Population and study design

The present study includes data from the 2015 Health Survey of São Paulo with a focus on nutrition (2015 ISA-Nutrition), a cross-sectional population-based study. People residing in permanent private households in the urban area of São Paulo city were randomly selected from September 2014 to December 2015.

Details about the 2015 Health Survey of São Paulo (ISA-Capital) sampling process were published elsewhere^{(Reference Alves, Escuder and Goldbaum15)}. In brief, the sample from São Paulo City, SP, Brazil, was stratified by clusters, performed in two stages: census tracts, stratified in five Regional Health Coordination of São Paulo City (north, mid-west, southeast, south and east) and domiciles. For statistical analysis, sampling survey weights were applied to account for the complex and stratified cluster sampling design. All the persons from the families who belonged to the demographic domain selected in the study and met the inclusion criteria (people of both genders, aged 12+ years and residing in the urban area of São Paulo during the survey period) were interviewed. The exclusion criteria were: individuals with chronic alcoholism and those on an enteral or parenteral diet.

To minimise the effects of losses and refusals, more extensive independent random selections were performed, securing a sample that allowed the estimation of proportions of the differences of 0⋅50. The sampling error was seven percentage points, considering a 95% confidence level and a delineation effect of 1⋅5.

Among the 4059 participants of the 2015 ISA-Capital, 1737 were randomly selected to participate in the first phase of 2015 ISA-Nutrition, a study that included diet and nutritional status data. More detailed information about 2015 ISA-Nutrition can be found elsewhere^{(Reference Fisberg, Sales and Fontanelli16)}.

The Research Ethics Committee of the School of Public Health of the University of São Paulo approved the 2015 ISA-Capital and 2015 ISA-Nutrition surveys (protocols 32344014.3.3001.0086 and 30848914.7.0000.5421, respectively). All individuals included in the survey had their data collected only after signing the Informed Consent Form.

Initially, the present study comprised adults (20–59 years) and older adults (age ≥60 years) of both genders who had their habitual intake obtained through two 24-h dietary recall (24HDR) in the 2015 ISA-Nutrition survey, totalling 1188 individuals. It was impossible to obtain the score of one of the indexes for one participant, resulting in his exclusion and thus, totalling 1187 participants in the present study. Supplementary Figure 1 brings the description of the sample in the 2015 Health Survey of São Paulo with Focus on Nutrition Study (2015 ISA-Nutrition) included in the present study.

Diet assessment

Food consumption data were obtained from two non-consecutive 24HDR, applied for a 1-year interval, covering all seasons and days of the week. The first 24HDR was collected from 1188 participants during a first domicile visit and followed the Multiple-Pass Method (MPM)^{(Reference Moshfegh, Rhodes and Baer17)} to reduce errors in dietary measurement^{(Reference Raper, Perloff and Ingwersen18)}. The second 24HDR was collected from a subsample of 367 adults and older adults using the Automated Multiple-Pass Method (AMPM)^{(Reference Moshfegh, Rhodes and Baer17,Reference Blanton, Moshfegh and Baer19)} through telephone calls using the interview system incorporated into the Nutrition Data System for Research (NDSR, version 2014, developed by the Nutrition Coordination Center, University of Minnesota, Minneapolis, USA), then entered into the NDSR software (version 2014). Subsequently, habitual dietary intake was estimated using the Multiple Source Method (MSM) programme^{(Reference Haubrock, Nothlings and Volatier20)}, a web-based tool accessed online at http://msm.dife.de/. This statistical method estimates the usual intake of food and nutrients consumed by the population reported in the two 24HDR^{(Reference Haubrock, Nothlings and Volatier20)}.

