Dietary Patterns and Risk of Developing Knee Osteoarthritis: Data from the Osteoarthritis Initiative

Abstract

Objective:

Some studies have examined the association between dietary factors and risk of knee osteoarthritis (OA). We aimed to examine the prospective association of major dietary patterns with the risk of developing knee OA.

Method:

We followed 2,842 participants in Osteoarthritis Initiative (OAI) aged 45 to 79 years and with at least one knee free from radiographic knee OA at baseline for up to 72 months. We defined knee OA incidence as Kellgren and Lawrence grade ≥ 2 during follow-up visits. Using principal component analysis, Western and prudent dietary patterns were derived. Cox proportional hazards models were used to assess the association between dietary patterns and incident knee OA.

Results:

Among study participants, 385 (418 knees) developed knee OA within 72 months. Following a Western dietary pattern was associated with an increased risk of knee OA (HR _{quartile 4 vs 1} = 1.69, 95% CI: 1.13 to 2.52, p trend: 0.03), while adherence to the prudent pattern was associated with a reduced risk of knee OA (HR _{quartile 4 vs 1} = 0.70, 95% CI: 0.50 to 0.98, p trend: 0.05). The observed associations attenuated after additionally adjusting for body mass index (BMI). The observed associations were mediated through BMI by approximately 30%.

Conclusion:

Following a Western diet was associated with increased risk of knee OA, whereas following a prudent pattern was associated with a reduced risk of knee OA. The associations were partially mediated through BMI.

Introduction

More than 32.5 million adults in the United States suffer from osteoarthritis (OA) – a slowly progressive heterogeneous joint disease characterized by joint pain, stiffness, deformity, and functional disability, which is a leading cause of huge earning losses and medical costs^1,2. Due to the lack of profound insights into knee OA’s underlying pathological mechanism, there is no drug therapy that can effectively delay disease progression^3,4. Current therapies mainly aim to relieve pain and alter the joint function or use surgical interventions, such as total knee replacement⁵. Therefore, similar to other chronic metabolic diseases, OA requires early disease intervention or prevention⁶.

Although the etiology of knee OA has not been entirely elucidated, several risk factors (i.e., older age, obesity, joint injury, and certain sports) have been found to be associated with an increased risk for incident OA⁷. Lifestyles, such as dietary factors, may also contribute to development of knee OA⁸. Accumulating evidence implies that the intake of specific foods or nutrients, such as soft drinks⁹, saturated fat¹⁰, dietary fiber¹¹, vitamins C, K, and D^12,13, may be associated with knee OA incidence or progression. However, individual food and nutrient analyses may be inadequate to consider complicated interactions among nutrients, and high level of inter-correlation makes it difficult to examine their separate effects. Dietary pattern analysis, which integrates individual foods into a scale, has become the emerging trend in nutritional epidemiological research as it allows for this complexity and potentially provides new insights into diet-disease associations¹⁴. Adherence to a Mediterranean diet measured by a Mediterranean dietary score based on consumption of some specific food groups has been associated with a lower risk of pain worsening and symptomatic forms of knee OA¹⁵. Principal component analysis (PCA) is a commonly used approach to incorporate all food groups for dietary pattern analysis. However, no published studies have examined the prospective association between PCA-derived dietary patterns and risk of knee OA. Our previous study showed that Western dietary pattern derived by PCA was associated with increased radiographic and symptomatic knee OA progression, while the prudent dietary pattern was associated inversely¹⁴. It is reasonable to deduce that dietary patterns may also be associated with the risk of developing knee OA. Therefore, we examined the prospective association between major dietary patterns derived by PCA and the risk of developing radiographic knee OA.

Method

The Osteoarthritis Initiative (OAI)

The OAI is a multi-center, prospective longitudinal cohort, which established a public data archive with identified and characterized information for OA. At each of its four clinical sites (Baltimore, MD; Columbus, OH; Pittsburgh, PA, and Pawtucket, RI), trained staff collected and recorded participants’ clinical features, biological specimens, as well as radiographs¹⁶. Since 2004, the cohort has included 4,796 participants aged 45 to 79 years in the US and followed them annually (60-month follow-up was by phone). During the first 48 months, the follow-up rate was > 90%. The study was fully compliant with the NIH guidelines. The cites had IRB approval, and all participants provided informed consent. Detailed OAI protocol is available elsewhere¹⁷.

