The document discusses the role of artificial intelligence (AI) in diabetes management, exploring its applications such as insulin bolus calculation and case-based reasoning models. It highlights the evolving clinical paradigm towards data-driven precision medicine and the importance of proper training for healthcare professionals in using AI technologies. The potential for AI to replace physicians is questioned, suggesting a complementary role rather than full autonomy in medical practice.

1. ARTIFICIAL INTELLIGENCE
AND DIABETES
Iris Thiele Isip Tan MD, MSc
Professor 3, UP College of Medicine
Chief, UP Medical Informatics Unit
Director, UP Manila Interactive Learning Center

2. NOTHING TO DISCLOSE
I give consent for the audience to tweet this talk
and give me feedback (@endocrine_witch).
Feel free take pictures of my slides (though the
deck will be at www.slideshare.net/isiptan).

3. What is AI?
Use of AI in diabetes
Will AI replace
physicians?

4. ARTIFICIAL
INTELLIGENCE
Allow machines to sense,
reason, act and adapt like
humans do - or in ways
beyond our abilities
https://newsroom.intel.com/news/many-ways-define-artificial-intelligence/#gs.0msuwc

5. More computing power
More data
Better algorithms
Broad investment
AI is not new …
https://newsroom.intel.com/news/many-ways-define-artificial-intelligence/#gs.0msuwc

6. https://newsroom.intel.com/news/many-ways-define-artificial-intelligence/#gs.0msuwc

7. https://newsroom.intel.com/news/many-ways-define-artificial-intelligence/#gs.0msuwc

8. https://newsroom.intel.com/news/many-ways-define-artificial-intelligence/#gs.0msuwc

9. What is AI?
Use of AI in diabetes
Will AI replace
physicians?

10. How to calculate insulin
bolus in type 1 diabetes

11. https://www.uni-trier.de/index.php?id=44502&L=2
CASE-BASED REASONING
Problem-solving paradigm in
artificial intelligence
Similar problems tend to
have similar solutions
Case base: collection of
memorized chunks of
experience (cases)
Adaptive systems: new problem
solving experience is retained and
outdated experience is removed

12. Brown D et al. Temporal case-based reasoning for type 1 diabetes mellitus bolus
insulin decision support. Artificial Intelligence in Medicine 2018;85:28-42.
Case-based reasoning model
for T1DM bolus insulin advice

13. Brown D et al. Temporal case-based reasoning for type 1 diabetes mellitus bolus
insulin decision support. Artificial Intelligence in Medicine 2018;85:28-42.
CASE FEATURES
Determine which parameters are
required by bolus calculators
Carbohydrate intake
Pre-meal blood glucose
Target blood glucose level
Insulin-on-board
Exercise
Time
Insulin Sensitivity Factor (ISF)
Carbohydrate-to-Insulin Ratio (CIR)

14. https://mysugr.com/mysugr-bolus-calculator/
Insulin on board
Blood glucose
Carbs
Previous injections
Carbs/insulin ratio
Insulin correction
factor
Blood glucose
target
MySugr Bolus
Calculator

15. RETRIEVE
Use the date/time of event to infer
ISF and CIR
Factors in preceding bolus doses
REUSE
Adaptation rule which resolves
differences between insulin-on-
board (IOB) in the problem and
retrieved case(s)
Brown D et al. Temporal case-based reasoning for type 1 diabetes mellitus bolus
insulin decision support. Artificial Intelligence in Medicine 2018;85:28-42.

16. Brown D et al. Temporal case-based reasoning for type 1 diabetes mellitus bolus
insulin decision support. Artificial Intelligence in Medicine 2018;85:28-42.
REUSE step: Average bolus prediction of
retrieved cases then adapt
Equation for averaging bolus prediction of retrieved cases
k = number of retrieved cases
in = bolus solution provided by a retrieved case

17. Brown D et al. Temporal case-based reasoning for type 1 diabetes mellitus bolus
insulin decision support. Artificial Intelligence in Medicine 2018;85:28-42.
REUSE step: Average bolus prediction of
retrieved cases then adapt
Equations for adapting bolus suggestion

