The OAI is a multi-center, longitudinal prospective cohort study. At baseline, 4796 US men and women (58.5%) age 47-79 years with, or at high risk of, knee OA were recruited from four sites, Columbus, Ohio; Providence, Rhode Island; Baltimore, Maryland and Pittsburgh, Pennsylvania. Annual radiographic assessments of knee OA were carried out until the 48-month OAI follow-up visit; therefore, in the present study we followed participants until 48 months.
Baseline assessment of inflammatory potential of diet
Habitual dietary intake of nutrients and foods was estimated at baseline using a validated dietary assessment tool, the Block Brief 2000 Food-Frequency Questionnaire (FFQ) (20). For each dietary component, the frequency of consumption was reported according to nine predetermined categories ranging from “never” to “everyday” with illustrations of standard portion sizes. Dietary intake of each component and energy was calculated based on nutrient composition values determined from the US Department of Agriculture nutrient database (21).
The DII, developed by Cavicchia et al (22) and updated by Shivappa et al (23), is a literature-derived population-based scoring algorithm to assess the inflammatory potential of diet as a whole based on 45 food parameters. A higher DII score indicates a greater inflammatory potential of diet (i.e., pro-inflammatory effect). Details regarding the development and calculation process are available elsewhere (23). The DII has been evaluated for validity in 27 studies and shown to predict inflammatory markers including C-reactive protein (CRP), interleukin (IL)-6, and tumor necrosis factor (TNF)-α (22, 24-27). Using data collected by the FFQ in the OAI, we calculated the energy-adjusted DII (E-DIITM) based on intake of 24 dietary components, defined as DII score per 4184 kJ (1000 kcal) of energy (28). Dietary components available in the OAI for calculating E-DII scores included vitamin B12, vitamin B6, β-carotene, caffeine, carbohydrate, cholesterol, fat, fiber, folic acid, iron, magnesium, monounsaturated fat acids, niacin, protein, polyunsaturated fatty acids, riboflavin, saturated fat acids, selenium, thiamin, vitamin A, vitamin C, vitamin E, vitamin D and zinc.
Assessment of incident knee OA outcomes
At baseline and at each annual follow-up visit, frequent knee pain was queried and defined as pain, aching or stiffness for more than half the days of a month during the past 12 months. Participants obtained weight-bearing fixed flexion posterior-anterior view radiographs of both knees. Central reading was carried out to assess Kellgren and Lawrence (KL) grade for each knee. Any disagreement as to whether the knee at any time point had radiographic OA was adjudicated by a panel of three experienced readers including the two primary readers and one other. We defined a knee as having incident radiographic knee OA (ROA) if it did not have ROA at baseline (i.e., K/L = 0 or 1) and developed ROA (i.e., K/L ≥ 2) over the follow-up time. Incident knee SxOA was defined as a new onset of a combination of frequent knee pain and ROA in the same knee during the follow-up period.
Assessment of other covariates and BMI as a potential mediator
At baseline, all participants were queried for age, sex, race, educational attainment, annual income and tobacco use. Physical activity was assessed using the Physical Activity Scale for the Elderly (PASE) capturing a broad spectrum of habitual physical activity and is summarized into a continuous score with higher scores indicating higher levels of physical activity. PASE was validated for the assessment of physical activity among older adults with knee pain and physical disability (29). At 12-month follow-up visit, body weight and height were measured in light clothing without shoes using calibrated devices. BMI was calculated using weight (kilogram) divided by the square of height (meter). We used BMI assessed at 12-month follow-up visit as the mediator in the current study.
We categorized E-DII scores into quartiles for men and women separately to account for sex differences in dietary intake of nutrients and foods. We examined the association of E-DII with the risk of knee ROA (or SxOA) by comparing higher E-DII quartiles (Q2, Q3 and Q4) with the lowest quartile (Q1: reference category) using generalized estimating equations (GEE) to account for the correlation between two knees for each participant. In the base model (model 1), we adjusted for age, sex (men vs women), race (White vs non-White) and total energy intake (kcal/day). In model 2, we further adjusted for educational attainment (below college vs college or above), annual income (<50,000 US$ vs ≥50,000 US$), tobacco use (non-smoker vs former and current smoker) and PASE score. We tested for linear trend using the median value of each quartile of E-DII score as a continuous variable in the regression model. We also conducted alternative analyses fitting E-DII as a continuous variable in model 1 and model 2. We conducted sensitivity analyses by counting total knee replacement due to knee OA during the follow-up period as incident knee ROA or SxOA, respectively.
We performed mediation analyses to assess to what extent the association of E-DII score with incident ROA (or SxOA) was mediated through BMI assessed at the 12-month follow-up visit. In these analyses, we grouped E-DII score into two categories using sex-specific median value as a cut-point. We decomposed the total effect of E-DII score on the risk of incident knee ROA (or SxOA) into two components (30, 31) (Figure 1), i.e., (1) the indirect effect (or mediated effect) representing the effect of E-DII on the risk of incident knee ROA (or SxOA) mediated via BMI, and (2) the direct effect representing the effect of E-DII score on the risk of incident knee ROA (or SxOA) that was not through BMI. We calculated the proportion of mediation to quantify the proportion of the effect of E-DII mediated by BMI (30, 31).
All statistical analyses were performed using Stata/SE 15.1 (StataCorp, Texas, USA). A P value <0.05 (two sided) was considered statistically significant.