regression parameters in () for each $$$ r\in \left\{1,\dots, R\right\} $$$, the corresponding estimate of the optimal decision rule for imputed data set $$$ r $$$ is $$$ {\hat{d}}_r^{\ast}\left(\boldsymbol{x}\right)=\operatorname{sign}\left({\hat{\beta}}_{2,r}+{\boldsymbol{x}}^T{\hat{\beta}}_{3,r}\right) $$$. One way to obtain a single recommended treatment from the $$$ R $$$ estimated decision rules is via majority voting which has been employed in several applications 42,43 . When a common linear model is specified for each $$$ {Q}_r\left(\boldsymbol{x},a\right) $$$, an alternative to majority voting is a model averaged decision rule given by $$$$ {\hat{d}}_{\mathrm{MA}}^{\ast}\left(\boldsymbol{x}\right)\triangleq \operatorname{sign}\left({\overline{\hat{\beta}}}_{2\cdotp }+{\boldsymbol{x}}^T{\overline{\hat{\beta}}}_{3\cdotp}\right), $$$$ where …”

“…One way to obtain a single recommended treatment from the R estimated decision rules is via majority voting which has been employed in several applications. 42,43 When a common linear model is specified for each Q r (x, a), an alternative to majority voting is a model averaged decision rule given by…”

“…We now describe the approach used to determine the number of imputed data sets for our analysis. Although MI has been proposed and applied in the context of developing optimal treatment strategies, 9,42 we are unaware of any existing guidance on how to choose an adequate number of imputed data sets when targeting estimation and evaluation of an optimal ITR or DTR. Until recently, $$$ R=2 $$$ to $$$ R=10 $$$ imputations were thought to be adequate, 52 particularly with respect to relative efficiency of point estimates.…”