RESUMO
BACKGROUND: Reduced rank regression (RRR) combines exploratory analysis with a-priori knowledge by including risk factors in the model. Dietary patterns, derived from RRR analysis, can be interpreted by the chosen risk factor profile and give an indication of positive or adverse health effects for a specific disease. Our aim was to assess the stability of dietary patterns derived by RRR over time. METHODS: We used data from 467 men, aged 64-85 years, participating in the 1985 and 1990 examination rounds of the Zutphen Elderly Study. Backwards regression on risk factors and food groups was applied prior to the RRR analysis to exclude food groups with low predictability (from 36 to 19 food groups) for the chosen risk factor profile. For the final RRR analysis, dietary intake data from 19 food groups as predictor variables and 6 established risk factors for cardiovascular diseases (body mass index, systolic and diastolic blood pressure, high density lipoprotein and total cholesterol levels, and uric acid) were used. RESULTS: Three RRR dietary patterns were derived for both examination years: a "(low in) cereal fibre pattern", an "alcohol pattern" and an "inconsistent pattern". The "(low in) cereal fibre pattern" was most stable over time, with a correlation coefficient of 0.47 (95% CI: 0.38-0.53) between 1985 and 1990 measurements. CONCLUSION: Dietary patterns as measured by RRR, after backwards regression, are reasonably stable over a period of five years. Thus, RRR appears to be an attractive method to measure long-term dietary exposure for nutritional epidemiological studies, with one dietary measurement at baseline.