Participants for the study were recruited through media advertisement and announcements, as well as invitation letters to memory clinic patients and participants from previous studies by the research team.

Participants were eligible for the study if they:

Potential participants were excluded if they:

All participants provided written informed consent. The study was approved by the Royal Perth Hospital Ethics committee, reference EC 2007/079 in August 2007.

Nine hundred potential participants responded to the recruitment strategies for the randomised controlled trial. Over half were excluded following screening by phone, with 374 attending the screening assessment, and 249 passing the screening assessment and had a sample of blood taken to confirm vitamin D levels. The final sample in this study was 249, with 155 men and 94 women.

An 107-item self administered food frequency questionnaire (FFQ), adapted from the validated 1995 National Nutrition Survey, assessed usual frequency of intake of food and beverages over the previous 4 weeks [13]. Each item had a choice of 9 frequency categories ranging from “never or less than once per month” to “6 or more times per day”, but did not include any information on serving size. The 107 items of the FFQ were grouped into 33 predefined food groups by adding food items within each group eg Other vegetables included 12 items: vegetable, tomato juice; stir-fried or mixed vegetables; vegetable casserole; pumpkin; sweet potato; carrots; zucchini, eggplant , squash; capsicum; sweetcorn, corn on the cob; mushrooms; celery, cucumber; onion or leeks. Frequency of consumption of the 33 predefined food groups were converted to daily equivalent frequencies and entered into the factor analysis. Dietary patterns were derived from the 33 food groups using factor analysis with factor loadings extracted using the principal component method and varimax rotation using data from the FFQ. Two dietary patterns were identified using multiple criteria: the diagram of eigenvalues, the scree plot, the interpretability of the factors and the percentage of factors explained by the factors (Table 1). The two dietary patterns were confirmed by a statistician. The factor score for each pattern was calculated by summing intakes of all food groups weighted by their factor loadings. Factor loadings represent correlation coefficients between the food groups and particular patterns. The first pattern “whole food” was heavily loaded by a high intake of vegetables, legumes, fruit, whole grain, fish, egg, low fat dairy, nuts and salad dressing. Although salad dressings are typically energy dense, it is likely that it was included in the whole food pattern because it is highly correlated with salad intake. The second pattern, labelled “processed food”, was heavily loaded by a high consumption of processed foods high in saturated fat (desserts, biscuits, fried foods, snacks, high fat takeaway, chocolate and sweets, processed meat and fish), high fat dairy, potatoes, refined grains, sugar beverages and red meat. Each participant received a factor score for the two identified patterns.

Cognition at baseline was measured with the Cambridge Cognitive Examination (CAMCOG) [14], a valid measure of global cognitive performance [15]. The total score can range from zero to 105, with higher scores indicating a higher degree of cognitive function. CAMCOG contains eleven subscales that evaluate different cognitive functions: remote memory, recent memory, executive function, orientation, new learning, comprehension, expression, attention and calculation, praxis, abstract thinking and perception. Levels of cognitive functioning were defined by tertiles of performance of the total CAMCOG score and scores of individual subscales.

Sociodemographic variables included age, gender, marital status (married, widowed, divorced/separated, single, defacto/cohabited), education (total years). Health behaviours included lifetime smoking (expressed in terms of pack-years, which was the mean number of cigarettes per day multiplied by the number of years smoking and divided by 20 [16]). Participants self reported if they did any physical activity, how many days they did physical activity and the average amount of physical activity they completed each day in minutes; there was no distinction made between moderate and vigorous physical activity. In a subset of 117 participants, the Human Activity Profile (HAP) was used to evaluate the participants’ physical activity [17]. The HAP is a self reported 96-item questionnaire, and items are ordered in increasing level of physical activity demand. Each item is marked as “still doing”, “have stopped doing” or “never did”. The Adjusted Activity Score was used in this study, which is the highest numbered item marked as “still doing” less the number of lower numbered items marked as “have stopped doing”. In order to determine if there was an association between the level of physical activity determined by the self reported and the HAP methods, a chi-square test of association was used. It was found that there was an association, χ² (1, n = 80) = 10.824, p < 0.05); self reported physical activity was used in the final analysis (categorised as low: up to 40 min/day; medium 40–50 min/day, and high >60 min/day on average). Health status variables included use of antidepressants.

Statistical analysis was performed using the Statistical Package for the Social Sciences software version 17.0 (SPSS, Inc., Chicago, IL, USA) and Statistical Analysis Systems statistical software package version 9.1 (SAS Institute, Cary, NC, USA). Continuous variables were summarised as means ± SD and categorical data as proportions. All continuous variables were normally distributed. Sample characteristics were compared between three levels of dietary pattern scores using chi-square analysis and analysis of variance for categorical and continuous variables, respectively. Pearson’s correlations were used to analyse associations between physical activity cognition and dietary patterns. Logistic regression analysis examined the association between the dietary pattern score and cognitive functioning. Identified confounding factors included in the models were age, gender, education level, smoking pack years, physical activity and use of antidepressants. The self-reported number of years attending school was used as a measurement of level of education, and the self-reported number of minutes spent exercising per day was used as an indicator of physical activity. Levels of significance were set at 5% and all probability tests reported are two-tailed.

Ninety-four (38%) women and 155 (62%) men completed the baseline assessment with a mean ± SD age of 73.4 ± 5.6 years and CAMCOG total score of 89.8 ± 5.9. There were no apparent differences in demographic characteristics between the whole food and processed food patterns according to tertiles of dietary pattern score, except there was a higher proportion of women in the lowest tertile of processed food pattern, which suggests that women consumed less processed food than men (Table 2).

There was no association between the whole food pattern and the overall CAMCOG score (r = 0.007, p = 0.908); however, there was a negative correlation between the processed food pattern and the overall CAMCOG score (r = −0.153, p < 0.05). The logistic regression models adjusted for age, gender, education, smoking status, antidepressant use and physical activity (Table 3, model 3) showed that participants in the highest tertile of the processed food pattern score were more likely to have cognitive functioning in the lowest tertile as assessed with the CAMCOG subscale for executive function. The test for a linear trend was consistent with the impression that increasing levels of processed food in the diet are associated with an increased risk of being in the lowest tertile of cognition, although for model 3 (full adjustment), these were only statistically significant for executive function.