点击显示 收起
From the Danish Epidemiology Science Centre, Institute of Preventive Medicine, Copenhagen University Hospital, Copenhagen, Denmark (JH and TIAS); The Danish Cancer Society, Institute of Cancer Epidemiology, Copenhagen, Denmark (JH, AT, and BLT); and the Department of Clinical Epidemiology, Aalborg Hospital, Aarhus University Hospital, Aalborg, Denmark (KO)
2 Supported by the National Danish Research Foundation and DiOGenes (Diet, Obesity and Genes), under contract FOOD-CT-2005-513946 with the European Community (www.diogenes-eu.org). 3 Address reprint requests to J Halkjær, The Danish Cancer Society, Institute of Cancer Epidemiology, Strandboulevarden 49, DK-2100 Copenhagen Ø, Denmark. E-mail: jytteh{at}cancer.dk.
ABSTRACT
Background:The diet may influence the development of abdominal obesity, but the few studies that have prospectively examined the relations between diet and changes in waist circumference (WC) have given inconsistent results.
Objective:Associations between total energy intake, energy intake from macronutrients, and energy intake from macronutrient subgroups based on different food sources and 5-y differences in WC (DWC) were investigated.
Design:A Danish cohort of 22 570 women and 20 126 men aged 50–64 y with baseline data on WC, diet, BMI, and potential confounders reported their WC 5 y later. Associations of baseline diet with DWC were assessed by multiple linear regression analysis.
Results:Neither total energy intake nor energy intake from each of the macronutrients was associated with DWC, except for an inverse association with protein, especially animal protein. In women, positive associations with DWC were seen for carbohydrate from refined grains and potatoes and from foods with simple sugars, whereas carbohydrate from fruit and vegetables was inversely associated and significantly different from any other carbohydrate subgroup. The results for men resembled those for women, although none were significant. Vegetable fat was positively associated with DWC for both men and women in a combined analysis. A U-shaped association between alcohol from wine and DWC was present for both sexes, and alcohol from spirits was positively associated with DWC in women.
Conclusions:Although no significant associations with total energy or energy from fat, carbohydrate, or alcohol were observed, protein intake was inversely related to DWC, and some macronutrient subgroups were significantly associated with DWC.
Key Words: Alcohol intake • carbohydrate intake • fat intake • protein intake • waist circumference • macronutrient composition • prospective study
INTRODUCTION
Abdominal obesity is associated with a high risk of developing type 2 diabetes, coronary artery disease, and stroke (1-4) and with a higher risk of mortality (5), and these associations seem to be present even after adjustments for general obesity (5). Part of the variation in abdominal obesity is explained by genetic factors (6-8), but studies investigating how modifiable factors, such as diet and alcohol intake, are associated with subsequent changes in abdominal fat are still sparse; ie, most studies of abdominal obesity measures and energy intake (9) and macronutrient intake (10-16), especially alcohol intake (11-15, 17-24), have been cross-sectional in design. This type of study does not allow inference about the temporal relation between diet and differences in abdominal obesity measures, which is needed to build up evidence for prevention.
Fewer studies have investigated the associations between intake of macronutrients and abdominal obesity measures in a prospective design (25-35), whereas fat (in 5 of the studies), and especially alcohol (in 8 of the studies), are the macronutrients studied most. However, different methods, including different study design and outcome variables, have been used, and the main focus has not always been on associations between diet and changes in measures of abdominal obesity. Two studies were based on participants' recall of whether they had gained weight, especially at the waist in the past 10 y (32, 33); others used waist-to-hip ratio (WHR), waist-to-hip breadth ratio (25, 26, 31), or waist circumference (WC) (27-30, 34, 35), of which 2 studies had information only on WC at follow-up (28, 35). One of the studies had a rather small sample size and a short follow-up time (27). These issues may have contributed to the inconsistent and oppositely directed results that are seen for fat intake and abdominal obesity measures, whereas both inverse (27), positive (25, 29), and no associations (26, 28, 29) have been reported.
Another reason for weak and inconsistent associations between macronutrients and changes in abdominal obesity measures might be differences in the effect of the specific subgroups contributing to the total macronutrient intake. If 2 subgroups within a macronutrient group, such as carbohydrate from refined grain products and carbohydrate from fruit and vegetables, were oppositely associated with obesity or abdominal fat accumulation, it would not show up as an effect in analyses of total macronutrients.
