Inflammatory potential of diet, weight gain, and incidence of overweight/obesity: The SUN cohort

Authors

  • Raúl Ramallal,

    1. Department of Preventive Medicine and Public Health, University of Navarra, Pamplona, Spain
    2. Department of Cardiology, Complejo Hospitalario de Navarra, Pamplona, Spain
    3. IdiSNA, Navarra Institute for Health Research, Pamplona, Spain
    4. Servicio Navarro de Salud, Pamplona, Spain
    Search for more papers by this author
  • Estefanía Toledo,

    1. Department of Preventive Medicine and Public Health, University of Navarra, Pamplona, Spain
    2. IdiSNA, Navarra Institute for Health Research, Pamplona, Spain
    3. Ciber de Fisiopatología de la Obesidad y Nutrición (CIBERObn), Instituto de Salud Carlos III, Madrid, Spain
    Search for more papers by this author
  • J. Alfredo Martínez,

    1. IdiSNA, Navarra Institute for Health Research, Pamplona, Spain
    2. Ciber de Fisiopatología de la Obesidad y Nutrición (CIBERObn), Instituto de Salud Carlos III, Madrid, Spain
    3. Centre for Nutrition Research, University of Navarra, Pamplona, Spain
    Search for more papers by this author
  • Nitin Shivappa,

    1. Cancer Prevention and Control Program, University of South Carolina, Columbia, South Carolina, USA
    2. Department of Epidemiology and Biostatistics, Arnold School of Public Health, University of South Carolina, Columbia, South Carolina, USA
    3. Connecting Health Innovations LLC, Columbia, South Carolina, USA
    Search for more papers by this author
  • James R. Hébert,

    1. Cancer Prevention and Control Program, University of South Carolina, Columbia, South Carolina, USA
    2. Department of Epidemiology and Biostatistics, Arnold School of Public Health, University of South Carolina, Columbia, South Carolina, USA
    3. Connecting Health Innovations LLC, Columbia, South Carolina, USA
    Search for more papers by this author
  • Miguel A. Martínez-González,

    1. Department of Preventive Medicine and Public Health, University of Navarra, Pamplona, Spain
    2. IdiSNA, Navarra Institute for Health Research, Pamplona, Spain
    3. Ciber de Fisiopatología de la Obesidad y Nutrición (CIBERObn), Instituto de Salud Carlos III, Madrid, Spain
    4. Department of Nutrition, Harvard T.H. Chan School of Public Health, Boston, Massachusetts, USA
    Search for more papers by this author
  • Miguel Ruiz-Canela

    Corresponding author
    1. Department of Preventive Medicine and Public Health, University of Navarra, Pamplona, Spain
    2. IdiSNA, Navarra Institute for Health Research, Pamplona, Spain
    3. Ciber de Fisiopatología de la Obesidad y Nutrición (CIBERObn), Instituto de Salud Carlos III, Madrid, Spain
    Search for more papers by this author

  • See Commentary, pg. 987.

  • Funding agencies: The Seguimiento Universidad de Navarra (SUN) study has received funding from the Spanish Ministry of Health and European Regional Development Fund (FEDER) (current grants PI10/02658, PI10/02293, PI13/00615, PI14/1668, PI14/01798, PI14/1754), RD06/0045, G03/140, the Navarra Regional Government (45/2011, 111/2014), and the University of Navarra. NS and JRH were supported by grant number R44DK103377 from the U.S. National Institute of Diabetes and Digestive and Kidney Diseases. Prof. Martínez-González was awarded with an Advanced Research Grant by the European Research Council (#340918). Funding sources had no role in the design, collection, analysis, and interpretation of the data, in the writing, or in the decision to submit the paper for publication.

  • Disclosure: JRH owns controlling interest in Connecting Health Innovations LLC, a company planning to license the right to his invention of the dietary inflammatory index from the University of South Carolina in order to develop computer and smartphone applications for patient counseling and dietary intervention in clinical settings. NS is an employee of Connecting Health Innovations LLC. The subject matter of this paper will not have any direct bearing on that work, nor has that activity exerted any influence on this project.

Abstract

Objective

This study prospectively assessed the association of the inflammatory potential of a diet using the dietary inflammatory index (DII) with average yearly weight changes and incident overweight/obesity.

