A Systematic Review of Unsupervised Anomaly Detection for Early Disease Identification in Electronic Health Records

Анотація

Electronic Health Records have become an invaluable source of patient data, holding the
potential to revolutionize healthcare through early disease identification.Unsupervised
anomaly detection methods offer a promising avenue to uncover hidden patterns and ir-
regularities within these complex datasets.This paper presents a systematic review of
unsupervised anomaly detection techniques applied to early disease identification within
EHRs. Our comprehensive analysis encompasses diverse methodologies, geographical dis-
tributions, subject areas, and temporal trends. The study design and methodologies vary
widely, showcasing the need for cross-disciplinary collaboration between data scientists and
healthcare professionals. The global significance of this research is evident in the geographic
distribution of studies, reflecting a concerted effort to develop solutions applicable across
diverse healthcare systems. The versatility of unsupervised methods is demonstrated across
a spectrum of subject areas, including clinical diagnoses and administrative processes. Our
temporal analysis reveals an evolving landscape characterized by advancements in ma-
chine learning techniques and increasing access to large-scale datasets.This systematic
review highlights the significance of combining technical expertise with domain knowledge
to address the challenges of early disease identification within EHRs. By examining global
trends, subject areas, and temporal patterns, we provide insights into the trajectory of
research efforts in this critical field. This work informs future directions for researchers and
practitioners seeking to leverage unsupervised anomaly detection for early disease identifi-
cation, ultimately contributing to improved patient care and enhanced healthcare systems.
Key words: Anomaly, Unsupervised, Disease, SLR, Health Record, EHR, Machine Learn-
ing, Deep Learning.