Study sample

To assess knee OA incidence, we included all participants with at least one knee without radiographic OA at baseline (Kellgren and Lawrence [KL] grade of 0 or 1). We excluded participants without follow-up visits and who had implausible dietary data (e.g. daily total energy intake <800 or >4200 kcal for men, <500 or >3500 kcal for women). The excluded sample had similar baseline characteristics with the original sample. Finally, 2,842 participants (4,573 knees) were followed up at 12, 24, 36, 48, and 72 months after the enrollment screening. More than 83% participants have at least 4 follow-up visits. No significant difference in rate of lost to follow-up was observed among quartile groups of each dietary pattern.

Assessment of dietary consumption

OAI mailed a Block Brief Food Frequency Questionnaire (FFQ) to every participant at baseline. Completed FFQ included 70 food items and the participants’ consumption frequency during the prior year for each of them based on standard units or portion sizes (categorized as never, a few times per year, once per month, 2–3 times per month, once per week, 3–4 times per week, 5–6 times per week, and every day)²¹. Then total calorie intake and nutrients intake were calculated according to USDA food composition data (evaluated by the NutritionQuest www.nutritionquest.com).

Assessment of knee OA onset

Participants had bilateral weight-bearing, fixed-flexion posteroanterior knee radiographs at baseline and follow-up visits. Central readers, who were blinded to the order of follow-up radiographs, scored the images for KL grades. The read-reread agreement for these readings was good (weighted κ [intrarater reliability] = 0.70–0.80). These KL grades are publicly accessible (files: kXR_SQ_BU##_SAS [versions 0.6, 1.6, 3.5, 5.5, and 6.3])¹⁷. We described the onset of knee OA with KL grade based radiographic changes. Adjudicated KL grades were noted at baseline and each follow-up visit. We defined any KL grade ≥ 2 during follow-up visits as the marker of newly developed OA for that specific knee which was free of knee OA at baseline.

Assessment on covariates

We extracted demographic and socioeconomic factors, such as age, gender, race/ethnicity, education level, and annual income, from the baseline visit. We characterized the race as African American, White, or “other” racial/ethnic groups, the educational level as high school or less, college, and above college. We also considered baseline KL grade, smoking status, physical activity, any traumatic knee injury and knee surgery, depression (defined as the CES-D 20 items scale >16¹⁸), non-steroidal anti-inflammatory drugs (NSAIDs) use, and daily total energy intake as potential confounding factors. We used the Physical Activity Scale for the Elderly (PASE), a validated questionnaire, to measure physical activity¹⁹. PASE score, body mass index (BMI) and daily energy intake (kcal) were analyzed as continuous variables.

Statistical Analyses

To derive dietary patterns, we gathered the 70 FFQ items into 25 food groups according to their nutrient profile and culinary methods. PCA is a multivariate statistical technique to identify common underlying dimensions of food consumption. It aggregates specific food groups based on correlations with one another. Briefly, each obtained dietary pattern is a linear combination of all food groups weighted by their factor loadings and explains as much inter-individual variation of the food groups as possible. Each subject receives a score for each dietary pattern (component), with a higher score indicating a higher adherence to the respective dietary pattern. Scree test was used to determine the dietary patterns to retain. The derived patterns were statistical independent with each other. Each individual has a separate score for each dietary pattern. Consistent with other studies^14,15, we identified two major dietary patterns (eigenvalues were 3.30 and 2.24 respectively, and 22% of variance was explained), Western pattern was delineated by high intakes of French fries, processed / red meats, refined grains, desserts and sweets, high-fat dairy products, and sugar-containing beverages, while the prudent pattern was portrayed by high intakes of vegetables, fruit, fish, whole grain, and legumes (Table 1). We then divided into quartile groups for each dietary pattern score.

Table 1.

