|
Abstract:
|
Precision nutrition is an emerging research field in nutritional sciences. Deriving a personal nutrition intervention scheme (PNIS) based on effects of treatment on health is a difficult problem in precision nutrition, where multiple benefit outcomes are often available. Two major challenges arise in the development of PNIS with multiple clinical outcomes, including heterogeneous multidimensional outcomes and complex missing data patterns. This paper is motivated by a clinical trial that aims to assess the effect of calcium supplementation for pregnant women on reducing infant's in utero exposure to lead. We propose to integrate different types of blood lead concentration measurements of varying clinical relevance and sample sizes in the training of PNIS. We develop an extended support vector machine (SVM), named Synergistic Self-learning (SS-learning), that allows us to synergize heterogeneous training data sources in a weighted self-learning paradigm. We establish the algorithmic convergence of the proposed SS-learning and illustrate the performance of this methodology through both simulation studies and real data analysis of the motivating calcium supplementation clinical trial.
|
