The traditional Mediterranean dietary pattern is characterised by a generous intake of vegetables, legumes, fruits, nuts, cereals and fish, as well as a high ratio of MUFA:SFA( Reference Trichopoulou, Kouris-Blazos and Wahlqvist 1 ). Other characteristics include moderate intakes of alcohol and dairy products mostly as yogurt and cheese, and low intakes of red or processed meats and sweets. The use of dietary scores stems from the idea that the overall dietary pattern may be more important for health and longevity than the individual nutritional or food components. The Mediterranean dietary score (MDS) was first introduced in 1995 on the basis of eight food/beverage components( Reference Trichopoulou, Kouris-Blazos and Wahlqvist 1 ) and subsequently refined( Reference Trichopoulou, Costacou and Bamia 2 ) in Greece. Investigators in Europe have extensively studied this and other variations of the Mediterranean diet score, which differed in the overall scale (ref: Trichopolou 2, Estruch, Papagiotakos) and in their measurement of individual dietary components such as assessment of olive oil intake as a separate item or, alternatively, the use of MUFA:SFA (ref: Estruch, Trichopolou 2). Researchers in the USA have subsequently modified the traditional MDS to reflect eating patterns in the USA( Reference Fung, McCullough and Newby 3 ). This alternate Mediterranean diet score (aMED), with scores ranging from 0 to 9, was created to accommodate researchers studying the association of diet with the risk of chronic disease in North American populations.
Both MDS and aMED have been consistently linked with a reduction in the incidence of CVD, stroke, cancer, diabetes and with a reduced risk of all-cause, CVD and cancer mortality( Reference Trichopoulou, Martinez-Gonzalez and Tong 4 – Reference Hoevenaar-Blom, Nooyens and Kromhout 19 ). Adjustment for estimated energy intake is usually performed in epidemiological studies to control confounding and reduce extraneous variation, and it can be achieved in a number of ways( Reference Willett, Howe and Kushi 20 ). The MDS standardises the sex-specific energy intake levels and adjusts for total energy intake. However, energy standardisation has been inconsistent for the aMED. Some studies( Reference Fung, McCullough and Newby 3 , Reference Fung, Hu and McCullough 6 – Reference Tobias, Hu and Chavarro 9 , Reference Lopez-Garcia, Rodriguez-Artalejo and Li 14 – Reference Hoevenaar-Blom, Nooyens and Kromhout 19 ) used raw component intakes to compute the total score and subsequently adjusted the models for energy intake, and others( Reference Li, Park and Wu 10 – Reference Li, Park and McGlynn 12 ) used energy adjustment and energy-standardised component intakes in the computation of the total score (aMED-e). It is unclear how the diet scores and their association with health outcomes differ between the two versions and whether studies using different aMED versions can be meaningfully compared.
The Dietary Patterns Methods Project (DPMP), established by the National Cancer Institute, was designed to strengthen scientific evidence relating diet to mortality through simultaneous identical analyses in three established US cohorts( Reference George, Ballard-Barbash and Manson 15 – Reference Liese, Krebs-Smith and Subar 18 ) of the association between four dietary indices and all-cause, CVD and cancer mortality. This seminal family of studies used aMED as one of the dietary indices. To date, an examination of aMED and aMED-e with regard to individuals being assigned the same score and whether the overall results yield similar risk estimates has not been performed. Such an analysis would provide guidance for comparison across studies using different aMED versions.
The Multiethnic Cohort (MEC) was part of the DPMP( Reference Harmon, Boushey and Shvetsov 17 ). In the present report, we use a larger sample of older men and women participating in the MEC to study the effect of energy standardisation on the Mediterranean diet score and how this effect varies among score components and different groups of cohort participants. We also prospectively examine and compare the association of aMED and aMED-e scores with mortality.
Methods
Study population and case ascertainment
The MEC is a prospective cohort study of adults from five racial/ethnic groups in Hawaii and Los Angeles, established to examine the association of lifestyle and genetic factors with the risk of cancer and other chronic diseases( Reference Kolonel, Henderson and Hankin 21 ). Over 215 000 men and women aged 45–75 years were recruited between 1993 and 1996. At cohort entry, participants completed a mail-in self-administered detailed baseline questionnaire (Qx1), which was treated as an informed consent and included information on demographics, anthropometric measures, medical history, reproductive history (women), occupational history, food intake and physical activity. Participants missing demographic and other essential information were excluded. The MEC study protocol was approved by the Institutional Review Boards of the University of Hawaii and the University of Southern California.
