The Inflammatory Potential of Diet and Pain Incidence: A Cohort Study in Older Adults

Abstract

Background

Despite its importance, evidence regarding pain prevention is inadequate. Leveraging the growing knowledge on how diet regulates inflammation, we examined the association of 3-year changes in the inflammatory potential of diet with pain incidence over the subsequent 3 years.

Methods

We used data from 819 individuals aged ≥60 years and free of pain in 2012, drawn from the Spanish Seniors-ENRICA-1 cohort. The inflammatory potential of diet was estimated via a validated diet history and 2 indices: the dietary inflammatory index (DII) and the empirical dietary inflammatory index (EDII). The frequency, severity, and number of locations of incident pain were combined into a scale that classified participants as suffering from no pain, intermediate pain, or highest pain.

Results

Shifting the diet toward a higher inflammatory potential was associated with subsequent increased risk of highest pain (fully-adjusted relative risk ratio [95% confidence interval] per 1-standard deviation increment in the DII and the EDII = 1.45 [1.16,1.80] and 1.21 [0.98,1.49], respectively) and intermediate pain (0.99 [0.75,1.31] and 1.37 [1.05,1.79]). The 3 components of the pain scale followed similar trends, the most consistent one being pain severity (moderate-to-severe pain: DII = 1.39 [1.11,1.74]; EDII = 1.35 [1.08,1.70]). The association of increasing DII with highest incident pain was only apparent among the less physically active participants (2.08 [1.53,2.83] vs 1.02 [0.76,1.37]; p-interaction = .002).

Conclusion

An increase in the inflammatory potential of diet was associated with higher pain incidence over the following years, especially among the less physically active participants. Future studies in older adults should assess the efficacy of pain prevention interventions targeting the inflammatory potential of diet.

Antiinflammatory, Arthritis, Dietary pattern, Disability, Prospective

Pain is a very common symptom that affects 25%–35% of adults and up to 60% of people older than 65 years (1–3). Years lived with disability caused by low back pain rose by 54% between 1990 and 2015 (4,5), while conditions characterized by the presence of pain accounted for 5 of the top 10 conditions responsible for most of the years lived with disability worldwide (6).

Effective response strategies are hence needed to minimize said burden (4,5,7). On the one hand, evidence on pain prevention is still insufficient, and few interventions―except for physical exercise―have a firm evidence base (7). On the other hand, pain management guidelines recommend nonpharmacological treatments as first-line choices, but this approach is not habitually followed in practice (7). Finally, pharmacological management of pain in older adults may be compromised by polypharmacy, excess toxicity, and risks on cognition and organ systems (8,9).

Since there is evidence that foods and food compounds play a role in the regulation of chronic inflammation (10,11), targeting the inflammatory potential of diet could be an adjunctive pain treatment and primary pain prevention strategy (7). A few studies have associated proinflammatory diets with higher pain in osteoarthritis and fibromyalgia, and higher symptomatic knee osteoarthritis incidence (12–14). Additional evidence supports a vegetarian diet, the Mediterranean diet, and caloric restriction as effective means to reduce pain in patients with rheumatoid arthritis, osteoarthritis, and fibromyalgia syndrome through reductions in chronic inflammation, irrespective of weight loss (15). Nevertheless, a few challenges still lie ahead: (a) more research on how diet modulates pain-related physiology―specifically inflammation―is needed to disentangle commonalities among healthy dietary patterns (16); (b) despite the inadequate evidence on pain prevention strategies, most published dietary studies focus on pain treatment (7); and (c) since pain may disrupt food hedonics, reverse causation may arise when examining diet and pain relationships, so cross-sectional designs might not be appropriate (17).

Objectives

Thus, this study aimed to examine whether changes in the inflammatory potential of diet were associated with subsequent pain incidence among older adults in Spain. To overcome the aforementioned challenges, we: (a) used evidence-based inflammatory dietary patterns (10,11); (b) restricted the analyses to the participants who were pain-free at baseline (18); (c) focused on changes in diet instead of single-time measurements (19); and (d) implemented a longitudinal design without overlapping intervals for diet and pain (20).

Our primary hypotheses were hence that (a) an increase in the inflammatory potential of diet would be associated with subsequent increased pain incidence; and (b) this association would be consistent across the three main components of pain: frequency, severity, and number of locations (2).

Method

Setting, Study Design, and Participants

The Seniors-ENRICA-1 (ClinicalTrials.gov Identifier: NCT01133093) is a cohort of community-dwelling individuals aged 60 years and older in Spain. Study participants were recruited by stratified cluster sampling from March 2008 to September 2010 (wave 0) and followed-up at wave 1 (February 2012 to November 2012) and wave 2 (November 2014 to June 2015) (21,22). We studied whether changes in the inflammatory potential of diet from wave 0 to wave 1 were associated with pain incidence from wave 1 to wave 2.

Trained personnel obtained home-based validated electronic diet histories at waves 0 and 1 and conducted comprehensive sets of physical examinations at all waves. Data on pain (at waves 1 and 2) and morbidity, sociodemographic, and lifestyle variables (at all waves) were gathered through computer-assisted telephone interviews (21,22). The Clinical Research Ethics Committee of “La Paz” University Hospital in Madrid approved the research protocol, and all participants gave written informed consent.

Variables

Inflammatory potential of diet

Food and nutrient consumption was assessed with a face-to-face electronic diet history (21,23), where participants could report up to 861 foods and recipes habitually consumed during a typical week of the previous year (Supplementary Appendix 1). Portion sizes were estimated with 127 digitized photographs and household measures. Nutrient and energy intake were estimated with Spanish and other standard food composition tables (23,24). A previous validation study comparing the results of this diet history against seven 24-hour recalls over 1 year showed a mean correlation coefficient of 0.53 for all 15 food groups considered, 0.76 for energy, and 0.55 for all 41 nutrients studied (23).

