This study was conducted as part of the Tasmanian Older Adult Cohort (TASOAC) study, an ongoing prospective, population-based study aimed at identifying the environmental, genetic, and biochemical factors associated with the development and progression of OA at multiple sites (hand, knee, hip, and spine). Subjects between the ages of 50 and 80 years were randomly selected from the electoral roll in Southern Tasmania (population, 229,000), with an equal number of men and women. The overall response rate was 57%. As TASOAC was designed to examine community-dwelling older adults, institutionalized older adults were excluded. Participants also were excluded if they reported contraindications for MRI. Of all initially eligible participants, 1,100 enrolled in the study, and 1,099 attended a baseline clinic between March 2002 and September 2004. Follow-up data were collected for 875 eligible participants at a subsequent clinic approximately 2 to 3 years later. The MRI machine was decommissioned halfway through the follow-up period; therefore, MRI scans were available for only approximately half of the follow-up participants (n = 425 of 875). The current study consists of a sample of 394 TASOAC participants who had MRI measures at baseline and follow-up and dietary and lipid measures at baseline.

All research conducted was in compliance with the Declaration of Helsinki and was approved by the Southern Tasmanian Health and Medical Human Research Ethics Committee. All subjects gave informed written consent.

Weight was measured to the nearest 0.1 kg (with shoes, socks, and bulky clothing removed) by using a single pair of electronic scales (Seca Delta Model 707). Height was measured to the nearest 0.1 cm (with shoes and socks removed) by using a stadiometer. Body mass index (BMI) was calculated as kilograms per square meter. Self-report of smoking status, statin use, and disease status, such as cardiovascular disease and diabetes, was recorded by questionnaire.

MRI of the right knee was acquired at baseline and follow-up with a 1.5-T whole-body magnetic resonance unit (Picker, Cleveland, OH, USA) by using a commercial transmit/receive extremity coil. Image sequences included the following: (a) a T₁-weighted fat-saturation three-dimensional (3D) gradient-recalled acquisition in the steady state; flip angle, 30 degrees; repetition time, 31 milliseconds; echo time, 6.71 ms; field of view, 16 cm; 60 partitions, 512 × 512-pixel matrix; acquisition time, 5 minutes 58 seconds; one acquisition; sagittal images were obtained at a slice thickness of 1.5 mm without a interslice gap; and (b) a T₂-weighted fat-saturation two-dimensional (2D) fast spin echo, flip angle, 90 degrees; repetition time, 3,067 milliseconds; echo time, 112 milliseconds; field of view, 16 cm, 15 partitions, 228 × 256-pixel matrix; sagittal images were obtained at a slice thickness of 4 mm with an interslice gap of 0.5 to 1.0 mm.

Subchondral BMLs were assessed on T₂-weighted MR images by using Osiris software and defined as areas of increased signal adjacent to the subcortical bone at the medial tibial, medial femoral, lateral tibial, and lateral femoral sites. One trained observer scored the BMLs by measuring the maximal area of the lesion at baseline and follow-up, as previously described 6. The observer manually selected the MRI slice with the greatest BML size. The BML with the highest score was used if more than one lesion was present at the same site. The maximal area was measured in square millimeters by using software cursors. Baseline and follow-up MRIs were read paired with the chronologic order known to the observer. Intraobserver repeatability of our areal BML measurement was assessed by randomly selecting 40 subjects with a BML to have their MRI scans re-read after at least a 2-week interval. The reader was blinded to the original BML measurements. Each BML present was remeasured at the medial tibial, medial femoral, lateral tibial, and lateral femoral sites. We compared the areal measurement 1 with the areal measurement 2 at each site where a BML was present. The intraclass correlation coefficient (ICC) was 0.97. At baseline and follow-up, participants were given a BML score (in square millimeters) at the medial tibial, medial femoral, lateral tibial, and lateral femoral sites. These were summed to create a total BML score. Figure 1(a and b) illustrates a change in BML size from baseline to follow-up.

Baseline dietary information was collected with the use of a self-administered, 74-item validated food-frequency questionnaire (FFQ) that was developed specifically for use in Australian adults 3031. Ten possible frequency responses were available for each food item, ranging from "never or less than once per month" to "3 or more times per day." This information was used to compute specific nutrient intakes, such as total energy (kJ/day), fatty acid (total fats, monounsaturated, polyunsaturated, and saturated (g/day)), carbohydrate (g/day), protein (g/day), and sugar (g/day).

Information regarding over-the-counter medication included information about vitamin and mineral supplementation and natural and herbal medications. We have chosen not to include these as part of nutrient intake in the current study because supplements have become more complex, with different brands containing highly variable ingredient combinations. As a result, evidence now suggests that brief and simple questionnaires (those that have been used in the current study) do not accurately reflect supplement intake, and more validated methods are necessary 32. Therefore, we limited our definition of nutrient intake to those data obtained from the FFQ.