Choice of the indexes of adherence to the Mediterranean diet

Firstly, we proceeded to an in-deep reading of two systematic reviews^{(Reference Burggraf, Teuber and Brosig21,Reference Zaragoza-Martí, Cabañero-Martínez and Hurtado-Sánchez22)} , recent publications at the beginning of this study, aiming to identify the criteria for the choice of an index and to evaluate the conceptual, applicability and psychometric properties of MD indexes. One of them included only Mediterranean indexes, citing twenty-eight tools. While the second evaluated a priori dietary indexes in general, citing twenty-one Mediterranean dietary indexes. Together, these two systematic reviews encompassed twenty-four different indexes evaluating the MD. Secondly, we selected from these indexes, excluding the ones with the following aspects: (i) to be a variation of the MDS or MedDietScore; (ii) to have a structured questionnaire with closed-ended questions, making it impossible its use with data from the 24HDR method; (iii) have a metric in which consumption was not transformed into normalised scoring range. After applying these criteria, nineteen dietary indexes (Supplementary Table 1) were excluded^{(Reference Haveman-Nies, Tucker and de Groot7–Reference Fung, McCullough and Newby9,Reference Sotos-Prieto, Moreno-Franco and Ordovás23–Reference Buckland, Travier and Cottet37)} , and five were selected for the present study.

The five selected indexes were calculated and compared concerning their similarities and differences, as detailed below.

Mediterranean diet scale (MDS)

The MDS was developed by Trichopoulou et al. ^{(Reference Trichopoulou, Costacou and Bamia38,Reference Trichopoulou, Kouris-Blazos and Wahlqvist39)} using dietary data from the Greek population. Individuals whose consumptions of fish, vegetables, legumes, fruits and nuts and cereals, as well as the MUFA:SFA ratio, were below or above the median of consumption, received the score of ‘zero’ or ‘one’, respectively. An inverse score was assigned for meat and dairy, components presumed to be detrimental, which are rarely non-fat or low-fat. All food components consumed were adjusted for total energy intake for both sexes. For alcohol, a value of one was assigned to moderate consumption by men (which corresponds to an amount of 10 and 50 g/d) and by women (5 and 25 g/d) and zero for consumption below or above these values for each sex. Thus, the total MDS ranged from 0 to 9, with higher scores indicating greater adherence to the Mediterranean dietary pattern (MDP).

Mediterranean Diet Score (MedDietscore)

The MedDietscore was proposed by Panagiotakos et al. ^{(Reference Panagiotakos, Pitsavos and Stefanadis40)} based on Greek Mediterranean traditional dietary habits. The scoring is based on the consumption of eleven main dietary components of the Mediterranean diet (unrefined cereals, fruits, vegetables, potatoes, legumes, olive oil, fish, red meat, poultry, whole dairy products and alcohol) according to their position in the pyramid of the traditional Mediterranean Diet. Each component scores from ‘zero’ to ‘five’ points. For the typical food components of the Mediterranean diet (unrefined cereals, fruits, vegetables, potatoes, legumes, olive oil and fish), the intake recommendation is either daily or more than four servings per week. Therefore, a score of ‘zero’ is given for no consumption, a score of ‘one’ when the consumption ranges from 1 to 4 servings/month, a score ‘two’ to 5 to 8 servings/month and a score of ‘three’ to 9 to12 servings/month, a score ‘four’ to 13 to18 servings/month and a score ‘five’ for more than 18 servings/month. The inverse score was applied for consuming foods that differed from this dietary pattern (meat and its derivatives, poultry, and whole dairy products). Especially for alcoholic beverages, points ranging from ‘zero’ to ‘five’ are assigned according to the amount of consumption in ml per day. A score ‘five’ is assigned to consumption <300 ml/d, a score of ‘zero’ to no consumption or consumption >700 ml/d and a score ‘four’ to ‘one’ for consumption of 300–400 ml/d, 400–500 ml/d, 500–600 ml/d, 600–700 ml/d, respectively (assuming 100 ml at 12 g ethanol concentration). As each of the eleven dietary components of the index are scored from 0 to 5, there was no need for truncation for the total score. However, there was need for truncation of individual food component scores. The cut-off points (recommendations in servings per month) were truncated to integers in each 0⋅5. Thus, the total MedDietscore score ranges from 0 to 55, with higher scores indicating greater adherence to the MDP.

