JMA Journal

Figure 1. Flowchart of the study subjects.

Figure 2. Single trajectories for each lifestyle behavior. A) shows the current smoking habits, B) indicates the frequency of alcohol drinking, and C) indicates regular exercise habits. We used a logit model because A) and C) were binary variables. The closer it is to 1, the greater the probability of the behavior, and vice versa for 0. B) used a Poisson regression model as it is a categorical variable. The y-axis represents the probability of being in each category. The “%” shown in the plot legend indicates the percentage of subjects belonging to each trajectory.

Figure 3. Multitrajectories of current smoking, frequency of alcohol consumption, and regular exercise. We applied the multitrajectory model to subjects other than those in Group 0 (nonsmokers, non-daily drinkers, and regular exercisers from the 1st to the 10th checkup). As a result, 2,423 subjects had no trajectory data in one or more models. This figure arranges the graph by health behavior in the rows and by the trajectory pattern group identified in this analysis in the columns. The y-axis of each plot represents the probability of a health behavior. We used a logit model as current smoking and regular exercise are binary variables. The closer to 1, the greater the probability of the behavior, and vice versa for 0. A Poisson regression model was used for drinking as it is a categorical variable. The y-axis denotes the probability of being in each category. As an example of interpretation, Group 1, which represents 9.9% of the analyzed population, tended to be almost all nonsmokers throughout the observation period. Approximately 40% of the subjects had regular exercise habits at the 4^th checkup.