The objectives of the present study were to investigate prospectively the associations between diet and subsequent changes in WC in a large population-based cohort of middle-aged women and men. The diet was analyzed at 3 levels of subdivisions: total energy intake, intake of the 4 energy-providing macronutrients, and macronutrient subgroups based on contributing food groups.
SUBJECTS AND METHODS
Subjects
The present study included data from the Danish Diet, Cancer and Health Study. From December 1993 to May 1997, all men and women aged 50–64 y, born in Denmark, living in the greater Copenhagen or Aarhus areas, and with no previous cancer diagnosis registered in the Danish Cancer Registry were invited to participate in the study (n = 160 725). The participants were identified from the computerized records of the Civil Registration System in Denmark. Thirty-five percent (57 053) of those invited participated in the study; 547 persons were excluded from the cohort because they had a cancer diagnosis before recruitment, which due to processing delays, had not been registered in the Danish Cancer Registry at the time of the invitation. At baseline, professional personnel examined all participants in clinics in either Aarhus or Copenhagen.
Between baseline and the time of invitation to the follow-up survey, which took place from September 1999 to October 2002, 1692 participants had died and 435 had emigrated, which left 54 379 participants eligible for invitation to the follow-up survey. Questionnaires to provide information regarding diet, lifestyle, and anthropometric information were mailed to the participants. Of the 54 379 individuals invited to the follow-up survey, 89 (0.02%) had died after the linkage to the Civil Registration System (information was given by the spouses), 25 (0.05%) had disappeared or had errors in the address, 2860 (5.3%) did not want to participate, and 5869 (10.8%) did not respond. Finally, 632 (1.1%) persons who had returned questionnaires with too many errors and 7 persons with missing values on exact follow-up time were excluded, which left a total of 44 897 persons in the follow-up study sample.
Anthropometric measures
All subjects were measured while wearing light underwear. At baseline, weight, height, WC, and hip circumference (HC) were measured at the clinics by technicians. WC and HC were measured with a rigid measuring tape and recorded to the nearest 0.5 cm. WC was measured at the smallest horizontal circumference between the ribs and iliac crest (the natural waist), or, in case of an indeterminable waist narrowing, halfway between the lower rib and the iliac crest. HC was measured at the largest horizontal expansion of the buttocks. Follow-up weight and WC were measured at home by the subjects and were reported in the questionnaires. A measuring tape was provided together with the questionnaire and to ease the identification of site of measurement, the participants were told to measure WC at the level of the umbilicus. Height and HC were not reported at follow-up. Body mass index (BMI) was calculated as weight (kg)/height2 (m2) at baseline by using the measured weight and height and at follow-up by using the reported weight and the height measured at baseline.
A validation study carried out on 408 men and women from the cohort in connection with the follow-up inquiry showed high correlations between both self-reported and technician-measured WC at the umbilicus and between technician-measured circumferences at the 2 measurement sites (36). However, the Bland-Altman plots indicated that some underreporting and wide limits of agreement were present in both men and women, and, as expected (37), larger circumferences were seen at the umbilicus than at the natural waist. Finally, the self-reported WC at the umbilicus and the technician-measured WC at the natural waist were highly correlated. The self-reported WC was slightly smaller for the men, whereas it was larger for the women than was the technician-measured WC at the natural waist. However, it was concluded that the self-reported WC at the umbilicus could be used as a proxy for technician-measured WC at the natural waist in regression analyses, if the analyses were adjusted for baseline BMI and WC (36).
Dietary intake
At baseline, the participants filled in a 192-item semiquantitative food-frequency questionnaire (FFQ), which was designed for the present study (38) and validated against two 7-d weighed diet records (39). In connection with scanning of the questionnaires, a dietitian checked the frequency and open-ended answers together with the participant. Dietary calculations were made by using the FOODCalc program (Internet: www.foodcalc.dk), which is based on values from the Danish national food tables 1996.The FFQ was designed with many global summary questions for each main food group (eg, cooked vegetables, hot dishes with beef, fresh fruit). Assuming that the summary questions gave the more valid estimates, this technique was used to adjust at the individual level the contributing food items for possible overestimation. The assumptions used for the dietary adjustment were made on the basis of unpublished work (40), whereas the frequency of the many specific vegetables and fruit in the FFQ were compared with the mean frequency from the global summary question and the sum from the same food items at the group level on the basis of information from one 24-h interview. No adjustment was made if the global summary frequency was equal to or exceeded the sum of the specific food items, because not all types of foods were asked about. However, if the sum of the specific foods was higher than the global frequency, all the specific foods were adjusted with the same weight so that the sum equaled the global value. After down-regulation of the frequency by the summary question, the daily frequency sum of vegetable and fruit from the FFQ was much closer to the daily summed frequency for fruit and vegetables from the 24-h recall. Furthermore, the agreement was generally better for the single fruit and vegetables.