Methods

Seven thousand and twenty-seven university graduates with body mass index <25 from the Seguimiento Universidad de Navarra (SUN) cohort were followed up during a median of 8.1 years. The DII, a validated tool based on scientific evidence to appraise the relationship between dietary parameters and inflammatory biomarkers, was used. A validated food-frequency questionnaire was used to assess intake of total energy, food, and nutrients, from which DII scores were calculated at baseline and after 10 years of follow-up.

Results

After a median follow-up of 8.1 years, 1,433 incident cases of overweight or obesity were observed. Hazard ratios for overweight/obesity were calculated, including multivariable time-dependent Cox regression models with repeated measures of diet. The hazard ratio for subjects in the highest quartile (most pro-inflammatory diet) was 1.32 (95% confidence interval 1.08-1.60) compared with participants in the lowest quartile (most anti-inflammatory diet), with a significant linear dose-response relationship (P = 0.004). Consistently, increases in average yearly weight gains were significantly associated with proinflammatory diets.

Conclusions

A proinflammatory diet was significantly associated with a higher annual weight gain and higher risk of developing new-onset overweight or obesity.

Introduction

Obesity is characterized by a state of chronic low-grade inflammation [1]. This low-grade inflammation has been suggested to underlie the link between obesity and the increased risk in chronic disease, especially type 2 diabetes and cardiovascular disease. Weight gain has also been associated with an increase in systemic inflammation [2]. However, a bidirectional association between inflammation and obesity has been hypothesized [3], and a low-grade inflammation might be likely to contribute to the development of obesity. Several studies have assessed the prospective association between initially elevated concentrations of inflammatory markers and future weight gain [4-9].

Lifestyle factors are associated with subclinical inflammation [10, 11]. Specifically, dietary patterns have been suggested to modulate inflammation status. Western dietary patterns have been related with a proinflammatory potential. Contrarily, inflammation status seems to be lower with increasing adherence to healthy dietary patterns such as the Mediterranean diet (MedDiet). Therefore, modulation of inflammation may be a major pathway to account for the observed relationships between diet and diet-related diseases [12-15].

The dietary inflammatory index (DII) is a new tool to quantify the inflammatory potential of a diet [16]. We have previously shown a cross-sectional association of a higher proinflammatory potential of diet and increased indices of general and abdominal obesity [17]. In this paper, we studied the prospective association between the DII and subsequent average yearly weight change and the incidence of overweight/obesity in a Mediterranean cohort. Our hypothesis was that a higher proinflammatory diet score would increase the average yearly weight gain and the risk of incident overweight/obesity.

Methods

Ethics statement

The Institutional Review Board of the University of Navarra approved the study protocol. Voluntary completion of the baseline questionnaire implied informed consent.

Study population

The SUN (Seguimiento Universidad de Navarra [University of Navarra follow-up]) is an ongoing, prospective, multipurpose, and dynamic cohort of Spanish university graduates aiming to study determinants of cardiovascular and other chronic diseases. The study design, methods, and the cohort profile have been published in detail elsewhere [18]. Briefly, beginning in December 1999, university graduates answered a questionnaire gathering information about lifestyle factors, sociodemographic variables, and clinical variables, including a detailed food-frequency questionnaire (FFQ). Enrollment is permanently open and participants are followed every 2 years by mailed questionnaires.

Up to November 2011, 21,374 participants had completed their baseline questionnaire. We excluded women pregnant at baseline or during the follow-up (n = 3,240), subjects who reported cardiovascular disease, diabetes, or cancer (at baseline or during follow-up, n = 1,915), subjects with overweight or obesity at baseline or participants with more than 5 kg weight change during the 5 years before entering the cohort (n = 7,202), participants > 65 years old (n = 109), and subjects with energy intake out of predefined values (women: 500-3,500 kcal/d; men: 800-4,000 kcal/d; n = 867). One or more of these reasons for exclusion were met by 8,041 participants. Additionally, we excluded subjects without any follow-up, without any weight measurement, or with an absolute weight change > 30% during follow-up. After exclusions, 7,027 participants remained available for analyses (Figure 1).

Figure 1.

Flowchart of participants: The SUN Project. Energy limits: women: 500-3,500 kcal/d; men 800-4,000 kcal/d [1]. CVD: cardiovascular disease.

Dietary assessment and the DII

Diet at baseline was assessed using 136-item a semiquantitative FFQ (136 food items) previously validated in Spain [19], and the nutrient data bank was updated by using the latest available information included in food composition tables for Spain [20].

