There is still great interest in the diet practiced around the Mediterranean Sea due to the observation of low disease incidences compared to Northern European countries. A broad search for common dietary practices in this region of the world identified a relatively high proportion of energy gained from cereals, vegetables, and fruits, relatively less meat consumption, and a greater reliance on vegetables than on animal fats compared to Northern Europe [1
]. These foods were recently highlighted by the Global Burden of Disease study on diet, which found that dietary factors have a high impact on disability-adjusted life years (DALYs) [2
However, the amounts and combinations of those foods either promoting or delaying the occurrence of diseases could vary between countries and within countries between regions. Such regional variations within a country were already well documented in The Seven Countries Study from the 1960s [3
], taking Italy as example [4
]. In that example, diet varied manifold in key food components such as milk and cheese, vegetables, fruits, meat, egg and fish, and even sugar and sweets taking three “Seven Countries Study” areas from North to South. Such regional dietary habits have prevailed up to now and even today understanding the preventive principles associated with regional dietary habits is still a challenge. In this spirit, the publicly funded research hospital IRCCS “S. de Bellis”, located in Castellana Grotte (Apulia, Italy) started in the 1980s a population-based cohort study, and has been measuring dietary intake in this Mediterranean population since that period [5
]. This interest has recently promoted the recruitment of a new cohort in order to gain a better understanding of the aging process, particularly related to neurodegeneration and functional decline [8
], with overlaps with the previous study populations. This was seen as opportunity to study diet and its implication not only cross-sectionally but also longitudinally [7
The aim of the current research was therefore to identify the existing study populations with their overlaps, to describe and analyze the dietary intake over the last two decades, and to link dietary intake longitudinally with a biomarker profile.
The characteristics of the study populations are shown in Table 2
. In 2005/2006 the MICOL3 participants were about 65 years old, with a slightly smaller percentage of women than men. There were no particular differences between the two study populations of the MICOL3 examination in 2005/2006. Likewise, the two “GreatAGE”-Study populations, which were aged 73.5 years on average at study examination in 2012/2018, did not show particular differences in regard to anthropometry, and other health indicators. When comparing the two time periods of examination (2005/2006 vs. 2012/2018), not much change could be observed. The difference over time seen in the education variable based on methodological differences (see method section). Of particular interest is the lack of an overall decrease in health conditions referred to the MICOL3/GreatAGE study population as well as the other population groups which served as control samples, not being affected by repeated study participation. Weight gain and a substantial increase in blood pressure did not occur within the time period and indeed, as regards the clinical chemical health indicators, some improvements could actually be observed. This was true in particular of cholesterol in blood. There was also a remarkable stability of energy intake among the study populations and over time, allowing a combined dietary analysis across all study populations. Of importance in the long run could be the observation of an increased fat component compared to the carbohydrate component observable in the fat/carbohydrate ratio, which rose from 0.18 to 0.22.
The dietary data, standardized to 2000 kcal, also showed stability regarding the intake of specific foods such as fruits, and other important foods characterizing the Mediterranean diet (Table 3
). Over the two time periods, the mean intake of root and other vegetables increased, as well as of olive oil, and water when taking the mean intake measurements of all 4 study groups and allowing a minimal difference of 3 g/day between the groups at each point in time (Table 3
). Likewise, the intake of grains (pasta and bread) and beer decreased. Red meat was partly replaced by white meat.
