Quantile regression based method for characterizing risk-specific behavioral patterns in relation to longitudinal left-censored biomarker data collected from heterogeneous populations.

There are many studies aimed at promoting positive lifestyle behaviors to reduce lifetime risk of cancer and related diseases. However, assessing these modifiable behaviors through statistical modeling is challenging because of the multidimensionality of interrelated measurements that may dramatically differ between at-risk individuals. Taking into account this heterogeneity while considering the multidimensionality of behavior changes is fundamental to tailoring interventions to their needs. Biomarkers that identify high-risk individuals may help validate proximal measures, but the number of validated methods that link biomarkers to multiple behavioral measurements by determining their dynamic relations with disease risks is limited to just a few, since it requires an advanced statistical methodology to address challenges in analyzing biomarker data, including left-censoring due to limits of detection. To address these challenges, we propose a method that constructs a quantile-specific weighted index of multiple behavioral measurements. Under the quantile regression framework, the proposed method renders a multidimensional view of risk-specific behavioral patterns by connecting them with biomarker levels to provide better insights into heterogeneous behavioral profiles among at-risk populations. We evaluate performances of the proposed method through simulations, and illustrate its applications to the Tu Salud ¡Sí Cuenta! data by examining behavior changes among Mexican-American adults.