Table 3 |
|||
|
Logistic regression predictive models results by using changes in biomarker levels from baseline to week 4 and clinical response at week 14. Model included training and test datasets combined |
|||
|
ACR 20 |
|||
|
|
|||
|
CRP-only modela (n = 308) |
|||
|
|
|||
|
True/Predicted |
NR |
R |
Accuracy |
|
NR |
35 |
115 |
Specificity = 23% |
|
R |
16 |
142 |
Sensitivity = 90% |
|
Predictive value |
NPV = 68% |
PPV = 55% |
|
|
Threshold = -0.35 |
|||
|
|
|||
|
Biomarker modela (n = 308) |
|||
|
|
|||
|
True/Predicted |
NR |
R |
Accuracy |
|
NR |
62 |
88 |
Specificity = 41% |
|
R |
16 |
142 |
Sensitivity = 90% |
|
Predictive value |
NPV = 79% |
PPV = 61% |
|
|
|
|||
|
Threshold = -0.35 |
|||
|
|
|||
|
Predictors from biomarker modelb |
Weight |
Odds ratio |
Multivariate P value |
|
|
|||
|
Hyaluronic acid |
0.60 |
1.82 |
0.0001 |
|
Apolipoprotein C3 |
-0.98 |
0.38 |
0.003 |
|
Rheumatoid factor positive at baseline |
-0.91 |
0.4 |
0.014 |
|
Basic fibroblast growth factor |
0.72 |
2.05 |
0.023 |
|
Interleukin-16 |
0.83 |
2.29 |
0.031 |
|
Interleukin-6 serum |
0.21 |
1.23 |
0.049 |
|
C-reactive protein |
0.17 |
1.19 |
0.062 |
|
|
|||
|
ACR 50 |
|||
|
|
|||
|
CRP-only modela (n = 308) |
|||
|
|
|||
|
True/Predicted |
NR |
R |
Accuracy |
|
NR |
108 |
116 |
Specificity = 48% |
|
R |
17 |
67 |
Sensitivity = 80% |
|
Predictive Value |
NPV = 86% |
PPV = 36% |
|
|
Threshold = -1.3 |
|||
|
|
|||
|
Biomarker modela (n = 308) |
|||
|
|
|||
|
True/Predicted |
NR |
R |
Accuracy |
|
NR |
137 |
87 |
Specificity = 61% |
|
R |
17 |
67 |
Sensitivity = 80% |
|
Predictive value |
NPV = 89% |
PPV = 44% |
|
|
Threshold = -1.2 |
|||
|
|
|||
|
Predictors from biomarker modelb |
Weight |
Odds ratio |
Multivariate P value |
|
|
|||
|
Hyaluronic acid |
0.53 |
1.70 |
0.001 |
|
Haptoglobin |
0.79 |
2.2 |
0.001 |
|
Apolipoprotein C3 |
-0.94 |
0.39 |
0.007 |
|
Interleukin-16 |
1 |
2.72 |
0.014 |
|
STJC at baseline |
0.17 |
1.19 |
0.015 |
|
Anti-CCP antibodies |
-0.82 |
0.44 |
0.044 |
|
|
|||
|
aAll biomarker values were log transformed before inclusion in the models. Model accuracy (sensitivity, specificity, NPV, PPV) was estimated by using cross validation. bWeights are the coefficients in the logistic regression model. Odds ratios are the exponential of the weights. Multivariate P values are the P values when all of the terms are included in the model. ACR, American College of Rheumatology; anti-CCP, cyclic citrullinated peptide; CRP, C-reactive protein; NPV, negative predictive value; NR, nonresponder; PPV, positive predictive value; R, responder; STJC, swollen and tender joint count. |
|||
|
Visvanathan et al. Arthritis Research & Therapy 2010 12:R211 doi:10.1186/ar3188 |
|||
