Back to Results

< Previous   Next >

Quantification of the whole-body burden of radiographic osteoarthritis using factor analysis

Creators: Nelson, Amanda E, DeVellis, Robert F, Renner, Jordan B, Schwartz, Todd A, Conaghan, Philip G, Kraus, Virginia B, Jordan, Joanne M

File Type: pdf | Filesize: 413 KB | Date Added: 2012-08-23 | Date Created: 2011-10-25

Abstract Introduction Although osteoarthritis (OA) commonly involves multiple joints, no widely accepted method for quantifying whole-body OA burden exists. Therefore, our aim was to apply factor analytic methods to radiographic OA (rOA) grades across multiple joint sites, representing both presence and severity, to quantify the burden of rOA. Methods We used cross-sectional data from the Johnston County Osteoarthritis Project. The sample (n = 2092) had a mean age of 65 &#177; 11 years, body mass index (BMI) 31 &#177; 7 kg/m2, with 33% men and 34% African Americans. A single expert reader (intra-rater &#954; = 0.89) provided radiographic grades based on standard atlases for the hands (30 joints, including bilateral distal and proximal interphalangeal [IP], thumb IP, metacarpophalangeal [MCP] and carpometacarpal [CMC] joints), knees (patellofemoral and tibiofemoral, 4 joints), hips (2 joints), and spine (5 levels [L1/2 to L5/S1]). All grades were entered into an exploratory common factor analysis as continuous variables. Stratified factor analyses were used to look for differences by gender, race, age, and cohort subgroups. Results Four factors were identified as follows: IP/CMC factor (20 joints), MCP factor (8 joints), Knee factor (4 joints), Spine factor (5 levels). These factors had high internal consistency reliability (Cronbach's &#945; range 0.80 to 0.95), were not collapsible into a single factor, and had moderate between-factor correlations (Pearson correlation coefficient r = 0.24 to 0.44). There were no major differences in factor structure when stratified by subgroup. Conclusions The 4 factors obtained in this analysis indicate that the variables contained within each factor share an underlying cause, but the 4 factors are distinct, suggesting that combining these joint sites into one overall measure is not appropriate. Using such factors to reflect multi-joint rOA in statistical models can reduce the number of variables needed and increase precision.