Abstract: Objective To understand dietary pattern and to explore the development of dietary pattern and the related metabolic indices for elderly people in Beijing. Methods The baseline information in 1992 and information in the follow-up study in 2000 for Beijing Longitudinal Study of Aging, which was collected by Xuanwu Hospital, Capital Medical University was used. There were 698 elderly subjects in this analysis who fulfilled the blood test and with full information on dietary habits both in the year of 1992 and 2000. Factor analysis and cluster analysis were used to reveal dietary pattern. Latent class analysis(LCA) was a technology that discriminate people to different latent class with different answering pattern for items(manifest variables or observed variables). LCA was first used in psychiatry, education or psychology. Recent, the use of LCA in dietary pattern analysis was reported. In this research, a self-made questionnaire for dietary habits was used to collect the information in recent 3 month. It covered the frequency for eating breakfast, the amount of different types of foods. The entire variable used were turned to be binary to get a clearer result. LCA was used to extract diet pattern: ANOVA, paired t test and chi-square test were used to analyze the difference of metabolic indicator among different patterns. Results Three types of diet were found: group 1(n=314, 45.0%) had sufficient nutrients; group 2(n=259, 37.1%) and group 3(n=125, 17.9%). As compared to 1992, had excessive intake of meat. In 2000, there was a rise in animal oil consumption, fresh fruit intake, fish or shrimp intake in group 2 and group 3. There was a decrease in egg intake in group 1 and group 3. Generally, the intake of milk fruit bean product was on an increase. There was statistically significant increase in BMI, SBP, FPG, TC and TG in 2000 compared with the value in 1992. Conclusion LCA might be a solution for dietary pattern extraction. There were mainly three type of dietary pattern that developed with age and impacted the related metabolic indices. Multivariable analysis was especially needed to exclude the impact of potential confounding factors.

Key words: dietary pattern, latent class analysis, metabolic disorder, longitudinal study