Association of dietary inflammatory potential with cardiometabolic risk factors and diseases: a systematic review and dose–response meta-analysis of observational studies

Context The association of dietary inflammatory index (DII®), as an index of inflammatory quality of diet, with cardiometabolic diseases (CMDs) and risk factors (CMRFs) has been inconsistent in previous studies. Objective The current systematic review and dose–response meta-analysis was performed to investigate the association of the DII score with CMDs and CMRFs. Data Sources All published observational studies (cohort, case–control and cross-sectional) using PubMed/Medline, Scopus, ISI Web of Science, and Google Scholar databases were retrieved from inception through November 2019. Data extraction Two reviewers independently extracted the data from included studies. Data analysis Pooled hazard ratio (HR) or odds ratio (OR) were calculated by using a random-effects model. Results Ten prospective cohort studies (total n = 291,968) with 31,069 CMDs-specific mortality, six prospective cohort studies (total n = 43,340) with 1311 CMDs-specific morbidity, two case–control studies with 2140 cases and 6246 controls and one cross-sectional study (total n = 15,613) with 1734 CMDs-specific morbidity were identified for CMDs. Meta-analyses of published observational studies demonstrated that the highest DII score category versus the lowest DII score category was associated with 29% increased risk of CMDs mortality (HR = 1.29; 95% confidence interval (CI) 1.18, 1.41). Moreover, there was a significant association between the DII score and risk of CMDs in cohort studies (HR = 1.35; 95% CI 1.13, 1.61) and non-cohort study (HR = 1.36; 95% CI 1.18, 1.57). We found a significant association between the DII score and metabolic syndrome (MetS) (OR: 1.13; 95% CI 1.03, 1.25), hyperglycemia and hypertension. None-linear dose response meta-analysis showed that there was a significant association between the DII score and risk of CMDs mortality (Pnonlinearity < 0.001). Moreover, evidence of none-linear association between the DII score and risk of CMDs was not observed (p-value = 0.1). Conclusions Adherence to pro-inflammatory diet was associated with increased risk of CMDs, mortality and MetS.

including high sensitivity C-reactive protein (hs-CRP), interleukin (IL)-6, and tumor necrosis factor-α (TNFα). This situation is associated with chronic outcomes including cardiovascular diseases (CVDs) [2], type 2 diabetes mellitus [3], cancer [4], obesity [5], and metabolic syndrome (MetS) and its components [6]. The association of diet with inflammation and CVDs is well demonstrated in previous studies. Adherence to Mediterranean diet, which is characterized by high intake of fruits and vegetables, whole grains, legumes, nuts, fish, and olive oil, decreases chronic inflammation and is associated with lower risk of CVDs [7][8][9][10][11], whereas intake of foods with high amount of sugar, refined grains, red and processed meat, foods with high saturated and trans fatty acids, and sodium (Western diet) is associated with higher levels of chronic inflammation and intermediate markers of CVDs [12].
The dietary inflammatory index (DII) is a novel and validated tool designed in 2009 [13] and updated in 2014 to estimate the inflammatory potential of an individual's diet [14]. According to this index, the food items, macronutrients, and micronutrients (45 food parameters) based on their effect on inflammatory biomarkers (IL-1β, IL-4, IL-6, IL-10, TNF-α, and CRP) were classified into pro-inflammatory, anti-inflammatory, and inflammatory neutral [14].
Given the inconsistent findings, this meta-analysis was conducted to summarize the association of DII with CMRFs and CMDs in observational studies.
Although recently some systematic reviews and metaanalyses have addressed the association between the DII score and CVDs morbidity and mortality [33][34][35] and MetS [34], none of them has evaluated the association of DII score with cardio-metabolic risk factors (e.g. lipid profile, glycemic indices, and anthropometric measures). Moreover, there is no comprehensive systematic review of assessing the association of both continuous and categorical DII score variables with CMRFs (e.g. lipid profile, glycemic indices, anthropometric measures, blood pressure (BP), and metabolic syndrome) and CMDs-specific morbidity and mortality. Therefore, the aim of this systematic review and meta-analysis study was to assess the association of both continuous and categorical DII score variables with risk of CMRFs and risk of CMDs and mortality.

Methods
We followed the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statement for reporting in the current systematic review and mateanalysis study (Additional file 1: Appendix S1).

Search strategy
Published reports with the aim of studying the association of DII score with CMRFs (e.g. glycemic indices, lipid profiles, anthropometric measures, MetS and its components) and CMDs (like MI, IHD, stroke, congestive heart failure, and coronary heart disease (CHD) according to the International Classification of Diseases (ICD)-9-390-465) were included through comprehensive searches on PubMed and the NLM Gateway (for MEDLINE), Scopus, and Institute of Scientific Information (ISI) electronic databases up to February 2020. The appropriate medical subject headings, Entry Terms, and Emtree options were applied to carry out the most sensitive search operations. The search strategy is presented in Additional file 2: Appendix S2. A manual search was performed on Google Scholar database and the references listed in relevant reviews.