The dietary variables, total energy including alcohol, the 4 macronutrients, and the subgroups were measured in megajoules. The food groups contributing to the subgroups of macronutrients are listed in Appendix A. The energy of the food subgroups was divided according to the contribution to the various macronutrients, and the energy intake from the macronutrient subgroups sums up to the total energy intake.
View this table:
Appendix A. Description of the dietary exposure variables included in the study
Other covariates
Information about baseline age, sporting activities (h/wk), and smoking habits (categorized as current, past, and never smoker) was included in the models.
Definition of the study population for analysis
Of the 44 897 persons being followed up, those with implausibly large WC values (>155 cm; n = 2), small WC values [<55 cm for women (n = 1) and <60 cm for men (n = 1)], or missing values for baseline WC (n = 29), baseline weight (n = 1), or baseline height (n = 3) were excluded. Follow-up WC was missing for 633 persons, 61 persons were excluded because of implausibly large WC values, and 53 were excluded because of implausibly small WC values (<60cm for both men and women).
Of the remaining 44 113 participants, 24 were excluded because they had missing values for dietary information. In addition, those in the highest and lowest 1% of the sex-specific distribution of total energy intake at baseline were excluded (n = 881). Those with missing information for the variables smoking status (n = 18) or sporting activities (n = 453) and those who reported doing sporting activities >21 h/wk (n = 41) were excluded. The final study population consisted of 42 696 persons (22 570 women and 20 126 men), who had complete information on all covariates of interest for the analyses. In additional subanalyses, 1286 women and 2109 men with prevalent diabetes and chronic heart disease at baseline were excluded to assess whether the associations between dietary factors and waist changes were biased by the presence of these diseases.
Statistical analyses
On the basis of the data from the validation study, calibration of self-reported anthropometric data were performed as in other similar cohort studies (41). Sex-specific calibrations were made by linear regression analysis, in which the technician-measured WC at the natural waist at follow-up was the dependent variable and the self-reported follow-up WC at the umbilicus and age were the independent variables. The same type of calibration equations were used for self-reported weight and BMI at follow-up. Information about both the self-reported and the calibrated follow-up values (medians and 5th and 95th percentiles) were used in the descriptive statistics, whereas the self-reported follow-up values were used when the associations between diet and waist changes were studied. Student's t test with unequal variance was used to test for sex differences in the descriptive analyses.
The associations between dietary components and waist changes were investigated in multiple linear regression analyses. The waist change calculated as the 5-y difference in WC was the dependent variable.
RESULTS
During the follow-up period the observed median WC increased in both men and women (Table 1), especially in women. In women, the calibrated median DWC was lower than the observed value; in men, the calibrated median DWC was slightly larger than the observed value. Although median weight or BMI at baseline and median self-reported weight or BMI at follow-up were similar, a minor decrease in weight and BMI was present, whereas the calibrated median weight and BMI at follow-up showed a minor increase in weight (Table 1). Also presented in Table 1 are the baseline characteristics of the remaining covariates. The median and 5th and 95th percentiles (MJ/d) for total energy intake, the 4 energy-providing macronutrient groups, and the subgroups are provided in Table 2.
View this table:
TABLE 1. Medians (and 5th to 95th percentiles) or percentages for anthropometric measurements and covariates at baseline and follow-up1
View this table:
TABLE 2. Baseline median intakes (and 5th to 95th percentiles) of total energy, the specific macronutrients, and the subgroups of macronutrients1
Total energy
Total energy intake (MJ/d) was not significantly associated with the subsequent DWC in either sex-specific or in combined analyses (P for sex interaction = 0.07). DWC (in cm) was estimated to be –0.04 (95% CI: –0.09, 0.01) in women, 0.01 (95% CI: –0.02, 0.04) in men, and –0.005 (95% CI: –0.03, 0.02) in the combined analyses, all as linear estimates per MJ/d.
