Skip to main content
. 2013 May 18;73(7):1340–1349. doi: 10.1136/annrheumdis-2013-203301

Figure 2.

Figure 2

Two-step decision tree for detection of pulmonary arterial hypertension in systemic sclerosis patients: the DETECT algorithm. Of the 408 SSc patients (87 PAH and 321 non-PH) at risk for PAH (SSc of >3 years’ duration, DLCO <60% of predicted, FVC ≥40% of predicted), data from 319 patients (72 PAH and 247 non-PH) were used for construction of the algorithm. All patients underwent right heart catheterisation. Sensitivity and specificity of the two steps of the algorithm (and the corresponding risk point cut-offs) were selected by the Study Scientific Committee with the aim of minimising the number of missed PAH diagnoses. Step 1: A complete dataset was available for 356 patients. The combined discriminatory ability of the six selected non-echocardiographic variables expressed as the AUC of the ROC curve was 84.4% (95% CI 79.5% to 89.8%) showing good discriminatory performance and no statistically significant lack of fit (see online supplementary appendix 5). At Step 1, a predefined sensitivity cut-off of 97% (corresponding to >300 risk points, compare figure 3), determined no referral to echocardiography in 52 patients. Among these, 50 were true negatives (patients without PAH on right heart catheterisation) and two were false negatives (PAH confirmed on right heart catheterisation). Step 2: A complete dataset was available for 267 patients. The AUC of the ROC curve for the total risk points from Step 1, plus the two selected echocardiographic variables, was 88.1% (95% CI 82.4% to 92.3%). A predefined specificity cut-off of 35% (corresponding to >35 risk points, compare figure 3), determined no referral to right heart catheterisation in 69 patients. Among these, 68 were true negatives and one was a false negative. Right heart catheterisation in the remaining 198 patients yielded 69 true positives (PAH confirmed) and 129 false positives. Thus, overall, the algorithm missed 3 (4%) out of the 72 PAH patients who had sufficient data to be included in the analysis. Note that the algorithm uses cut-offs for the risk points of the two steps only but not for individual parameters. ACA, anticentromere antibody; AUC, area under the curve; DLCO, pulmonary diffusing capacity for carbon monoxide; FVC, forced vital capacity; NTproBNP, N-terminal probrain natriuretic peptide; PAH, pulmonary arterial hypertension; PH, pulmonary hypertension; ROC, receiver operating characteristic; SSc, systemic sclerosis; TR, tricuspid regurgitant jet.