Blood samples were collected at baseline after a 12-hour overnight fast, and assays were conducted to measure enzymatically the total cholesterol, high-density lipoprotein (HDL) cholesterol, and triglycerides by using an Olympus AU5400 automated analyzer. The concentration of LDL cholesterol was calculated by using the Friedewald formula 33. Assays were performed on thawed blood samples.

A standing anteroposterior semiflexed view of the right knee with 15 degrees of fixed knee flexion was performed at baseline and scored individually for osteophytes and joint-space narrowing (JSN) on a scale of 0 to 3 (0 = normal, and 3 = severe) according to the Altman atlas 34, as previously described 35. The presence of radiographic OA (ROA) was defined as any score ≥1 for JSN or osteophytes.

Cartilage defects were assessed on T₁-weighted MR images (score range, 0 to 4) at the medial tibial, medial femoral, lateral tibial, and lateral femoral sites, as previously described 36, as follows: grade 0 = normal cartilage; grade 1 = focal blistering and intracartilaginous low-signal-intensity area with an intact surface and base; grade 2 = irregularities on the surface or base and loss of thickness < 50%; grade 3 = deep ulceration with loss of thickness > 50%; and grade 4 = full-thickness chondral wear with exposure of subchondral bone. A cartilage defect also had to be present on at least two consecutive slices. The cartilage was considered to be normal if the band of intermediate signal intensity had a uniform thickness. The highest score was used if more than one defect was present on the same site.

Meniscal damage evaluation was performed by a trained observer, as previously described 37. In brief, the proportion of the menisci affected by a tear or a partial or full extrusion was separately scored (yes/no) on the medial and lateral edges of the tibiofemoral joint at the anterior, middle, and posterior horns.

PA was assessed as steps per day, as determined by pedometer (Omron HJ-003 and HJ-102; Omron Healthcare, Kyoto, Japan), as previously described 35. In brief, each participant was told to wear the pedometer for 7 consecutive days and to record the number of steps each day and the duration and type of PA for any activities in which the pedometer could not be worn (for example, swimming). An average of the 7 days was used to give each participant a mean steps-per-day value. Participants were then mailed a second pedometer after a 6-month period and repeated the process. A steps-per-day value, which was an average of the steps per day in summer and winter, was calculated.

The t tests and χ² tests were used to compare differences in means and proportions, as appropriate. Logistic regression was used to examine the associations between baseline BMLs (absent versus present) with baseline dietary factors and lipids after adjustment for age, sex, BMI, smoking, cardiovascular diseases, diabetes, ROA, and statin use in the lipids model.

Mixed-effect models were used to examine the association between change in total BML area and baseline dietary factors and lipids after adjustment for age, sex, BMI, baseline BMLs, smoking, cardiovascular diseases, diabetes, ROA, time to follow-up, and statin use in the lipids model. Standard diagnostic checks of model adequacy and unusual observations were performed and revealed that change in BML area was not normally distributed. The data are clumped at zero because a large proportion of participants did not have a BML at baseline or follow-up. However, the distribution remains not normal even after box-cox transformation. As a result, we have analyzed change in BML area as it was, but performed two separate analyses examining (a) BML size change at all four sites (medial tibial, medial femoral, lateral tibial, and lateral femoral); and (b) total BML size change (all four sites combined). This was done to check the consistency of our results.

Logistic regression analysis was used to examine the association between BMLs that have completely resolved and dietary factors and lipids after adjustment for age, sex, BMI, ROA, and statin use in the lipids model.

To compare with a previous study 17, a separate analysis examined only incident BMLs with baseline dietary factors and lipids. As we used a continuous, quantitative measure of BML size, we defined incident BMLs based on the least-significant criterion (LSC) 38. We used this equation in a recent article 6 to define significant changes in BML size when assessing BMLs as a continuous measure. The LSC takes into account measurement error and the correlation between the BML measurements at baseline and follow-up. The formula was as follows:

L S C = 1 . 96 × σ 2 ( 1 - ρ )

where σ is the standard error of the mean, and ρ is the serial correlation. LSC was calculated to be 25 mm² (where σ = 11.67 and ρ = 0.38). Therefore, incident BMLs were defined as any new BML in those with no BMLs at baseline larger than 25 mm². Logistic regression analysis was performed to examine incident BMLs with dietary factors and lipids after adjustment for age, sex, BMI, ROA, and statin use in the lipids model.

Standard diagnostic checks of model adequacy and unusual observations were performed for all models. Hosmer-Lemeshow tests were performed to assess goodness-of-fit for the logistic regression models. The result for each logistic regression was > 0.05, indicating that the model fits were adequate. Each dietary factor was entered into a separate multivariable model to avoid collinearity. A P value less than 0.05 (two-tailed) was considered statistically significant. All statistical analyses were performed on Intercooled Stata 10.0 for Windows (StataCorp LP).