Macronutrients
In a model in which the dietary variables were included linearly, the effects of the 4 macronutrients on DWC did not differ significantly from each other (P = 0.13 for women and P = 0.42 for men). There were no overall interactions between sex and the 4 macronutrients on DWC (P for interaction = 0.18). In a partition model, no statistically significant associations were observed between the 4 macronutrients and DWC, when included linearly, other than an inverse association with protein in a combined analysis of men and women (P = 0.02; Table 3). However, for alcohol and fat, the assumption of linearity was not appropriate; there was a tendency toward a nonlinear U-shaped association with DWC. After allowing for different slopes > and <300 kJ alcohol intake and > and <3.3 MJ fat (data not shown), there was still no sign of a sex difference (P for interaction = 0.81).
View this table:
TABLE 3. Associations between energy intake from a macronutrient or subgroup of macronutrients and estimates (and 95% CIs) of 5-y differences in waist circumference (DWC; in cm) per MJ/d of each macronutrient based on multiple linear regression analyses in partition models1
Subgroups of carbohydrate
Significantly different effects of the different subgroups of carbohydrate on subsequent DWC were observed for women (P < 0.0001) but not for men (P = 0.33). Although the estimates pointed in the same direction for both men and women, an overall significant interaction (P = 0.002) between sex and the carbohydrate subgroups was observed (Table 3). In women, a partition model showed that DWC was inversely related to intake of carbohydrate energy from fruit and vegetables, whereas intake of carbohydrate energy from food sources with simple sugars or from added sugar was positively associated with DWC (Table 3). A clear and significant positive association with intake of carbohydrates from refined grain and potatoes was observed and a similar but insignificant pattern was found for carbohydrates from whole grain. The results for men resembled those for women, but with nonsignificant estimates.
The associations with carbohydrate energy from fruit and vegetables deviated clearly from the other carbohydrate subgroups, and a pertinent substitution model showed for women that for a given intake of total carbohydrate energy, substitution of carbohydrate energy from fruit and vegetables with any of other carbohydrate subgroups was significantly positively associated with subsequent DWC (P < 0.001 for carbohydrate from refined grains, whole grains, and simple sugars; P = 0.008 for carbohydrate from other sources). Similar nearly significant results were observed for men when carbohydrate energy from fruit and vegetables was substituted with carbohydrate energy from food items with simple sugars (P = 0.06) or from whole grains (P = 0.07).
Subgroups of protein
There were no significantly different effects on DWC for vegetable and animal protein (P = 0.66 for women and P = 0.06 for men), and there was no overall significant interaction between sex and the subgroups of protein (P = 0.26). The partition model showed tendencies toward nearly significant inverse associations for animal protein in both sexes (Table 3), and, in an otherwise sex-specific model, the combined estimate for men and women for animal protein was inversely and significantly associated with DWC (P = 0.02), whereas no association was seen for vegetable protein.
Subgroups of fat
There was no overall significant interaction between sex and the subgroups of fat on the effect on DWC (P = 0.26). In an otherwise sex-specific partition model, the combined linear estimate for men and women for vegetable fat was positively associated with DWC (P = 0.02), whereas animal fat was inversely but not significantly associated with DWC (Table 3). For animal fat, however, the assumption of linearity was not appropriate, because there was a tendency toward a U-shaped association. An inverse association with DWC was seen for intakes 2 MJ, whereas a nonsignificant positive association between animal fat and DWC was present at intakes >2 MJ (data not shown).
Subgroups of alcohol
The different sources of alcohol energy had significantly different effects on DWC (P = 0.007 for women and P = 0.0005 for men) when included linearly in the model. There was an overall significant interaction, mainly because of alcohol from wine, between sex and the alcohol subgroups in a model, where the alcohol subgroups were entered as linear estimates (Table 3; P = 0.0002). For wine, an inverse association was seen for women, whereas a positive association was seen for men, but the assumption of linearity was not appropriate (Figure 1). However, the pattern of the overall nonlinear association between wine alcohol and DWC was very similar for men and women (Figure 1), and only a borderline significant sex differences was present for the association between 52 and 182 kJ intake (Table 3; alternative model). In a reduced model that allows 2 different slopes > and <182 kJ intake, a sex interaction was present at low intakes of wine (data not shown). For women, a positive association between alcohol from spirits and DWC was present (Table 3).
