Prediction models for the risk of total knee replacement: development and validation using data from multicentre...

Area of Science:

• Orthopedics
• Radiology
• Epidemiology

Background:

• Prognostic models for total knee replacement (TKR) are limited.
• The predictive value of radiographic findings for TKR is unclear.

Purpose of the Study:

• Develop and validate predictive models for TKR risk.
• Assess if incorporating radiographic findings enhances predictive performance.

Main Methods:

• Cox proportional hazard models were used on data from the Multicenter Osteoarthritis Study (MOST) and Osteoarthritis Initiative (OAI).
• Two models were developed: one with clinical factors and another including radiographic findings (Kellgren-Lawrence grade).
• Model performance was evaluated using C-statistic, calibration, net reclassification improvement (NRI), and integrated discrimination improvement (IDI).

Main Results:

• The clinical model included age, sex, race, knee pain, and WOMAC score (C-statistic: 0.79).
• The radiographic model added Kellgren-Lawrence grade (C-statistic: 0.87), showing improved discrimination.
• Radiographic findings significantly improved prediction (NRI: 0.43, IDI: 0.14).

Conclusions:

• Predictive models using common risk factors demonstrate good discrimination and calibration for TKR.
• Incorporating radiographic findings substantially improves the predictive performance of TKR risk models.