Prognostic model to predict the incidence of radiographic knee osteoarthritis

Abstract

Objective: Early diagnosis of knee osteoarthritis (KOA) in asymptomatic stages is essential for the timely management of patients using preventative strategies. We develop and validate a prognostic model useful for predicting the incidence of radiographic KOA (rKOA) in non-radiographic osteoarthritic subjects and stratify individuals at high risk of developing the disease.

Methods: Subjects without radiographic signs of KOA according to the Kellgren and Lawrence (KL) classification scale (KL=0 in both knees) were enrolled in the OA initiative (OAI) cohort and the Prospective Cohort of A Coruña (PROCOAC). Prognostic models were developed to predict rKOA incidence during a 96-month follow-up period among OAI participants based on clinical variables and serum levels of the candidate protein biomarkers APOA1, APOA4, ZA2G and A2AP. The predictive capability of the biomarkers was assessed based on area under the curve (AUC), and internal validation was performed to correct for overfitting. A nomogram was plotted based on the regression parameters. Model performance was externally validated in the PROCOAC.

Results: 282 participants from the OAI were included in the development dataset. The model built with demographic, anthropometric and clinical data (age, sex, body mass index and WOMAC pain score) showed an AUC=0.702 for predicting rKOA incidence during the follow-up. The inclusion of ZA2G, A2AP and APOA1 data significantly improved the model's sensitivity and predictive performance (AUC=0.831). The simplest model, including only clinical covariates and ZA2G and A2AP serum levels, achieved an AUC=0.826. Both models were internally cross-validated. Predictive performance was externally validated in an independent dataset of 100 individuals from the PROCOAC (AUC=0.713).

Conclusion: A novel prognostic model based on common clinical variables and protein biomarkers was developed and externally validated to predict rKOA incidence over a 96-month period in individuals without any radiographic signs of disease. The resulting nomogram is a useful tool for stratifying high-risk populations and could potentially lead to personalised medicine strategies for treating OA.

Keywords: Chondrocytes; Incidence; Knee Osteoarthritis; Osteoarthritis; Outcome Assessment, Health Care.

Figure 1

Overlap of ROC curves and area under the curve (AUC) values for the clinical model and the clinical model including biomarkers to predict KOA incidence in patients with baseline KL=0 in the development phase (OAI cohort). (A) Model with the highest AUC (best performance), including serum levels of the biomarkers ZA2G, A2AP and APOA1. (B) Optimal model, including only the biomarkers ZA2G and A2AP. Inclusion of the biomarkers significantly improved the predictive power of the model built exclusively with the clinical variables age, sex, BMI and WOMAC pain score. No significant differences between the two models were found in terms of AUC. BMI, body mass index; KOA, knee osteoarthritis; KL, Kellgren and Lawrence; OAI, osteoarthritis initiative; WOMAC, Western Ontario and McMaster University Osteoarthritis pain index.

Figure 2

Nomogram developed as clinical tool for predicting radiographic knee osteoarthritis (rKOA) incidence in individuals with baseline Kellgren and Lawrence (KL)=0. (A) Including only clinical variables (age, sex, body mass index (BMI) and WOMAC score) and (B) considering the optimal model developed in this study, which also included serum levels of ZA2G and A2AP. In using the nomogram, the scale at the top of the figure indicates the points that correspond to the score for each predictor. Once all the points for each predictor are summed, the total points scale at the bottom of the nomogram is aligned with the risk of incidence scale to determine the probability a particular individual will develop rKOA within a period of 96 months. A histogram of the data for each variable recorded in the development phase is shown to provide an overall landscape of the study population. As an example, a woman of 53 years of age with a BMI of 35.5 kg/m² and WOMAC pain score of 5 shows a total of 172 points (29.85% risk of incidence) when applying the nomogram of the clinical model. For this individual, the nomogram of the optimal model provided a risk of incidence 85 points higher than the clinical model, with a probability of 52.22%. WOMAC, Western Ontario and McMaster University Osteoarthritis pain index.