Factor-loading (correlation) matrix for two major dietary patterns derived by principal component analysis¹

Food or food group	Western	Prudent
French fries	0.59
Processed meats	0.59
Refined grains	0.55
Red meats	0.53
Poultry	0.44
Pizza	0.43
Snacks	0.39
Margarine	0.39
Eggs	0.38
Desserts and sweets	0.37
Sugar-containing beverages	0.36
High-fat dairy products	0.34
Salad dressings	0.33
Butter	0.30
Vegetables		0.77
Legumes		0.66
Fruit		0.53
Fish		0.48
Whole grains		0.48
Potatoes	0.33	0.41
Tomatoes	0.30	0.39

¹Absolute correlation of < 0.3 were excluded from the table for simplicity.

We applied Cox proportional hazards models to evaluate the associations between dietary patterns and OA onset with quartiles of dietary pattern scores as exposures. For each knee, we defined the follow-up period as the length from baseline to knee OA onset (KL grade first increased to 2 or above), death, loss to follow up, or 72 months, whichever came first. We employed the discrete likelihood method for tied event times and used robust covariance estimates to deal with the intra-cluster correlation between two knees from the same participant. We adjusted the models with demographic and socioeconomic factors, baseline KL grades, depression, the indication of any traumatic knee injury and knee surgery, PASE score, NSAIDs use, and total daily energy intakes. We omitted other pain relief medications due to the lack of evidence of their relationship with the onset of knee OA. Less than 1% of the participants had missing BMI or PASE score; we refilled the missing values with a sex-specific median. Hazard ratio (HR) and 95% confidence interval (CI) were calculated to describe the strength of associations. We also used the median value for each quartile of dietary pattern score as a continuous variable to test linear trend. Additionally, we included pattern scores as continuous exposures in the models and calculated the HR for one standard deviation increase of pattern scores. The proportional hazard assumption was evaluated by including an interaction term between dietary patterns and logarithm of follow-up time and was met in all analyses.

As mentioned in previous studies, BMI may be a potential mediator for dietary factors²⁰. In separate models, we additionally adjusted for baseline BMI. To better address this issue, we performed mediation analyses for Cox model using a well-established approach of Lin et al^21,22. Separated models were developed: incorporating BMI or not, and adjusting for the same covariates used in the previous models. The mediation (indirect) effect was estimated as the proportion of change in exposure effect due to BMI.

We conducted several sensitivity analyses to test the robustness of the primary findings. First, we examined the relationship between dietary patterns and OA risk incorporating TKR as endpoints (6 participants). Second, we developed a parsimonious model (excluding baseline KL grade, NSAIDs use, PASE score). Then, to better address the confounding from knee OA in the contralateral knee, we also conducted two sensitivity analyses: 1. Keeping only participants without baseline OA for both knees at baseline; 2. Adjusting for OA status of the other knee in knee level analysis. Data analyses were conducted with SAS 9.4 (SAS Institute, NC).

Results

We included a total of 2,842 participants (4,573 knees) from OAI in the present study (Figure 1). The mean age was 60.5 ± 9.2 years, 42.7% were male, 84.5% were white. Among all the participants, 385 (418 knees) developed knee OA within 72-month follow-up period (mean follow-up time of 55.4 ± 21.2 months). Table 1 shows a detailed description of the Western and prudent dietary patterns. The strength of correlation between each food group and diet patterns can be interpreted as the loading values. Positive loading values are representative of a food group’s positive correlation with a dietary pattern, and a negative value represents a negative correlation, with the magnitude of values indicating the degree to which each food contributes to the dietary pattern. For simplicity, absolute values of <0.30 were omitted from the table. French fries, processed / red meats, refined grains, poultry, pizza, and snack intake account heavily for the Western dietary pattern, while vegetables, beans, fruits, fish, whole grains, and legumes mainly constitute the prudent dietary pattern.

Figure 1.

Study population and exclusions in the Osteoarthritis Initiative. After baseline, participants were lost to follow-up due to death, total knee replacement, or nonresponse: 120 in year 2, 109 in year 3, 146 in year 4, and 469 in year 6. KL, Kellgren–Lawrence; OA, osteoarthritis; OAI, Osteoarthritis Initiative.

Table 2 shows the baseline characteristics by the increasing quartiles of the Western and prudent dietary pattern scores. Individuals most adhered to the Western diet (i.e., included in Q4) tend to be younger, more frequently depressed, and less educated than those in the lowest quartile (Q1). They are also more likely to be African American, NSAIDs users, and current smokers. They tended to have more energy intake per day and be overweight or obese. Meanwhile, participants who were more following prudent dietary pattern (i.e., included in Q4) had higher education levels and fewer suffering from depression. They were less likely to be current smokers, and more likely to exercise more, and consume more total calories per day.

