Article Text

PDF
SAT0319 Prediction models for progression of knee osteoarthritis in the multicenter osteoarthritis study (MOST)
  1. B. Wise1,
  2. Y. Zhang2,
  3. N.E. Lane1,
  4. C. McCulloch3,
  5. D.T. Felson2,
  6. M. Nevitt3,
  7. J. Torner4,
  8. C.E. Lewis5,
  9. A. Sadosky6,
  10. J. Niu2
  1. 1Medicine, University of California, Davis, Sacramento, CA
  2. 2Clinical Epidemiology Unit, Boston Univ School of Medicine, Boston
  3. 3Epidemiology, UC San Francisco, San Francisco
  4. 4Epidemiology, University of Iowa, Iowa City
  5. 5Preventive Medicine, University of Alabama, Birmingham
  6. 6Pfizer, Inc., New York, United States

Abstract

Background Although individual risk factors for incident and progressive knee osteoarthritis (OA) have been identified, less is known about models to predict progression of knee OA.

Objectives A prospective cohort study was conducted to identify predictors for progression of radiographic knee OA over 30 months.

Methods The NIH-funded Multicenter Osteoarthritis Study (MOST) is an observational study of 3026 persons age 50 to 79 years with either symptomatic knee OA or at high risk of disease. Weight-bearing, fixed flexion, in-frame posterior-anterior knee radiographs were taken at baseline and 30-month follow-up clinic visits between 2003 and 2007. Knees were included with Kellgren/Lawrence (K/L) grades of 2 or 3 at baseline. Two outcome definitions were prespecified: (1) progression to K/L grade 3 or 4 over 30 months, or incident first knee replacement (KR); (2) tibiofemoral joint space narrowing (JSN) progression by 1 grade or more, or incident first KR. Fourteen potential predictors were considered for inclusion: body mass index (BMI), height, weight, race (white vs non-white), education (partial graduate/graduate, partial college or college, vs. high school or below), occupation (labor vs. non-labor vs. other), Western Ontario McMaster (WOMAC) knee pain subscale, WOMAC function subscale, baseline K/L grade or baseline maximal tibiofemoral JSN score, malalignment (varus vs. valgus vs. neutral), history of knee injury, depressive symptoms. Additionally, age and sex were forced into all models. We used a 10-fold cross-validation procedure to select logistic regression models using as criteria: (1) higher area under curve (AUC) measured as c-statistics; (2) higher positive predictive value (PPV).

Results 1603 knees without missing data in any outcome or predictor parameters were included in this analysis. The best model to predict K/L progression based on AUC included age, sex, race, malalignment and WOMAC pain (c-statistic 0.59). The best model for K/L progression based on PPV using cut point of 0.25 included age, sex, BMI, weight and weight squared, and WOMAC pain (PPV 0.32). The best model to predict JSN progression based on AUC included age, sex, race, malalignment, WOMAC pain, and baseline maximal JSN score (c-statistic 0.55). The best model to predict JSN progression based on maximizing PPV at a cut point of 0.25 included age, sex, BMI, race, occupation, depression, WOMAC pain and WOMAC function (PPV=0.36).

Conclusions Models including known risk factors have moderate power to predict progression of radiographic knee OA over a 30 month period.

Disclosure of Interest B. Wise Grant/Research support from: Pfizer, Inc., Y. Zhang Grant/Research support from: Pfizer, Inc., N. Lane: None Declared, C. McCulloch: None Declared, D. Felson: None Declared, M. Nevitt: None Declared, J. Torner: None Declared, C. Lewis: None Declared, A. Sadosky Employee of: Pfizer, Inc., J. Niu Grant/Research support from: Pfizer, Inc.

Statistics from Altmetric.com

Request permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.