Since there is no consensus on how to better estimate the inflammatory potential of diet, we used 2 of the most frequent, evidence-based approaches: the dietary inflammatory index (DII) and the empirical dietary inflammatory index (EDII) (10,11). This allowed us to check the robustness of study results, for the DII is mostly based on a priori knowledge on nutrients, food compounds, and inflammation, while the EDII was empirically derived over food groups, using a more limited set of inflammatory biomarkers (10,11). For the analyses, we calculated changes in each index from wave 0 to wave 1, so that positive values indicated a shift to a more proinflammatory diet.

Dietary inflammatory index

The DII is a literature-derived index designed to compare diverse populations based on the inflammatory potential of their diets, which was built by Shivappa et al. as follows (10). First, a review of 1 943 articles identified 45 food components that either increased, decreased, or had no effect on 6 inflammatory biomarkers (interleukin-1β, interleukin-4, interleukin-6, interleukin-10, tumor-necrosis-factor-α, and C-reactive protein). Second, every food component was assigned a weighted overall inflammatory effect score, ranging from −0.663 (maximally antiinflammatory) to 0.373 (maximally proinflammatory).

To obtain the DII, we multiplied the overall inflammatory effect scores of the 32 food components available in our study by their corresponding standardized intakes (note that individuals’ intakes were normalized using a compilation of eleven food consumption datasets from countries around the world) and then summed them, so that higher values indicate a more proinflammatory diet (10). The inflammatory effect scores and the global versus our study intakes (waves 0 and 1) of the components of the DII are shown in Supplementary Table 1.

Empirical dietary inflammatory index

The EDII is a hypothesis-driven, empirically derived index that assesses diet quality based on its inflammatory potential. It was developed―and further validated―by Tabung et al. (11). The authors entered 39 predefined food groups in reduced rank regression models followed by stepwise linear regression analyses to identify the dietary pattern most predictive of 3 plasma inflammatory markers (interleukin-6, tumor-necrosis-factor-α receptor 2, and C-reactive protein). Eighteen food groups were retained in the final stepwise linear regression model and their regression coefficients―ranging from −1 175 (maximally antiinflammatory) to 252 (maximally proinflammatory)―were recorded.

To compute the EDII, we first weighted and then summed the consumption of such 18 food groups by their corresponding regression coefficients. Second, this weighted sum was standardized by subtracting its mean and dividing by its standard deviation. As with the DII, higher values indicate a more proinflammatory diet (11). Said regression coefficients (inflammatory effect scores) and the consumption of the components of the EDII in our study (waves 0 and 1) are presented in Supplementary Table 2.

Pain

Self-reported pain in the previous 6 months was assessed at waves 1 and 2 via a pain scale developed from the Survey on Chronic Pain in Europe (2,25,26), consisting of 3 components: (a) pain frequency, classified as either sporadic (pain happening <1 time/month, 1–3 times/month, or weekly; which was scored 1 point) or persistent pain (≥2 times/week, every day, or at all times; scored 2 points―please note that it does not necessarily reflect chronic pain, as we lacked data on the onset of pain); (b) pain severity, classified as light (pain troubling a little or nothing on daily activities; which was scored 1 point) or moderate-to-severe pain (troubling moderately, a lot, or completely; scored 2 points); and (c) pain locations out of the 6 considered (head and neck, back, bones and joints, arms, legs, and other sites―abdomen, chest, or any other site), further classified as 1–2 pain sites (scored 1 point) or ≥3 pain sites (scored 2 points).

The pain scale was built as the summed score of said three components (ie, pain frequency, pain severity, and pain locations). It ranged from 0 to 6 points and was categorized as follows: (a) no pain in the previous 6 months (0 points in the pain scale); (b) intermediate pain (3 or 4 points); and (c) highest pain (5 or 6 points). Finally, we defined incident pain as the presence of intermediate or highest pain at wave 2 among the participants who had scored 0 points on the pain scale at wave 1. Despite being developed from a widely used pain survey (2), our pain questionnaire and scale have not been validated.

Potential confounders

We considered several possible confounders of the association between the inflammatory potential of diet and pain incidence. First, sociodemographic characteristics, namely sex, age, living conditions (ie, we asked study participants to grade how difficult it was for them to make ends meet, from very difficult to very easy), and educational level (primary or less, secondary, or university).

Second, lifestyle variables, specifically tobacco smoking (never, former, or current), alcohol consumption (never, former, moderate [≤10 g/day in women and ≤20 g/day in men], or heavy), energy intake (kcal/day), and diet quality (estimated with the Alternate Healthy Eating Index) (27). Recreational physical activity was assessed with the validated questionnaire developed in the EPIC-cohort study in Spain and expressed as metabolic equivalents of task-hours/week (MET-hours/week) (28), while time spent watching television (hours/day) was assessed with the Nurses’ Health Study questionnaire, validated in Spain (29). Finally, body mass index (BMI) was calculated as weight (kg) divided by height (m) squared, both measured under standardized conditions (30).

Third, we considered several comorbidities: (a) diabetes―either treatment with antidiabetic drugs or self-reported medical diagnosis; (b) self-reported cardiovascular disease (coronary heart disease, stroke, or heart failure), respiratory disease, musculoskeletal disease (osteoarthritis, arthritis, or hip fracture), cancer, and depression requiring medical treatment; and (c) cognitive impairment (Mini-Mental State Examination score <24 points) (31).

Statistical Methods

Study size

From the 2 519 participants who were followed-up at wave 1 (year 2012), we excluded 1 086 (43.1%) prevalent pain cases and 94 (3.7%) participants with no information on pain at wave 1. From the remaining 1 339 individuals free of pain at wave 1, 44 participants (3.3%) had died, and 306 (22.9%) had been lost to follow-up at wave 2 (year 2015). From these 989 participants who were followed-up at wave 2, we further excluded 170 (17.2%) with inadequate data (101 participants had no information on diet at wave 0 or wave 1, 74 on pain at wave 2, and 100 on potential confounders at wave 1; note that one individual may lack data in more than one variable). Hence, the analytical sample comprised 819 individuals (Supplementary Figure 1).

