158. Development of a phenotyping algorithm for Hypertensive Disorders of Pregnancy (HDP) in large-scale prospective genome cohort study

Background To elucidate genetic and environmental factors causing hypertensive disorders of pregnancy (HDP) in large-scale cohort study, we need to conduct precise phenotyping on subjects in cohort study. However, large-scale phenotyping by medical doctors’ diagnoses is impracticable with regard to cost-effectiveness. Here, we developed a rule-based HDP phenotyping algorithm. Our algorithm enables us to identify HDP patients and classify them into subtypes using large-scale cohort dataset. Methods We developed a phenotyping algorithm according to the HDP diagnosis criteria of Japan Society of the Study of Hypertension in Pregnancy (JSSHP). Our phenotyping algorithm was applied to 22,257 mothers with pregnancy recruited by the BirThree Cohort Study of Tohoku Medical Megabank. To validate consistency between our algorithm-based phenotype and medical-doctor’s diagnoses, an independent medical doctor examined medical records and made clinical diagnoses on 50 subjects. Results By conducting our phenotyping algorithm, 1939 (8.71%) subjects were phenotyped with HDP patients. Among them, HDP patients were phenotyped with subtypes as follows; 995 (4.47%) patients with gestational hypertension (GH), 318 (1.43%) patients with super-imposed pre-eclampsia (SuPE), and 626 (2.81%) patients with pre-eclampsia (PE). As for consistency between our algorithm-based phenotype and diagnose, 36 cases shows consistentency for subtypes and timing of onset, whereas 14 cases shows no consistency; 10 cases were inconsistent in onset of HDP, 2 cases were in subtypes and the remaining 2 cases were in timing of onset. Discussion In this study, we developed high-performance HDP phenotyping algorithm. The major part of the reasons of inconsistency was that postpartum hypertension was phenotyped in non-HDP group by our algorithm. This misphenotyping was caused by that postpartum blood pressure was not included in cohort data but included in medical records. In future, we will develop not only rule-based but also machine-learning based phenotyping algorism to perform more precise phenotyping in large scale clinical information.