Alexander W Hirsch1, Michael C Monuteaux2, Mark I Neuman2, Richard G Bachur2. 1. Division of Emergency Medicine, Boston Children's Hospital and Harvard Medical School, Boston, MA. Electronic address: alexander.hirsch@childrens.harvard.edu. 2. Division of Emergency Medicine, Boston Children's Hospital and Harvard Medical School, Boston, MA.
Abstract
OBJECTIVE: To improve the prediction of pediatric pneumonia by developing a series of models based on clinically distinct subgroups. We hypothesized that these subgroup models would provide superior estimates of pneumonia risk compared with a single pediatric model. STUDY DESIGN: We conducted a secondary analysis of a prospective cohort being evaluated for radiographic pneumonia in an urban pediatric emergency department (ED). Using multivariate modeling, we created 4 models across subgroups stratified by age and presence of wheezing to predict the risk of pneumonia. RESULTS: A total of 2351 patients were included in the study. In this series, the prevalence of pneumonia was 8.5%, and 21.6% were hospitalized. The highest prevalence of pneumonia was in children aged >2 years without wheezing (13.3%). Children aged <2 years with wheezing had the lowest prevalence of pneumonia (4.0%). The most accurate model was for children aged <2 years with wheezing (area under the curve [AUC], 0.80), and the poorest performing model was for those aged <2 years without wheezing (AUC, 0.64). The AUC of a combination of the 4 subgroup models was 0.76 (95% CI, 0.72-0.80). The precision of the models' estimates (expected vs observed) was ± 3.7%. CONCLUSIONS: Using 4 complementary prediction models for pediatric pneumonia, an accurate risk of pneumonia can be calculated. These models can provide the basis for clinical decision making support to guide the use of chest radiographs and promote antibiotic stewardship.
OBJECTIVE: To improve the prediction of pediatric pneumonia by developing a series of models based on clinically distinct subgroups. We hypothesized that these subgroup models would provide superior estimates of pneumonia risk compared with a single pediatric model. STUDY DESIGN: We conducted a secondary analysis of a prospective cohort being evaluated for radiographic pneumonia in an urban pediatric emergency department (ED). Using multivariate modeling, we created 4 models across subgroups stratified by age and presence of wheezing to predict the risk of pneumonia. RESULTS: A total of 2351 patients were included in the study. In this series, the prevalence of pneumonia was 8.5%, and 21.6% were hospitalized. The highest prevalence of pneumonia was in children aged >2 years without wheezing (13.3%). Children aged <2 years with wheezing had the lowest prevalence of pneumonia (4.0%). The most accurate model was for children aged <2 years with wheezing (area under the curve [AUC], 0.80), and the poorest performing model was for those aged <2 years without wheezing (AUC, 0.64). The AUC of a combination of the 4 subgroup models was 0.76 (95% CI, 0.72-0.80). The precision of the models' estimates (expected vs observed) was ± 3.7%. CONCLUSIONS: Using 4 complementary prediction models for pediatric pneumonia, an accurate risk of pneumonia can be calculated. These models can provide the basis for clinical decision making support to guide the use of chest radiographs and promote antibiotic stewardship.