FIGURE 1.. Associations between intakes of alcohol energy from wine and 5-y differences in waist circumference (DWC) in men (•) and women (
Additional analyses
In some studies, participants with prevalent or incident disease (such as cancer, diabetes, or coronary artery disease) developed between baseline and follow-up, which might influence diet, physical activity habits, weight, or WC, were excluded to avoid bias (29). Our cohort was sampled as a cancer-free cohort at baseline. Further exclusion of participants with prevalent diabetes and heart disease (1286 women and 2109 men) at baseline did not change the associations between diet and waist changes.
DISCUSSION
This study showed that total energy or energy from each of the 3 macronutrient groups—carbohydrates, fat, and alcohol—were not associated with subsequent DWC in either women or men, whereas an inverse association with protein intake was observed. When the macronutrients were divided into subgroups on the basis of the contributing food sources, significantly different associations were seen for some of these groups. There was an inverse association with animal protein and a direct association with vegetable fat. In women, substitution of carbohydrate from fruit and vegetables with any other type of carbohydrate was significantly positive associated with DWC, and the same, though nonsignificant, trends were observed in men. The alcohol subgroups showed different relations with waist changes with a U-shaped relation with alcohol from wine as a distinct result.
Strengths and limitations
The prospective study design used allowed us to analyze the association between baseline diet and subsequent changes in WC rather than the concurrent changes in diet and WC, as done in some other studies (27, 29). Thereby, we separated the exposure from the possible effects and avoided the difficulties due to a possible modification of the diet as a consequence of a change in WC, so-called reverse causality.
However, the estimates in a prospective study design may still be biased because of the misreporting of energy intake or of specific food groups especially if the misreporting is related to baseline variables that also influence the DWC, such as degree of obesity, weight history, social status, or the propensity toward socially desirable answers (45-48). Inclusion of summary questions in the present FFQ may to some extent regulate for a possible overreporting of socially desirable food items such as fruit and vegetables.
The self-measurement of WC at the level of the umbilicus at follow-up compared with technician-measured WC at the natural waist at baseline could have introduced bias. However, even though absolute differences were present, the 2 measures were highly correlated (36), which is the important aspect when the significances of associations between diet and waist changes are evaluated. Possible measurement errors due to the use of different measurement sites was assumed to be randomly distributed across potential factors associated with selective misreporting, and any association with previous diet is likely to be captured by the adjustment for baseline WC.
Interpretation of the results
We found no clear association between total energy intake and DWC. Furthermore, energy from different subtypes of a macronutrient showed associations in opposite directions, which indicated that the sources of energy and not total energy may be important.
Total energy intake from carbohydrate was not associated with DWC in the present study, which agrees with the results of another prospective study (26). However, the association with total carbohydrate is a weighted average of the association with different types of carbohydrate, with weights depending on the composition of the diet of the specific population. In the present study, carbohydrate energy from fruit and vegetables was inversely associated, whereas carbohydrate energy from all other food groups was positively associated with subsequent DWC. Although the associations with DWC were significantly stronger in women than in the men, they were in the same direction for both sexes. In agreement with the findings for women in the present study, we previously found that a high intake of refined bread was associated with WC gain in women, whereas a tendency toward an inverse association was seen for men (30). Food items with a high glycemic index, such as refined grains or potatoes, are suggested to promote fat accumulation (49-51). In contrast, whole grains, which are high in fiber, were found to be inversely associated with WC or WHR in cross-sectional studies (52-54), and overall fiber intake has also been found to protect against gains in WHR or WC (26, 29). However, the division into whole-grain and refined-grain cereals in the present study may not have been sophisticated enough to detect the true differences or the possible protective effect of carbohydrate from whole grains (similar to that of carbohydrate from fruit and vegetables), because intakes were based on standard recipes without any information on the specific brands.