The design and the development of the DII have been described elsewhere [16]. Briefly, the DII is computed using a scoring algorithm based on a review of 1,943 articles linking 45 food parameters with six inflammatory biomarkers (IL-1β, IL-4, IL-6, IL-10, TNF-α, and C-reactive protein). Dietary parameters were scored positively (+1) if the effect was proinflammatory, negatively (1) if the effect was anti-inflammatory, or zero if these parameters produced no significant change in inflammatory biomarkers. A food-parameter-specific overall inflammatory score was calculated and multiplied by a centered percentile value of the mean intake for each dietary parameter. These values were used to create a DII score for each participant (individual DII score). Positive DII scores represented the overall dietary patterns with a more proinflammatory potential, whereas negative DII values represented more anti-inflammatory diets. Construct validity of the DII was assessed by analyzing the correlation between the DII and inflammatory biomarkers [16, 21].

For this analysis, a total of 28 dietary parameters obtained from our FFQ could be used to calculate the DII. These dietary parameters were the following: energy intake, protein, carbohydrates, cholesterol, total fat, saturated fat, monounsaturated fat, polyunsaturated fat, trans-fat, omega-3 fatty acids, omega-6 fatty acids, alcohol, fiber, niacin, thiamine, riboflavin, vitamin B12, vitamin B6, vitamin A, vitamin C, vitamin D, vitamin E, folic acid, iron, magnesium, selenium, zinc, and caffeine. The scoring for each food parameter used to calculate the DII is shown in Supporting Information Figure S1.

Adherence to the MedDiet was appraised according to the score proposed by Trichopoulou et al. [22]. Accordingly, one point was assigned to persons whose consumption was at or above the sex-specific median of six components in agreement with the traditional MedDiet (vegetables, fruits/nuts, legumes, fish/seafood, cereals, and monounsaturated to saturated lipid ratio). The participant also received one point if intake was below the median for the two components not in line with the traditional MedDiet (meat or meat products and dairy products). For ethanol, one point was assigned only for moderate amounts of intake (5-25 g/d for women or 10-50 g/d for men). Therefore, this score could range from the highest possible (nine points, reflecting maximum adherence) to the minimum possible (zero points, reflecting no adherence at all).

Assessment of other variables

A baseline questionnaire gathered information about sociodemographic and anthropometric characteristics, health-related habits, and personal and family history of disease. Physical activity was assessed with a previously validated questionnaire, which included 17 different activities during leisure time [23]. Metabolic equivalents (METs) were estimated to yield MET-hours per week scores for each participant. Self-reported weight and height in the SUN cohort were previously validated [24].

Statistical analysis

Multiple linear regression models were fitted to assess the relationship between DII and average yearly weight change during follow-up. Successive degrees of adjustment were used: 1) adjusted for sex and age; 2) additionally adjusted for body mass index (BMI), physical activity (quartiles), hours of TV watching, hours spent sitting down, smoking status (current smoker, never smoker, former smoker), snacking between meals, following a special diet at baseline, parental history of obesity, total energy intake (quartiles), midday nap (siesta), depression (previous or incident), and alcohol intake; 3) additionally adjusted for analgesic use, antidepressant/anxiolytic therapy, and antihypertensive therapy. We also analyzed the association between the DII and the risk of clinically relevant weight gain (with two alternatives, > 3 kg or > 5 kg) during the first 2 years of follow-up using logistic regression models. We adjusted for the same variables included in previous model 2.

We assessed the association between DII score (quartiles) and weight changes within 2-year periods over 10 years of follow-up using generalized estimating equations with an unstructured correlation matrix to account for within-individual repeated measures across follow-up. We adjusted for baseline weight, age, sex, energy intake (quartiles), physical activity (quartiles), time sitting (quartiles), time watching TV (quartiles), smoking status, snacking, special diets at baseline, parental history of obesity, siesta, and depression (both at baseline and during follow-up).

To assess the risk of overweight or obesity across quartiles of DII, we used Cox proportional hazard models to calculate hazard ratios (HRs) and their 95% confidence intervals (CIs). The lowest quartile of DII (most anti-inflammatory) was considered the reference category. We calculated person-years of follow-up for each category from the date of return of the baseline questionnaire to the date of reported overweight (BMI 25 to < 30 kg/m2) or obesity (BMI ≥ 30 kg/m2), death, or the end of follow-up, whichever occurred first. We fitted three models, adjusting for the same variables included in the linear regression models previously described. Tests of linear trends (likelihood ratio test) across increasing quartiles of DII were conducted by assigning the medians to each quartile of the DII and using it as a continuous variable.

