Artificial intelligence and machine learning techniques like logistic regression, collaborative filtering, and deep learning can be applied in healthcare for tasks like predicting patient outcomes, automating care coordination, and assisting disease diagnosis. However, challenges remain around limited data availability in healthcare, the complexity of medical data, and the lack of data scientists currently working in the field. With more healthcare data and interdisciplinary collaboration between data scientists and medical experts, artificial intelligence has strong potential to positively impact areas like early detection of sepsis and reducing insurance claim rejections.
26. Extract and
Interpret Data
Extracting value from data
Extraction
Secure file transfers protocol
Anonymizing P.H.I.
Gigabytes, Terabytes, &
Petabytes
The bigger the better!
Interpretation
Categorical variables:
‘male’,‘yellow’,‘jaundice’,‘IVIg’
Numerical variables:
82, 1.325, 1.024 E^31
Sparsity:
Null, NaN, 0
How many variables?
Data has Dimensions!
27. 80%
The proportion of our time
that’s spent cleaning the
data.
Before:
(Nonsense)
After:
(Insight!)
Cleaning and
Munging
the Data
Extract and
Interpret Data
Extracting value from data
28. Statistical inference:
Deducing properties of an underlying
probability distribution by analysis of data.
Before:
After:
Cleaning and
Munging
the Data
Extract and
Interpret Data
Multivariate Inferential
Statistics
𝛼
Parameters
Extracting value from data
29. Cleaning and
Munging
the Data
Extract and
Interpret Data
Multivariate Inferential
Statistics
Machine Learning
Algorithm
s
Machine learning:
The latest field of computer science.
Giving computers the ability to learn
without being explicitly programmed
How does it work?
Algorithms iteratively and implicitly
learn the data.
The Result:
A model is developed.
Insights made go deeper, beyond
basic descriptive analysis
Extracting value from data
43. Logistic regression: Unsimplified
Sex Weight Ethnic City BP Smoker Diabetes
F 142 H M 130/80 Y Y
M 178 A F 170/90 N N
F 203 C M 130/90 N Y
M 187 A P 170/90 Y Y
F 162 A P 170/90 Y N
F 120 H M 80/50 N Y
M 263 A F 80/50 Y Y
M 207 C P 130/80 N
53. Collaborative filtering: Unsimplified
Sex Weight Ethnic City BP Smoker COPD Diabetes Heart Disease GI
F 142 H M 130/80 Y 2 4 3
M 178 A F 170/90 N 5 1
F 203 C M 130/90 N 1 3 2
M 187 A P 170/90 Y 3 2
F 162 A P 170/90 Y 3 4 1
F 120 H M 80/50 N 4 2
M 263 A F 80/50 Y 2 5 2 4
M 207 C P 130/80 N 5 4 3
57. Identify deductibles, copays, and
patient’s share of medical costs,
even when there are a wide
variety of insurance coverage
options
Insurance coverage
62. Deep learning
Deep learning is inspired by how our brain
works and is based on learned, rather than
programmed, parameters which can be in the
order of few thousands to 100+ million
80. Listening to the data is important… but so is
experience and intuition. After all, what is intuition
at its best but large amounts of data of all kinds
filtered through a human brain rather than a math
model? – Steve Lohr
89. If you are reading this, there is a
99% chance you are not looking at
your phone right now.
some guy
Editor's Notes
What if could quickly generate a list of evidence based treatment options tailored for each patient, and provide them the best care while optimizing costs? Instead of having to manually search and evaluate the lastet data for hours.
What if you could know the costs and insurance coverage of every patient before they walk in the door? Instead of spending hours coordinating with carriers and giving patients surprise bills.
What if you could coordinate the care and individual support that each of your patients need by harnessing the data you already have about them? Instead of having to manually create and assign tasks to nurses, doctors, home health aids, social services, and transportation companies?
When will the path from sickness to health become clear?
How do you bring together the disparate data points that sit in your organization together to help your doctors, caregivers, administrators, and most importantly the patients. [use the data pictures]
Spreadsheets
EHR data
Log files
Unstructured corpus of text
You are each uniquely positioned to remove barriers and increase access to care, driven by your human expertise and the tools of data science.
Data science can allow you to turn what you’ve done to what you do. It helps you to create value from data by identifying hidden, valuable and actionable insights. Allowing you to answer questions before they’re asked. Now some of you have probably heard this before
And you hear words like AI
Big Data
And start knowing what they mean for you and your business so you can start taking your business from the present of healthcare to the future.
Questionable cause logical fallacy!
Computers have not achieved the *singularity*. Computers are not self-aware
Computers cannot predict with 100% accuracy what tomorrow holds.
Computers cannot make innovative, creative solutions
Computers struggle with the underlying context.
Data Science cannot solve ‘complex systems’ problems (i.e. give solution to economic depression)
Data Science cannot solve ill-defined problems
Automate care coordination by assigning patient-specific tasks
Is apnoea predictive of hypertension, allowing for age, sex and body mass index?
Determine potential efficacy of treatment
optimize scheduling
Limited Outcome Variables
Logistic regression could not be used to determine how high an influenza patient's fever will rise, because the scale of measurement -- temperature -- is continuous. Researchers could attempt to convert the measurement of temperature into discrete categories like "high fever" or "low fever," but doing so would sacrifice the precision of the data set.
Reduce the costs of having to adjudicate claims and accelerate approval and speed up the revenue cycle
Efficiently allocate resources to ensure patients medication compliance, consider drug costs, increase medication compliance, and reduce repeat visits or rehospitilization
Certify that procedures are covered and that the correct revenue is being collected and prevent sticker shock for the patient
Quickly generate a list of ranked potential and non-recommended treatment options for oncologists
Image recognition supports the dermatologist’s efforts; as well as radiologists, oncologists, cardiologists
-Reduced nurse staff and manual screening by close to 70 percent; being used at hospitals across the southeast
Prioritize and streamline claim submissions; change processes to prevent rejections; being used at hospitals all over the U.S.
field of medical study that aims to extract large amount of quantitative features from medical images, uncover disease characteristics that fail to be appreciated by the naked eye. Prediction of clinical outcomes; Prediction risk of distant metastasis; it’s being used at Memorial Sloan Kettering
field of medical study that aims to extract large amount of quantitative features from medical images, uncover disease characteristics that fail to be appreciated by the naked eye. Prediction of clinical outcomes; Prediction risk of distant metastasis
field of medical study that aims to extract large amount of quantitative features from medical images, uncover disease characteristics that fail to be appreciated by the naked eye. Prediction of clinical outcomes; Prediction risk of distant metastasis
field of medical study that aims to extract large amount of quantitative features from medical images, uncover disease characteristics that fail to be appreciated by the naked eye. Prediction of clinical outcomes; Prediction risk of distant metastasis