Deaths from causes other than cancer were identified using state death files and the National Death Index. Deaths from CVD were identified and classified as International Classification of Diseases, Ninth Revision (ICD-9) codes 390–448 or ICD-10 codes I00–I78 and G45( 22 , 23 ). Cancer deaths were identified through linkages to the Hawaii Tumor Registry, the Cancer Surveillance Program for Los Angeles County and the California State Cancer Registry, which are part of the US National Cancer Institute’s Surveillance, Epidemiology, and End Results Program, and the US National Death Index. Cancer deaths were classified using ICD-9 codes 140–208 or ICD-10 codes C00–C97( 22 , 23 ). All-cause mortality included CVD and cancer deaths, as well as deaths from other causes, including accidents and suicides. All death files were current up to 31 December 2011 for participants in Hawaii and 31 October 2010 for Los Angeles participants. Participants with no recorded death as of these dates were censored.
Dietary assessment and score calculation
The Qx1 included a 182-item quantitative FFQ (QFFQ), which has been described in detail elsewhere( Reference Kolonel, Henderson and Hankin 21 , Reference Stram, Hankin and Wilkens 24 , Reference Sharma, Murphy and Wilkens 25 ). Usual intake over the past 12 months was assessed using eight categories, from ‘never or hardly ever’ to ‘two or more times a day’, and nine categories, from ‘never or hardly ever’ to ‘four or more times a day’, for some beverage items. Quantities of foods were assessed using three portion sizes specific to each food item, which were shown as representative images. The QFFQ was validated and calibrated in each ethnic–sex group using data from 1606 participants and three randomly scheduled 24-h dietary recalls( Reference Stram, Hankin and Wilkens 24 ). The MEC QFFQ has several unique attributes, including the presence of ethnic-specific foods, reliance on a food composition table specific to the MEC and use of a large recipe database( Reference Murphy 26 ).
The foods and beverages in the QFFQ were disaggregated using the MEC food composition tables( Reference Sharma, Murphy and Wilkens 25 ), to create the major food groups and subgroups that make up the MyPyramid Equivalents Database (MPED), a standardised food-grouping system developed by the US Department of Agriculture that disaggregates most foods into their ingredients and allocates each ingredient to one of thirty-two food groups( Reference Bowman, Friday and Moshfegh 27 ). Amounts of foods reported were converted from ‘portions’ to ‘cup equivalents’ or ‘ounce equivalents’. The MPED groups and subgroups were used to construct each component contributing to aMED and aMED-e. The component scores of aMED were based on total component intake, whereas those of aMED-e were based on intake standardised to 8368 kJ (2000 kcal) in women and 10 460 kJ (2500 kcal) in men( Reference Li, Park and Wu 10 – Reference Li, Park and McGlynn 12 ). One point was assigned for intake above the sex-specific median for the healthful components or below the sex-specific median for the fat ratio component:red meat component. Specific intakes for men (10–25 g/d) and women (5–15 g/d) were used for the alcohol component. The total score was calculated as the sum of all component scores and ranged between 0 and 9. The components of aMED and aMED-e, their composition and scoring criteria for MEC participants are listed in Table 1.
Components, optimal quantities and scoring standards for the alternate Mediterranean diet score (aMED) variations among the Multiethnic Cohort participants
aMED-e, energy-standardised alternate Mediterranean diet; MEC, Multiethnic Cohort.
* Component score =1, if the criterion is met; 0, otherwise.
† Median (servings/d) for all components except alcohol and MUFA:SFA ratio. Specified range (g/d) for alcohol. Medians for all components were established separately for men and women in the MEC.
Statistical analysis
Analyses were limited to cohort participants from five main MEC ethnic groups (white, African-American, Japanese-American, Native Hawaiian, Latino) who had valid dietary assessment information. To better represent the general population, individuals with prior history of cancer or heart disease at baseline were not excluded. A total of 87 338 men and 106 189 women were included in the analyses.