Inclusion criteria
Two reviewers (ZA and HA) independently reviewed and screened the appropriate published papers based on title, abstract, and full text. The third reviewer (MQ) resolved any discrepancy in choosing eligible records. All observational studies (cross-sectional, case-control, and cohort) on human subjects without restriction of age group, gender, year of publication, and language examining the association between the DII score with CMRFs (e.g. glycemic indices, lipid profiles, anthropometric measures, MetS and its components) and CMDs were included in the current study.

Exclusion criteria
The papers with the following conditions were excluded: (1) studies that considered the DII as a dependent variable, (2) letters, abstracts and reviews, and (3) duplicated publications. For multiple publications of the same population, only the article with the largest sample size was included.
The participants, intervention, comparators, outcomes, study design criteria are listed in Table 1.

Data extraction
Two investigators (ZA and SD) independently extracted the following information from each qualified study: first author, year of publication, study design, country, age range/mean age, gender, sample size, diet assessment tool, the number of subjects with abnormal CMRFs/ the number of subjects with CMDs, follow-up duration, exposure variable (DII/E-DII), and the number of food items used to calculate it, the type and definition of outcome, outcome assessment method, the type of DII score variable (categorical/continuous), and effect size, study quality, and confounders. Any disagreements were removed by the third author (MQ). Studies which reported correlation or beta coefficient, were included in the systematic review and they were not entered the meta-analyses.

Quality assessment
The quality assessment of included studies was performed by two independent reviewer using Newcastle-Ottawa Scale (NOS) [36]. This scale consists of three portions of the selection, comparability and outcomes/ exposures, and the studies earned maximum nine points. In the present study, the reports with seven or more stars were assumed to have high quality. Any discrepancy between reviewers was resolved by the third reviewer (MQ).

Statistical analysis
All observational studies with any reported effect size (odds ratio (OR), hazard ratio (HR), correlation, or Beta coefficient) were included in qualitative synthesis. Metaanalysis was performed only for studies which reported OR and HR.
In meta-analysis, we examined association of all types of DII [continuous (per one-unit increment), categorical (highest/lowest level) and dose-response association] with CMRFs and CMDs. Meta-analyses were performed separately for CMRFs morbidity, CMDs morbidity, and CMDs related mortality.
We performed random/fixed effects meta-analysis using maximally adjusted OR/HR with 95% confidence interval (CI). Heterogeneity among studies was assessed by I 2 [37][38][39]. There was between-study heterogeneity if I 2 > 50% and p < 0.1 for the result of Q test. If the results showed the heterogeneity, a random-effects model (the DerSimonian-Laird estimator) was applied to assess the pooled OR/HR. The results of the meta-analyses were schematically presented by forest plots.
Dose-response meta-analysis was performed using a method suggested by Greenland and Orsini [40] to assess the dose-response association between DII score and CMDs related morbidity and mortality. The natural logs of the HRs and their CIs across categories of the DII score were used to compute study-specific slopes (linear trends). In this method, the distribution of cases and the HRs with the variance estimates for ≥ 3 quantitative categories of exposure were required. We considered the median or mean values of the DII scores in each category to the corresponding HR for each study. For studies that reported the scores as ranges, the midpoint was estimated in each category by calculating the mean of the lower and upper bound. When the highest and lowest categories were open-ended, the length of these openended intervals was assumed to be the same as that of the adjacent intervals. Restricted cubic splines (three knots at fixed percentiles of 10%, 50%, and 90% of the distribution [41]) was used to examine potential nonlinear doseresponse associations of the DII score with risk of CMDs and mortality.
Publication bias was examined using Egger test and funnel plots. Subgroup analysis according to the type of study design was used to examine the association between the DII score with risk of CMDs and mortality. Sensitivity analysis was performed to assess the effect of removing any of the studies or group of studies on CMDs and CMRFs. All statistical analyses were performed using Stata software version 12 (Stata Corp, College Station, Texas, USA) and p-value < 0.05 was considered statistically significant.

Search results and study selection
A flow diagram for the process of study selection is shown in Fig. 1. The initial search recognized 1,535 papers, and 708 of them remained after duplicate exclusion. Then 653 papers were removed after examining title/abstract and full text of records. The papers were investigated according to the inclusion and exclusion criteria. Eventually, 55 studies were included in the systematic review [15-17, 19-32, 42-79] and 32 records (16 records for CMRFs [17, 19, 20, 23-26, 28, 58, 61, 62, 68, 70, 72, 73, 76] and 18 records for CMDs [15-17, 19, 29-32, 51-57, 77-79]) were selected for meta-analysis. Two studies addressed the association between the DII score and both CMRFs and CMDs outcomes [17,19]. Due to various outcomes of CMRFs, we considered only studies reporting OR along with 95% confidence interval (CI) for MetS or its components in the meta-analysis.