Next, the role of food intake per health indicators, as shown in Table 1
, was prospectively investigated. The intake of the 29 food groups in 2005/2006 was examined in relation to the health indicators in 2012/2018 by forming subgroups without impaired health indicators at the 2005/2006 examination (see method section). In Table 4
, the results from the regression models are shown as p-values and the direction of relations. BMI gain (>1.5 units) between the two time periods and the values of SBP, DBP, GLY, total and HDL cholesterol, and TG in 2012/2018 were investigated for their relation with food group intake in 2005/2006 by calculating two models, one without adjustment for other food groups and one with adjustment for all other food groups. Overall, each regression model with a biomarker as dependent variable at the second examination was adjusted for gender, age, educational score, smoking, BMI, and value of the dependent variable at the first examination, and medication at the second examination (for details of the models see legend Table 4
). Ideally, both analyses should come to the same conclusion, which was primarily the case for the many non-significant results. Significant results with agreement between the two models regarded low fat dairy and reduced weight gain, juices and reduced SBP, olive oil and reduced HDL-cholesterol, and water and reduced cholesterol (total and HDL) levels. Next, the significant results without adjustments for other food groups which lost significance after adjustment for other foods are described. We found an inverse relation between seafood/shellfish and total cholesterol and a positive relation between sugary foods and HDL-cholesterol. Finally, we like to mention the relations that became significant after adjustment for other foods. For most health indicators, one or two food groups subsumed all the variance of the food groups and became significant, with the exception of the health indicator DBP. We found a direct relation between leafy vegetables and an inverse relation between other vegetables and SBP, an inverse relation of the same food group with glucose, a direct relation between sugar and an inverse relation between coffee and total cholesterol, and increased triglycerides with eggs and red meat, but decreased triglycerides with ready-to-eat dishes.
Next, we investigated the healthy diet indices. The indices were constructed with the medians for the total group, allowing analysis of changes over time and between groups. The DASH diet index score in 2005/2006 was 5.3 and 5.1, (MICOL3/GreatAGE and MICOL3 only) and in 2012/2018 it was 5.1 and 4.9 (MICOL3/GreatAGE and GreatAGE only); the Meddietscore in 2005/2006 was 5.7 and 5.5, and in 2012/2018 it was 5.2 and 5.3; and the MIND diet index in 2005/2006 was 7.4 and 6.8, and in 2012/2018 it was 6.7 and 6.6, taking into account the 4 subgroups. Thus, we could observe a slightly decreased adherence to the Meddietscore between the two time periods, and to a lesser extent also for the MIND diet index. None of the indices showed a significant relation to the health indicators (data not shown).
Finally, in Figure 2
, Figure 3
and Figure 4
, the correlations between the healthy diet indices and the food groups are shown, combined across all the study populations Overall, the correlations between the food groups and the indices were not very strong, but often several food groups stood out. Negative correlations did not reach a value < −0.2. It must be noted that foods could be correlated with the indices without these foods necessarily forming the index. For the DASH-diet index the leading positive correlations (>0.3) were with fruiting vegetables, fruits, and grains. For the Meddietscore the leading positive correlations (>0.3) were with leafy vegetables, fruiting vegetables, other vegetables, legumes, fruit, and borderline potatoes. For the MIND diet index, the leading positive correlations (>0.3) were with leafy vegetables, fruiting vegetables, other vegetables, and borderline legumes. As regards beverages, only the positive relation between wine and the DASH index stood out, whereas the other two indices were not well correlated with beverages. It could also be observed that the diet healthy indices scores were mostly independent of the intake of sweet food groups. In terms of dairy products, weak correlations were found, with the exception of a positive relation between the DASH diet index and low fat dairy. The DASH diet index also differs from the other indices in terms of the slight negative correlations with meat food groups.
This study evaluated the dietary intake of the population of Castellana Grotte, located in a rural area in Apulia, at two different time periods and related the dietary intake to health indicators and healthy diet indices. Within the observation period of 7 years from 2005/2006 to 2012/2018, covering the lifetime age periods between 65 to 73 years on average, the source population studied showed a remarkable stability in terms of dietary intake of the 29 food groups considered, and in terms of health indicators such as anthropometry, blood pressure and clinical biochemistry. Only the intake of bread and pasta (grains) was partly replaced by olive oil, slightly increasing the fat-to-carbohydrate ratio. It could also be seen that dietary intake related prospectively to health indicators, such as olive oil that was inversely correlated to HDL-cholesterol. Three established healthy diet indices were constructed and correlated with food group intake, showing that the DASH-index represents particular fruit and grain intake and the MIND-index vegetable intake.