Descriptive data

Differences in characteristics of study participants at wave 1 across categories of change in the DII or the EDII from wave 0 to wave 1 were evaluated with Pearson’s chi-squared tests for discrete variables and Wilcoxon rank-sum tests for continuous variables. To investigate how prevalent pain, loss to follow-up, and inadequate data may have affected our findings, we also used such tests to compare the wave 1 characteristics between the participants included in the analyses and those excluded for each of these three reasons.

Main statistical methods and control for confounding

Relative risk ratios (RRR) and their 95% confidence interval (CI) for intermediate pain (3 or 4 points in the pain scale) and highest pain incidence (5 or 6 points in the pain scale)―as well as those for the 3 pain components―were calculated using multinomial (polytomous) logistic regression models. Changes in the inflammatory potential of diet from wave 0 to wave 1, estimated via the DII or the EDII, were modeled in the analyses as: (a) a continuous variable (per 1-standard deviation [SD] increment); (b) a dichotomous variable (decrease [negative values; reference] vs increase [positive values]); and (c) a restricted cubic spline (knots located at the 10th, 50th, and 90th percentiles) (32). We used 2 a priori incrementally adjusted models to control for potential confounding at wave 1: the first, adjusted for sociodemographic characteristics, and the second, additionally adjusted for lifestyle variables and comorbidities.

Interactions and sensitivity analyses

Since pain prevalence has been found to vary according to sex (2,3), while age, tobacco smoking, recreational physical activity, and BMI are important determinants of pain incidence (1,4,5), we examined if these variables modified the main study associations by using likelihood-ratio tests that compared models with and without interaction terms, defined as the product of said variables by the continuous DII or EDII variables. Since a significant interaction between the DII and recreational physical activity was found, stratified results are also presented.

To check the robustness of the study associations, we also conducted 6 sensitivity analyses (Supplementary Appendix 2).

Analyses were performed with Stata (StataCorp, College Station, TX), version 14.

Results

Descriptive and Outcome Data

Characteristics of study participants are shown in Table 1. Compared with individuals who decreased their DII from wave 0 to wave 1, those who increased it had a lower educational level, energy intake, diet quality, BMI, and prevalence of cognitive impairment at wave 1. Participants who increased their EDII were less often heavy drinkers and suffered less often from depression at wave 1. Characteristics of the individuals included in the analyses, those excluded because of prevalent pain at wave 1, those who were not followed at wave 2, and those with inadequate data are shown in Supplementary Table 3.

Table 1.

Characteristics of 819 Older Adults Free of Pain, by Categories of Change in the Dietary Inflammatory Index (DII) and the Empirical Dietary Inflammatory Index (EDII) Over the Previous 3.2 Years

	Change in the DII†		Change in the EDII‡
	Decrease	Increase	Decrease	Increase
n	393	426	416	403
Change in the DII	−1.70 (1.30)	1.61 (1.22)*	0.05 (2.14)	−0.00 (2.01)
Change in the EDII	−0.02 (1.20)	−0.07 (1.13)	−0.86 (0.90)	0.80 (0.71)*
Sex―Men (%)	212 (53.9)	248 (58.2)	236 (56.7)	224 (55.6)
Age (years)	71.8 (5.76)	71.7 (6.02)	71.9 (6.03)	71.5 (5.75)
Living conditions (make ends meet)§	4.57 (1.03)	4.49 (1.04)	4.49 (1.06)	4.57 (1.00)
Educational level (%)
Primary or less	162 (41.2)	214 (50.2)*	192 (46.2)	184 (45.7)
Secondary	113 (28.8)	127 (29.8)	112 (26.9)	128 (31.8)
University	118 (30.0)	85 (20.0)	112 (26.9)	91 (22.6)
Tobacco smoking (%)
Never	222 (56.5)	234 (54.9)	231 (55.5)	225 (55.8)
Former	136 (34.6)	153 (35.9)	147 (35.3)	142 (35.2)
Current	35 (8.91)	39 (9.15)	38 (9.13)	36 (8.93)
Alcohol consumption (%)
Never	71 (18.1)	63 (14.8)	58 (13.9)	76 (18.9)*
Former	57 (14.5)	59 (13.8)	50 (12.0)	66 (16.4)
Moderate‖	159 (40.5)	201 (47.2)	175 (42.1)	185 (45.9)
Heavy	106 (27.0)	103 (24.2)	133 (32.0)	76 (18.9)
Recreational physical activity (MET-hours/week)	24.0 (14.4)	22.7 (15.4)	22.5 (14.2)	24.2 (15.6)
Sedentary behavior (TV hours/day)	2.58 (1.41)	2.61 (1.36)	2.64 (1.43)	2.55 (1.33)
Energy intake (kcal/day)	2 102 (501)	1 957 (383)*	2 005 (461)	2 049 (436)
Diet quality (AHEI)¶	65.0 (9.50)	61.0 (9.43)*	63.4 (9.73)	62.3 (9.58)
Body mass index (kg/m2)	28.3 (4.21)	27.4 (3.90)*	27.8 (3.96)	27.9 (4.20)
Diabetes (%)#	79 (20.1)	99 (23.2)	87 (20.9)	91 (22.6)
Cardiovascular disease (%)	34 (8.65)	31 (7.28)	29 (6.97)	36 (8.93)
Chronic lung disease (%)	30 (7.63)	39 (9.15)	31 (7.45)	38 (9.43)
Musculoskeletal disease (%)	184 (46.8)	207 (48.6)	197 (47.4)	194 (48.1)
Cancer (%)	12 (3.05)	21 (4.93)	15 (3.61)	18 (4.47)
Depression (%)	32 (8.14)	33 (7.75)	41 (9.86)	24 (5.96)*
Cognitive impairment (%)**	14 (3.56)	6 (1.41)*	10 (2.40)	10 (2.48)

Notes: Values are numbers (%) or means (standard deviations). AHEI = Alternate Healthy Eating Index; MET-hours/week = metabolic equivalents of task-hours/week.

