POS0966 DEVELOPMENT AND VALIDATION OF AN ANKYLOSING SPONDYLITIS RISK PREDICTION MODEL USING LONGITUDINAL REAL-WORLD DATA FROM A LARGE INTEGRATED HEALTHCARE DELIVERY SYSTEM

Abstract

BackgroundEarly detection and diagnosis of Ankylosing spondylitis (AS) is challenging due to heterogeneity in disease presentation, lack of specific biomarkers and high prevalence of mechanical back-pain that is difficult to distinguish from inflammatory back-pain. However, if diagnosed and treated early, the risk of AS complications and disease progression can be slowed.ObjectivesTo develop and validate a risk prediction model for early identification of patients at high risk of AS, using a large longitudinal real-world clinical data in the US.MethodsThis retrospective study included all members aged ≥ 21 years with back pain symptoms who were enrolled in the Kaiser Permanente Southern California health plan between 01/2009-12/2013. Patients who presented with back pain symptoms at a physician visit were followed until 12/2020 to see if they subsequently developed AS. The cohort was randomly divided into a training (60%) and a validation (40%) sample. A proportional odds model was specified to create a risk score for AS in the training sample. Best fit model was determined based on Area Under the Curve (AUC) and Akaike Information Criterion (AIC). The cut off threshold of “high-risk” was based on Youden’s (1950) index.1 We assessed the model performance for internal validity using split-samples. The model was further validated using manual chart review of 900 patient records. These 900 records were selected such that 70% (N=630) met the high-risk cut-off and the remainder had scores below the cut-off. We also derived the probability of AS in each chart reviewed case using the method proposed by Feldtkeller et al. (2013) and Rudwaleit et al. (2006).2,3ResultsThe cohort comprised 527,509 members with mean age 54 years and majority female (58%). Sixty-six percent were White race and 33% were Hispanic ethnicity. The crude incidence of AS was 1% and increased steadily during the follow-up period (Figure 1). The final risk prediction model included 15 risk factors and had an AUC of 0.72 (Table 1). Using Youden’s index, a cut-off value of 11 or higher was identified as the threshold to define high-risk. No evidence of overfitting to our training sample was observed based on the split-sample analysis. The model validation based on manual chart review of 900 records showed sufficient ability to discriminate between those at high-risk vs those not identified to be high-risk. When a concrete rule out or rule in determination could be made using Feldtkeller et al. approach, our model correctly classified 75% of such records.Table 1.Final Model Coefficients and Derived Risk ScoreModel CoefficientPr(>|z|)Risk ScoreAge above 45 years0.1460.0111.46Male Sex-0.292<0.001-2.92White Race0.259<0.0012.59Non-Hispanic Ethnicity0.177<0.0011.77Corticosteroid Use (Yes/No)0.235<0.0012.35*NSAID User (Yes/No)0.202<0.0012.02Opioid Analgesic User (Yes/No)0.416<0.0014.16Had Enthesitis (Yes/No)0.298<0.0012.98Had Disorders of the Back (Yes/No)0.973<0.0019.73Had Fatigue and/or Malaise (Yes/No)0.1550.0561.55Had Psoriasis (Yes/No)0.2880.0082.88Had Spondylosis (Yes/No)0.838<0.0018.38Had Synovitis (Yes/No)0.1430.0661.43Had Uveitis (Yes/No)0.6210.0076.21Depression Diagnosis (Yes/No)0.0990.0170.99* NSAID: Non-steroidal anti-inflammatory drugsFigure 1.Cumulative Incidence Over Time (in Days)ConclusionTo aid early detection, we have developed and validated an AS risk prediction model with an easy to implement scoring system using demographics, medication use and diagnosis data that is routinely collected in clinical practice.References:[1]Youden WJ. Index for rating diagnostic tests. Cancer. 1950;3(1):32-35.[2]Feldtkeller E, Rudwaleit M, Zeidler H. Easy probability estimation of the diagnosis of early axial spondyloarthritis by summing up scores. Rheumatology (Oxford, England). 2013;52(9):1648-1650[3]Rudwaleit M, Feldtkeller E, Sieper J. Easy assessment of axial spondyloarthritis (early ankylosing spondylitis) at the bedside. Ann Rheum Dis 2006;65:1251-2.Disclosure of InterestsAniket Kawatkar Grant/research support from: Novartis, Medac, Esther Yi Employee of: Novartis, Erika Estrada Grant/research support from: Novartis, Jose Pio Grant/research support from: Novartis, Cecilia Portugal Grant/research support from: Novartis, David Yi Grant/research support from: Novartis, Talar Habeshian Grant/research support from: Novartis, David Wei Employee of: Novartis, Steven Lee Grant/research support from: Novartis