"Algorithm development to estimate pregnancy data from Electronic Health Records. A Use Case from Argentina"
Carolina Mengoni Goñalons Juliana Reves Szemere María Cristina Nanton
Access our slides in this LINK
In this work we present a natural language processing algorithm designed to detect and characterize pregnancies among patients that receive care in the public healthcare system of Buenos Aires. Our aim is to enhance statistical analysis, patient health tracking, and public health monitoring and management.
Since the implementation of Electronic Health Records (EHR) in 2016 as part of the Hospital Information Management System (SIGEHOS) in Buenos Aires City, the digitization of medical records has enabled large-scale data collection. However, pregnancy status is not explicitly recorded in a dedicated EHR section or in any other SIGEHOS module, a limitation which requires this and other key information to be inferred from structured and unstructured fields across EHR.
To achieve this goal, we developed an algorithm using open-source tools, primarily R. The algorithm applies data mining techniques, rule-based processing, and regular expressions to extract and classify pregnancy-related information from SIGEHOS. It infers key attributes such as pregnancy start and finish dates, estimated delivery date, and gestational age at the first visit.
Our presentation will be particularly useful for researchers and professionals working with large-scale population data who need to develop projects to track time-evolving processes in cases where information is non structured, dispersed across multiple fields and difficult to locate.