The economy of Castellana Grotte and the surrounding area is based on agricultural production, mainly olives, grapes, cherries, and cattle. The local diet makes use of the existing agricultural infrastructure and, for example, did not include a lot of fish and seafood, available 15 km away at the coastal sites around Monopoli. The local production of most of the food products eaten, in addition to the dietary tradition of a small town, resulted in a remarkable stability of dietary habits, characterized by high intakes of vegetables, fruit, and local pasta. The slightly lower energy intake of the older subgroups is in line with the biologically reduced energy requirement due to the reduced physical activity and muscle mass [19
]. The study also showed that the population with repeated examinations in the study did not differ from the study populations only invited once, indicating a negligible influence of a healthy participant bias in terms of the dietary estimates. Whereas vegetables and fruits are established characteristics of the Mediterranean diet, the Apulian diet differs from that of other regions around the Mediterranean Sea in regard to the type of grain products. Local grain products are based on semolina meal, which is richer in dietary fiber, vitamins of the B-complex, and minerals, as compared to the flours used in industrial products [20
], especially if a brown, non-refined version of Semolina flour has been used. A typical bread is called “Altamura bread”, first made in Altamura, a city near Bari and Castellana. It is obtained from milled durum wheat semolina from the Alta Murgia in the Province of Bari, produced with a natural sourdough. In 2003, “Altamura bread” was recognized as DOP (Protected Designation of Origin) by the European Community. Semolina meal is not only used for bread produced locally by small enterprises, or homemade, but more importantly, for fresh pasta (Orecchiette), which constitutes a traditional typical food product accompanied by vegetable consumption.
There is continuous debate as to whether local dietary habits will survive the next decades. The current study results offer a more positive view on this issue. We show that this elderly population has followed stable dietary habits over the last decade. Recently, it was stated that the intake of cereals (including pasta, bread and similar), whole grain bread, vegetables, legumes and fish significantly increased between 1985–1986 and 2005–2006 in this study population, whilst the consumption of fruits, meat, poultry, dairy, olive oil and alcohol significantly decreased [6
]. The most dramatic decrease was observed for olive oil, that declined by 2.35 points in younger people and by 0.89 in older age groups [6
]. These findings between 1985 and 2005/2006 are based on a broader type of dietary assessment [7
]. Currently, these trends seem to be partially reversed, in particular in regard to olive oil and vegetables. However, our study did not address a younger population, which seems to be more sensitive to dietary changes, as the study by Veronese et al. showed. The decreased intake of red meat is in line with the regional trends of consumption, using data from the National Food Consumption Survey from 2005–2006 [21
]. Our study could confirm the finding by Veronese et al. [6
] that over time, adherence to the Mediterranean diet in general is slightly decreasing. However, in terms of dietary change it should be borne in mind that in Southern Italy, adherence to the Mediterranean Diet also depends on the general economic situation, as exemplified by the 2007/2010 crisis [22
The stable consumption of vegetables and fruit could be one of the reasons why the study populations as a whole did not show increases in BMI, and also only a small increase in blood pressure within the observation period, despite being in a critical age class for the deterioration of health indicators. Of further note is the slight decrease of basic clinical chemistry parameters, such as cholesterol and fasting glucose, along with weight stability. The laboratory which analyzed the samples is also involved in clinical practice and performs continuous quality control, which should exclude such a drift due to methodological changes. The only explanation could be the increased use of medication, such as statins, that improve the clinical chemical profile. We found that about one fifth of the subjects were taking this type of medication, confirming the overall trend of high use in Italy [23
In regard to the prospective relation between food intake and health indicators, we would like firstly to discuss those associations which showed consistency between the statistical models. These regard low fat dairy, juices, olive oil, and water. The finding of a direct association between low-fat dairy products but not high fat dairy consumption and weight gain (BMI > 1.5) in about 7 years is not supported by findings in the literature [24
], and is thus a surprising result. We can only consider the role of fat in the discussion about obesity, and messages to circumvent this relation by eating low-fat products. Thus, our finding could be indicative for populations exposed to such concepts. However, low-fat products are not low energy products, since most of them provide similar energy to dairy products with regular fat, as was recognized more than 10 years ago [25
