External Validation of a Prognostic Model for Predicting Nonresponse Following Knee Arthroplasty

• Published in: The Journal of arthroplasty 2017-04, Vol.32 (4), p.1153-1158.e1
• Main Authors: Riddle, Daniel L., PT, PhD, FAPTA, Golladay, Gregory J., MD, Jiranek, William A., MD, Perera, Robert A., PhD
• Format: Article
• Language: English
• Subjects: Aged; arthroplasty; Arthroplasty, Replacement, Knee; Decision Support Techniques; Female; function; Humans; knee; Male; Middle Aged; nomogram; Nomograms; Obesity - complications; Orthopedics; Osteoarthritis, Knee - diagnosis; Osteoarthritis, Knee - surgery; pain; Pain - surgery; Pain Measurement; Prognosis; Recovery of Function; Treatment Failure
• ISSN: 0883-5403; 1532-8406
• DOI: 10.1016/j.arth.2016.11.007

Abstract Background Instruments designed to predict extent of pain and function following knee arthroplasty (KA) recovery have strong potential to guide patients and clinicians in shared decision-making. Our purpose was to test the external validity of a recently developed prognostic instrument designed to estimate the probability of nonresponse following knee arthroplasty. Methods We used data from the Osteoarthritis Initiative (OAI), a nine-year multi-site National Institutes of Health study designed to examine the natural history of knee osteoarthritis in 4,796 subjects. A total of 427 subjects underwent KA over the study period. Dowsey and colleagues examined the prognostic role of obesity, general mental health, pain and function, and Kellgren and Lawrence knee osteoarthritis (OA) grades. Calibration of the prognostic model was determined using a calibration curve. The c-statistic was used to indicate discrimination of the model. Results In the primary analysis 63 (19.3%) of 326 subjects in OAI were classified as non-responders. The calibration curve generated from OAI data indicated poor calibration relative to the recently developed instrument. Discrimination as measured by the c-statistic was 0.76. Conclusion The external validity of the prognostic instrument was partially supported. While discrimination of the model was very similar to the recently developed instrument, calibration was poor indicating poor agreement between actual versus predicted probabilities of nonresponse. WOMAC and Kellgren and Lawrence grades show strong potential for use in future prognostic model development. Measurements of general mental health and obesity were not prognostic for nonresponse.