Research Article
Kunihiko Takahashi, Hiroyuki N
Abstract
In epidemiological studies that measure the risk at different levels of exposure, the data is often only available for the analyses that summarized the response data in grouped exposure intervals. In typical methods, the midpoints are used as the assigned exposure levels for each interval. Results of the analysis with grouped data may be sensitive to the assignment of the exposure levels. In this paper, we propose a procedure for assessing J-shaped associations based on the likelihood-based assignment of values to grouped intervals of exposure, and applying the cubic spline regression models. Numerical illustrations and comparisons based on simulations showed that the proposed procedure can yield better estimates for curves than those obtained using the typical assignment method based on the midpoints of each interval.