The present study aimed at assessing dietary intake and patterns, as well as associations with socio-economic cofactors and health outcomes such as overall nutritional status and blood pressure, among Tunisian adolescents.
As for overall energy intake, within the limits of the estimation of dietary intake using a semi-quantitative food frequency questionnaire, we found high overall intake by both males and females, largely exceeding the requirements based on their physical expenditure level, and more so in females than in males. Indeed, we found close average energy intake levels between boys and girls, contrary to what has been observed over the last decades in Europe, where, whatever the methodology employed, females always displayed much lower energy intakes than males . In this context, changes in energy expenditure for girls vs. boys were also accompanied by a modification of the desirable body image for girls which resulted in an even more drastic decrease in energy intake over past decades than for boys [54, 55]. Obviously, this change is not yet fully at work in Tunisia where girls still maintain, relative to boys, a high level of food energy consumption despite a quite sedentary lifestyle; also, it has been shown that if there is indeed a shift from the cultural attitude favouring plumpness towards preference for a normal corpulence, younger women do not systematically value slimness even in urban settings . Although it is difficult to derive longitudinal inferences from cross sectional data especially given the rapid economic development and societal changes in Tunisia in the last 30 years, these observations are coherent with obesity in Tunisia being much more prevalent among adult women vs. men [13, 17] while obesity figures are much more gender independent in European countries.
Though not far from recommendations, on average the diet was a little high in fat as a % of energy, as is common in European Mediterranean countries, with a favourable ratio of unsaturated fats due to the consumption of olive oil . As the consumption of fruit and vegetables continued at a rather high level, well above the recommended threshold of 400 g/day , the fibre and micronutrient contents of the overall diet appear to be relatively satisfactory with the exception of calcium. However, only 38% of adolescents displayed a satisfactory diet quality score (DQI-I), generally due to a good level of variety and adequacy but a low score for moderation and balance.
Concerning food based descriptive analysis of dietary intake by computation of dietary patterns, the main dimension underlying food consumption data was found to be a traditional to modern gradient, characterised from the food point of view by a linear decrease in consumption of long-established foods such as vegetable oil, cereals, grains, legumes and vegetables and an increase in foods made more widely available in the last decades such as white bread, dairy products, sugars and confectionery, added fats, fried potatoes and fruits. It is of course always difficult from a methodological standpoint to discuss longitudinal inferences based on cross sectional data. But anyway, this finding is in line with a chronological food transition process  that would be actively underway as strong unconfounded associations were observed with environmental and socio-economic cofactors: indeed, in our study the diet was all the more 'modernised' in urban or the more affluent households, or for a higher level of education of the mother, indicating that there was, at the time of the study among Tunisian adolescents a gradient of nutritional transition correlated to a gradient of socio-economic development within the country. It may be hypothesised that this dietary gradient is likely to recede if/when socio-economic development, lifestyle and dietary practices uniformise and give way, within a globally "modernised" food consumption, to different patterns such as those observed in studies among youth of industrialised countries [44, 58, 59], where such a gradient was obviously not observed. But as for now, from a descriptive point of view, this main traditional to modern gradient is a characteristic dimension underlying the variability of food based dietary intakes among Tunisian youth.
A second gradient, though of less importance, was observed, characterised by a monotonous and increasing relationship with energy, meat and fish consumption and somewhat decreasing with white bread, milk, sugars and sweet beverages and added fats. Interestingly, it revealed another path to a modified diet accompanying economic improvement, which yielded a better quality diet (DQI-I) for the last tertile than the modernised diet, with still a better ratio of unsaturated to saturated fats due to an increase in olive oil consumption, despite its richness in animal proteins; however, it also featured an excess of energy. It is apparently rather specific to the coastal Middle Eastern Region in Tunisia and to intermediate economic level households, without much independent effect of other socio-economic cofactors.
Even after adjustment (including for physical activity), with the exception of WC for boys, there was no straightforward relationship between anthropometry and energy intake or imbalance in neither gender. However, such a null association is rather frequently observed in either cross-sectional or even prospective studies. Underreporting, methodological errors, insufficient measurement precision or accuracy of energy intake and expenditure or dieting have successively been invoked [60, 61]. On the other hand, there was a marked association between the modern diet score and both BMI and WC, but only in males. In Iranian adolescents a linear relationship between the consumption of high fat and salty snacks and carbohydrates (bread, rice pasta) and BMI was also observed, though total energy did not differ between overweight or obese and non-obese subjects . Among females, even after adjustment for age, total energy intake, physical activity and anthropometry, modernisation of the diet was associated with a decrease of high blood pressure and hypertension (though somewhat levelling off after the 2nd tertile). It is to be noted that the association was somewhat reduced from a statistical point of view by adjustment on socio-economic factors; nevertheless, it is debatable whether this third type of adjusted estimate is the relevant one for assessing association of diet patterns with health outcomes. Indeed, some authors do not adjust for socio-economic factors when assessing relationship between health outcomes and dietary patterns, which is coherent with a conceptual framework such as that presented in Figure 1, where these latter factors are the more distal ones, whose effect on health outcomes is hypothesised to be mostly mediated through the proximal ones i.e. dietary intake and physical activity. Our tentative adjustment on socio-economic factors in the third type of models was an attempt to try and take into account psychological and/or sociocultural differences that might be linked to socio-economic status in the Tunisian context, but some kind of over adjustment cannot be ruled out. From our first type of adjusted estimate (i.e. adjustment for age, total energy intake, physical activity and anthropometry only) it would seem that girls with a somewhat modernised diet have a reduced risk of arterial hypertension. In adults, it has been shown that diets high in potassium, vitamin C, or low fat dairy products, and low in sodium, were effective to reduce BP significantly [63–66]. However, previous studies have not revealed always consistent associations in adolescents between systolic BP and diet components [44, 67]. Nevertheless, in northern Greece, decreased systolic BP was found to be associated with decreases in potassium and increase in calcium intakes among adolescents . Although the evidence regarding associations between calcium intake and BP is somewhat mixed , other studies also mentioned the potentially significant role of calcium: dairy foods were an important component of the DASH (Dietary Approaches to Stop Hypertension) diet, and calcium supplementation has been shown to be associated with a modest reduction of BP, even in the absence of a decrease in sodium, specifically when calcium intakes were initially rather low [63, 70–72]. Indeed, in the present study, the level of added salt decreased along the modern diet pattern axis, but not the total sodium intake, with however a slight difference between males and females. Potassium and vitamin C also decreased, while calcium rose significantly along with modern diet score tertiles. Zinc, B12 and in particular folates increased significantly with a modern diet, these nutrients being generally associated with a better blood pressure status, even in children .
