Presentation Information
[P055]Use of Change-point Regression Models to Analyze the Effects of Functional Foods ~Comparison with subgroup analysis~
*Suzuka Oyama1, Kohsuke Hayamizu2, Naotaka Yoshida2, Kosei Nishikawa1, Yuki Asahina1, Hirohito Ishikawa1, Yosuke Takimoto1, Hiroe Shinohara3, Masahiro Nakano3 (1. Healthcare Systems Co., Ltd., 2. Lab. Food Chem., Yokohama Univ. Pharm., 3. Shin-Oyama City Hosp., Prev. Med. Ctr.)
Keywords:
Change-point Regression Models,Functional food,Adlay,Stratum corneum water,Human clinical trial
【Purpose】
In functional food intervention studies, the study population is often a mixed population of healthy individuals and/or less healthy but not sick people. For such populations, the intervention effects of food are not constant, and the effectiveness in healthy individuals tends to be smaller than in those in the borderline range. Conventional t-tests and similar methods assume homogeneity of effects, but Change-point Regression Models (CPRMs) have been proposed to address this heterogeneity.
CPRM is an analytical method that takes into account the point of change by the treatment. In this study, we examined the results obtained when subgroup analysis using conventional t-tests was performed around the point of change calculated by CPRM.
【Methods】
We analyzed stratum corneum water content of the study subjects using the following CPRM model formula.
yi = α + β1xi + β2 I(xi > xcp)(xi - xcp)(1 - gi) + εi
Here, xi; the pre value of the ith subject, yi ; the post value, gi; group information (Adlay tea group = 1, Placebo group = 0) and xcp; the point of change, I(·); an indicator function (1 when xi > xcp is true and 0 otherwise).
【Results】
Subgroup analysis conducted around the CPRM-estimated point of change (24.8 (a.u.)) did not reveal any association, such as the p-value was minimum when the subgroup analysis was performed at the point of change. Consequently, it was considered difficult to estimate the change point based on the results of the subgroup analysis. This is likely due to CPRM and the t-test examine different objects and the p-value is affected by multiple factors such as sample size. These results suggest that CPRM could be a powerful tool for a deeper understanding of the intervention effects of functional foods.
In functional food intervention studies, the study population is often a mixed population of healthy individuals and/or less healthy but not sick people. For such populations, the intervention effects of food are not constant, and the effectiveness in healthy individuals tends to be smaller than in those in the borderline range. Conventional t-tests and similar methods assume homogeneity of effects, but Change-point Regression Models (CPRMs) have been proposed to address this heterogeneity.
CPRM is an analytical method that takes into account the point of change by the treatment. In this study, we examined the results obtained when subgroup analysis using conventional t-tests was performed around the point of change calculated by CPRM.
【Methods】
We analyzed stratum corneum water content of the study subjects using the following CPRM model formula.
yi = α + β1xi + β2 I(xi > xcp)(xi - xcp)(1 - gi) + εi
Here, xi; the pre value of the ith subject, yi ; the post value, gi; group information (Adlay tea group = 1, Placebo group = 0) and xcp; the point of change, I(·); an indicator function (1 when xi > xcp is true and 0 otherwise).
【Results】
Subgroup analysis conducted around the CPRM-estimated point of change (24.8 (a.u.)) did not reveal any association, such as the p-value was minimum when the subgroup analysis was performed at the point of change. Consequently, it was considered difficult to estimate the change point based on the results of the subgroup analysis. This is likely due to CPRM and the t-test examine different objects and the p-value is affected by multiple factors such as sample size. These results suggest that CPRM could be a powerful tool for a deeper understanding of the intervention effects of functional foods.
Comment
To browse or post comments, you must log in.Log in