Using Unsupervised Learning to Identify Clinical Subtypes of Alzheimer's Disease in Electronic Health Records

Nonie Alexander, Daniel C Alexander, Frederik Barkhof, Spiros Denaxas

Research output: Contribution to journalArticleAcademicpeer-review

8 Citations (Scopus)


Identifying subtypes of Alzheimer's Disease (AD) can lead towards the creation of personalized interventions and potentially improve outcomes. In this study, we use UK primary care electronic health records (EHR) from the CALIBER resource to identify and characterize clinically-meaningful clusters patients using unsupervised learning approaches of MCA and K-means. We discovered and characterized five clusters with different profiles (mental health, non-typical AD, typical AD, CVD and men with cancer). The mental health cluster had faster rate of progression than all the other clusters making it a target for future research and intervention. Our results demonstrate that unsupervised learning approaches can be utilized on EHR to identify subtypes of heterogeneous conditions.

Original languageEnglish
Pages (from-to)499-503
Number of pages5
JournalStudies in health technology and informatics
Publication statusPublished - 16 Jun 2020


  • Alzheimer's disease
  • Electronic health records
  • Machine learning
  • Phenotyping

Cite this