Diet scores were divided into quintiles using separate cut-off points for men and women. BMI was categorised as normal weight (<25 kg/m2), overweight (25–29·9 kg/m2) and obese (≥30 kg/m2) using self-reported height and weight. The effect of energy standardisation on the diet score was assessed by examining the distributions of aMED and aMED-e scores by sex, age group, ethnicity and BMI. We calculated the percentage of participants whose aMED-e score changed or remained the same, compared with the aMED score. We also examined individual scores for the dietary components comprising the Mediterranean diet, and computed percentage of participants scoring 1 point according to the component score guidelines before and after energy standardisation, as well as the percentage of those whose component scores changed. To characterise participants who were more likely to experience component score change, we calculated mean age at cohort entry, education level, BMI, physical activity level and total energy intake by component score and total score change level (decrease, unchanged, increase). Linear regression was used to estimate the linear trend in component score change by each of these factors.
The associations of aMED and aMED-e diet scores with all-cause, CVD and cancer mortality were examined using Cox’s regression with years since study entry as the time metric. For CVD and cancer models, study participants who died of other causes were considered censored at the time of death. Hazard ratios and 95 % CI were calculated for diet score quintiles, represented by four indicator variables. The lowest quintile was used as a reference category. Linear trend was evaluated based on the median dietary score within each quintile. Analyses were adjusted for age at baseline (continuous), ethnicity (as indicator variables), BMI (normal weight, overweight, obese), moderate-to-vigourous physical activity (<2·5, ≥2·5 h/week), smoking (current smoker, past smoker, never smoked), education (<12, 12, 13–15, 16 or more years) as a proxy of socio-economic status, marital status (married, not married), hormone replacement therapy (yes, no – women only) and history of diabetes, heart disease and cancer (yes, no). Continuous measures (age at baseline and total energy intake) had no missing values because of our inclusion criteria. Categorical covariates with missing values were modelled with a separate missing value category. Missing values ranged from <1 to 2·3 % of the total sample. Models were fit with and without additional adjustment for the total energy intake (continuous) and were stratified by sex and ethnicity. Analyses were repeated with participants’ age as time metric and with follow-up restricted to 5 and 10 years. The proportional hazard assumption for Cox’s models was verified by plotting scaled Schoenfeld residuals against time to event( Reference Grambsch and Therneau 28 ). All analyses were conducted with SAS version 9.3 (SAS Institute Inc.). All P values were two-sided and P<0·05 was defined as significant.
Results
At the end of 13–18 years of follow-up, a total of 51 702 deaths (27 744 among men, 23 958 among women) were recorded, of which 19 000 (10 433 men, 8567 women) were from CVD and 16 414 (8811 men, 7603 women) from cancer.
The majority of cohort participants scored between 2 and 6 on both aMED and aMED-e (Table 2). Total scores of 1 and 7 were less common (<7 and <8 % participants, respectively), scores of 0 (<1 %) and 8 (about 2 %) were rare and a score of 9 (<0·3 %) was very rare. On average, energy intake standardisation had the effect of shifting total scores towards the middle values of 3–5 among both men and women.
Distribution (%) of alternate Mediterranean diet (aMED) and energy-standardised alternate Mediterranean diet (aMED-e) scores by sex in the Multiethnic Cohort
To assess how the effect of energy standardisation changes by age and ethnicity, we compared the correlation between the aMED and aMED-e total scores and the percentage of participants who scored the same on both scores (Table 3). There was little difference by age group in the correlation between the aMED and aMED-e. Men were slightly more likely than women to score the same on aMED and aMED-e. The percentage of participants whose aMED-e score was higher than aMED steadily increased with age, with approximately 31 % of older men and women scoring higher on aMED-e. White and Japanese American participants had higher correlations between the two scores. Native Hawaiians had the lowest percentage and African-Americans the highest percentage of those who had higher aMED-e scores. These two ethnic groups also had the lowest fraction of those who had identical aMED and aMED-e scores. The correlation between the total scores, as well as the percentage of those scoring higher on aMED-e, was lower among participants with higher BMI.
