Nomogram for Predicting Pulmonary Hypertension in Patients without Pulmonary Embolism

Editor:

We read with considerable interest the article by Dr Aviram and colleagues in the October 2015 issue of Radiology entitled “Pulmonary Hypertension: A Nomogram Based on CT Pulmonary Angiographic Data for Prediction in Patients without Pulmonary Embolism” (1). They showed that cardiovascular data derived from computed tomographic (CT) pulmonary angiography were associated with pulmonary hypertension (PH), and they created a nomogram based on patient age and CT pulmonary angiographic data that may facilitate identification of PH in patients who do not have acute pulmonary embolism.

The nomogram is a mathematically simple approach with the ability to reduce statistical predictive models into a single numeric estimate of the probability of an outcome (eg, presence of a disease), facilitating clinical decision making. The generation of these estimates enables the nomogram to be tailored to the personalized profile of a patient (2). Dr Aviram and colleagues reported a good predicting performance of their constructed model; however, the traditional biostatistical methods they used for evaluating models focus solely on accuracy, calibration, and discrimination using metrics such as sensitivity, specificity, or area under the receiver operating characteristic curve (AUC). Although these methods are mathematically simple and generally have an intuitive interpretation, they have little clinical relevance. For example, one cannot tell how high an AUC must be to justify clinical use of a prediction model (3). Its clinical usefulness should be prospectively validated in an independent cohort before its clinical implementation; however, the large sample size needed for prospective validation may make relative research daunting.

To address this issue, decision curve analysis (DCA) might be the desirable method for evaluating the clinical usefulness of a predictive model. DCA is a method for evaluating the benefits of a diagnostic test across a range of patient preferences for accepting risk of undertreatment (false-negative findings) and overtreatment (false-positive findings) to facilitate decisions about test selection and use. The key concept of DCA is that of a “probability threshold,” which indicates a level of diagnostic certainty above which the patient would choose to be treated (4). It can be applied directly to a data set without the need for the sort of external data on costs, benefits, and preferences typically required by traditional decision analytic techniques (5).

References