Table 2.

Baseline characteristics of participants according to quartiles of Western and prudent dietary pattern scores¹

		Quartiles of Western dietary pattern					Quartiles of Prudent dietary pattern
	Total N = 2842	Q1 N = 710	Q2 N = 711	Q3 N = 711	Q4 N = 710	p-value	Q1 N = 710	Q2 N = 711	Q3 N = 711	Q4 N = 710	p-value
Variable
Age, y	60.5 (9.2)	63.2 (9.0)	61.0 (9.2)	59.6 (9.2)	58.3 (8.6)	< 0.01	58.9 (8.8)	60.9 (9.3)	60.7 (9.3)	61.4 (9.2)	< 0.01
Male	42.7	42.7	42.6	42.6	42.7	1.00	42.7	42.6	42.6	42.7	1.00
Race
White	84.5	85.1	88.6	85.5	78.9	< 0.01	80.0	85.9	88.2	83.9	< 0.01
African American	12.7	11.5	8.3	12.8	18.3		17.6	11.0	9.3	13.1
Other	2.7	3.4	3.1	1.7	2.8		2.4	3.1	2.5	3.0
Education
≤ High School	13.4	10.4	9.0	15.6	18.7	< 0.01	18.2	13.4	11.5	10.7	< 0.01
College	45.0	41.1	45.1	45.3	48.6		43.9	45.9	45.1	45.2
> College	41.5	48.5	45.9	39.1	32.5		37.7	40.8	43.3	44.1
Missing	0.04	0.0	0.0	0.0	0.1		0.1	0.0	0.0	0.0
Family income
≤ 25 k	11.1	10.7	9.7	10.3	13.8	0.44	12.3	11.4	9.8	11.0	0.53
25–50 k	22.7	22.5	21.2	24.5	22.4		22.8	24.3	22.8	20.7
50–100 k	35.8	36.5	37.0	34.5	35.1		33.9	34.6	36.7	37.7
> 100 k	24.9	23.9	26.7	25.6	23.4		24.9	24.3	26.4	23.9
Missing	5.6	6.3	5.3	5.2	5.4		6.1	5.3	4.2	6.6
Depressed	7.9	6.2	5.8	8.4	11.0	< 0.01	10.7	8.4	5.3	6.9	< 0.01
Smoking status
Never	52.9	57.2	55.4	49.9	49.2	< 0.01	53.0	53.6	51.5	53.7	< 0.01
Current	6.4	2.3	3.8	8.3	11.1		10.0	6.6	4.9	3.9
Past	40.7	40.6	40.8	41.8	39.7		37.0	39.8	43.6	42.4
PASE	164.2 (81.3)	158.7 (78.5)	164.8 (80.8)	164.5 (80.4)	168.9 (85.4)	0.13	158.1 (82.8)	162.5 (76.4)	163.3 (80.2)	173.1 (85.2)	< 0.01
BMI, kg/m2	27.9 (4.5)	26.6 (4.2)	27.3 (4.2)	28.3 (4.5)	29.2 (4.8)	< 0.01	28.3 (4.6)	27.7 (4.3)	27.7 (4.6)	27.6 (4.5)	0.02
KL grade = 12	31.92	30.50	32.87	31.87	32.43	0.64	30.12	32.47	31.74	33.39	0.39
NSAIDs use3	19.0	15.9	19.3	17.9	22.7	< 0.01	19.5	17.3	18.0	21.0	0.29
Total calories, 1000 kcal/day	1.4 (0.5)	1.1 (0.4)	1.2 (0.4)	1.4 (0.4)	1.9 (0.6)	< 0.01	1.2 (0.5)	1.3 (0.4)	1.4 (0.5)	1.8 (0.6)	< 0.01

¹Values are mean (SD) for continuous variables, percentages for categorical variables. Tests for significant difference across quartiles of dietary pattern intake included the Chi-square test, Mantel–Haenszel test, and ANOVA. NSAIDs, nonsteroidal anti-inflammatory drug; PASE, Physical Activity Scale for the Elderly; Q, quartile; KL, Kellgren–Lawrence,

²KL was in knee level.