Mediterranean Adequacy Index (MAI)

Constructed with data from the population of Italy, the MAI is based on the percentage of the total daily energy intake of four food groups: (i) carbohydrate group (bread, cereals, dried pulses, potatoes); (ii) protective group (vegetables, pulses, fresh, fruits, fish, alcoholic beverages such as red wine, vegetable oils); (iii) land animal group (milk, cheese, meat, eggs, animal fats and margarine) and (iv) sweets group (drinks, sweets, cakes, pies and cookies, sugar). The MAI score is obtained by dividing the sum of the percentage of the total daily energy intake of groups one and two by the sum of the total daily energy intake of groups three and four. Thus, its final score is interpreted as a ratio that can range from 0 to 100, with higher ratios meaning greater adherence to the MD.^{(Reference Alberti-Fidanza, Fidanza and Chiuchiù41)}

Mediterranean-Style Dietary Pattern Score (MSDPS)

Developed using data from the population of North America, the MSDPS calculation includes thirteen components: whole grains, fruits, vegetables, wine, fish and seafood, olives, legumes and oilseeds, potatoes and other starchy roots, olive oil, chicken, red meat, sweets, eggs and dairy products. Each component (except olive oil) is scored from 0 to 10 according to the Mediterranean Diet Pyramid recommendation. One point is also deducted proportionally to the number of servings consumed above the recommended intake for the group. Negative scores default to zero. The olive oil score includes only 0, 5 or 10 points, equivalent to non-consumption of olive oil, use of olive oil and other vegetable oils, and exclusive use of olive oil, respectively.

Subsequently, the thirteen component scores are summed, and the total is standardised on a scale of 0–100. For this, the total score is divided by the maximum score of 130 points and multiplied by 100. The total standardised score is multiplied by a factor ranging from zero to one (reflecting 0–100% of energy intake attributed to foods that are part of the MDP). Thus, for each individual, the MSDPS was calculated by the following equation^{(Reference Rumawas, Dwyer and McKeown10)}:

$${\rm MSDPS} = \left[{\left({\displaystyle{{\mathop \sum \nolimits_{i = 1}^{ 13} {\rm S}i} \over { 130}}} \right)\times 100} \right]\times P$$

where Si is the individual item score; P is the proportion of total energy intake from Mediterranean diet pyramid foods.

Mediterranean dietary pattern (MDP)

Developed based on the consumption of the population of Spain, MDP calculation is carried out considering the following steps: the daily energy consumption of cereals, fruits, vegetables, legumes, meat and dairy products; the adjusted energy intake of each of these foods is standardised to z-scores; before standardisation, centralised alcohol consumption for men (30 − |30 − absolute alcohol intake|) and for women (20 − |20 − total alcohol consumption|) to obtain the highest value for men consuming 30 g/d or women consuming 20 g/d, and progressively lower values for more distant consumption. Moreover, the ratio of monounsaturated to saturated fatty acids (MUFA:SFA) and the intake of trans-fatty acids (TRANS) was also directly standardised to z-scores. To obtain the total score for the MDP, the consumption of legumes, cereals, fruits, and vegetables and moderate consumption of alcohol and MUFA:SFA ratio are positively weighted (summed), for being considered beneficial components, while the consumption of trans-fatty acids, meat and dairy products are negatively weighted (subtracted), for being considered harmful components. Thus, higher scores configure higher adherence to MD assessed by the MDP, according to the following formula:

$$\sum Zi = Z{\rm legumes} + Z{\rm cereals} + Z{\rm fruits} + Z{\rm vegetables} + Z{\rm alcohol} + Z{\rm MUFA/SFA}\ndash Z{\rm TRANS}\ndash Z{\rm meat}\ndash Z{\rm dairy}$$

where Zi is the individual total standardised z-scores of nine components

Then, the total MDP score is converted into a percentage of adherence based on the sample variation range, as described in the following formula^{(Reference Sánchez-Villegas, Martínez and De Irala42)}:

$${\rm Adherence}( {{\rm Percentage}\,i} ) = \left({\sum Zi\ndash \sum Z{\rm min}} \right)\times 100/\left({\sum Z{\rm max}\ndash \sum Z{\rm min}} \right)$$

where Zmin is the lowest z-score obtained in the sample; Zmax is the highest z-score obtained in the sample.