*p Value < .05 for differences in means (Wilcoxon rank-sum) or proportions (Pearson’s chi-squared) across the categories of change in the DII and the EDII.

^†DII change categories: decrease, −6.57 to <0; increase, >0 to 5.95; note that no participant had a DII change equal to zero.

^‡EDII change categories: decrease, −9.52 to <0; increase, >0 to 5.37; note that no participant had a EDII change equal to zero.

^§Living conditions: difficulty to make ends meet, from very difficult (1) to very easy (6).

^‖Moderate drinking: ≤10 g/day in women and ≤20 g/day in men.

^¶AHEI: higher scores indicate better diet quality (range: 25.3 to 93.3).

^#Diabetes: treated with antidiabetic drugs or diabetes diagnosis.

^**Cognitive impairment: Mini-Mental State Examination score <24 points.

During the mean 3.2-year follow-up from wave 0 to wave 1, 393 participants (48.0%) decreased and 426 (52.0%) increased their DII (mean change [SD] = 0.02 [2.07]); corresponding figures for the EDII were 416 (50.8%) and 403 (49.2%; −0.04 [1.16]; Supplementary Table 3). Throughout the subsequent mean 2.8 years of follow-up from wave 1 to wave 2, 71 individuals developed intermediate pain (8.7%) and 140 highest pain (17.1%; Table 2).

Table 2.

Relative Risk Ratios (95% confidence interval) for the Association of 3.2-Year Changes in the Dietary Inflammatory Index (DII) and the Empirical Dietary Inflammatory Index (EDII) With Pain Incidence and That of Its Components Over the Subsequent 2.8 Years in 819 Older Adults

	Change in the DII†			Change in the EDII‡
	Decrease	Increase	Per 1-SD Increment	Decrease	Increase	Per 1-SD Increment
Pain scale
Intermediate pain versus no pain
Cases/n	36/393	35/426	71/819	34/416	37/403	71/819
Model 1§	Ref.	1.00 (0.60,1.64)	0.98 (0.77,1.25)	Ref.	1.16 (0.71,1.91)	1.23 (0.96,1.58)
Model 2‖	Ref.	1.00 (0.58,1.73)	0.99 (0.75,1.31)	Ref.	1.33 (0.79,2.24)	1.37 (1.05,1.79)*
Highest pain versus no pain
Cases/n	56/393	84/426	140/819	67/416	73/403	140/819
Model 1§	Ref.	1.56 (1.06,2.29)*	1.21 (1.00,1.45)*	Ref.	1.21 (0.83,1.76)	1.16 (0.96,1.40)
Model 2‖	Ref.	2.13 (1.37,3.33)***	1.45 (1.16,1.80)**	Ref.	1.19 (0.79,1.79)	1.21 (0.98,1.49)
Components of the pain scale
Pain frequency
Sporadic pain versus no pain
Cases/n	23/393	29/426	52/819	29/416	23/403	52/819
Model 1§	Ref.	1.36 (0.76,2.42)	1.05 (0.79,1.40)	Ref.	0.87 (0.49,1.54)	1.12 (0.84,1.49)
Model 2‖	Ref.	1.50 (0.80,2.83)	1.12 (0.81,1.55)	Ref.	0.93 (0.51,1.71)	1.20 (0.89,1.63)
Persistent pain versus no pain
Cases/n	69/393	90/426	159/819	72/416	87/403	159/819
Model 1§	Ref.	1.33 (0.93,1.90)	1.14 (0.96,1.36)	Ref.	1.33 (0.93,1.89)	1.21 (1.01,1.45)*
Model 2‖	Ref.	1.64 (1.09,2.46)*	1.32 (1.08,1.62)**	Ref.	1.38 (0.94,2.02)	1.28 (1.05,1.57)*
Pain severity
Light pain versus no pain
Cases/n	42/393	42/426	84/819	43/416	41/403	84/819
Model 1§	Ref.	1.03 (0.65,1.64)	1.04 (0.83,1.31)	Ref.	1.04 (0.65,1.64)	1.09 (0.87,1.36)
Model 2‖	Ref.	1.10 (0.66,1.82)	1.11 (0.85,1.44)	Ref.	1.11 (0.69,1.80)	1.16 (0.91,1.48)
Moderate-to-severe pain versus no pain
Cases/n	50/393	77/426	127/819	58/416	69/403	127/819
Model 1§	Ref.	1.60 (1.07,2.39)*	1.18 (0.98,1.43)	Ref.	1.31 (0.89,1.94)	1.26 (1.03,1.55)*
Model 2‖	Ref.	2.16 (1.36,3.42)**	1.39 (1.11,1.74)**	Ref.	1.35 (0.89,2.07)	1.35 (1.08,1.70)**
Pain locations
1–2 pain sites versus no pain
Cases/n	43/393	59/426	102/819	45/416	57/403	102/819
Model 1§	Ref.	1.37 (0.89,2.11)	1.10 (0.90,1.36)	Ref.	1.35 (0.88,2.07)	1.35 (1.09,1.68)**
Model 2‖	Ref.	1.49 (0.94,2.38)	1.17 (0.93,1.47)	Ref.	1.45 (0.93,2.26)	1.45 (1.15,1.83)**
≥3 pain sites versus no pain
Cases/n	49/393	60/426	109/819	56/416	53/403	109/819
Model 1§	Ref.	1.29 (0.85,1.97)	1.14 (0.93,1.40)	Ref.	1.05 (0.69,1.59)	1.04 (0.85,1.28)
Model 2‖	Ref.	1.76 (1.08,2.87)*	1.39 (1.08,1.78)**	Ref.	1.03 (0.65,1.64)	1.08 (0.86,1.35)

Notes: *p < .05; **p < .01. MET-hours/week = metabolic equivalents of task-hours/week; SD = standard deviation.

