Welcome to Health Care Data Analytics, Data Analytics in Clinical Settings.
This is Lecture b, the component, Health Care Data Analytics,
covers the topic of health care data analytics which applies the use of data,
statistical and quantitative analysis, and explanatory and
predictive models to drive decisions and actions in health care.
The learning objectives for data analytics and clinical settings are to describe
the current state of analytics in clinical settings, identify key tools and
approaches to improve analytics capabilities in clinical settings,
describe different governance and
operations strategies in analytics in clinical settings.
Discuss value based payment systems and
the role of data analytics in achieving their potential, and
analyze data used in population management and value-based care systems.
In this lecture, we'll provide some examples of how to address complex issues
related to data and measurement for data analytics for clinical settings.
First, we'll describe a case where data is diverse and fragmented, and
how data needs to be integrated to perform effective analysis.
Then, we'll describe the importance of using detailed measure specifications and
standard measures to ensure that your analytics are predicting what is important
and related to broader health and health care goals.
This messy diagram is intended to show the information flow, as Ms.
Viera, a person with five complex chronic conditions,
interacts with the health care system over the course of one year.
If we were to estimate outcomes for, we will find out that she 90 times the rate
of hospitalization as someone with no chronic conditions.
That she takes on average 14 medications,
and that she'll see 12 specialists during that one year.
How can analytics help us measure her risks?
And whether her care is of high quality and safe.
Where does the system break down when we're trying to measure
the quality of her care and predict her needs?
First, the more interaction with the health care system she has,
the more likely her data is to be fragmented.
Often, when someone is hospitalized,
the hospital's information system is not the same as the out-patient system.
So, the information about her previous care is not available to the hospital.
And if the hospital changes to drugs,
then the primary care setting may not get a notification about that.
In fact, the primary care setting may never know she's in the hospital.
The only complete source of data about her healthcare utilization maybe in the bills
that go to health insurance companies or the federal government for medicare.
These bills are stored in so called claims databases and
are not integrated back to EHRs in most cases.
This can cause serious problems as we try to
analyze data to understand Miss Viera's care and health outcomes.
Let's explore this data measurement issue in a different way, by data sets.
EHR data in our example is stored by the primary care providers.
It has a longitudinal record of all the data collected at the clinic and
associated laboratory.
This information can be useful in analysis when on going patient factors are needed.
For instance, when we are predicting risk of re-admissions,
comorbidities are helpful to understand the patient's health is complicated by
multiple other diseases.
The claims data, which are the set of bills for healthcare,
may contain other equally important information.
Information about ED visits, the acuity of the hospitalization, and
final diagnosis may only be available through these claims since the hospital
may be on a different system.
Finally, other data usually not stored in either the EHR or
the claims data may be helpful such as social, behavioral and environmental data.
Environmental in this case, means information about where people live.
The availability of healthy food for
instance may be low in certain urban settings.
The lack of integration of this data leads to poor prediction in health care
settings.
And Miss Viera may be re-admitted because the system did not adapt.
It is possible however to integrate data from these sources or
at least exchange it as in health information exchange programs.
To truly analyze risk, the data can be integrated into an analytic data warehouse
where all relevant and available data can be processed then, a more accurate and
complete risk score can be calculated, and programs like transitional care
coaching or the care transitions intervention can be targeted to her needs.
These programs can reduce re-admission risk by working with patients
more closely before and upon discharge, this is an ideal version.
In the real world, there are still many issues to be dealt with.
For instance, claims data often takes a long time to be available.
Sometimes more than a year.
Quicker access to the information is needed in many cases and so
communities may exchange information electronically.
Similarly, data that means the same thing is not stored in the same way across
settings.
We must strive for semantic interoperability,
where we can recognize when things mean the same thing.
Finally, we have the issue that, even electronically, most data is stored
as unstructured notes or reports, which makes it difficult to predict issues.
Another kind of data that is increasingly important but
is not integrated is genomic data.
We have rapidly increased our ability to process genomic information but we can't
exchange it easily and it's not integrated into most analytic or health care systems.
The Implementing Genomics in Practice, or IGNITE Network, was funded by
the National Institutes of Health to improve the integration of this data.
Shown are the more than a dozen sites involved in the Network.
Two examples are given from the Network.
Mt. Sinai in New York is looking at why
African-Americans with hypertension have more kidney failure or
end stage renal disease.
They found that particular alleles from the APOL one locus vary by race and
may help predict risk and provide insight into different approaches to treatments.
Similarly, Indiana is using what we know about how people
process drugs based on their genetics.
Known as pharmacogenomics, to help people pick both the right drugs, and
the right doses to avoid costly complications.
The IGNITE network wants to see if providing this information
reduces overall health care costs.
Data about where a person lives and how they live can be more important than
the health care they receive in many cases.
For instance, your day to day behaviors,
such as how much you walk or what you eat, may effect your risk
of hospitalization from heart failure more than medications of health care.
Similarly, the walkability of your neighborhood and availability of healthy
foods may also affect your health more than particular treatments or procedures.
We have not previously had good ways to integrate this data into health and
health care, but that's changing as people recognize this data's value.
Similarly, the outcomes they perceive, such as the amount of pain they feel or
the things they can do, are more important than many lab tests or
other study results.
There are a number of new standards and methods to record and
integrate information about patient outcomes and environmental data.
For instance, patient reported outcomes are stored and
can be retrieved from a measurement information system known as PROMIS.
This system provides structured, validated ways for
patients to report on outcomes that may be important to them,
such as their depression symptoms or their overall function for daily activities.
Now, let's think about measures of healthcare quality and
how they help us to better analyze the care we deliver and the help it provides.
Let's think about a particular person, Mr. Smythe.
At 68 years old, he's out gardening in 2001 when suddenly he has chest pain.
He is rushed to the hospital where the EKG shown is demonstrating ST wave
elevations that definitely show something going on with his heart.
Within echocardiogram that shows his heart is not pumping well and
with other ongoing symptoms.
The medical team notice having an active heat attack, many people assume that he'll
get the exactly the treatment he needs in the vast majority of cases.
What evidence do we have about what are effective treatments for heart attack?
And what about his congestive heart failure, where his heart can no longer
pump efficiently enough to meet the needs of his body.
There is very good evidence from a variety of clinical trials
that he needs medications to limit the damage from the heart attack and
improve the function of his heart.
The medications are shown here, along with their related studies and
reduction in deaths from affectively using them.
Other treatments are important too, such as cardiac rehabilitation.
With this number of life saving treatments available,
we anticipate he'll get the great care he needs.
But to be sure, we must measure how well we did.
Now pause, and think to yourself, how often would a patient like Mr.
Smthe receive these treatments in 2001?