18. Brown D et al. Temporal case-based reasoning for type 1 diabetes mellitus bolus
insulin decision support. Artificial Intelligence in Medicine 2018;85:28-42.
REVISE
If postprandial BG is equal
or close to target BG then
recommendation is optimal
and not revised

19. Brown D et al. Temporal case-based reasoning for type 1 diabetes mellitus bolus
insulin decision support. Artificial Intelligence in Medicine 2018;85:28-42.
Advanced Bolus Calculator for Diabetes (ABC4D)
CBR approach: tuning of ISF
and CIR for a small set of meal
scenarios
ISF and CIR from the most
similar case used in a standard
bolus calculator to suggest a
bolus dose
No temporal approach

20. Brown D et al. Temporal case-based reasoning for type 1 diabetes mellitus bolus
insulin decision support. Artificial Intelligence in Medicine 2018;85:28-42.
Focus on helping patient directly
(instead of aiding the clinician)
RETAINS all
successful cases
Derives bolus
suggestion from
similar cases

21. Brown D et al. Temporal case-based reasoning for type 1 diabetes mellitus bolus insulin
decision support. Artificial Intelligence in Medicine 2018;85:28-42.
CBR method can be
adopted by insulin
pumps, blood glucose
monitors, PCs and as
a web service

22. CBR service in the cloud opens possibility
of case sharing between subjects
Brown D et al. Temporal case-based reasoning for type 1 diabetes mellitus bolus insulin
decision support. Artificial Intelligence in Medicine 2018;85:28-42.

23. OBJECTIVE
Apply gradient forest analysis to identify subgroups of
ACCORD participants with increased/decreased risk of all-
cause mortality attributable to intensive therapy

24. Action to Control Cardiovascular Risk in Diabetes
(ACCORD) Trial halted due to increase in all-cause
mortality in intensive therapy arm
Median A1c
Intensive: 6.4%
Standard: 7.5%
ACCORD study group. N Engl J Med 2008; 358:2545-2559

25. Which subgroup in ACCORD was
most likely to benefit or have increased
mortality?
Heterogeneous
treatment effects

26. Basu et al. Diabetes Care 2017
doi.org/10.2337/dc17-2252

27. Basu et al. Diabetes Care 2017
doi.org/10.2337/dc17-2252
Summary risk stratification decision tree
Hemoglobin Glycosylation Index
(Observed - Predicted A1c)
Predicted A1c
0.009 x FPG [mg/dl] + 6.8

28. Basu et al. Diabetes Care 2017 doi.org/10.2337/dc17-2252
Survival curves for all-cause mortality among
subsets identified by each subgroup in the decision tree
A: HGI <0.44, BMI <30 kg/m2 and age <61 years
B: HGI <0.44, BMI <30 kg/m2 and age >61 years

29. Basu et al. Diabetes Care 2017 doi.org/10.2337/dc17-2252
Survival curves for all-cause mortality among
subsets identified by each subgroup in the decision tree
C: HGI <0.44 and BMI >30 kg/m2
D: HGI >0.44

30. What is AI?
Use of AI in diabetes
Will AI replace
physicians?

31. Machine learning represents a
shifting clinical paradigm from rigidly
defined management strategies to
data-driven precision
medicine.
Buch et al. Diabet Med 2018;35:495-7.

32. Buch et al. Diabet Med 2018;35:495-7.
Clinical guidelines will be
delivered through apps
rather than static documents.

33. Buch et al. Diabet Med 2018;35:495-7.
Healthcare professionals will require adequate
training to operate AI-based solutions
Appreciate the limitations of technology
Over-reliance on AI risks de-skilling the profession

34. “The pinnacle of AI is being fully
autonomous. But I don’t think
that will happen in medicine;
AI will always need human
backup.
- Eric Topol MD

35. A robot may not injure a human
being or, through inaction, allow a
human being to come to harm.
A robot must obey orders given it
by human beings except where
such orders would conflict with
the First Law.
A robot must protect its own
existence as long as such
protection does not conflict with
the First or Second Law.
Isaac Asimov’s Three Laws of Robotics