Correlation and differences in score quintiles between alternate Mediterranean diet (aMED) and energy-standardised alternate Mediterranean diet (aMED-e) scores among the Multiethnic Cohort participants
* The percentage of participants reclassified into a different quintile in aMED-e, compared with aMED. Decrease is −1 or more quintiles; unchanged is the same quintile, increase is +1 or more quintiles.
We also examined individual scores of the nine components of the Mediterranean diet among the MEC men and women (Table 4). One component, MUFA:SFA ratio, was unaffected by energy standardisation. In seven components, nearly half of the cohort participants scored ‘1’. A notable exception was alcohol, with <15 % men and <8 % women scoring ‘1’ on either aMED or aMED-e. Although the percentage of participants scoring ‘1’ was similar between aMED and aMED-e for all components, there was substantial difference in the individual scores, as evidenced in the percent discordant scores – that is, ‘1’ for one version of the diet score and ‘0’ for the other. The highest fraction of discordant component scores was observed for vegetables (24 % men and women), followed by red and processed meat (20 % men and 19 % women).
Percent scoring 1 by alternate Mediterranean diet (aMED) and energy-standardised alternate Mediterranean diet (aMED-e) food components by sex in the Multiethnic Cohort
* Percentage of the participants whose component score changed (from 0 to 1 or from 1 to 0) in aMED-e, compared with aMED.
The groups of cohort participants whose component scores changed after energy standardisation are characterised in Table 5. We computed mean age at baseline, education, BMI, physical activity and total energy intake separately among three groups of participants: those whose component scores decreased, increased or remained the same. For all components, except alcohol and red and processed meat, men and women whose component scores increased after energy standardisation tended to be older, leaner and more educated and to have lower energy intake. Men whose component scores decreased were more physically active, whereas women with decreased component scores were slightly less or equally active than whose component scores improved or did not change. The opposite trends were observed for red and processed meat component scores. Little difference among the three participant groups was found for alcohol scores.
Characteristics of participants by alternate Mediterranean diet (aMED) to energy-standardised alternate Mediterranean diet (aMED-e) component score change, by food component and sex in the Multiethnic Cohort
MET, metabolic equivalents.
* Change in component or total score or in total score quintile, from aMED to aMED-e.
† MET of activity per day.
‡ P for trend according to the t test.
Both versions of the Mediterranean diet score showed inverse associations with all-cause, CVD and cancer mortality among men and women in all ethnicity analyses (Table 6; online Supplementary Table S1). Estimates for aMED-e with and without additional adjustment for energy intake were identical (data not shown). Compared with aMED with energy adjustment in the model, aMED-e exhibited the same or slightly stronger associations with all-cause and CVD mortality and slightly weaker associations with cancer mortality for most sex–ethnic groups. The majority of statistically significant associations were observed among White and African-American participants. Although the estimates for Native Hawaiians are mostly not statistically significant, they are still very similar between aMED and aMED-e. Without energy adjustment in the model, the estimates for highest v. lowest quintile of aMED were also in the negative direction, but of smaller magnitude (attenuated towards 1) than those from energy-adjusted models, and for African-American women it resulted in the loss of statistical significance. Analyses with participants’ age as time metric and with follow-up restricted to 5 and 10 years yielded estimates 3, 6 and 8 %, respectively, of those reported in Table 6 (data not shown).
Associations of alternate Mediterranean diet (aMED) and energy-standardised alternate Mediterranean diet (aMED-e) with all-cause, CVD and cancer mortality in the Multiethnic Cohort, by sex and ethnicity(Hazard ratios (HR) and 95 % confidence intervals)
* Statistically significant (P<0·05).
† HR for the highest (5th) v. the lowest (1st) quintile. All models adjusted for age at baseline (continuous), ethnicity (as indicator variables), BMI (normal weight, overweight, obese), moderate-to-vigourous physical activity (<2·5, ≥2·5 h/week), smoking (current smoker, past smoker, never smoked), education (<12, 12, 13–15, 16 or more years), marital status (married, not married), hormone replacement therapy (yes, no – women only) and history of diabetes, heart disease and cancer (yes, no).
‡ No additional adjustment for the total energy intake.
