Augmented intelligence facilitates concept mapping across different electronic health records.

Area of Science:

• Medical Informatics
• Artificial Intelligence
• Data Science

Background:

• Secondary use of electronic healthcare records (EHR) data is growing with AI.
• Inconsistent naming conventions across EHR systems hinder scalable model development.
• Ontology alignment is crucial but labor-intensive, necessitating automated solutions.

Purpose of the Study:

• To propose an augmented intelligence approach for automated ontology alignment.
• To predict correct medical concepts from raw EHR parameter names.
• To facilitate scalable model development and clinical implementation.

Main Methods:

• Trained machine learning models using manually mapped EHR parameter names from a multicenter Dutch ICU data warehouse.
• Utilized Natural Language Toolkit (NLTK) with WordNet synonyms and TF-IDF for preprocessing parameter names.
• Employed linear classifiers with stochastic gradient descent and fine-tuned predictions using data distribution similarity and skewness.

Main Results:

• Achieved an initial top 1 precision of 0.744, recall of 0.771, and F1-score of 0.737.
• Leave-one-hospital-out analysis showed top 5 recall increased to 0.905.
• Reducing training data to relevant parameters yielded top 1 recall of 0.811 and top 5 recall of 0.914.

Conclusions:

• Augmented intelligence offers a promising solution for concept mapping in EHR data.
• The proposed method robustly maps data across domains, aiding diverse data source integration.
• Manual validation remains essential due to imperfect recall rates.