For men and women combined, total protein was inversely associated with DWC, which is in contrast with the findings of another prospective study (26). However, more recently published intervention studies show that a protein-rich diet during both a weight-loss and a weight-maintenance period (55), or the addition of extra protein to a diet during a weight-maintenance period (56), resulted in the better maintenance of previous total or abdominal weight loss (or both). An inverse association between animal protein and DWC was seen in the present study for both sexes combined, whereas no clear associations were seen for energy intake from vegetable protein. A possible explanation of this effect may be that protein, compared with other macronutrients, and perhaps especially animal protein, induces a greater degree of satiety and possibly a greater energy expenditure and diet-induced thermogenesis (57-59).
Previous studies have given inconsistent result on the relation between energy intake from total fat and waist changes; inverse (27), direct (25, 29), and no associations (26, 28, 29) have been found. The studies are difficult to compare because of differences in the way that fat intake is included in proportion to the waist changes in the model (eg, expressed as previous changes in intake, in baseline intake, or concurrent changes in intake), energy-adjustment methods, and adjustment for other anthropometric measures. We found no association for total fat energy for both sexes, and neither vegetable nor animal fat showed any strong linear association with DWC, although vegetable fat showed a direct relation when both sexes were combined. Linear modeling of truly nonlinear effects may have obscured the associations. Subdivisions of dietary fat other than animal and vegetable fats, which were used in the present study, such as fatty acids, may also be relevant.
Studies that investigated the association between changes in abdominal obesity measures and alcohol in gram/d (29, 34), in drinks/wk (30, 31, 35), or in terms of drinking status (27) found positive or no clear associations (27, 29-31, 34, 35). There was no overall association between energy intake from alcohol and DWC in the present study, but the association, especially that due to wine, was not linear. There was a clear U-shaped relation, which indicated that those who drank slightly less than one drink per day may have had the smallest WC gain. Alcohol energy analyzed in subgroups showed different associations with DWC, which also seemed to be the case for alcohol types measured in grams or as frequencies in some studies (30, 32, 33, 35), but not in others (34); however, mostly no specific tests for heterogeneity of the estimates were performed. Our previous study showed tendencies toward sex differences, with no associations between any alcohol type and waist changes for men and positive associations between high intakes of beer and spirits and WC gain for women (30). In the present study, the sex differences were only significant for alcohol from wine, but the overall U-shape of the association was very similar for men and women.
The recently published article in the Journal of the American Medical Association (60) has given new life to the ongoing debate about the effect of fat versus that of carbohydrate in regard to obesity treatment and prevention. The findings of our study indicate that the debate should not be simplified only to total macronutrients. Our results do not clearly support the most commonly used guidelines in the 1990s, which recommend a diet low in total fat and high in carbohydrates from fruit and vegetables and grain products, as investigated previously (60), because only a high intake of carbohydrates from fruit and vegetables showed a clearly beneficial effect on the subsequent waist changes, whereas carbohydrates from grains and potatoes were either not associated (whole grain) or promoted WC gain (refined grains and potatoes). Total fat intake at baseline had no clear effect on subsequent changes in WC, whereas different effects seemed to be present for subgroups of fat. Furthermore, our findings suggest that protein might play a role that, until now, has not been given so much attention. The results suggest that the focus should not necessarily be on the ratio between types of fat and carbohydrate but also on the types of protein.
In conclusion, no clear associations between total energy intake or energy intake from each of the macronutrients and waist differences were present, except for an inverse association with protein intake, whereas some of the subgroups of macronutrients were associated with DWC. Animal protein was inversely and vegetable fat positively associated with DWC. A strong protective effect in women of carbohydrate energy from fruit and vegetables on WC gain was observed, relative to that of intake from any other type of carbohydrate, and a similar, though insignificant tendency, was seen in men. The results may form a good basis for investigating the biological mechanisms and pathways related to weight changes that may be influenced differently by specific dietary components. Gene-diet interaction analyses might become a strong tool in these future investigations.
ACKNOWLEDGMENTS
We thank Katja Boll (Data Manager) and Jytte Fogh Larsen (Secretary) for assistance with the data collection and Connie Stripp (Dietitian) for assistance with the dietary data.
AT and KO contributed to the original design and data collection for the Diet, Cancer and Health project. JH was responsible for planning the analysis strategies, but all authors contributed to the process. JH performed the statistical analyses, under the supervision of BLT, and drafted the manuscript. TIAS contributed to the conception of the study. All authors contributed to the interpretation of the results, made critical comments during the preparation of the manuscript, and accept responsibility for the work. None of the authors had personal or professional conflicts of interest.
REFERENCES