講演情報
[II-OR18-04]先天性心疾患自動分類アルゴリズムを用いたリアルタイム予後分析システム
○鳥羽 修平1,4,8, Smith Taylor M.1,5, Sperotto Francesca1,5, Carreon Chrystalle Katte3,6, Saengsin Kwannapas1,5,9, Casella Samuel1,5, Sanders Stephen P.4,5, Feins Eric N.2,7, Colan Steven D.1,5, Mayer, Jr. John E.2,7, Kheir John N.1,5 (1.Department of Cardiology, Boston Children's Hospital, 2.Department of Cardiovascular Surgery, Boston Children's Hospital, 3.Department of Pathology, Boston Children's Hospital, 4.The Cardiac Registry, Boston Children's Hospital, 5.Department of Pediatrics, Harvard Medical School, 6.Department of Pathology, Harvard Medical School, 7.Department of Surgery, Harvard Medical School, 8.三重大学大学院 医学系研究科 胸部心臓血管外科, 9.Faculty of Medicine, Chiang Mai University)
キーワード：
Phenotyping、Outcome reporting、Prognosis
Introduction
Identifying patients with a specific phenotype of congenital heart disease (CHD) is fundamental for research, clinical decision-making, and policy-making. However, despite the accumulating data digitally recorded, the identification of phenotypes of CHD is usually performed manually due to the complexity of CHDs. In this study, we aimed to develop an automated algorithm to identify phenotypes of patients with CHD directly based on clinical databases and provide a basis for interactive reporting of outcomes.
Methods
Patients who underwent echocardiograms between January 1st, 1981, and March 31st, 2020, at Boston Children’s Hospital were included. A structured hierarchy and relationships among 7,500 diagnostic and treatment codes were re-defined based on their clinical significance, incorporating 129 defining anatomies. The algorithm to identify a patient's phenotype based on codes recorded in echocardiograms was developed. Five hundred unseen patients randomly assigned were used for external validation.
Results
We included 514,541 echocardiographs performed on 161,735 patients. Phenotypes of CHD were assigned to 84,285 patients (52%), while the others were considered to have normal cardiovascular anatomy. The accuracy of the algorithm was 96.4% (482 of 500 patients). Based on the phenotypes identified, an interactive tool to report outcomes was developed.
Conclusions
An automated algorithm could identify diagnostic phenotypes in patients with CHD and provide a basis for interactive real-time outcome reporting.
Identifying patients with a specific phenotype of congenital heart disease (CHD) is fundamental for research, clinical decision-making, and policy-making. However, despite the accumulating data digitally recorded, the identification of phenotypes of CHD is usually performed manually due to the complexity of CHDs. In this study, we aimed to develop an automated algorithm to identify phenotypes of patients with CHD directly based on clinical databases and provide a basis for interactive reporting of outcomes.
Methods
Patients who underwent echocardiograms between January 1st, 1981, and March 31st, 2020, at Boston Children’s Hospital were included. A structured hierarchy and relationships among 7,500 diagnostic and treatment codes were re-defined based on their clinical significance, incorporating 129 defining anatomies. The algorithm to identify a patient's phenotype based on codes recorded in echocardiograms was developed. Five hundred unseen patients randomly assigned were used for external validation.
Results
We included 514,541 echocardiographs performed on 161,735 patients. Phenotypes of CHD were assigned to 84,285 patients (52%), while the others were considered to have normal cardiovascular anatomy. The accuracy of the algorithm was 96.4% (482 of 500 patients). Based on the phenotypes identified, an interactive tool to report outcomes was developed.
Conclusions
An automated algorithm could identify diagnostic phenotypes in patients with CHD and provide a basis for interactive real-time outcome reporting.