Participants
Osteoarthritis initiative (OAI) is a publicly available multicentre observational cohort study of knee OA (https://oai.nih.gov). The OAI comprises data of 4796 participants aged 45–79 years at baseline. OAI exclusion criteria included inflammatory arthritis, severe joint space narrowing in both knees, unilateral knee joint replacement and severe joint space narrowing in the contralateral knee, inability to undergo magnetic resonance imaging (MRI) or to provide a blood sample, required use of walking aids excepting a single straight cane ≤ 50% of the time, or unwilling to provide informed consent. Participants were recruited at four clinical sites (University of Maryland School of Medicine and Johns Hopkins University, Baltimore, Maryland; Ohio State University, Columbus, Ohio; University of Pittsburgh, Pittsburgh, Pennsylvania; Memorial Hospital of Rhode Island/Brown University, Pawtucket, Rhode Island), and the study was approved by the institutional review boards at each of the sites. All participants gave informed consent.
Bilateral standing posteroanterior fixed-flexion knee radiographs were assessed for baseline Kellgren- Lawrence (K-L) grading (0–4) (n = 4369). If only one knee had radiographic OA, this was selected for analyses. If both knees had radiographic OA, the most severe knee (i.e. highest K-L grade) was selected for analyses. When the severity was equal between sides, the most painful knee was selected for analyses. In the case of equal pain in both knees, the dominant knee was selected for analyses. The current study included participants with radiographic knee OA (K-L grade > 2) who were obese (body mass index (BMI) > 30 kg/m2), as K-L grade > 2 has been used as the definition of radiographic knee OA and has been included in typical entry criteria of knee OA clinical trials.
Metformin use and diabetes
OAI participants were required to bring their medications when attending the baseline and yearly follow-up study visits, with medication name, duration and frequency of use (as needed or regular) recorded. Participants were classified as metformin users if they self-reported regular metformin use at baseline, 1-year and 2-year follow-up. Non-users of metformin were defined as participants who did not report the use of metformin at any visit from baseline to 4-year follow-up. Data on self-reported diabetes was collected using a questionnaire at baseline, 2-year and 4-year follow-up.
Knee pain
At baseline and yearly follow-up, knee pain was assessed using the Western Ontario and McMaster Universities Osteoarthritis Index (WOMAC) pain subscale [12] Likert scale version. It consists of five items, with the score ranging 0–20.
MRI and cartilage volume measurement
MRI was performed for the target knee using a 3 T scanner (Magnetom Trio, Siemens, Erlangen, Germany) and the exam consisted of a sagittal double-echo in steady-state (DESS) sequence. Knee cartilage volume was measured for medial and lateral tibiofemoral compartments (i.e. femoral condyle and tibial plateau) using the automatic human knee cartilage segmentation as previously described and validated, delineated as previously described and implemented in the automated segmentation [13]. The correlation between the automatic and the semi-automatic segmentations was excellent with Pearson correlations of r = 0.96 (p < 0.0001) for the global knee, r = 0.95 (p < 0.0001) for the femur, and r = 0.83 (p < 0.0001) for the tibia. The test-retest revealed an excellent measurement error of 0.3 ± 1.6% for the global knee (0.14 ± 1.7% for the femur and 1.12 ± 3.8% for the tibial plateau), corresponding to 30.3 ± 126.2 mm3 [13]. The annual rate of cartilage volume loss over 4 years was obtained by (4-year follow-up volume - baseline volume)/baseline volume/4, expressed as a percentage.
Total knee replacement
At each available follow-up, participants indicated whether they had received total knee replacement surgery. It was defined as any knee with patient-reported total knee replacement which was confirmed on subsequent radiograph between baseline and 6-year follow-up visit. Missing data for a knee replacement was treated conservatively by assuming that the participant had not undergone knee replacement surgery.
Statistical analyses
Demographic, clinical, radiological and MRI data were systematically entered into a computerised database. Descriptive statistics of participant characteristics were tabulated and compared between metformin users and non-users using independent samples t test or chi-square test, as appropriate. The rates of cartilage volume loss over 4 years were compared between metformin users and non-users using F-test (general linear model) with estimated marginal means (standard error, SE), adjusted for gender, baseline age, BMI, K-L grade, WOMAC pain score, diabetes, and weight change over 4 years. Binary logistic regression was used to examine the association between metformin use and risk of total knee replacement over 6 years, adjusted for gender, baseline age, BMI, K-L grade, WOMAC pain score, and diabetes. All tests were two-sided and a p value < 0.05 was considered statistically significant. Statistical analyses were performed using Stata 13.0 (StataCorp LP., College Station, TX, USA).