^†DII change categories: decrease, −6.57 to <0; increase, >0 to 5.95; note that no participant had a DII change equal to zero.

^‡EDII change categories: decrease, −9.52 to <0; increase, >0 to 5.37; note that no participant had a EDII change equal to zero.

^§Model 1: multinomial logistic regression model adjusted for sex, age, living conditions (make ends meet), and educational level (primary or less, secondary, or university).

^‖Model 2: as Model 1 and additionally adjusted for smoking status (never, former, or current), alcohol consumption (never, former, moderate, or heavy), leisure-time physical activity (MET-hours/week), sedentary behavior (TV hours/day), energy intake (kcal/day), diet quality (Alternate Healthy Eating Index), body mass index (kg/m²), diabetes, cardiovascular disease, chronic lung disease, musculoskeletal disease, cancer, depression, and cognitive impairment at wave 1.

Main Results

The associations of changes in the DII and EDII (modeled as continuous and dichotomous variables) with pain incidence are shown in Table 2. Shifting the diet to a higher inflammatory potential was associated (via either the DII or the EDII) with subsequent increased risk of highest pain (model 2 RRR [95% CI] per 1-SD increment in the DII = 1.45 [1.16,1.80] and per 1-SD increment in the EDII = 1.21 [0.98,1.49]) and of intermediate pain (0.99 [0.75,1.31] and 1.37 [1.05,1.79], respectively). Consistent results were observed when modeling changes in the DII and EDII as restricted cubic splines (Figure 1).

Figure 1.

Relative risk ratios (RRR) for the association of 3.2-year changes in the dietary inflammatory index (DII) and the empirical dietary inflammatory index (EDII) with pain incidence over the subsequent 2.8 years in 819 older adults. Notes: Plotted values are RRR (95% confidence interval) from a multinomial logistic regression model as Model 2 in Table 2, adjusted for sex, age, living conditions (make ends meet), educational level (primary or less, secondary, or university), smoking status (never, former, or current), alcohol consumption (never, former, moderate, or heavy), leisure-time physical activity (metabolic equivalents of task-hours/week), sedentary behavior (TV hours/day), energy intake (kcal/day), diet quality (Alternate Healthy Eating Index), body mass index (kg/m²), diabetes, cardiovascular disease, chronic lung disease, musculoskeletal disease, cancer, depression, and cognitive impairment at wave 1. Restricted cubic spline knots for the change in the DII: −2.56, 0.10, 2.56. Reference: −2.32. Restricted cubic spline knots for the change in the EDII: −1.28, −0.02, and 1.33. Reference: −1.17.

The 3 components of the pain scale followed trends comparable to those of the whole pain scale: a 1-SD increment in both the DII and the EDII was associated with higher risk of moderate-to-severe pain (1.39 [1.11,1.74] and 1.35 [1.08,1.70], respectively); persistent pain (model 2 RRR [95% CI] per 1-SD increment in the DII = 1.32 [1.08,1.62] and per 1-SD increment in the EDII = 1.28 [1.05,1.57]); pain in 1–2 locations (1.17 [0.93,1.47] and 1.45 [1.15,1.83], respectively); and pain in ≥3 locations (1.39 [1.08,1.78] and 1.08 [0.86,1.35], respectively).

The association of increasing DII with highest pain incidence was only apparent among the participants who were less physically active (≤18 MET-hours/week: model 2 RRR [95% CI] per 1-SD increment in the DII = 2.08 [1.53,2.83]; >18 MET-hours/week: 1.02 [0.76,1.37]; p for interaction = .002; Table 3). This differential association was also observed for the three components of the pain scale (Table 3). However, no interaction between the EDII and physical activity was evident (p for interaction = .64). Neither age, nor sex, tobacco smoking, or BMI significantly modified the study associations (data not shown).

Table 3.

Relative Risk Ratios (95% confidence interval) for the Association of 3.2-Year Changes in the Dietary Inflammatory Index (DII) With Pain Incidence and That of Its Components Over the Subsequent 2.8 Years in 819 Older Adults, Stratified by Recreational Physical Activity Levels