§ With additional adjustment for the total energy intake.
Discussion
The aMED and aMED-e appear to be distinctly different diet scores, which are more closely correlated among those of older age, lower BMI and higher physical activity, and differ more widely among those with higher BMI and higher total energy intake. We have observed that although the overall distribution of the two scores appears very close the individual aMED and aMED-e scores may differ substantially within individuals. This suggests that a choice between raw and energy-standardised Mediterranean diet score deserves some attention.
We observed that among individuals whose total aMED-e score decreased, compared with aMED, average energy intakes were higher. This is not surprising. The unstandardised aMED score measures total consumption of food components and the aMED-e score reflects the proportion of these components within a standardised amount of energy. Therefore, higher-energy-intake individuals may score well on the total consumption of component foods but lag behind in the proportion of these foods, if they are consuming even greater amounts of other, less beneficial foods. This observation underscores that consuming high amounts of healthy foods does not in itself constitute a healthy diet.
Diet scores increased after energy standardisation among older participants, which may reflect their healthier overall eating habits or reduced total food intake. Energy standardisation also tended to reduce the diet score among physically active men. It is conceivable that physically active individuals would eat more to satisfy their increased energy needs( Reference Westerterp 29 ), and thus the above observation on high-energy-intake individuals may explain this finding. We note, however, that this finding was not observed in women. The correlation of physical activity and total food intake was also weaker in women compared with men in our study (data not shown). Potential reasons for this may include under-reporting of food intake by physically active women, or a possibility that women undertaking weight-loss effort may increase their physical activity but not their food intake( Reference Westerterp 29 , Reference Blundell and King 30 ).
The problem of estimating energy intake from self-reported dietary data has received much attention( Reference Freedman, Commins and Moler 31 , Reference Subar, Kipnis and Troiano 32 ). Most FFQ are not considered appropriate dietary assessment tools to estimate true energy intake( Reference Willett 33 ). Energy intakes derived from FFQ have been found to be underestimated primarily because of the specific set of foods and beverages included in the FFQ that may not cover an individual’s entire diet( Reference Subar, Freedman and Tooze 34 ). At the same time, it has been observed that because the error in self-reported energy correlates with that in the reported intake of foods and beverages adjustment for the self-reported energy intake can actually correct for measurement error in other dietary components( Reference Kipnis, Subar and Midthune 35 ). In a recent commentary, Subar et al. ( Reference Subar, Freedman and Tooze 34 ) presented a thorough discussion on the limitations and merits of energy intakes derived from FFQ and advocated the use of energy intakes from self-reported dietary data for studies of dietary patterns, as a means of estimating contribution of specific foods to the overall diet. Adjustment for the total energy intake in models has been advocated for both standard and nutrient density approaches in nutrient intake assessment( Reference Willett, Howe and Kushi 20 ).
In the estimation of an effect of Mediterranean diet on mortality risk, with additional adjustment for the total energy intake, risk estimates for the aMED score were very close to those for aMED-e among men and women across all ethnic groups. The estimates for aMED without energy adjustment in the model were not as close to those for adjusted aMED or aMED-e, although they were in the same direction. At the same time, estimates for aMED-e with and without additional energy adjustment were nearly identical (data not shown). This suggests that either aMED combined with total energy intake or aMED-e alone could be used to gauge a person’s diet in terms of its effect on the risk of death. An investigator preferring a single quantity for this purpose may choose aMED-e, whereas someone striving for uniform methodology with prior research may opt for aMED. For the same reasons, comparison of results can be performed across studies using different versions of the score.
We note that Mediterranean diet indices, both traditional (MDS) and aMED, are based on scoring most food components relative to the population medians. Thus, they are population dependent, whereby different populations under study would entail varying score cut-off points. This approach has its advantages and drawbacks. One rationale for such a population-based scoring method is its flexibility: populations may differ in a variety of ways, such as prevailing attitudes towards eating and exercise, regional cuisine, availability of certain food items and not others, etc. Because of these and other factors, beneficial effects of consuming specific groups of foods may differ, and the composition of a diet score may need to be adjusted. Diet scoring based on population medians accomplishes such fine tuning of a dietary index. On the other hand, this approach complicates direct comparison across studies, because the same score in two populations may not represent the same level of intake or the same diet composition. In addition, median-based diet scores are not stable over time: if the general pattern in a population shifts towards increased consumption of healthy foods and therefore higher median intakes, individuals with no change in the diet may likely see their diet score drop. This is counterintuitive and may bias the results of studies temporally spaced apart.