³NSAIDs (including aspirin, ibuprofen, etc.) use for joint pain or arthritis in past 30 days.

As shown in Table 3, individuals in the highest Western quartile reported significantly increased risk of knee OA (HR _{Q4 vs. Q1} =1.69, 95% CI: 1.13–2.52, P trend = 0.03). Meanwhile, being in the highest prudent quartile was associated with reduced risk of knee OA (HR _{Q4 vs. Q1}= 0.70, 95% CI: 0.50–0.98, P trend = 0.05). When we used continuous pattern scores, HR for one standard deviation increase was associated with 22% risk increase for Western pattern score, and 6% risk reduction for the prudent pattern score. After adjusting for baseline BMI, the observed associations attenuated. The significant association remained between Western pattern and risk of knee OA (HR _{Q4 vs. Q1}= 1.46, 95% CI: 0.96–2.21, p-trend=0.18). However, a non-significant association was observed between the prudent pattern and incident knee OA (HR _{Q4 vs. Q1}= 0.74, 95% CI: 0.53–1.04, p-trend=0.10). No effect modifications were observed between age, sex and each dietary patterns. In mediation analyses, the proportion of exposure effect explained by BMI was 31.4% (CI: 9.0%-67.8%, p < 0.01) for the Western pattern, 27.3% (CI: 2.1%-86.7%, p = 0.02) for the prudent pattern. It indicated that BMI might explain about 30% of the association between dietary patterns and knee OA risk.

Table 3.

Hazard ratios (HR) and 95% confidence intervals (CI) for incident radiographic knee osteoarthritis according to quartiles of dietary pattern scores

			Model 11		Model 22
	Cases, n	Person-years	HR (95% CI)	p-trend4	HR (95% CI)	p-trend4
Quartiles of Western pattern3
Q1	88	5471	1.00	0.03	1	0.18
Q2	114	5565	1.29 (0.95, 1.74)		1.25 (0.92, 1.69)
Q3	99	5449	1.22 (0.87, 1.72)		1.10 (0.78, 1.56)
Q4	117	5346	1.69 (1.13, 2.52)		1.46 (0.96, 2.21)
Quartiles of Prudent pattern3
Q1	115	5472	1.00	0.05	1	0.10
Q2	102	5418	0.86 (0.63, 1.15)		0.88 (0.65, 1.18)
Q3	112	5569	0.93 (0.69, 1.25)		0.95 (0.71, 1.29)
Q4	89	5372	0.70 (0.50, 0.98)		0.74 (0.53, 1.04)

¹Adjusted for age, sex, race (African American, white, other), baseline Kellgren–Lawrence grades (0 or 1), injury/surgery (yes, no), Physical Activity Scale for the Elderly (PASE) score, nonsteroidal anti-inflammatory drug use (yes, no), depression (yes, no), and total energy intake (kcal/d, continuous). Additional adjustment for income, education, smoking, and alcohol did not significantly alter results.

²Additionally adjusted for BMI (continuous).

³Western pattern: Q1 is more healthy, Q4 is less healthy; Prudent pattern: Q1 is less healthy, Q4 is more healthy.

⁴p values for testing linear trend across quartiles.

In sensitivity analyses, we examined the relationship between dietary patterns and OA risk additionally including TKR as the endpoint. The result was highly consistent with the primary analysis (results are not shown). Meanwhile, after controlling for less confounding variables (exclude baseline KL grade, NSAIDs use, PASE score), the findings generally remained (Supplementary Table 1, model 1). When we only kept participants without baseline OA for both knees, the findings were consistent (Results were not shown). Finally, after adjusting for the OA status in other knee, the observed associations remained but attenuated (Supplementary Table 1, model 2).

Discussion

In this large longitudinal study with a 6-year follow-up period, our findings indicated that more adherence to a Western dietary pattern was associated with an increased risk of knee OA. In contrast, following a prudent dietary pattern was associated with a decreased risk of knee OA. The observed associations may be partially mediated through BMI.

