Table 2

Logistic regression models

 Univariable logistic regression models*Multivariable logistic regression models (two-step decision tree)
ROC AUC,% (95% CI)Estimated coefficient (95% CI)p valueROC AUC,% (95% CI)
Step 184.4 (79.5 to 89.8)
 Intercept−12.488 (−16.372 to −8.603)
 FVC % pred./DLCO % pred.71.5 (65.6 to 77.4)1.149 (0.566 to 1.731)<0.001
 Current/past telangiectasias59.7 (55.4 to 64.1)1.156 (0.336 to 1.975)0.006
 Serum ACA (presence)62.4 (56.4 to 68.4)0.753 (0.133 to 1.373)0.017
 Serum NTproBNP (log10)67.5 (60.9 to 74.2)0.915 (0.308 to 1.521)0.003
 Serum urate71.9 (65.9 to 77.9)1.247 (0.497 to 1.997)<0.001
 Serum urate (spline component†)−1.132 (−2.048 to −0.215)
 Right axis deviation (presence)54.9 (51.1 to 58.7)1.850 (0.507 to 3.193)0.007
Step 288.1 (82.4 to 92.3)
 Intercept−2.452 (−5.747 to 0.844)
 Linear predictor Step 10.891 (0.559 to 1.224)<0.001
 Right atrium area‡71.2 (65.0 to 77.3)0.075 (−0.004 to 0.154)0.062
 TR velocity79.5 (73.7 to 85.3)0.209 (−1.117 to 1.534)<0.001
 TR velocity (spline component†)2.656 (0.380 to 4.933)
  • *The ROC AUC values of the corresponding univariable logistic regression models are added for the sake of comparison.

  • †Serum urate and TR velocity were included in the respective models using restricted cubic splines with three knots. Knots for serum urate were selected at 3.3, 4.7 and 7.1 mg/100 ml and for TR velocity were selected at 2, 2.5 and 3.4 m/s. For each of these two variables only one p value is presented indicating its overall effect.

  • ‡When right ventricle area was used instead of right atrium area, the p value was 0.035; the rest of the p values remained with similar results.

  • ACA, anticentromere antibody; DLCO, pulmonary diffusing capacity for carbon monoxide; FVC, forced vital capacity; NTproBNP, N-terminal probrain natriuretic peptide; ROC AUC, area under the receiver operating characteristic curve; TR, tricuspid regurgitant jet.