Skip to main content

Table 3 Characteristics of diagnostic algorithms for diagnosis and typing of early-stage arthritis: predictive algorithm outcomes for twofold cross-validation

From: Glycation marker glucosepane increases with the progression of osteoarthritis and correlates with morphological and functional changes of cartilage in vivo

Algorithm features Algorithm 1: plasma Hyp and GSP, G-H1, MG-H1, 3DG-H, CEL, CMA, MetSO, 3-NT, NFK, and DT free adducts Algorithm 2: anti-CCP-Ab positivity assessment and plasma GSP, FL, 3DG-H, CML, CEL, MetSO, and 3-NT free adducts
Classification Disease vs control eOA vs non-RA and eRA eRA vs non-RA and eOA Non-RA vs eRA and eOA
Accuracy (%) 88.4 (86.9–90.0) 95.5 (93.7–97.3) 78.1 (74.3–82.0) 78.9 (74.7–83.2)
Sensitivity (%) 90.4 (88.7–92.1) 94.0 (88.8–99.3) 69.1 (54.4–83.7) 67.5 (46.2–88.8)
Specificity (%) 83.2 (77.4–89.0) 96.1 (92.6–99.6) 83.2 (71.8–94.6) 84.6 (77.1–92.1)
AUROC 0.93 (0.92–0.94) 0.98 (0.97–0.99) 0.86 (0.81–0.90) 0.88 (0.86–0.90)
Positive likelihood ratio 8.26 (5.77–10.75) 16.11 (9.56–22.66) 7.66 (2.94–12.37) 4.96 (3.31–6.60)
Negative likelihood ratio 0.11 (0.10–0.13) 0.06 (0.01–0.11) 0.34 (0.20–0.48) 0.36 (0.14–0.57)
Positive predictive value (%) 93.6 (91.7–95.6) 92.8 (86.5–99.0) 76.3 (65.1–87.4) 71.8 (62.7–80.8)
Negative predictive value (%) 77.3 (74.4–80.2) 97.5 (95.3–99.6) 84.4 (79.0–89.8) 86.2 (78.9–93.4)
F-score 0.92 (0.91–0.93) 0.93 (0.90–0.96) 0.69 (0.62–0.75) 0.64 (0.49–0.79)
  1. Abbreviations: MetSO Methionine sulfoxide, FL Nε-fructosyl-lysine, G-H1 Glyoxal-derived hydroimidazolone, MG-H1 Methylglyoxal-derived hydroimidazolone, CMA Nω-carboxymethylarginine, 3DG-H 3-Deoxyglucosone-derived hydroimidazolone isomers, GSP Glucosepane, NFK N-formylkynurenine, 3-NT 3-Nitrotyrosine, Hyp Hydroxyproline, eOA Early-stage osteoarthritis, TKR Total knee replacement, RA Rheumatoid arthritis, eRA Early-stage rheumatoid arthritis, DT Dityrosine, CCP Cyclic citrullinated peptide, CML Nε-carboxymethyl-lysine, CEL Nε-carboxyethyl-lysine
  2. Data are mean (95% CI). Analyte data other than GSP (Table 2) employed in algorithm development were reported previously [5]