A carregar...
Publicação
Creation of Synthetic Patient Data for Health Care Research
| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 5.57 MB | Adobe PDF |
Autores
Orientador(es)
Resumo(s)
In recent years, with the digitalization of the healthcare field, the use of Electronic Health
Records (EHRs) has significantly increased. Their ability to store a wide range of patient
information, combined with ease of access and the capacity for smooth data sharing
among hospitals and medical providers, has made them an essential tool in modern
healthcare. While EHRs are not yet perfect, the data they contain is highly valuable
for professionals and researchers in the healthcare field. This data has the potential
to drive advancements and breakthroughs in medical research. However, since EHRs
contain personal and sensitive information about patients, their content must be protected,
leading to challenges in protecting privacy while making it available for research purposes.
Synthetic data, a field that has also been rapidly growing in recent years, primarily driven
by advancements in machine learning, is a potential solution for this problem. By
generating synthetic datasets that replicate real world data, it is possible to preserve the
characteristics and quality of the original dataset while addressing privacy concerns. In
order to protect patient privacy, synthetic data must not resemble the original dataset or
allow re-identification.
The goal of this work is to explore the potential of synthetic data generation when
applied to a dataset of clinical records referred to as High Users (HU). This dataset
contains both demographic information about patients and a four year history of medical
appointments, emergency episodes and hospitalizations. Using the High Users dataset,
the objective is to create synthetic data that replicates the original dataset’s characteristics
and quality through machine learning techniques, while ensuring privacy, meaning that
the synthetic data should not allow the re-establishment of the original dataset. The study
analyzes multiple approaches for generating synthetic data that retain the properties of
real world data. The experiments conducted determine whether the selected synthetic
data generation techniques are suitable for the HU dataset, but the results were largely
negative, with standard evaluation metrics also proving disappointing in assessing effec-
tiveness in preserving properties while ensuring privacy.
Descrição
Palavras-chave
Synthetic Data Healthcare Privacy Machine Learning Electronic Health Records