The creation of a novel pediatric-automated respiratory score (pARS) that leverages machine learning techniques to analyze simple vital signs and limited clinical data is a feasible option to standardize pediatric intensive care unit (PICU) management of patients with acute asthma exacerbations. A single-center study was conducted at a large quaternary children’s hospital in Colorado, and published in Pediatric Pulmonology.
Recognizing the manual pediatric asthma score (PAS) as the current standard used for acute asthma clinical care pathways, investigators sought to develop an automated system with the ability to evaluate disease severity, time course, and effect of treatment in patients in the PICU with severe asthma exacerbations. They combined continuous monitoring of vital signs (eg, heart rate, respiratory rate, and pulse oximetry) with health record data, including a provider-determined PAS in children aged 2 to 18 years admitted to the PICU for status asthmaticus. They used a cascaded artificial neural network (ANN) to create an automated respiratory score and validated 2 approaches. The ANN was then compared with the Poisson and Normal regression models.
From an initial group of 186 patients, a total of 128 individuals met study inclusion criteria. Overall, 50.8% of patients in the inclusion cohort were boys, with 34.4% of Hispanic/Latino ethnicity and 19.5% of African American ethnicity. When physiologic data were merged with clinical data, >37,000 data points were available for model training. Use of the pARS score demonstrated good predictive accuracy, with 80% of the pARS values within ±2 points of the provider-determined PAS, particularly over the mid-range of PASs (ie, 6-9). Poisson and Normal distribution regressions both yielded a smaller overall median absolute error.
The investigators concluded that the pARS actually reproduced the manually recorded PAS. Once it has been validated and studied prospectively as a research tool and as support for physician decisions, the pARS methodology can be used in the PICU to help guide treatment decisions.
Messinger AI, Bui N, Wagner BD, Szefler SJ, Vu T, Deterding RR. Novel pediatric-automated respiratory score using physiologic data and machine learning in asthma [published online April 21, 2019]. Pediatr Pulmonol. doi:10.1002/ppul.24342