A Lightweight API-Based Approach for Building Flexible Clinical NLP Systems.

Area of Science:

• Biomedical Informatics
• Natural Language Processing
• Health Data Science

Background:

• Clinical natural language processing (NLP) is crucial for secondary data use but existing systems lack scalability and require extensive customization.
• Current NLP solutions are often tied to specific datasets and clinical settings, posing challenges for broader application and resource-intensive maintenance.

Purpose of the Study:

• To present a lightweight, composable, and extensible architecture for developing clinical NLP systems with limited resources.
• To demonstrate the feasibility of this approach through a web-based prototype for clinical concept extraction.

Main Methods:

• Employed a design science research approach to develop a novel, lightweight clinical NLP architecture.
• Integrated NLP as external components accessed via web APIs, orchestrated in a pipeline.
• Developed and evaluated a web-based prototype using six NLP APIs on three distinct clinical datasets.

Main Results:

• Achieved high F1 scores (0.861, 0.724, 0.805) on three clinical datasets, comparable to benchmarks.
• Observed a lower F1 score (0.373) on a small test dataset, likely due to its size.
• The prototype demonstrated the potential for effective clinical NLP system development with resource constraints.

Conclusions:

• The proposed lightweight architecture offers a scalable and configurable solution for clinical NLP.
• This approach significantly reduces the resources and time needed for developing and maintaining clinical NLP systems.
• The findings suggest a promising pathway for democratizing clinical NLP applications across diverse settings.