The adapted REHIPED Reveal Lite score achieved a C-statistic of 0.83 for 1-year and 0.82 for 3-year survival, effectively stratifying pediatric PH patients by mortality risk.
Does the REHIPED risk score accurately stratify 1-year and 3-year mortality risk in pediatric patients with pulmonary hypertension?
The REHIPED risk score, an AI-adapted version of the Reveal Lite score, provides good discrimination for predicting 1-year and 3-year survival in pediatric patients with pulmonary hypertension.
Tasa de eventos absoluta: 0% vs 0%
Abstract Risk stratification has been proven fundamental in managing patients with pulmonary hypertension (PH). However, the current risk scores are less validated in pediatric populations. Our aim is to design a pediatric risk score for PH based on the Reveal Lite score. REHIPED registry is a voluntary, multicenter registry including patients under 18 years of age with PH, from 22 Spanish centers, and 4 Colombian centres. The variables used for training the model were heart rate, systolic blood pressure, functional class, glomerular filtration rate, 6-minute walking test distance, natriuretic peptides, the age at diagnosis, weight and height, and the TAPSE. All variables were normalized according to age, weight and height. A Gradient Boosting machine learning scheme was set up using a model composed of single branch decisions with a single cut-off on one of the available variables. The Gradient Boosting algorithm generates a sequence of cut-offs that, on each training round, minimizes the negative log-likehood of an accelerated failure time model of a patient’s survival time. To avoid overfitting, a Monte Carlo analysis was carried out over 32 random splits between an 80% training and 20% validation populations, finding the average optimal stopping point at 7800 training rounds. The full population was finally used for training, obtaining 7800 decisions, and adding up the cut-off scores, the risk score function was obtained. Finally, a risk stratification analysis was performed. To estimate the risk for each score value, the patients in a neighbourhood of the score value are used. The following risk levels were obtained: low risk (0-5% 1 year mortality), intermediate risk (5-20% 1 year mortality) and high risk ( 20% 1 year mortality). 420 children and adolescents were analysed. 309 patients had group 1 PH (pulmonary arterial hypertension) and 111 had group 3 PH (associated with lung diseases and/or hypoxia). The median age was 1.9 years and the interquartile range was 0.3-7.4 years. The model designs a function that returns a floating-point score which is inversely proportional to the mortality risk. The contribution of each variable to the risk score function is listed in table 1: negative values increase risk; and the higher the absolute number, the higher the impact. In the table, it is also shown the assigned contribution when the value is missing. The height wasn’t assigned a different factor at any cut-off, this is, it does not impact the result. The C statistic of the risk score function is 0.83 for 1 year survival and 0.82 for 3 years survival. The estimated 1-year and 3-year survival rates, free from death or transplantation, of the entire cohort were 84% (CI 95% 80-97%) and 76 % (CI 95% 71-80 %), respectively. The survival rate by risk strata can be found in figure 1. This REHIPED risk score, adapting the Reveal Lite score to pediatrics, provides good discrimination for low, intermediate and high-risk pediatric patients.Table 1
Escribano et al. (Sat,) reported a other. The adapted REHIPED Reveal Lite score achieved a C-statistic of 0.83 for 1-year and 0.82 for 3-year survival, effectively stratifying pediatric PH patients by mortality risk.