Page 29 - Assessing right ventricular function and the pulmonary circulation in pulmonary hypertension Onno Anthonius Spruijt

To test whether adding ventricular measurements to the PA/AAAX-ratio would improve the diagnostic model of CTPA for precapillary pulmonary hypertension, we compared three different diagnostic models: Model 1: PA/AAAX (standard); Model 2: PA/AAAX + RV/LVAX; and Model 3: PA/AAAX + RV/LV4CH (table 1).
Chapter 2
Prediction models Model 1
Model 2
Model 3
PA/AAAX
PA/AAAX + RV/LVAX PA/AAAX + RV/LV4CH
Table 1: Prediction models. PA/AAAX = ratio between pulmonary artery and ascending aorta. RV/LVAX = ratio between RV and LV in the axial plane. RV/LV4CH = ratio between RV and LV in the 4CH view.
The statistical approach to test the predictive value for precapillary PH of the three diagnostic models contained three different steps.
First we tested the predictive value of the three different models using multivariable binary logistic regression analysis. Second, the predictive value of the three different diagnostic models were tested using the area under the curve (AUC) derived from the Receiver Operating Characteristic curves. The AUC from the different models were compared using the DeLong method.
Third, to test the predictive value of the different diagnostic models within the clinical context of this study, we used decision curve analysis. With decision curve analysis it is possible to evaluate the clinical net benefit of the different prediction models [21, 22]. The net benefit is defined as the sum of benefits (true positives) minus the harms (false positives). Importantly, the threshold probability of the outcome determines the weights given to the true positives and false positives. The threshold probability is defined as the minimum probability of precapillary PH where a physician would decide to act. In this study it means that on the basis of the CTPA scan, it is decided to do further diagnostic tests to confirm the diagnosis. Since the exact threshold probability is unknown and will vary among physicians, we calculated the net benefit over a variety of probabilities. These net benefits can be calculated from the net benefit when nobody has precapillary PH (no positives) or from the net benefit when everybody has precapillary PH (no negatives). In this study, we focused on a range of low threshold probabilities (1-20%) since the weight assigned to false negatives (missing the diagnosis) is considerably larger than to false positives (further diagnostic evaluation).
For clinical purposes of the diagnostic models, a cut-off value to define precapillary PH is demanded. An established cut-off value to define PH is PA/AAAX > 1 [12]. A well-recognized cut-off value for the
27
2