ID-5184 Wonca Abstracts supplement A-K 13-10-23 - Flipbook - Page 41
WONCA 2023 Supplement 1: WONCA 2023 abstracts (A–K)
A
B
C
International Classification of Primary Care (ICPC) and
natural language processing: Automating medical coding
Dr Vinicius Anjos De Almeida1, Luis Fernandez Lopez1, Christopher Paul Ley2
D
E
F
G
H
I
J
K
1
Department of Legal Medicine, Bioethics, Occupational Medicine, and Physical Medicine and
Rehabilitation, Faculty of Medicine, University of São Paulo, 2Center for Mathematical Modeling,
University of Chile
International classifications are a crucial tool for various aspects of healthcare, including general
medical practice, population health analysis and research. However, coding health records can prove
to be a tedious and inaccurate task, given the complexity of the classifications, limited time available,
need for previous training and dependence on professional experience. Natural language processing
encompasses a range of technologies that provide insight into text data and can be utilised for various
purposes, including text data classification.
The aim of this research is twofold: (1) to develop a neural network that can analyse primary care
medical records and classify the reasons for patient visits using the International Classification of Primary
Care (ICPC), second edition; and (2) to compare the results of automated coding with those coded
by family physicians with varying levels of experience in ICPC for benchmarking purposes. The main
hypothesis is that a natural language processing-based neural network will perform comparably to a
general practitioner with some experience in ICPC coding.
This technology has the potential to provide value by reducing distractions and saving healthcare
providers time and cognitive effort for more patient-centred activities. Furthermore, it can generate
consistent data that will contribute to future research and general practice. The primary objectives of this
presentation are to discuss the problems that this technology can solve, present preliminary research
results and highlight expected limitations and challenges.
39
