The cohort
The MCCS is a prospective cohort study of 41,528 people (17,049 men) aged between 27 and 75 years at baseline, 99.3% of whom were aged 40 - 69 years [15]. Participants were recruited via Electoral Rolls (registration to vote is compulsory for Australian adults), advertisements, and community announcements in the local media (e.g., television, radio, newspapers), between 1990 and 1994. Southern European migrants to Australia (including 5,425 from Italy and 4,535 from Greece) were deliberately over-sampled to extend the range of lifestyle exposures and to increase genetic variation. The study protocol was approved by The Cancer Council Victoria's Human Research Ethics Committee. All participants gave written consent to participate and for the investigators to obtain access to their medical records.
Follow-up was conducted by record linkage to Electoral Rolls, electronic phone books and the Victorian Cancer Registry and death records. To update lifestyle exposures, the cohort was followed up by mailed questionnaire and where necessary by telephone from 1995 to 1998 (first follow-up) and by face-to-face interviews from 2003 to 2007 (second follow-up). From 2003 onwards, 28,046 study participants (68% of the original MCCS participants) took part in the second follow-up.
Dietary assessment
At baseline, dietary intake over the previous 12 months was estimated using a 121-item food frequency questionnaire (FFQ) specifically developed for the MCCS. The FFQ was developed from a study of weighed food records in a sample of 810 Melbourne residents of similar age and ethnic origin to the MCCS cohort [16]. The FFQ included 22 items on intake of fresh red meat, processed meat, chicken, and fish. Fresh red meat was defined as veal or beef schnitzel, roast beef or veal, beef steak, rissoles (meat balls) or meatloaf, mixed dishes with beef, roast lamb or lamb chops, mixed dishes with lamb, roast pork or pork chops, and rabbit or other game (rarely consumed). Processed meat included salami or continental sausages, sausages or frankfurters, bacon, ham including prosciutto, corned beef, and manufactured luncheon meats. Chicken included roast or fried chicken, boiled or steamed chicken, and mixed dishes with chicken. Fish included steamed, grilled, or baked fish, fried fish, smoked fish, and canned fish including tuna, salmon, and sardines. At the MCCS first follow up, basic questions (not complete FFQ as used at MCCS baseline) about the frequency of meat consumption (fresh red meat, chicken, and fish) over the last year were asked.
Nutrient intakes were calculated using standard sex-specific portion sizes from the weighed food records [16]. The energy and nutrient contents in food were computed using Australian food composition tables [17]. Energy from alcoholic beverages was added to that calculated from the FFQ. Fatty acid composition of foods was obtained from the Royal Melbourne Institute of Technology fatty acid database [18]. Carotenoid data were obtained from the 1998 United States Department of Agriculture database [19].
To estimate the reproducibility of the FFQ, between July 1992 and June 1993, 275 participants were invited to participate in a study that required completing a second FFQ 12 months after baseline. Of these, 242 (88%) completed the second FFQ. The weighted kappa statistics for the reproducibility of the quartiles of meat intake were 0.42 (95% CI, 0.30-0.55) for fresh red meat, 0.60 (0.48-0.73) for processed meat, 0.44 (0.32-0.56) for chicken, and 0.48 (0.35-0.61) for fish.
Assessment of demographic, lifestyle and anthropometric factors
At baseline, a structured interview schedule was used to obtain demographic and lifestyle information including date of birth, country of birth, smoking, alcohol consumption, current physical activity during leisure time, and education. Height and weight were measured once at baseline attendance for each participant according to written protocols based on standard procedures [20]. Weight was measured to the nearest 0.1 kg using digital electronic scales, height was measured to the nearest 1 mm using a stadiometer. Body mass index (BMI) was calculated as weight in kilograms divided by the square of height in meters.
At the MCCS second follow-up, the participants were asked questions enquiring about their first joint replacement surgery: Have you ever had a hip replacement? When did you have your first hip replacement? Have you ever had a knee replacement? When did you have your first knee replacement?
Study participants
Of the 41,528 participants recruited, 6197 (14.9%) were excluded from analysis because they: reported extreme energy intake (<1st percentile or >99th percentile); reported an acute myocardial infarct, angina or diabetes at baseline and were likely to have changed their diet; had missing dietary data; died or left Australia prior to January 1, 2001; at the MCCS second follow-up had reported a primary joint replacement prior to January 1, 2001; had left Australia before the date of having a primary joint replacement; or had the first recorded procedure being a revision joint replacement as recorded in the Australian Orthopaedic Association National Joint Replacement Registry (AOA NJRR), thus leaving 35,331 participants available for analysis.