The association between diet and knee OA has been discussed for a long time²³. Still, most previous studies focused on the associations between single nutrients or food intake and knee OA progression. A protective association was observed between higher-dose vitamin C intake and the progress of knee OA in several studies^12,24. Long-term vitamin D supplementation was inversely correlated with dysfunction and increased pain in patients with knee OA²⁵. Vitamin K deficiency was associated with an increased risk of developing radiographic knee OA and MRI-based cartilage lesions¹³. Data from OAI and Framingham Offspring Osteoarthritis Study consistently showed that higher total fiber intake was related to a lower risk of symptomatic OA, while the relation to radiographic OA was unclear¹¹. Moreover, emerging evidence suggested that overall dietary pattern or dietary quality analysis may comprehensively reveal the association between diet and disease risk^26,27. In the previous study, we demonstrated that following a Western dietary pattern was associated with increased radiographic and symptomatic knee OA progression, while adherence to a prudent dietary pattern was associated with a reduced progression¹⁴. A recent 4-year longitudinal follow-up cohort study revealed that participants with a higher adherence to the Mediterranean diet reported a lower risk of pain worsening and symptomatic knee OA¹⁵. Our findings add to the existing studies which have examined individual food components, nutrients, and the Mediterranean diet in relation to knee OA risk.

Numerous studies have demonstrated that an unhealthy dietary pattern is an important risk factor of overweight and obesity^28,29. BMI, as a general measure of body adiposity and a significant risk factor for OA incidence and progression^30,31, may be both a confounder and mediator in studies of diet–disease relationships^20,32. In our study, after adjusting for BMI, the associations of two dietary patterns with knee OA risk attenuated, indicating BMI may partially mediate the observed associations. Currently, OA is recognized to be an inflammatory disease affected by a variety of factors, including obesity³³, synovitis³⁴, and systemic inflammatory mediators³⁵. Previous studies have demonstrated the association between dietary patterns and inflammation markers^36–38. For instance, the Western diet has been reported to be related to the increased production of inflammatory cytokines such as interleukin (IL)-6 and tumor necrosis factor-α (TNFα)³⁹. Another example was that the Western diet promotes inflammation that arises from both structural and behavioral changes in the resident microbiome in metabolic disease⁴⁰. By contrast, recent studies have revealed that increased consumption of fruits, vegetables, and whole grains are associated with lower levels of inflammation^41,42. The observed association on knee OA risk might be attributed to inflammatory mediator variations affected by different dietary patterns. Future studies are needed to examine the potential mediation effect of inflammatory biomarkers.

This study’s strengths include the use of a large amount of carefully collected dietary and covariate data from a multi-center prospective study, with repeated knee radiographs over 72 months in individuals at risk of knee OA. In terms of limitations, first, the dietary data were only obtained at baseline; therefore, we could not update the change of their dietary habits during follow-up years. However, FFQ measures a long-term diet which is less likely to change significantly within six years based on previous studies⁴³. Second, due to the multidimensionality of diet, one major limitation with regard to PCA being used with dietary data is that the retained dietary patterns explain only a small amount of the variance in the diet.^44,45 The literature has not established the ideal percentage of variance explained by the components retained in studies on the identification of dietary patterns. However, our derived dietary patterns were highly consistent with those derived from many other populations^46–48. For studies of dietary patterns and disease we may be more prudent to rely on interpretability of the patterns, rather than variance explained. Moreover, the effect of a single diet or nutrient may be diluted in dietary pattern analysis. Inevitably, though we adjusted for several potential confounding factors associated with knee OA and dietary intake, residual confounding is possible. In addition, there is an argument of whether or not the derivation of dietary patterns is based on original (unadjusted) or energy-adjusted food group data. A recent study compared these two approaches and observed little differences in the dietary pattern solutions⁴⁹. Using energy-adjusted food groups, which are relative rather than actual food intake, makes it difficult to interpret results. Therefore, we adjusted the total energy intake at a later stage of the analytical process, which was commonly used in other studies^47,48.

In conclusion, the results from this large prospective study demonstrated that higher adherence to the Western dietary pattern was associated with an increased risk of future knee OA while following a prudent dietary pattern was associated with a decreased risk of knee OA onset. The observed associations may be partially mediated through BMI. Further studies should be replicated in other cohorts.