The applications of machine learning in electronic health records

1. Electronic Health Records
and Machine Learning
Eman Abdelrazik
Bioinformatics Research Assistant, Center of Informatics Science, Nile University
H3ABioNet Teaching Assistant

2. Grey Areas in Clinical Medicine

3. Grey Areas in Clinical Medicine

4. Real World Data (RWD)
● RWD are information reported and collected from real-world
medical settings that may show the effectiveness and safety
of a medicine for an approved indication in patients.
● These data may be collected in one country or collected from
all over the world (heterogeneous patient population in real-
world settings).

6. Sources of RWD

7. HOW ARE RWD USED?
● RWD analyses generate insights about the effectiveness and safety
profile of a medicine.
● RWD studies can help explore additional research questions,
complement clinical trial findings, and fill gaps related to how a
medicine is used in real-world medical settings.
● RWD may also be used to inform future investigational plans to
increase efficiency and reduce cost.

8. Real world evidence (RWE)
● Obtained from real world data (RWD), which are observational data
obtained outside the context of randomized controlled trials (RCTs) and
generated during routine clinical practice.
○ In order to assess patient outcomes and to ensure that patients get
treatment that is right for them, real world data needs to be utilized.
● RWE is generated by analyzing data which is stored in electronic health
records (EHR), medical claims or billing activities databases, registries,
patient-generated data, mobile devices, etc.
○ It may be derived from retrospective or prospective observational studies and
observational registries

9. The journey from Data to Real-world evidence (RWE)

10. Use cases for RWE in the drug development cycle

11. Electronic Health Record (EHR)
● EHRs are systematic collections of longitudinal health information of patient
and population electronically-stored health information in a digital format.
● These records can be shared across different health care settings.
● Records are shared through network-connected, enterprise-wide information
systems or other information networks and exchanges.

12. EHR vs. Paper Charts

13. EHR
There are two types of data contained in
patient EHRs:
1) Structured Data
● fields that contain data using existing
lexicons, such as demographics,
diagnosis, laboratory tests,
medications, and procedures.
2) Unstructured Data
● free text documents such as clinical
notes from physicians and nurses.
EHR Components

14. EHR is the GPS of Health Care!
EHR & Pathology Reports!

15. Diagnosis

16. Diagnosis : why it is hard to?
● Phenotype is a spectrum
○ Physical characteristics (signs; tumor markers, grade,
tumor mass)
○ Diseases (stroke, lung cancer, Rheumatoid Arthritis “RA” )
● A correct diagnosis refers to the presence, absence and
severity grade of pathology and determines decision making in
prognosis, risk assessment and treatment.
● Deciding cut points for who has or does not have disease.

17. Misdiagnosis!!
● Misdiagnosis in clinic
○ Major implications for patient
● Misclassification in research
○ Reduces ability to identify relationships at population level
■ Effectiveness of new therapies
● Side effects
■ Risk factors
Algorithms are only as good as training set and reviewers!

18. Artificial Intelligence (AI): Machine Learning
● Artificial Intelligence
○ Intelligence demonstrated by machines.
■ Contrast to human intelligence.
● Machine Learning (ML)
○ Require data to train
■ Develops mathematical models to predict the data of
interest
■ ML abstracts rules from the data, similar to what a
physician might experience during his residency

19. Machine Learning Models
● Choice of a distinct model for a given problem is determined by:
1) Features of the data
● Fewer unique data points indicate that classical techniques:
○ linear regression, or decision-tree methods
● which segment data sets into regions according to fixed rules
2) Data types
● a collection of images (histopathology data, radiography)
● a time-series signal (longitudinal studies)
● general descriptive data (demographics & narrative data)

20. Machine Learning Models
Pattern Recognition Vs. annotated data

21. Example:

22. Develop ML model

23. Develop ML model

24. Develop ML model

25. ML Model to identify who has what disease?!

26. ML applications in clinical medicine

27. Machine learning for medical ultrasound
Suffers from inter-
and intra-observer
variability + image
quality
❏ Region of Interest (ROI);
liver fibrosis staging
Credit: Brattain, L.J., Telfer, B.A., Dhyani, M. et al. Abdom Radiol (2018) 43: 786.

28. Machine learning for medical ultrasound
Credit: Brattain, L.J., Telfer, B.A., Dhyani, M. et al. Abdom Radiol (2018) 43: 786.

30. Proposed framework for
machine learning-based
intelligent diagnostic
assistant

32. Publications
relevant to ML
methods used for
cancer
susceptibility
prediction.

33. Publications relevant
to ML methods used
for cancer
recurrence
prediction.

34. Publications relevant
to ML methods used
for cancer survival
prediction.

35. Future challenges
● Selecting training data set
● Setting gold standard
● Adapting new diagnosis and new treatments
● Responsible party for the final decision for diagnosis and
treatment

36. Summary

37. References
1. Kourou, Konstantina, et al. “Machine Learning Applications in Cancer
Prognosis and Prediction.” Computational and Structural Biotechnology
Journal, Elsevier. 2014; 13: 8-17.
2. L.J., Telfer, B.A., Dhyani, M. et al. Abdom Radiol. 2018; 43: 786.
3. Nichols JA, Herbert Chan HW, Baker MAB. Machine learning: applications of
artificial intelligence to imaging and diagnosis. Biophys Rev. Biophysical
Reviews. 2019;11:111–8.
4. Wang F, Preininger A. AI in Health: State of the Art, Challenges, and Future
Directions. Yearb Med Inform. 2019;28:16–26.