]. It might not be surprising that subjects concerned about weight gain would start eating low-fat dairies without considering the energy provided by such products. More support from the literature is available for the finding that juices were associated with lower systolic blood pressure. In a recent meta-analysis of observational studies, Zheng et al. came to the same conclusion and discussed the biological mechanism underlying this finding [26
]. Also the finding regarding olive oil and an inverse association with HDL-cholesterol is supported by a recent meta-analysis of intervention studies [27
]. The authors of this meta-analysis pointed out that the effect is only seen with virgin olive oil, rich in phenols. The population in Castellana mainly consumes this type of olive oil because it is produced locally. Also, the inverse association between water and total HDL-cholesterol is supported by the only (to the best of our knowledge) intervention study in which water with a high content of minerals reduced LDL and HDL-cholesterol [28
For the other relationships, some doubts arise as to the significance of the statistical behavior. If the initial relation disappears after adjustment, we need to assume that statistical interferences of other foods have taken place. If significance only appeared after adjustment for other foods it is not obvious why. For the statistical findings with no agreements between the models, the interrelations among food intake play an important role firstly attributable to statistical properties and only secondly to biological mechanisms. Therefore, we discuss here only some results from the final regression models adjusted for other foods which are widely discussed in the literature. For triglycerides, we observed a direct relation with eggs and red meat, and an inverse relation with ready-to-eat dishes. The first two findings were not supported by meta-analyses of intervention studies [29
]. The finding regarding eggs is also in contrast to a recent publication from Spain that found the contrary [31
]. For the second finding, ready-to-eat dishes could contain fats that decrease triglycerides if eaten at the expense of carbohydrates [32
]. Finally, we would like to draw attention to coffee, which is consumed in this area mostly as expresso. The Dutch food-based dietary guidelines recommend not to use unfiltered coffee due to intervention studies showing an increase in cholesterol compared to filtered coffee [33
]. In Apulia, as in all Italy, coffee is produced by expresso machines which do not use a filter. Fortunately, we did not observe any cholesterol increase with increasing consumption of coffee; indeed, the contrary was observed when considering the fully adjusted model.
The correlations between complex healthy diet indices and single food groups could help to understand the significance of the indices themselves in the context of the overall diet and of diet-health associations. Indices are currently widely used in the nutritional epidemiological literature due to the conviction that health effects are best addressed by a number of dietary principles [34
]. However, indices are based only on a proportion of the dietary information for scoring, and it is of interest to know the whole picture when using such indices in studies.
It was not surprising that the indices were correlated with the intake of fruit and vegetables, with some differences in number and type of food group, in accordance with the general principles of a healthy diet. Nevertheless, the correlation analyses show that each of the indices has a slightly different significance in terms of overall food group intake. At this stage of the analysis and in view of the low to moderate correlations, it would be premature to predict the performance of the indices regarding disease outcomes with different dietary hypotheses. However, we can now be sure that the application of the indices in statistical risk analyses is not guided by side effects of other foods, not being part of the scoring algorithm.
In this context, the study by Castello et al. [13
] in Spain is interesting because they used a similar approach, relating food group intake to dietary health indices. Their results regarding the aMEDindex (Alternative Mediterranean Diet Index) can be directly compared to our results regarding the Meddietscore. There seems to be little difference in the overall profile between their study and ours, demonstrating the validity of the Mediterranean diet concept across regions.
The strengths of our investigation are the large number of subjects, the concurrent investigation of a control group for the group of subjects with the two measurements, and the use of a validated dietary questionnaire. A further advantage of the study is the high participation rate at the start of the cohort and at the different subsequent examinations. Despite the losses to follow-up at each examination time, we can assume that the population was still representative of the local population at each time period. The internal validity of the data can, therefore, be considered high, since—at each time period—the subgroups did not show large differences. Nevertheless, we admit that the questionnaire was not designed to capture, in particular, changes of diet. Like many others, we consider the FFQ more as a ranking tool with less focus on the true quantitative intakes, owing to the use of a fixed number of foods and an often arbitrary definition of the portion sizes.