	Change in the DII†
	Lower Physical Activity‡			Higher Physical Activity‡
	Decrease	Increase	Per 1-SD Increment	Decrease	Increase	Per 1-SD Increment
Pain scale
Intermediate pain versus no pain
Cases/n	15/171	19/214	34/385	21/222	16/212	37/434
Model 1§	Ref.	1.30 (0.63,2.70)	1.08 (0.76,1.54)	Ref.	0.78 (0.39,1.55)	0.88 (0.62,1.24)
Model 2‖	Ref.	1.42 (0.66,3.04)	1.11 (0.76,1.64)	Ref.	0.72 (0.35,1.51)	0.90 (0.63,1.31)
Highest pain versus no pain
Cases/n	22/171	58/214	80/385	34/222	26/212	60/434
Model 1§	Ref.	3.06 (1.74,5.38)***	1.63 (1.24,2.13)***	Ref.	0.74 (0.42,1.30)	0.85 (0.65,1.12)
Model 2‖	Ref.	4.68 (2.48,8.83)***	2.08 (1.53,2.83)***	Ref.	0.96 (0.52,1.77)	1.02 (0.76,1.37)
Components of the pain scale
Pain frequency
Sporadic pain versus no pain
Cases/n	6/171	18/214	24/385	17/222	11/212	28/434
Model 1§	Ref.	3.37 (1.28,8.86)*	1.54 (1.01,2.37)*	Ref.	0.68 (0.31,1.50)	0.76 (0.51,1.12)
Model 2‖	Ref.	3.91 (1.43,10.7)**	1.64 (1.03,2.60)*	Ref.	0.70 (0.30,1.64)	0.82 (0.54,1.26)
Persistent pain versus no pain
Cases/n	31/171	59/214	90/385	38/222	31/212	69/434
Model 1§	Ref.	2.09 (1.25,3.47)**	1.39 (1.08,1.78)**	Ref.	0.79 (0.47,1.34)	0.91 (0.70,1.17)
Model 2‖	Ref.	2.77 (1.59,4.83)***	1.67 (1.27,2.21)***	Ref.	0.93 (0.53,1.64)	1.05 (0.80,1.39)
Pain severity
Light pain versus no pain
Cases/n	17/171	23/214	40/385	25/222	19/212	44/434
Model 1§	Ref.	1.40 (0.71,2.77)	1.10 (0.78,1.53)	Ref.	0.79 (0.42,1.49)	0.99 (0.72,1.36)
Model 2‖	Ref.	1.59 (0.78,3.25)	1.19 (0.82,1.72)	Ref.	0.78 (0.39,1.54)	1.06 (0.75,1.50)
Moderate-to-severe pain versus no pain
Cases/n	20/171	54/214	74/385	30/222	23/212	53/434
Model 1§	Ref.	3.16 (1.76,5.67)***	1.67 (1.26,2.20)***	Ref.	0.73 (0.40,1.33)	0.78 (0.59,1.04)
Model 2‖	Ref.	4.70 (2.45,9.02)***	2.07 (1.51,2.84)***	Ref.	0.94 (0.49,1.79)	0.92 (0.67,1.26)
Pain locations
1–2 pain sites versus no pain
Cases/n	18/171	35/214	53/385	25/222	24/212	49/434
Model 1§	Ref.	2.04 (1.09,3.81)*	1.33 (0.99,1.80)	Ref.	0.93 (0.51,1.71)	0.90 (0.67,1.21)
Model 2‖	Ref.	2.31 (1.20,4.43)*	1.44 (1.05,1.98)*	Ref.	0.98 (0.52,1.86)	0.96 (0.70,1.32)
≥3 pain sites versus no pain
Cases/n	19/171	42/214	61/385	30/222	18/212	48/434
Model 1§	Ref.	2.57 (1.40,4.73)**	1.51 (1.12,2.03)**	Ref.	0.60 (0.32,1.13)	0.83 (0.61,1.12)
Model 2‖	Ref.	4.11 (2.05,8.26)***	2.02 (1.43,2.85)***	Ref.	0.73 (0.36,1.46)	0.99 (0.70,1.38)

Notes: *p < .05; **p < .01; ***p < .001. MET-hours/week = metabolic equivalents of task-hours/week; SD = standard deviation.

^†DII change categories: decrease, −6.57 to <0; increase, >0 to 5.95; note that no participant had a DII change equal to zero.

^‡Recreational physical activity categories: lower, 0 to ≤18 MET-hours/week; higher, >18 to 101 MET-hours/week.

^§Model 1: multinomial logistic regression model adjusted for sex, age, living conditions (make ends meet), and educational level (primary or less, secondary, or university).

^‖Model 2: as Model 1 and additionally adjusted for smoking status (never, former, or current), alcohol consumption (never, former, moderate, or heavy), sedentary behavior (TV hours/day), energy intake (kcal/day), diet quality (Alternate Healthy Eating Index), body mass index (kg/m²), diabetes, cardiovascular disease, chronic lung disease, musculoskeletal disease, cancer, depression, and cognitive impairment at wave 1.

Sensitivity Analyses

When calculating the DII without considering alcohol intake and the EDII without the beer and wine items, the association of the former with highest pain incidence remained, but that of the latter with increased intermediate pain was all but lost. When computing alternate versions of the EDII with (a) proinflammatory scoring for snacks, fruit juice, and pizza; and (b) antiinflammatory scoring for fish, tomatoes, and other vegetables, its association with increased intermediate pain incidence vanished (Supplementary Tables 4 and 5).

The associations of changes in the inflammatory potential of diet with incident pain would generally hold: (a) when assessing pain intensity with a numeric rating scale; (b) if all the participants lost to follow-up had remained free of pain, developed intermediate pain, or developed highest pain at wave 2; and (c) when analyzing changes in the pain scale from wave 1 to wave 2 (Supplementary Tables 4 and 5).

Nevertheless, changes in either the DII or EDII were not related to prevalent pain at wave 1 (Supplementary Tables 4 and 5) and, even though the association between changes in the DII and highest pain incidence (both in the whole sample and among the less physically active participants) would remain significant when using a false discovery rate of 5%, that between changes in the EDII and intermediate pain incidence would not (Supplementary Table 6).

Discussion

Key Results

In this cohort study of older adults in Spain, we found that an increase in the inflammatory potential of diet was associated with subsequent higher pain incidence. The 3 components of the pain scale followed similar trends, the most consistent one being pain severity (moderate-to-severe pain). The association of increasing DII with highest incident pain was only apparent among the participants who were less physically active.

Interpretation

Relevant findings from other published studies

Our results are mostly in line with the few studies that have directly examined the association between the inflammatory potential of diet and pain-related outcomes. First, a cross-sectional study on 220 patients with knee osteoarthritis from Iran found that an increase in the DII was associated with higher pain intensity, according to the Visual Analogue Scale, and with lower pain-related quality of life, evaluated via the 36-item short-form health survey (12). Second, in a case-control study on 95 Spanish patients with fibromyalgia syndrome, no association between the DII and the Visual Analogue Scale was found in either cases or controls, but a consistent association of a higher DII with lower pressure pain thresholds―a measure of pain hypersensitivity―on 8 tender point sites was observed among cases (13). Third, in a U.S. cohort study including 2 940 participants, a higher DII at baseline was associated with increased symptomatic knee osteoarthritis incidence (ie, a combination of frequent knee pain and radiographic knee osteoarthritis) over 4 years (14).