To minimize the potential effect of a variation in diet during follow-up, we calculated the average of the DII and energy intake using updated DII scores and energy intake after 10 years of follow-up. We fitted Cox proportional hazard models with repeated measures using the updated data.

Nelson-Aalen survival curves were used to describe the incidence of overweight or obesity over time across quartiles of the DII. We used inverse probability weighting to adjust the Nelson-Aalen curves for potential confounders.

In order to determine the effect of the DII independently of the adherence to the MedDiet, we calculated the residuals of DII by regressing the DII on the MedDiet score. We calculated HRs and 95% CIs for the risk of overweight or obesity using the lowest quartile of these residuals of DII as the reference category.

Subgroup analyses and tests for interactions were conducted according to sex, age (≤40 y or > 40 y), smoking status, family history of obesity, and energy intake (≤2,000 kcal/d or > 2,000 kcal/d). We used sensitivity analyses to estimate HRs and 95% CIs under several assumptions: using laxer energy limits (percentiles 1% to 99%), excluding participants with baseline BMI < 18.5 kg/m2, excluding postmenopausal women, excluding participants with depression at baseline, and restricting our follow-up period to 5 years.

The statistical analyses were performed with Stata® version 12 SE (StataCorp, College Station, Texas). All P values were two-tailed, and significance was set at P < 0.05.

Results

After exclusions, 7,027 participants of the SUN cohort were included in our analysis (Figure 1). Table 1 shows the main baseline characteristics of these participants according to quartiles of DII. The median DII score was 1.74 and ranged from a maximum anti-inflammatory value of 5.12 to a maximum proinflammatory value of +3.96. The mean age of participants was 37.4 years (SD: 10.5), and 35% were men. The mean baseline BMI was 21.9 kg/m2 (SD: 1.9). Age, physical activity (MET-h/wk), hyperlipidemia, total energy intake, alcohol intake, and adherence to the MedDiet tended to decrease with increasing categories of DII. The proportions of men and current smokers were higher in the most proinflammatory group. Supporting Information Table S1 shows the distribution of nutrient intake across categories of the DII. Total, saturated, monounsaturated, and trans-fat intake tended to increase across quartiles of the DII, while protein carbohydrates, and polyunsaturated fat tended to decrease. Vegetables, fruits, cereals, legumes, and overall vitamins, minerals, and fiber intake were higher in the most anti-inflammatory categories of the DII.

Table 1. Characteristics (mean [SD] or percentage) of participants according to quartiles of the dietary inflammatory index (DII) score: The SUN Project
 Q1 (most anti-inflammatory)Q2Q3Q4 (most proinflammatory)P value
N1,7571,7571,7571,756 
Median DII score−3.1−2.1−1.30.6 
(DII: minimum, maximum)(−5.1, −2.5)(−2.49, -1.7)(−1.69, 0.6)(−0.59, 4.0) 
Age (y)39.1 (11.0)38.0 (10.6)36.7 (10.1)35.6 (9.9)<0.001
Sex (%)     
Male34.230.434.442.8<0.001
Marital status (%)     
Married46.845.248.152.30.003
BMI (kg/m2)21.9 (1.8)21.8 (1.9)21.8 (1.9)21.9 (2.0)0.270
Prevalent disease at baseline (%)     
Depression10.210.18.88.90.320
Hypertension4.13.52.63.30.073
Hyperlipidemia22.519.919.718.40.023
Physical activity (MET-h/wk)28.9 (29.8)22.5 (22.1)22.0 (21.1)19.3 (21.0)<0.001
Smoking status (%)    <0.001
Former smoker27.925.124.523.0 
Current smoker17.320.822.325.6 
Time spent sitting down (h/d)5.0 (2.0)5.2 (2.1)5.3 (2.0)5.4 (2.0)<0.001
Time spent watching TV (h/d)1.5 (1.2)1.5 (1.2)1.5 (1.2)1.6 (1.1)0.039
Midday nap (siesta) (%)52.852.050.348.80.090
Total energy intake (kcal/d)2752 (564)2522 (544)2313 (503)1949 (527)<0.001
Alcohol intake (g)7.0 (10.0)5.7 (8.1)5.6 (7.4)5.4 (7.6)<0.001
Adherence to the Mediterranean diet5.2 (1.5)4.4 (1.5)3.6 (1.5)2.7 (1.4)<0.001
Between-meal snacking (%)29.129.430.031.80.300
Special diets at baseline (%)6.44.95.24.00.014
Medication use (%)     
Analgesics9.710.310.89.60.580
Antidepressants/anxiolytics4.15.95.65.30.080
Antihypertensives2.21.41.51.40.230
Total protein intake (% energy)18.0 (3.1)17.9 (2.9)17.8 (2.8)17.6 (3.6)<0.001
Total fat intake (% energy)35.4 (6.7)35.4 (6.1)36.6 (6.0)38.5 (6.7)<0.001
Carbohydrate intake (% energy)44.7 (7.2)45.1 (6.7)43.9 (6.8)41.9 (7.6)<0.001

