Sign in Subscribe
Machine learning

Diabetes Research Reveals Insights into Unstructured Electronic Health Records and Machine Learning

Strategic Insights

— 1 min read

Researchers at the Hospital Clinic of Barcelona in Spain have conducted a study on the potential of unstructured electronic health records (EHRs) data to describe the prevalence and clinical spectrum of diabetes mellitus (DM) in hospitals. The study used natural language processing (NLP) and machine learning (ML) to analyze EHRs from eight Spanish hospitals between 2013 and 2018. The findings have provided valuable insights into the clinical profile and prevalence of people with diabetes attended in the hospital setting.

Key Takeaways:

  • The study analyzed 56,181,954 EHRs of 2,582,778 individuals and identified 638,730 with diabetes: 75.4% with unregistered type DM (UrDM), 21.3% with type 2 DM (T2DM), and 3.3% with type 1 DM (T1DM).
  • The machine learning model reclassified 93.5% of T2DM and 6.5% of T1DM cases, highlighting the potential of ML in accurately identifying diabetes types.
  • Major comorbidities included hypertension, dyslipidemia, chronic kidney disease (CKD), ischemic heart disease, and chronic heart failure (CHF), with CKD and CHF being the most frequent complications for T1DM/T2DM.
  • The study concluded that NLP and ML for profiling DM using EHRs unstructured data are helpful, but additional data and better EHR documentation are crucial.
  • The prevalence of T1DM/T2DM was 2.6%/38.4% in the hospital setting, with over 50% of relevant variables like anthropometrics, laboratory values, and treatments missing.
  • The study highlights the importance of using unstructured healthcare data and NLP to improve clinical profile and prevalence estimates of diabetics.

Statistics:

  • 638,730 individuals identified with diabetes out of 56,181,954 EHRs.
  • 75.4% of individuals with diabetes had unregistered type DM (UrDM).
  • 21.3% of individuals with diabetes had type 2 DM (T2DM).
  • 3.3% of individuals with diabetes had type 1 DM (T1DM).
  • 93.5% of T2DM cases were reclassified by the machine learning model.
  • 6.5% of T1DM cases were reclassified by the machine learning model.
  • 50% of relevant variables like anthropometrics, laboratory values, and treatments were missing for T1DM/T2DM.

Sources:

  • "Characterizing the Clinical Profile and Prevalence of People With Diabetes Attended In the Hospital Setting By Using Unstructured Healthcare Data and Natural Language Processing: the Diabetic@ Study" published in Diabetes Research and Clinical Practice, 2025;226.
  • Diabetes Research and Clinical Practice, Elsevier Ireland Ltd, Elsevier House, Brookvale Plaza, East Park Shannon, Co, Clare, 00000, Ireland.

Read more