In total, 1,099 subjects (51% female) aged between 51 and 81 years (mean, 63 years) participated in the TASOAC study. The current study consists of a sample of 394 TASOAC participants who had MRI measures at baseline and follow-up (approximately 2.7 years) and dietary and lipid measures at baseline. The range of follow-up times was 2.0 to 4.7 years. The majority of participants (90%) were followed up between 2.2 and 3.2 years. No significant differences were found in demographics, baseline energy intake, fatty acid intake, carbohydrate intake, protein intake, sugar intake, total cholesterol, triglycerides, LDL cholesterol, HDL cholesterol, ROA, or statin use between the rest of the cohort (n = 705) and the subjects included in the current study (n = 394). The characteristics of the study population are presented in Table 1. Those participants with a BML present at baseline (n = 0.168) were more likely to be male (P = 0.02) and had a higher BMI (P = 0.01) and a lower total and HDL cholesterol (P = 0.01). Total BML area at baseline ranged from 0 to 727 mm². The change in total BML area ranged from -710 to 1,344 mm².

Statin use was not associated with baseline BMLs (OR, 1.22; P = 0.452); however, those taking statin medication had a significantly lower total cholesterol level (OR, 0.51; P < 0.001), higher triglyceride level (OR, 1.64; P < 0.001), and lower LDL cholesterol level (OR, 0.23; P < 0.001); therefore, we adjusted for statin use in the multivariable analysis examining lipids.

Table 2 documents the associations between baseline BMLs with dietary factors and lipids. Dietary factors and lipids were not significantly associated with baseline BMLs in a multivariable model.

The association between baseline dietary factors and lipids with change in total BML size is presented in Table 3. Change in BMLs was significantly positively associated with energy, carbohydrate, and sugar intake in a multivariable model, adjusting for age, sex, BMI, baseline BMLs, smoking, cardiovascular diseases, diabetes, ROA, and time to follow-up. No significant associations were found between total fat intake or individual fats (monounsaturated, polyunsaturated, and saturated fat) and BML change. After adjustment for age, sex, BMI, baseline BMLs, smoking, cardiovascular diseases, diabetes, ROA, time to follow-up, and statin use, HDL cholesterol tended to be negatively associated with change in BMLs.

Similar results were seen when area change at all four sites (medial tibial, medial femoral, lateral tibial, and lateral femoral) was used instead of total area change. This demonstrates consistency in our findings, showing that energy, carbohydrate, and sugar intake were positively associated with a BML change; and HDL cholesterol tended to be negatively associated with BML change.

No differences were found in the relation between dietary factors and lipids with change in BMLs in those with and without ROA, as no interaction terms with ROA were significant.

Twenty-one subjects had complete BML resolution, and 147 subjects did not have complete BML resolution in those with a baseline BML. In a univariate analysis, HDL cholesterol was associated with resolving BMLs (OR, 1.67; P = 0.041), and after adjustment for age, sex, BMI, ROA, and statin use, this remained significant (OR, 2.00; P = 0.027). No dietary factors were associated with BML resolution.

The relation between baseline dietary factors and lipids and incident BMLs is presented in Table 4. Fourteen incident BMLs occurred across all four sites (medial tibial, medial femoral, lateral tibial, and lateral femoral). Total fat intake was protective against incident BMLs after adjustment for age, sex, BMI, and ROA. Saturated fat was also protective against incident BMLs in a multivariable model. In regard to serum lipids, HDL cholesterol was protective against incident BMLs after adjustment for age, sex, BMI, ROA, and statin use.

In a separate analysis that did not define incident BMLs by 25 mm² (that is, including all those who developed a BML of any size; n = 24), our results are largely consistent but appeared weaker, suggesting that measurement error increased when we did not define incident BMLs by 25 mm².

Additional data on cartilage defects and meniscal pathology were available in this study. The dietary factors found to be associated with BML changes (energy, carbohydrate, and sugar intake) were not associated with cartilage defects, meniscal extrusion, or meniscal tears cross-sectionally (all P > 0.05) or longitudinally (all P > 0.05) in univariate and multivariable analysis, adjusting for age, sex, BMI, baseline BMLs, smoking, cardiovascular diseases, diabetes, and ROA. HDL cholesterol was also not associated with cartilage defects, meniscal extrusion, or meniscal tears cross-sectionally (all P > 0.05) or longitudinally (all P > 0.05) in univariate and multivariable analysis adjusting for age, sex, BMI, baseline BMLs, smoking, cardiovascular diseases, diabetes, ROA, and statin use.

Information also was available about physical activity. When the BML models were further adjusted for baseline steps/day, the results were largely unchanged.