Augmented Intelligence for Clinical Discovery in Hypertensive Disorders of Pregnancy Using Outlier Analysis

Area of Science:

• Medical Informatics
• Obstetrics and Gynecology
• Data Science

Background:

• Clinical discoveries often arise from identifying rare and unusual patient cases.
• The process of identifying such cases is time-consuming for busy clinicians.
• Preeclampsia and hypertensive disorders of pregnancy have seen limited advancements in clinical management.

Purpose of the Study:

• To assess the feasibility and applicability of an augmented intelligence framework.
• To accelerate the rate of clinical discovery in preeclampsia and hypertensive disorders of pregnancy.
• To compare the effectiveness of different outlier analysis methods in identifying potential novelties.

Main Methods:

• Retrospective exploratory outlier analysis of two cohorts: the folic acid clinical trial (FACT) and the Ottawa and Kingston birth cohort (OaK).
• Applied two outlier detection methods: extreme misclassification contextual outlier and isolation forest point outlier.
• Content experts reviewed identified outliers for potential clinical novelty.

Main Results:

• In FACT (N=2,301), extreme misclassification identified 10 potential novelties (76.9%) versus 3 (15.8%) from isolation forest.
• In OaK (N=8,085), extreme misclassification identified 32 potential novelties (32.7%) versus 4 (2.3%) from isolation forest.
• Overall, 49 out of 302 identified outliers were deemed potential novelties by experts.

Conclusions:

• Augmented intelligence, particularly with extreme misclassification outlier analysis, is a feasible approach to accelerate clinical discovery.
• Extreme misclassification yielded a higher proportion of potential novelties compared to isolation forest in both study datasets.
• This AI-driven outlier analysis can be integrated into electronic medical records to expedite the identification of clinical insights across disciplines.