Supplementary Figure 2 summarises the criteria adopted to construct the five selected dietary indexes, considering the method of calculation, dietary components, cut-off point, scoring system, scoring range, inclusion or not of non-Mediterranean foods and energy adjustment.

Covariates

2015 ISA-Capital data were obtained through a structured questionnaire organised by thematic areas. In the present study, socio-demographic and anthropometric variables were included as follows: sex (male or female), age (20–59 and ≥60 years old), years of formal education (≤9, 10–12 or >12), per capita household income (≤1 minimum wage and >1 minimum wage), and self-reported weight (kg) and height (m), from which we calculated the body mass index (BMI; weight/height²) and classified as below or above normality range: ≥18⋅5 and <25 kg/m² for adults⁽⁴³⁾ and >23 and <28 kg/m² for older adults⁽⁴⁴⁾.

Statistical analyses

The categorical variables were described as relative frequency (percentage), and continuous variables were presented as a mean and standard deviation. For the continuous variables, the type of distribution was investigated by the Kolmogorov–Smirnov test, histogram examination, and measures of skewness and kurtosis. Median and interquartile ranges (IQRs) were calculated to describe the scores for each dietary index presented by the study sample. The multicollinearity between multiple independent variables was tested in a multiple regression model. The variance inflation factor (VIF) was obtained to measure the amount of multicollinearity. Analyses considered the sampling weight for representativeness of the population of São Paulo.

The correlation between the total scores of the five calculated indexes (MDS, MedDietscore, MDP, MSDPS and MAI) was verified using Spearman's correlation test. A correlation coefficient ≥0⋅70 was considered a very strong correlation; from 0⋅40 to 0⋅69 as a strong correlation; from 0⋅30 to 0⋅39 as a moderate correlation; from 0⋅20 to 0⋅29, a weak correlation and from 0⋅00 to 0⋅19, no or negligible correlation^{(Reference Dancey and Reidy45)}. In addition, considering the data metric as interval scales, after classifying the participants into tertiles (data driven using the Stata's command xtile) of adherence to the MD, linearly weighted Cohen's Kappa coefficient was estimated to measure the agreement between the indexes. Its interpretation followed the classification: fair concordance when the coefficient was ≤0⋅40; moderate concordance for coefficients from 0⋅41 to 0⋅60 and substantial or almost perfect concordance when the coefficient ranged from 0⋅61 to 0⋅80 and >0⋅80, respectively^{(Reference Landis and Koch46)}.

Considering the differences between the indexes and the difficulty of identifying the most appropriate ones to assess the adherence to the MD, the convergent validity and model fit measures were assessed using confirmatory factor analysis (CFA). In this analysis, scores obtained in each food component of dietary indexes were adopted. The following goodness-of-fit (GOF) indices were applied: Comparative fit index (CFI >0⋅95), Tucker–Lewis Index (TLI > 0⋅95), Residual mean square error of approximation with 90% CI (RMSEA ≤ 0⋅06 and upper limit of the 90% CI < 0⋅08) and Standardised root mean square residual (SRMR ≤ 0⋅05)^{(Reference Hu and Bentler47)}. Robust maximum likelihood (MLR) and robust weighted least squares (WLSMV) estimators were used when observed variables were continuous and categorical, respectively. Factor loadings ≥0⋅50 were considered to identify relevant correlations between each model's observed variables (food components intake) and latent variables^{(Reference Igbaria, Zinatelli and Cragg48)}. The internal consistency reliability of each model was also evaluated by calculating McDonald's omega with values equal to or higher than 0⋅7, which was considered acceptable^{(Reference McDonald49)}.

The statistical significance was set for all the variables at P < 0⋅05. The statistical package Stata v.14 (Stata Corp., LP, College Station, TX, USA) was used to perform the descriptive and comparative analysis, while the Mplus Version 8⋅7 (Los. Angeles, CA) to fit the CFA models and their reliability.