DII and weight change

Absolute average yearly weight change (g/y) increased across quartiles of DII. Estimated annual weight change ranged from +207.2 g in the most anti-inflammatory group to +264.5 g in the most pro-inflammatory group (Table 2). In the fully adjusted model, participants in the most proinflammatory quartile presented a yearly weight change of +57.3 g (95% CI: 12.5-102.1) higher than those in the lowest quartile (Table 2). Figure 2 shows the adjusted weight over the years of follow-up according to quartiles of DII. A larger weight increase over time was observed in the highest quartile of DII (P for time × DII quartile interaction in the generalized estimating equation model ≤ 0.001).

Figure 2.

Adjusted estimates of average weight during follow-up according to quartiles of dietary inflammatory index (DII) (generalized estimated equation): The SUN Project. Quartiles 2 and 3 were merged into a single (medium) category of DII. Adjusted for sex, age, basal weight, total energy intake, physical activity, hours of TV watching, hours spent sitting down, smoking status, snacking between meals, following a special diet at baseline, siesta, family history of obesity (parents), and depression (previous or incident). P for time × DII interaction < 0.001.

Table 2. Multivariable-adjusted differences (95% confidence intervals) in average yearly weight change (g/y): The SUN Project
 Quartiles of dietary inflammatory index score
 Q1 (most anti-inflammatory)Q2Q3Q4 (most proinflammatory)P for trend
  1. a

    Adjusted for sex, age, baseline BMI, physical activity (quartiles), hours of TV watching, hours spent sitting down, smoking status (current smoker, never smoker, former smoker), snacking between meals, following a special diet at baseline, family history of obesity (parents), total energy intake (quartiles), depression (previous or incident), analgesics use, antidepressant/anxiolytic therapy, antihypertensive therapy, and alcohol intake.

  2. b

    Adjusted for sex, age, baseline BMI, physical activity (quartiles), hours of TV watching, hours spent sitting down, smoking status (current smoker, never smoker, former smoker), snacking between meals, following a special diet at baseline, family history of obesity (parents), total energy intake (quartiles), siesta, depression (previous or incident), and alcohol intake.

  3. c

    Additionally adjusted for analgesics use, antidepressant/anxiolytic therapy, and antihypertensive therapy.

Absolute yearly weight change (g), adjusted meana207.2 (178.0 to 236.4)204.6 (177.1 to 232.2)220.7 (193.2 to 248.1)264.5 (234.5 to 294.5) 
Crude0 (ref)1.8 (−37.0 to 40.6)18.7 (−20.1 to 57.5)62.4 (23.6 to 101.1)0.001
Adjusted for sex and age0 (ref)0.2 (−38.6 to 38.9)11.7 (−27.1 to 50.6)48.9 (9.8 to 88.0)0.007
Multivariable adjustedb0 (ref)−0.7 (−39.8 to 38.4)15.9 (−24.9 to 56.7)59.9 (15.1 to 104.7)0.004
Multivariable adjustedc0 (ref)−2.6 (−41.8 to 36.5)13.4 (−27.4 to 54.2)57.3 (12.5 to 102.1)0.006

The risk of experiencing a relevant weight gain with two alternatives, >3 kg or > 5 kg, within the first 2 years was higher in participants in the highest DII quartile (OR = 1.29 [1.05-1.59] for ≥ 3 kg and OR = 1.43 [1.05-1.95] for ≥ 5 kg) after multivariable adjustment as compared with participants in the lowest quartile of the DII.