Identification of incident primary knee and hip joint replacement
Cases were identified from the AOA NJRR. The implementation of the AOA NJRR commenced in 1999 and was introduced in a staged state by state approach which was completed nationally by mid 2002. Victorian data collection commenced in 2001. The Registry monitors the performance and outcome of both hip and knee replacement surgery in Australia. It has detailed information on the prostheses and surgical technique used and the clinical situation that it was used in for both primary and revision joint replacement [21]. By using detailed matching technology it is able to determine the success or otherwise of the joint replacement surgery. Although data collection for the Registry is voluntary, it receives cooperation from all hospitals undertaking joint replacement surgery [21].
The AOA NJRR validates its data by using both internal systems and external data sources. The most important external data source is state health department data. Validation of registry data against health department recorded data involves a sequential multilevel matching process. Following the validation process and the retrieval of unreported records, the Registry collects the most complete set of data relating to hip and knee replacement in Australia [21].
Identifying information of MCCS participants, including first name, last name, date of birth, and gender, was provided to the staff at the AOA NJRR in order to identify those MCCS participants who had had a primary or revision joint replacement between January 1, 2001 which is when the Registry commenced Victorian data collection, and December 31, 2005. The matching was performed on these data provided using U.S. Bureau of the Census Record Linkage Software. Exact matches were identified and probabilistic matches were reviewed. The staff from the AOA NJRR forwarded this information to MCCS and it was then added to the MCCS database.
The study was approved by The Cancer Council Victoria's Human Research Ethics Committee and the Standing Committee on Ethics in Research Involving Humans of Monash University.
Statistical analysis
Cox proportional hazards regression models were used to estimate the hazard ratios (HR) for first recorded primary joint replacement associated with individual meat consumption after adjustment for confounding variables. Follow-up for primary joint replacement (i.e. calculation of person-time) began at January 1, 2001, and ended at date of first primary joint replacement for OA or date of censoring. Participants were censored at either the date of first primary joint replacement performed for indications other than OA, the date of death, the date left Australia, or end of follow-up (i.e. December 31, 2005 (the date that ascertainment of joint replacement by NJRR was complete), whichever came first.
All meat consumption variables were analysed categorically, based on approximate quartiles of the distribution of weekly frequency of consumption, with the first quartile used as the reference category. Linear associations between meat consumption and the risk of joint replacement were investigated by comparing the regression models with meat consumption as a categorical variable and a pseudo-continuous variable using the likelihood ratio test. Tests for trend across categories of meat consumption were calculated using meat consumption as a pseudo-continuous variable, assuming that, within each quarter all participants consumed at its median frequency. We also calculated the ratio of frequency of consumption of fresh red meat to the combined frequency of consumption of chicken and fish and divided it into groups based on quartiles. To estimate HR separately for knee and hip replacement risk and to test for heterogeneity, Cox models based on competing risks were fitted using a data duplication method [22].
Age, gender, BMI, country of birth, and energy intake (kj/d) were included in all models. Other potential confounding variables were included in all the definitive analyses if they changed the HR of any of the meat consumption variables for either hip or knee joint replacement risk by at least 5%. First, education, current level of physical activity, smoking (current/past/never), and alcohol consumption (g/d) were added. The HR changed < 5%, thus, none of these variables were retained for further analysis. Second, multi-vitamins and fish oil supplement were added. HR changed < 5%. Third, fruit, vegetable, vitamin C, vitamin E, beta-carotene, and polyunsaturated fatty acids were added to the model, one at a time. HR changed < 5%. Thus none of the dietary variables were retained for further analysis.
To test whether associations between meat consumption and the risk of joint replacement were modified by country of birth, gender or educational level, interactions between country of birth, gender or educational level and meat consumption were fitted, and tested using the likelihood ratio test. Tests based on Schoenfeld residuals and graphical methods using Kaplan-Meier curves showed no evidence that proportional hazard assumptions were violated for any analysis. All statistical analyses were performed using Stata (Intercooled Stata 9.2 for Windows, StataCorp LP., College Station, TX, USA).