Results of qualitative synthesis Association between the DII score with risk of CMDs and mortality
The positive association between the DII score (as a continuous variable) and risk of CMDs and mortality was observed in three [16,19,57] and six [29,30,54,56,78,79] studies, respectively. Moreover, three records did not indicate the significant association between the DII score and risk of CMDs [31,32,52]. In addition, two studies failed to find any significant association between the DII score and risk of CMDs mortality [31,55]. The DII score (as a categorical variable) was associated significantly with the risk of CMDs in six studies [15-17, 19, 53, 57] and seven reports showed the positive association between the index and risk of CMDs mortality [29,51,54,56,[77][78][79]. Furthermore, three studies did not demonstrate any significant association between the DII score and risk of CMDs [31,32,52]. Moreover, three studies reported no significant association between this index and risk of CMDs mortality [30,31,55]. In one study, a significant association was observed between the DII score and risk of CMDs mortality only in normal and pre-diabetic participants [51].

Results of meta-analysis DII score and risk of CMDs and mortality
Thirteen studies that investigated the association between the DII score (as a continuous variable) and risk of CMDs and mortality were included in this meta-analysis [16, 19, 29-32, 52, 54-57, 78, 79] (Figs. 2 and 3). Subgroup analysis was performed according to the type of outcome (morbidity/mortality) and study design (cohort/ non-cohort) ( Table 4). Results of fixed effect meta-analysis showed that per one-unit increment in the DII score the risk of CMDs mortality increased significantly by 4% (HR = 1.04; 95% CI 1.03, 1.05). Also, a significant association was observed between the continuous DII and risk of CMDs in cohort (HR = 1.06; 95% CI 1.03, 1.09) and non-cohort studies (HR = 1.06; 95% CI 1.03, 1.10). We also assessed the association between the categorical DII score and risk of CMDs and mortality using 18 observational studies [15-17, 19, 29-32, 51-57, 77-79]. Meta-analysis of cohort studies showed that the most pro-inflammatory diet category (the highest DII score group) compared to the most anti-inflammatory diet category (the lowest DII score group), increases the risk of CMDs mortality by 29% (HR = 1.29; 95% CI 1.18, 1.41) (Fig. 4). Also, the association between the DII and risk of CMDs was statistically significant in cohort (HR = 1.35; 95% CI 1.13, 1.61) and non-cohort studies (HR = 1.36; 95% CI 1.18, 1.57) (Fig. 5).

Results of dose-response meta-analysis
In the terms of risk of CMDs mortality in relation to the DII score, nine cohort studies [29,31,51,[54][55][56][77][78][79] were included in dose-response analysis. A significant non-linear positive association was found between the DII score and CMDs mortality (P nonlinearity < 0.001).
Unlike the overall association, the DII score was inversely associated with CMDs mortality from score of − 5 to − 2 (P nonlinearity = 0.01). However, the risk was significantly increased when increasing the score of DII from − 2 to 1.5 (P nonlinearity < 0.001). The slope was slightly flattening from DII score of 1.5 to upper levels (Additional file 3: Figure S1).
Six studies (four cohorts [16,17,31,52], one casecontrol [57] and one cross-sectional study [19]) were included in dose-response analysis assessing the association between the DII score and risk of CMDs (Additional file 4: Figure S2). No significant non-linear association was found in this regard (p-value = 0.1). Such non-significant association was also seen after considering only cohort studies and excluding case-control and cross-sectional studies (p-value = 0.2) (Additional file 5: Figure S3).

Publication bias
No publication bias was observed between studies of MetS according to Egger test results (p-value = 0.323). Moreover, the results of Egger test for studies evaluated the association between the continuous DII score and risk of CMDs and mortality showed that there was no evidence of publication bias between studies (p-value = 0.114, p-value = 0.745, respectively)   Figure  S5). When we considered studies with the categorical DII score, the publication bias was observed in our analysis (P Egger = 0.001 for risk of CMDs and P Egger = 0.04 for risk of CMDs mortaliy) (Additional file 8: Figure S6 and Additional file 9: Figure S7).

Sensitivity analysis
Sensitivity analysis showed that removing any of the studies or a group of studies could not significantly change the effect of DII score (as a continuous or categorical variable) on risk of CMDs and mortality. In terms of MetS and its components, the results of sensitivity analysis demonstrated that neither an individual study nor group of studies had a remarkable effect on our results.