DII and incidence of overweight or obesity

The association between quartiles of the DII and the risk of overweight/obesity during follow-up is shown in Table 3. Figure 3 shows the adjusted Nelson-Aalen curves for the incidence of overweight or obesity across baseline quartiles of the DII. During a median follow-up of 8.1 years, we observed 1,433 incident cases (1,409 cases of overweight and 24 cases of obesity). In all models, we found a direct association between the baseline DII and the risk of developing new-onset overweight or obesity during follow-up. In the multivariable fully adjusted model, participants with the highest inflammatory diet had a significant 32% higher relative risk of developing overweight or obesity (95% CI: 8%-60%). A significant linear dose-response relationship was found (P = 0.011). Updated DII and energy intake with information gathered after 10 years of follow-up did not change the reported association.

Figure 3.

Nelson-Aalen estimates of incidence of overweight/obesity across quartiles of the dietary inflammatory index (DII): The SUN Project. Adjusted for age, sex, baseline BMI, physical activity (quartiles), hours of TV watching, hours spent sitting down, smoking status (current smoker, never smoker, former smoker), snacking between meals, following a special diet at baseline, family history of obesity (parents), total energy intake (quartiles), depression (previous or incident), alcohol intake, and siesta using inverse probability weighting. The two intermediate quartiles (Q2 and Q3) were merged to build the medium category for graphical purposes.

Table 3. Hazard ratios (HR) and 95% confidence intervals of incident overweight/obesity according to baseline dietary inflammatory index score in participants of the SUN Project
 Quartiles of dietary inflammatory index score 
 Q1 (most anti-inflammatory)Q2Q3Q4 (most pro-inflammatory)P for trend
  1. a

    Age as underlying time variable.

  2. b

    Adjusted for sex, age, baseline BMI, physical activity (quartiles), hours of TV watching, hours spent sitting down, smoking status (current smoker, never smoker, former smoker), snacking between meals, following a special diet at baseline, family history of obesity (parents), siesta, total energy intake (quartiles), depression (previous or incident), and alcohol intake.

  3. c

    Additionally adjusted for analgesics use, antidepressant/anxiolytic therapy, and antihypertensive therapy.

Cases339344358392 
Person-years of follow-up13,68113,68513,58313,763 
Rate 100/person-years2.482.512.632.85 
Crude HRa1 (ref)1.05 (0.91-1.22)1.12 (0.97-1.30)1.24 (1.07-1.43)0.002
Repeated measures1 (ref)1.02 (0.88-1.19)1.14 (0.99-1.33)1.23 (1.07-1.43)0.001
HR adjusted for sexa1 (ref)1.08 (0.93-1.25)1.11 (0.96-1.29)1.16 (1.00-1.34)0.051
Repeated measures1 (ref)1.05 (0.90-1.21)1.14 (0.99-1.32)1.14 (0.99-1.32)0.047
Multivariable-adjusted HRa, b1 (ref)1.21 (1.02-1.44)1.24 (1.04-1.50)1.33 (1.09-1.61)0.009
Repeated measures1 (ref)1.16 (0.98-1.38)1.31 (1.10-1.56)1.32 (1.09-1.59)0.004
Multivariable-adjusted HRa, c1 (ref)1.21 (1.01-1.44)1.24 (1.03-1.49)1.32 (1.08-1.60)0.011
Repeated measures1 (ref)1.16 (0.98-1.38)1.31 (1.10-1.56)1.32 (1.09-1.59)0.004

Figure 4 shows the association between the DII score and overweight or obesity risk according to subgroup analysis. None of the interaction terms was statistically significant.

Figure 4.

Subgroup analyses: hazard ratio (HR) of overweight or obesity for the highest versus the lowest quartile of the dietary inflammatory index: The SUN Project.

Several sensitivity analyses were conducted to assess the robustness of our results under different assumptions (Table 4). A slight attenuation was found when laxer energy limits were used or when participants with depression at baseline were excluded. A stronger association was found when the follow-up period was restricted to 5 years (HR = 1.53; 95% CI: 1.17-2.01). Similar results were found when additional exclusion criteria were applied.

Table 4. Sensitivity analyses: Hazard ratios (95% confidence intervals)a for the risk of overweight/obesity between extreme quartiles of dietary inflammatory index score (the SUN Project)
 CasesnQ4 vs. Q1 (ref.)P for trendb
  1. a

    Adjusted for sex, age, baseline BMI, physical activity (quartiles), hours of TV watching, hours spent sitting down, smoking status (current smoker, never smoker, former smoker), snacking between meals, following a special diet at baseline, family history of obesity (parents), siesta, total energy intake (quartiles), depression at baseline, and alcohol intake. Age as underlying time variable.