Not so many studies have tried to distinguish dietary patterns in adolescents in relation with their health status in industrialised countries [44, 74–78] and even fewer in emerging countries [79, 80]: the study , though using cluster analysis instead of a method related to principal components or factor analysis, showed in Korea results comparable to those shown here, in that the transition to a modernised western-type diet had both positive and negative potential health consequences, improving the variety of the diet but increasing the level of fat or sugar for instance .
As for the strong and weak points of the study, regarding design and sampling, the cross-sectional design with retrospective estimates of exposure has known limitations when trying to interpret observed associations as causal. The study carried out in 2005 is quite recent but given the rapid pace of societal and socio-economic changes in emerging countries such as Tunisia it cannot be ruled out that further changes may have occurred since. The stratified, clustered random sample of 1019 subjects, although not national, was representative of three regions that feature a panel of the socio-economic and nutritional situations in Tunisia, as assessed by observed distributions of socio-economic factors and nutritional status indicators with reference to a national survey ; there was nevertheless a somewhat lower response rate for males vs. females which is not unusual in this context; we used weighting factors to partly control for non response but this does not preclude some amount of selection bias.
Although widely used in other fields, the MCA technique itself has previously been rarely used for computation of dietary patterns compared to PCA or factor analysis, or only with dichotomous coding of food group intakes [40, 42]. In our study, use of MCA applied to food groups intakes in quintiles enabled a non parametric and robust assessment of associations between consumptions of the 43 food groups and deriving dietary patterns without relying on the assumptions on which methods based on correlation coefficients depend and which rarely hold in data from large food consumption surveys. Moreover, one common advantage of using methods related to PCA (i.e. PCA itself but also factor analysis and MCA), deriving patterns which are linear combination of the initial variables vs. methods related to cluster analysis, is that it results in scores, for which a dose-effect relationship with health outcomes can be assessed. We could also have tried to derive dietary pattern predictive of the studied health outcomes [37, 81], but did not do so as our main objective was to assess dietary practices among Tunisian adolescents from a descriptive point of view.
Regarding measurements, assessment of dietary intake by a retrospective frequency questionnaire does raise a number of concerns  especially accuracy of estimates of subject level absolute intakes in energy and other nutrients, although at population level individual variation might be mitigated; also these drawbacks are likely less of an issue regarding computations of food list based diet scores. Although memory bias is known to increase with the length of the recall period, the one month retrospective period was an advantage regarding estimates of usual intakes (vs. instruments measuring either prospective or retrospective intakes over shorter periods); also the questionnaire was validated and adapted to the local context and - contrary to some other studies- included portion size data. Some seasonal difference probably affected availability of some foods during the survey period, although the most important seasonal change is between the winter season and the rest of the year. Regarding assessment of diet quality, some authors do not approve of the universal use of the DQI-I based on U.S. guidelines, preferring to derive specific indices, e.g. in West Africa based on WHO and FAO recommendations [83, 84]; nevertheless which of those indices should be used among those already proposed is an open question as all those indices do have specific limitations and often fulfill different purposes .
A specific strength of the study was to feature, through a validated questionnaire, a direct and detailed assessment of physical activity and thus control of confounding related to this dimension.
Regarding assessment of overweight and obesity, there are, indeed, a number of issues regarding the choice of international references and/or the use of a national reference [86–88]. Our choice was made for comparability issues with previously published data on the whole country from which the subjects of the food consumption survey are a subsample , taking into account that descriptive anthropometric status data was not the main focus of the present study.
If the nutrient related hypotheses put forward to explain decrease of BP with modernisation of the diet among girls are coherent with the observed associations, some amount of confounding by factors not taken into account in the study cannot be ruled out. Indeed, the modernisation of the diet could be a correlate of a modernised lifestyle which, all the more for girls in the Tunisian context, could have both indirect effects through dietary intake and physical activity, but also, direct effects on measured health outcomes. It should then be of interest to assess dimensions such as health related quality of life, perceived stress or family psychosocial factors [89, 90]. More research is likely needed on these issues which would likely be more appropriately dealt with by a mixed quantitative and qualitative approach.