Discussion
The present meta-analysis showed evidences of the association between increasing the inflammatory potential of diet and risk of CMDs and mortality. Also, individuals with the highest pro-inflammatory diet had 13%, 21%, and 17% higher risk for MetS, hyperglycemia and HTN than those with the lowest pro-inflammatory diet.
Subgroup analysis showed that the association of DII (as continuous and categorical variable) with risk of CMDs did not change appreciably in the cohort and non-cohort studies. One important issue in studies on the association of the dietary indices and chronic diseases is the sample size. We can find more precise results using larger sample sizes. Similar findings in the cohort and non-cohort studies can be probably explained by the larger sample size of non-cohort studies.
In the current study, there was a significant association between the DII score and risk of CMDs and mortality. There are some theories that explain the relationship between the DII score and risk of CVDs. Findings of studies showed that higher consumption of pro-inflammatory foods such as red and processed meat, sugar, and refined grains increases level of IL-6, TNF-a, and hs-CRP [12]. Higher levels of these inflammatory biomarkers is the main etiologic factor in CMDs development [80][81][82][83][84]. Since the DII score was calculated using dietary factors (nutrients and specific food items) which show the diet-associated inflammation [14], it was anticipated to observe an association between the DII score and risk of CMDs.  A population-based study including 1,363 men aged 18 years and older (the Geelong osteoporosis study) showed that the adjusted OR (95% CI) for CVDs was 2 (1.01-3.96) for those with pro-inflammatory diet compared with anti-inflammatory diet [53]. The PREDIMED study investigated 7,216 men aged 55-80 years and women aged 60-80 years at high risk of CVDs. A total of 277 CVDs events were considered. The adjusted hazard ratio (95% CI) for CVDs was 1.73 (1.15-2.60) for participants with pro-inflammatory diet. A stronger relationship was showed when cases occurring during the first year of follow-up were excluded from the analysis [16]. Moreover, the in SU.VI.MAX study included 7743 women aged 35-60 years and men aged 45-60 years with 11.4 years follow-up, no statistically significant association was found between the DII score and the composite CVDs outcome. However, a significant relationship was shown for MI when the highest quartile was compared with the lowest quartile of the DII score [52]. Moreover, another cross-sectional study carried out on Sweden men and women aged 30-73 years showed a positive association between the DII score and risk of CMDs [15]. A cohort study on a large sample size of Sweden women indicated that there is not association between the DII score and risk of CVDs mortality [55]. This finding may related to the low number of used dietary factors in DII calculation. In another cohort study on diabetic patients, results showed that there is not any association between the DII score and risk of CVDs mortality that it is not in line our study. This finding can be related to the low sample size and dietary factors used for DII calculation [51].
This meta-analysis of 16 studies examining the association between the DII score and MetS or at least one of its components [17, 19, 20, 23-26, 28, 58, 61, 62, 68, 70, 72, 73, 76], showed a significant association between the DII score and MetS, hyperglycemia and HTN. Several population-based studies carried out in France, Ireland, USA and Iran demonstrated the significant association between the DII score and MetS [23,62,70,73]. However, other studies failed to find this association [20,25,26,28,61,76]. Ramallal et al. [17] in a cohort study on 18,794 Spanish men and women showed the higher DII score is associated with greater incidence of HTN. Also, other studies indicated this significant association [19,24,70,76]. In the regard of hyperglycemia, studies carried out in USA and Iran indicated a positive association between the DII score and hyperglycemia [20,73]. However, some studies did not demonstrate this association [25,26,28,58].
The meta-analysis of 14 studies revealed that subjects in the highest versus the lowest DII score category showed 36% increased risk of CVDs incidence and mortality [33]. Another meta-analysis found that participants with higher DII score had a higher risk of cardiovascular and cancer mortality [30]. The strengths of our study against the other two meta-analyses include the evaluation of the association between the DII score and CMRFs and the dose-response association between the DII score and risk of CMDs and mortality. In addition, we assessed the risk of CMDs separately in all cohort and non-cohort studies. The current study had several limitations. Absence of a specific cut-off point for the association of the DII score and occurrence of morbidity or mortality of CMDs is the first limitation. Most of studies included in the MetS and its components analyses had a cross-sectional design, so the limitations of this type of study should be considered and the results should be interpreted with cautious. Other limitations include different numbers of dietary factors used in the DII score calculation and applying different adjustment models in the analyses. Evidence of publication bias, the other limitation, was observed when the DII score was considered as a categorical variable in the analyses.

Conclusion
The current meta-analysis study showed a positive association between the DII score and risk of CMDs and mortality. Also, we find a significant association between adherence to pro-inflammatory diet and MetS, hyperglycemia, and HTN. More studies with prospective designs and in different societies are needed to confirm the findings.