  2. b

    Across the four quartiles of the dietary inflammatory index.

Energy limits: percentiles 1% to 99%1,5337,6251.22 (1.00-1.47)0.061
Excluding participants with basal BMI <18.51,4316,7141.36 (1.11-1.65)0.003
Excluding postmenopausal women1,2545,8291.31 (1.05-1.63)0.007
Excluding participants with depression at baseline1,2476,1751.20 (0.96-1.49)0.104
Restricting follow-up to 5 years7427,0271.53 (1.17-2.01)0.002
Including older subjects1,4577,1131.27 (1.05-1.54)0.029

DII and MedDiet

A moderate inverse association between the Trichopoulou score and the DII was apparent with Pearson's r = 0.55. In order to determine the association between the DII and overweight/obesity independently of the MedDiet, we computed the DII residuals by regressing the DII on the MedDiet score. When using the lowest quartile as reference, the multivariable-adjusted HRs for overweight/obesity across successive quartiles of residuals of the DII were 1.27 (95% CI: 1.07-1.52), 1.28 (95% CI: 1.07-1.55), and 1.47 (95% CI: 1.19-1.80).

In a joint analysis, we observed a higher although not significant risk of overweight/obesity among those participants with high adherence to the MedDiet but high (more proinflammatory) DII (HR = 1.06, 95% CI: 0.85-1.34) when using the high adherence to the MedDiet and low DII as reference group (Supporting Information Figure S2). The highest risk was found among those with low adherence to the MedDiet and high DII (HR = 1.16, 95% CI: 0.99-1.36). The P value for interaction between the DII (quartiles) and the MedDiet (dichotomous) was not statistically significant (P = 0.0578).

Discussion

In this prospective Mediterranean cohort study, we found that a higher proinflammatory diet was associated with a greater risk of clinically relevant weight gain ( > 3 or > 5 kg) and a greater average yearly weight gain. Moreover, a higher proinflammatory diet was associated with a higher risk of overweight or obesity, as compared with those in the lowest quartile of DII (anti-inflammatory diet). In this healthy population, the observed effect of proinflammatory diet was small in terms of average yearly weight change, and most incident cases were overweight instead of obesity. However, these results suggest that a proinflammatory diet can be a risk factor prior to the occurrence of overt obesity independently of other potential confounders such as total energy intake, physical activity, parental history of obesity, and baseline weight. We also found that the DII captures the inflammatory effect of diet independently of the MedDiet. However, a joint analysis suggested that the adverse effect of the proinflammatory capacity of diet may be counterbalanced by a higher adherence to the MedDiet.

The relation concerning obesity and inflammation has been hypothesized to be bidirectional. However, little evidence has been published about a previous inflammatory status as a cause of subsequent obesity. To our knowledge, this is the first study to assess the prospective association between the inflammatory potential of the overall dietary pattern and weight gain or overweight/obesity during a long follow-up period. Previously, we have found in the PREvention with MEDiterranean Diet (PREDIMED) study that a higher (i.e., more proinflammatory) DII was associated with increased indices of general and abdominal obesity [17]. However, the cross-sectional nature of that study prevented identification of the direction of the association between inflammation and adiposity.

Previous findings have reported that high levels of inflammatory biomarkers are found early in the process leading to weight gain. Duncan et al. found in the Atherosclerosis Risk in Communities (ARIC) study that fibrinogen, leukocytes, and other markers of chronic low-grade inflammation promoted weight gain or over a 3-year period in middle-aged adults [4, 5]. Elevated concentrations of inflammation-sensitive plasma proteins predicted weight gain in middle-aged men from the Malmö Preventive Study cohort [6]. Similarly, inflammatory markers were directly associated with weight gain in older [7] and middle-aged participants [8]. This association persisted in the latter study over a long follow-up period of approximately 10 years.