These disadvantages could be overcome by using the same cut-off points (e.g. baseline medians) throughout repeat dietary assessments, as implemented by Hoevenaar-Blom et al.( Reference Hoevenaar-Blom, Spijkerman and Boshuizen 36 ), or by applying a scoring method with fixed cut-off points for component scores, similar to other diet scores such as Healthy Eating Index( Reference Guenther, Reedy and Krebs-Smith 37 ) or Alternative Healthy Eating Index( Reference Guenther, Casavale and Reedy 38 ). In the Prevención con Dieta Mediterránea (PREDIMED) study based in Spain, Estruch et al.( Reference Estruch, Martinez-Gonzalez and Corella 39 ) used normative cut-off points and a fourteen-item MeDiet screener tool to quantify adherence to the Mediterranean diet. Sofi et al.( Reference Sofi, Macchi and Abbate 5 ) recently proposed a literature-based adherence score as a modification of the MDS dietary index. They reviewed a total of twenty-seven cohort studies from across three continents (Europe, North America and Australia) and defined the scoring cut-off points, separately for men and women, as means of the weighted medians from all individual studies. These are commendable efforts, even if the resulting cut-off points reflect only the included studies and may not be applicable in populations geographically and ethnically different from those surveyed. Fixed scoring cut-off points based on biological and epidemiological evidence would be better justified and more desirable.
Strengths of the present study include its large sample size, long follow-up and multi-ethnic composition, which makes our findings generalisable across multiple ethnic groups. The inclusion of ethnic-specific foods and calibration of the QFFQ within ethnic groups allowed us to better estimate intake levels by food group and the overall energy intake. Among the limitations of the present study, we note that only a baseline dietary assessment was available in our cohort. As a consequence, we were unable to examine diet change patterns and to investigate how diet change potentially affects mortality risk. In summary, although the individual aMED and aMED-e diet scores may substantially differ, both aMED versions, with energy standardisation built in and with energy adjustment at the analytic stage, show very similar reductions in all-cause, CVD and cancer mortality for individuals scoring high on the index scale. Therefore, either version of the score can be used in studies examining the association of diet with mortality. Moreover, results across such studies using different versions of the aMED score could be meaningfully compared in future meta-analyses.
Acknowledgements
The authors thank all the participants in the Multiethnic Cohort Study. We want to recognise Anne Tome of the University of Hawaii Cancer Center for her assistance in applying the indexes’ criteria to the MEC food composition table. The authors thank Jill Reedy for helpful comments and guidance with the manuscript. The authors wish to acknowledge the Dietary Patterns Methods Project Working Group, including MEC: C. J. B., R. E., B. E. H.; Women’s Health Initiative: Stephanie George, Marian Neuhouser; National Institutes of Health-AARP Diet and Health Study: Susan M. Krebs-Smith, Paige E. Miller, Jill Reedy, TusaRebecca Schap, Amy F. Subar; and University of South Carolina: Angela D. Liese.
Support for this work comes from the National Cancer Institute (HHSN261201200423P). The Multiethnic Cohort Study is supported by NIH/NCI 4R37 CA 54281. B. E. H. and R. E. were recipients of the post-doctoral fellowship R25 CA 90956. Partial support for this work also comes from the NIH/NCI P30 CA071789.
L. L. M., L. N. K. and C. J. B. were responsible for conception and design, acquisition of funding and overseeing the study; Y. B. S., L. R. W. and C. J. B. developed methodology; Y. B. S. analysed and interpreted the data; Y. B. S. and C. J. B. wrote the manuscript; B. E. H., R. E. and L. R. W. contributed to manuscript writing and editing.
The authors declare that there are no conflicts of interest.
Supplementary material
For supplementary material/s referred to in this article, please visit http://dx.doi.org/10.1017/S0007114516003482