Possible mechanisms and explanations

The relationship of the DII and EDII with pain may be partially mediated by chronic, low-grade, systemic inflammation. This may be supported by the fact that certain components of these dietary inflammatory indices, such as foods (eg, garlic, onion, and tea), nutrients (eg, saturated fatty acids, omega-3 fatty acids, and vitamin D), and bioactive compounds (like caffeine and flavonoids) have shown various associations with inflammatory markers, as interleukin-1β, interleukin-4, interleukin-6, interleukin-10, tumor-necrosis-factor-α, and C-reactive protein (10,11,33). A thorough description of the possible mechanisms behind said associations arguably lies outside the scope of this section. To cite a few, saturated fatty acids may lead to increased inflammation due to accumulation of diacylglycerol and ceramide, activation of mitogen-activated protein kinases and subsequent induction of inflammatory genes, decreased oxidation of glucose and fatty acids, and recruitment of immune cells to white adipose tissue and muscle (34). Conversely, omega-3 fatty acids might decrease the production of proinflammatory eicosanoids, reactive oxygen and nitrogen species, and cytokines, additionally generating antiinflammatory mediators (35).

Credible explanations for local and systemic inflammation being potential mediators of acute and chronic pain exist as well (1). Specifically, acute noxious stimulation seems to result in the activation of the immune system and the release of inflammatory cytokines (15). Such molecules work to maintain homeostasis and attract cells from both the innate and adaptive immune systems, which are involved in the recognition, repair, and removal of damaged cells (15). It is then plausible that chronic pain may arise through a sustained increase in proinflammatory and consequent inhibition of antiinflammatory cytokines, which, in turn, might sensitize nociceptors and lower the threshold required to activate them, hence perpetuating the pain experience (15). Not in vain, one commonality to most chronic pain conditions is the presence of elevated levels of proinflammatory cytokines (C-reactive protein, interleukin-6, interleukin-8, and tumor-necrosis-factor-α, among others) (14,15).

The distinct association of increasing DII with highest incident pain among more and less physically active participants is in line with the studies showing that sedentary behavior leads to higher inflammatory and lower antiinflammatory cytokine concentrations in both local and systemic circulation, contributing to the perpetuation of chronic pain (1). On the contrary, physical activity has well-established antiinflammatory effects, and exercise has preliminarily shown to reduce systemic inflammation, which in turn might reduce chronic pain (1). Specifically, regular exercise might normalize neuroimmune signaling in the central nervous system, which can prevent and even revert hyperalgesia (1). When taking these findings together, it is then possible that the association of physical activity with inflammation may compensate that of diet, and hence is the reason why the deleterious association of increasing the DII with incident pain was only apparent among the participants who were less physically active―note that, in our study, recreational physical activity at wave 1 was indeed associated with a lower risk for subsequent highest incident pain, irrespective of the previous DII change.

Regarding sensitivity analyses, the weaker association of the EDII without the beer and wine items with intermediate pain incidence is in line with the mechanistic evidence linking low-to-moderate alcohol intake with reduced inflammation (10,11). Specifically, such a drinking pattern has been associated with favorable biological markers of inflammation and hemostasis (increased HDL cholesterol and adiponectin, and decreased C-reactive protein, certain interleukins, fibrinogen, C-terminal proendothelin-1, and pentraxin-3 levels) (36,37). We also observed weaker associations with intermediate pain incidence for the alternate EDII versions with (a) proinflammatory scoring for snacks, fruit juice, and pizza, and (b) antiinflammatory scoring for fish, tomatoes, and other vegetables. Any explanation for these findings must be conjectural, though possibly reflects food preparation methods (11). On the one hand, well-done or browned fried, grilled, or barbecued fish may be proinflammatory―due to the oxidation of long-chain polyunsaturated fatty acids and the generation of heterocyclic amines and benzopyrene (38,39). On the other hand, while the effects of net tomato consumption on concentrations of inflammatory markers are conflictive (40–42), tomato paste contains 2.5- to 4-fold higher bioavailable lycopene than fresh tomatoes (43), which could explain the inverse association of pizza with inflammatory biomarkers, given the antiinflammatory properties of lycopene (44). These sensitivity analyses suggest that a maximally antiinflammatory dietary pattern might not necessarily be optimal from the overall health perspective, so these inflammatory indices―especially the EDII―may benefit from slight modifications if they are to be used for primary pain prevention, for (a) the effects of alcohol intake on chronic disease burden, disability, and death are greatly controversial (18,45); and (b) snacks, fruit juice, and pizza consumption may be associated with increased risk for obesity, diabetes, and cardiovascular disease, contrary to most fish and vegetables (including tomato) (27,46–48).

Limitations

Restricting the analyses to the participants who were free of pain at wave 1 may have mitigated reverse causation and allowed the study of diet as a potential pain prevention strategy (7,17,18), yet 2 limitations in this regard should be acknowledged. First, because of the high prevalence of pain at wave 1 (43.1%), we identified relatively few incident pain cases, and our analytical sample size was, thus, rather small. When altering the study design to include prevalent pain cases, an increase in the DII (among the least physically active participants) would still be associated with a detrimental change in the pain scale from wave 1 to wave 2, though not with prevalent pain at wave 1. Even though the study hypotheses were prespecified based on the effects of the inflammatory potential of diet on pain, the combination of reduced precision and several subgroup and interaction tests may have led to false positives. Specifically, when using a false discovery rate of 5%, the association between the EDII and intermediate pain incidence did not remain significant.

Second, some prevalent and incident pain cases―particularly in their milder forms―were likely missed, because most episodes of pain are short-lasting with little or no consequences (7). Nevertheless, on the one hand, the study associations were consistent when adjusting for musculoskeletal disease―note that while these participants reported no pain, they possibly suffered from some―and, on the other hand, missed cases of chronic and severe pain were unlikely, as we observed a predominance of persistent versus sporadic pain incident cases (159 vs 52) and of moderate-to-severe versus light pain (127 vs 84 cases).