The mechanisms through which a proinflammatory diet induces obesity are unclear. Several animal models have supported the role of inflammatory cytokines in the predisposition of weight gain [25, 26]. Moreover, proinflammatory cytokines, including IL-6, IL-1, and TNF-α, could stimulate appetite, thereby increasing energy intake and fat deposition [27]. Weight gain also could be promoted by β-adrenegic desensitization due to chronic stimulation of the peripheral sympathetic nervous system caused by adiposity signals, such as leptin and insulin, which are related to the inflammatory process [28, 29]. Excess of certain nutrients can also trigger hypothalamic inflammation, which has the potential to cause obesity [30, 31]. Another possible explanation is the effect of diet on changes in the intestinal microbiota, which precede the low-grade inflammation that promotes adiposity [32, 33].

Some potential limitations of our study need to be acknowledged. The use of an FFQ at baseline to assess nutritional information may entail some degree of error in dietary assessment. However, validity and reliability of the FFQ have been previously evaluated, showing good correlation with nutritional intake using repeated food records. In order to avoid possible bias due to changes in diet during follow-up, we conducted repeated-measure analyses updating DII with nutritional information from a new FFQ obtained in the year 10 of follow-up of participants and obtained comparable estimates. In any case, misclassification would be more likely nondifferential and therefore would have biased the results toward the null. Another limitation is that BMI was used as the anthropometric adiposity measure. BMI includes muscle mass, which has been linked to an anti-inflammatory state and may not be a valid measure to evaluate the association between the inflammatory potential of the diet and overweight in healthy young people. However, BMI is a common tool; it is the most frequently used index in epidemiological studies as a surrogate of visceral adipose tissue, and it is highly correlated with waist circumference [34]. Moreover, in a cross-sectional study, the DII showed stronger association with waist circumference and with the waist-to-height ratio than with BMI [17]. Therefore, a greater risk of overweight/obesity would be expected with other alternative indexes that better capture central adiposity. An additional limitation is that other foods or nutrients not included in the DII are associated with a higher risk of overweight/obesity. For example, excess consumption of antioxidant food supplements has been associated with inflammation related to obesity through reactive oxygen species dysregulation [35, 36]. The intake of vitamins or food supplements in the SUN cohort was low, but this factor should be taken into account in future assessments of this association in populations with higher prevalence of supplement use.

Several strengths of our study also deserve to be mentioned. These include its large sample size, a prospective cohort design, long-term follow-up, updated nutritional data, and adjustment for several potential confounders and several sensitivity analyses. The high education level of our participants and the use of validated questionnaires improve the reliability of the self-reported information provided by our participants and ensure a fair degree of homogeneity in our cohort, which reinforces the internal validity of our study by reducing sources of confounding related to socioeconomic or education-related variables. Finally, the exclusion from our analysis of participants with any inflammation-related disease reduces the potential for reverse causality bias.

The DII has been associated with several other inflammatory diseases such as cardiovascular disease [37, 38], depression [39], and cancer [40]. Our results reinforce the concept that the anti-inflammatory potential of the overall dietary pattern may have a substantial value in the prevention against inflammation-related conditions, including overweight/obesity. Further analyses in cohort studies in which participants experienced higher average weight gains and a higher number of obesity incident cases, as well as studies in non-Mediterranean cohorts, are warranted to confirm the association between a proinflammatory diet and overweight and obesity. Randomized experimental studies assessing diet-induced changes in inflammatory biomarkers would be of interest to confirm this prospective association.

Conclusion

A more proinflammatory diet (expressed as a higher DII) was directly associated with the risk of developing overweight or obesity and with a higher average weight gain in a healthy, middle-aged Mediterranean cohort.

Acknowledgments

We are indebted to the participants of the SUN Project for their continued cooperation and participation. We thank the other members of the SUN Group: Alonso A, Barrio López MT, Basterra-Gortari FJ, Benito Corchón S, Bes-Rastrollo M, Beunza JJ, Carlos Chillerón S, Carmona L, Cervantes S, de Irala Estévez J, De la Fuente C, de la Rosa PA, Delgado Rodríguez M, Donat Vargas CL, Donázar M, Fernández Montero A, Galbete Ciáurriz C, García López M, Gea A, Goñi Ochandorena E, Guillén Grima F, Lahortiga F, Llorca J, López del Burgo C, Marí Sanchís A, Martí del Moral A, Martín Calvo N, Núñez-Córdoba JM, Péez de Ciriza P, Pimenta AM, Rico Campa A, Ruiz Zambrana A, Sánchez Adán D, Sayón Orea C, Toledo Atucha J, Vázquez Ruiz Z, Zazpe García I.

Ancillary