Table 3

Logistic regression model comparing the ability of baseline CRP and baseline biomarker levels for the prediction of clinical improvement at week 14

Model (true/predicted)*ASAS 20 responseBASDAI 50 response≥2 Unit improvement in BASFI score
Baseline CRP only* NRRAccuracyNRRAccuracyNRRAccuracy
 Non-responder1411Specificity: 56%2011Specificity: 65%2110Specificity: 68%
 Responder1139Sensitivity: 78%2421Sensitivity: 47%1827Sensitivity: 60%
 Predictive valueNPV 56%PPV 78%NPV 45%PPV 66%NPV 54%PPV 73%
Baseline biomarker*NRRAccuracy*NRRAccuracy*NRAccuracy*
 Non-responder178Specificity: 65%2618Specificity: 59%1722Specificity: 61%
 Responder545Sensitivity: 90%329Sensitivity: 91%334Sensitivity: 85%
 Predictive valueNPV 77%PPV 85%NPV 90%PPV 65%NPV 85%PPV 61%
PredictorsWeightORp valueWeightORp valueWeightORp value
 Immunoglobulin M1.243.460.006
 Insulin-0.500.610.007
 Leptin-0.690.500.0004
 N-terminal propeptide of type I collagen1.987.240.0011.414.100.007
 Tissue inhibitor of metalloproteinases 13.3528.500.002
 Vascular endothelial growth factor0.792.200.026
  • * All biomarker values were log2-transformed before inclusion in the models. Both models are based on 76 patients.

  • Model accuracy (sensitivity, specificity, NPV, PPV) was estimated using cross-validation.

  • Weights are the coefficients in the logistic regression model. OR are the exponential of the weights. Multivariate p values are based on the inclusion of all terms in the model.

  • ASAS 20, at least 20% improvement in Assessment in SpondyloArthritis international Society response criteria; BASDAI 50, ≥50% improvement in the Bath ankylosing spondylitis disease activity index; BASFI, Bath ankylosing spondylitis functional index; CRP, C-reactive protein; NPV, negative predictive value; NR, non-responders; PPV, positive predictive value; R, responders.