A Virtual Knee Joint Replacement Clinical Endpoint Based On Longitudinal Trends and Thresholds in Koos Knee Pain and Function in Osteoarthritis Initiative Participants

Robert M. Boudreau¹, David J. Hunter², Zhijie Wang³, Frank Roemer⁴, Felix Eckstein⁵, Michael J. Hannon³, Ali Guermazi⁶ and C. Kent Kwoh⁷, ¹Epidemiology, University of Pittsburgh, Pittsburgh, PA, ²Rheumatology, Institute of Bone and Joint Research, Kolling Institute, University of Sydney, Sydney, Australia, ³Department of Medicine, University of Pittsburgh School of Medicine, Pittsburgh, PA, ⁴Radiology, Boston University School of Medicine, Boston, MA, ⁵Anatomy & Musculoskeletal Research, Paracelsus Medical University, Salzburg, Austria, ⁶Boston University, Boston, MA, ⁷School of Medicine, Department of Medicine, Division of Rheumatology and Clinical Immunology, University of Pittsburgh, Pittsburgh, PA

Meeting: ACR/ARHP Annual Meeting 2012

Keywords: Knee, longitudinal studies, Osteoarthritis, outcome measures and total joint replacement

Session Information

Title: Osteoarthritis - Clinical Aspects II: Structural Risks for Osteoarthritis End-points and Potential Treatments

Session Type: Abstract Submissions (ACR)

Background/Purpose: Knee joint replacement (KR) is a cost-effective procedure with good long-term outcomes, but has limitations as an endpoint in OA intervention trials. A definition of virtual total joint replacement (vTKR) based on clinical measures may be a more useful, efficient and practical outcome measure. The aim of this study was therefore to develop and evaluate a potential vTKR definition based on clinical measures in the OAI. Methods: The sample included 8204 knees from 4143 individuals (42% male, age 61.6 ± 9.2) in the Progression, Incident and Control cohorts. A total of 187 individuals underwent KR (216 knees) over the five years after baseline. Individuals with continuous health care coverage during the five years were included, except for individuals who underwent KR. Clinical measures individually predictive of KR, thus considered for inclusion in the model, were: WOMAC Knee Pain, Stiffness, Disability and Total scores; KOOS Knee Pain (KOOSKP), Symptoms, ADL, Quality of Life (QOL) and Total scores; Knee pain severity in past 30 days and PASE. Several other clinical measures were excluded as not predictive. At each annual visit, logistic regression models were developed predicting actual KR during the following year. Baseline, the two timepoints prior to KR (and changes between them) entered models with up to three clinical measures. Sensitivity, specificity, ROCs and AUCs were evaluated. Adapting a propensity-score approach, the 5% of non-KR knees with the highest predicted probability of being a KR were eligible vTKR knees. We then further required that clinical measures at the incident vTKR timepoint be as poor or worse than baseline and also did not improve through year 5. Results: A combination of KOOS Knee Pain and QOL scores was identified as the best KR-predicting vTKR model. vTKR criterion was based on three factors that increased the predicted odds of KR: crossing a threshold of severe knee pain (KOOS), a threshold of sufficiently poor quality of life, a weighted combination of these two, and/or a marked increase in knee pain over the year prior. Cross-validated AUCs, distinguishing between KR and non-KR knees across annual visits ranged from 0.867 to 0.921. A parity-inducing probability cutpoint correctly separated KR from non-KR knees with sensitivity and specificity of 0.85. A total of 399 knees maintained or got worse after their vTKR. Of these, 143 were as bad or worse than baseline. Conclusion: We have developed a promising vTKR criterion based on a combination of threshold KOOS Knee Pain and QOL scores, and/or worsening KOOS Knee Pain. This vTKR criterion may be a useful outcome measure for OA intervention trials.

Final clinical variable vTKR prediction model *Odds-ratios (95% C.I.’s) for KOOSKP and KOOSQOL (during preceding two periods) Predicting Actual KR**
vTKR month	KOOSKP Threshold (OR per 1 unit lower)	KOOSQOL Threshold (OR per 1 unit lower)	Additional effect if KOOSKP got worse during preceding two periods (OR per 1 unit worsening)
24†	1.047 (1.028,1.065)	1.032 (1.013,1.051)	1.018 (0.997,1.039)**
36	1.043 (1.025,1.062)	1.028 (1.011,1.044)	1.054 (1.035,1.073)
48	1.053 (1.035,1.072)	1.027 (1.010,1.044)	1.060 (1.041,1.079)
60	1.048 (1.030,1.066)	1.023 (1.008,1.039)	1.048 (1.029,1.067)
† vTKR month 24: Model predicting KRs during year following 12 month OAI contact * All OR’s were statistically significant at p < 0.003, except KOOSKP worsening at 24m ** KOOSKP worsening at 24m: p=0.0903

Disclosure:

R. M. Boudreau,
None;

D. J. Hunter,
None;

Z. Wang,
None;

F. Roemer,
None;

F. Eckstein,
None;

M. J. Hannon,
None;

A. Guermazi,
None;

C. K. Kwoh,
None.

ACR Meeting Abstracts - https://acrabstracts.org/abstract/a-virtual-knee-joint-replacement-clinical-endpoint-based-on-longitudinal-trends-and-thresholds-in-koos-knee-pain-and-function-in-osteoarthritis-initiative-participants/