A further limitation is the somewhat high loss to follow-up rate from wave 1 to wave 2 (roughly 22.9 % of the 1 339 pain-free participants). This led to a selection of participants with distinctive characteristics, which may have biased some of the study results―although not in a high degree, as shown in sensitivity analyses.

Two remarks on how outcomes, confounders, and exposures were measured are also warranted, as that may leave room for misclassification and residual confounding. First, our pain questionnaire and scale have not been validated, and they do not make distinctions between etiologies and types of pain (eg, neuropathic and nociceptive), whose biological mechanisms may differ. Nevertheless, the scale items were similar to those used in the Survey of Chronic Pain in Europe (2), and the study associations were relatively consistent when using other widely used instruments for pain assessment, such as the Numeric Rating Scale (49). Second, despite mostly using validated instruments for their collection (21,23,28,29), data on many potential confounders and diet were also self-reported. Our estimates remained anyway similar when adjusting the analyses for cognitive impairment, which was probably related to increased risk for recall bias.

Finally, our diet history did not collect dietary supplement use (23) and we lacked data on 13 DII food components (ginger, saffron, turmeric, pepper, thyme/oregano, rosemary, isoflavones, anthocyanidins, flavan-3-ol, flavones, flavonols, flavonones, flavan-3-ol, and eugenol) (10), both of which may have biased the associations of the DII with pain incidence in any direction.

Generalizability

First, even though the associations of the EDII and DII with pain incidence were somewhat consistent, there were still some differences, possibly because each inflammatory index likely estimates different dietary traits, as shown by the lack of correlation between the change in the EDII and that in the DII (r = −0.04). On the one hand, the EDII was only significantly associated with intermediate pain incidence, and the DII with highest pain incidence. This may be explained by the latter being based on nutrients/food compounds and the former on food groups, whose single foods’ and food components’ actions may be partially neutralized by each other, which could hence attenuate the study associations. Moreover, a posteriori dietary indices, such as the EDII, may underperform when transferred from the population in which they were derived to other populations (50). On the other hand, the pain incidence interaction between the inflammatory potential of diet and physical activity was apparent for the DII, but not the EDII. A possible explanation may be related to the components of both dietary inflammatory indices. While an increase in the DII may come together with increased energy, carbohydrate, fat, and protein intake, that in the EDII may be accompanied by higher processed meat, red meat, refined grains, and high-energy beverages consumption. Arguably, the first nutrient cluster may have more deleterious consequences over health―and possibly pain―in less versus more physically active participants, for they likely have concomitant lower energy requirements (51), while the second food group cluster has been associated with excess risk of mortality and several chronic conditions, regardless of physical activity levels (52).

Second, we were not able to examine whether the relationship between the inflammatory potential of diet and pain incidence was mediated via inflammatory biomarkers―as we lacked the corresponding data at wave 1. Both dietary inflammatory indices have nevertheless shown to predict inflammation (interleukin-6, tumor-necrosis-factor-α receptor 2, and C-reactive protein) (10,11), which may therefore be responsible for the observed higher pain incidence―note that our analyses were adjusted for the Alternate Healthy Eating Index, as diet quality have been associated with favorable pain changes (25,26). Indeed, though not specifically validated in the Spanish older adult population, the EDII has gone through validation in 3 independent populations of U.S. men and women up to 75 years (11), and the DII has been standardized to a representative range of dietary intakes based on 11 populations from all around the world (including European and Latin American) and further validated in racially diverse populations of men and women up to 79 years (10,53,54).

Third, our study population was 60 years and older. On the one hand, inflammation steadily increases with age, regardless of diet and even in apparently healthy individuals (55). On the other hand, pain etiology may be different in younger and older adults. Namely, while pain might be more prevalent in workers than in nonworking populations, that pain accompanied by activity limitation seems to increase with age (5). We nevertheless found no evidence that age modified the studied associations.

Finally, the Seniors-ENRICA-1 population was entirely white (99.2%), which warrants caution when extrapolating our results to multiethnic/multiracial populations.

Conclusion

An increase in the inflammatory potential of diet was associated with subsequent higher pain incidence among older adults in Spain, casting some light on its potential role as an adjunctive primary pain prevention strategy. The three main components of pain followed similar trends, the most consistent one being pain severity (moderate-to-severe pain). The association of increasing DII with highest incident pain was only apparent among the less physically active participants.

Larger studies with repeated pain measurements are warranted to address whether the link between the inflammatory potential of diet and incident pain is generalizable to younger and ethnically diverse populations. More research assessing the presumed mediation via inflammatory markers of the study associations is also needed. Finally, future studies in older adults should assess the efficacy of pain prevention interventions targeting the inflammatory potential of diet.

Funding

The present study was supported by Instituto de Salud Carlos III, State Secretary of R+D+I and FEDER/FSE (FIS grants 16/1512, 18/287, and 19/319), as well as the Funding REACT EU Program (Comunidad de Madrid and the European Regional Development Fund. ERDF. European Union) (FACINGLCOVID-CM project). A.C.-C. has an FPI fellowship from the Universidad Autónoma de Madrid. The funding agencies played no role in study design, data collection and analysis, interpretation of results, manuscript preparation, or in the decision to submit this manuscript for publication.

Conflict of Interest

None declared.

Author Contributions

F.R.-A. and R.O. conceived the study. A.C.-C. and R.O. performed the statistical analyses. A.C.-C., F.R.-A., and R.O. drafted the manuscript. All authors contributed to results interpretation. All authors reviewed the manuscript for important intellectual content, read, and approved the final manuscript. All authors have agreed both to be personally accountable for the author’s own contributions and to ensure that questions related to the accuracy or integrity of any part of the work, even ones in which the author was not personally involved, have been appropriately investigated, resolved, and the resolution documented in the literature.