|
Abstract:
|
Suppose we have some measurements of a response variable for many people taken over a period of time. We hypothesized that there might be a finite number of distinct groups of people with different mean trajectories over times. For example, we can have a few dozen systolic blood pressure (SBP) measurements taken over the course of a day for several dozens of people, resulting in hundreds or thousands of total data points. In this context, we might hypothesize there are a small number of daily patterns of blood pressure. Given the data and the number of groups the user requires, we want to identify the patterns and the probability that each person belongs to each pattern. We developed an algorithm that will fit a mixture of nonparametric regressions and apply it to data containing repeated measurements over time. The algorithm is a pseudo-EM algorithm using kernel regression.
|
