A new model for predicting disease likelihood at the time of hospital admission can
help healthcare providers generate a better estimate of a patient’s hospital stay
and more effectively allocate scarce hospital resources. It can also provide insurance
companies with a more realistic estimate of the eventual length of hospital stay for
the patient, ensuring appropriate treatment is started earlier and resulting in better
overall health outcomes.
The research, entitled “When will I get out of the hospital? Modeling Length of Stay
using Comorbidity Networks,” was co-authored by Pankush Kalgotra, Assistant Professor,
Business Analytics in the Department of Systems and Technology at Auburn University’s
Harbert College of Business, and Ramesh Sharda, Vice Dean, Watson Graduate School
of Management and Regents Professor of Management Science and Information Systems
in the Spears School of Business at Oklahoma State University. The peer-reviewed study
was recently accepted for publication in the Journal of Management Information Systems,
a top journal in the information systems discipline.
The Challenge
A key determinant of length of stay (LOS) when a patient enters a hospital is the
level of information available to admissions staff at the time of intake – it defines
the patient’s initial treatment plan as well as any insurance coverage originally
authorized. But current intake processes typically consider only a small portion of
critical patient information presented at admissions – current patient symptoms, their
medical history (where available) and almost no information regarding how patients
with similar symptoms end up progressing during their stays. This can be severely
limiting, given that patients often develop new diseases during their stay in the
hospital that expand the treatment needed and increase insurance coverage extensions
required.
But what if patterns of additional diseases acquired during hospital stays of similar patients could be collected and considered during the hospital
admissions process? Could new disease expectations be accounted for, better treatment
plans prescribed and more realistic cost estimates made right from the start?
These are the questions Pankush and his co-author posed, and the answers they found
present a host of opportunities for significant improvements in patient care, reductions
in treatment costs and better allocations of scarce healthcare resources all around,
particularly in COVID-19 times where ICU beds and critical equipment such as ventilators
are in short supply.
The Harbert College of Business sat down with Pankush to find out what prompted their
research and what the healthcare industry needs to do to make practical, productive
use of the valuable insight their study revealed.
| HCOB: |
What prompted you to apply the power of Big Data and advanced business analytics techniques
to this particular problem among the many that could have been chosen? |
| Pankush: |
The specific focus of this project came out of a personal experience I and the co-author
had during the initial phases of this research, which was actually a part of my dissertation.
My academic advisor had a friend who was admitted to a hospital for what was supposed
to be a fairly routine surgery. As it turned out, this friend contracted pneumonia
after the surgery, the pneumonia got complicated and he died in the hospital. We discovered
that this was far from unusual – hospital patients often contract diseases that heretofore
had been considered unrelated to the reason they were admitted into the hospital in
the first place. |
| HCOB: |
We’ve all heard that hospitals can be breeding grounds for disease, and you’ve discovered
hard data to support that view – with data on millions of patients backing you up.
But you went further in that you found patterns of specific diseases whose likelihood
could be predicted at the time of admission using advanced business analytics, correct? |
| Pankush: |
That’s right. My advisor’s friend entered the hospital for a specific surgical procedure,
but then new medical issues arose that weren’t considered at the time of admission
which ultimately became life-determining. We were already working on a number of related
healthcare issues and had published research on those, and this incident motivated
us to apply advanced business analytics to addressing this increasingly important
dilemma. |
| HCOB: |
Can you walk us through a typical hospital intake process – what’s done and what’s
not done – and how your analysis of all potential factors you’ve identified can be
leveraged to produce what are clearly beneficial outcomes of hospitals employing your
model? |
| Pankush: |
Sure. So, when a patient goes to a hospital, he or she goes in with some primary symptoms,
"I'm having headaches." We discovered from our assessment of more than 20 million
hospital patient records that, over the course of their hospital stay, many patients
contract multiple other diseases. We proposed that a robust network of historical
diseases contracted in hospitals could help identify which diseases come together
in patients admitted to hospitals. For example, diabetes and hypertension are strongly
related to each other.
We then compiled a large network database of more than 15,000 diseases – and then
mapped which ones go together – using historical values. That was Step One.
But the second step was to use a patient’s symptoms at the time of admission and find
the probable comorbidities that are not diagnosed yet but may show up during the hospital
stay. Let's say a new patient comes into the hospital with just a headache. We then
access the network we created and found that if a patient comes in with a headache,
he or she is likely to contract something else during their hospital stay. We are
predicting, based on our database, what he or she might contract during their hospital
stay – some other, allegedly unrelated, disease or diseases.
Using these three types of data – from the past, the present and the future – we can
produce a predictive model that will postulate what other diseases are likely to arise
in that patient. This information can augment their initial treatment plan as early
in the process as possible. That way, preventative measures can be taken before the expected disease presentation occurs. Not only does this benefit the patient,
but it also shortens hospital stays and facilitates the allocation and management
of critical hospital resources.
|
| HCOB: |
So, there are really two primary benefits of your research – first, and perhaps foremost
from a healthcare outcomes perspective – is the ability your model offers for predicting
the onset of additional diseases that are likely to occur during a patient’s hospital
stay before they present themselves. This enables healthcare workers to include these likely new
diseases in their initial treatment plans and their initial authorization requests
for insurance coverage. That’s the health outcomes benefit.
But, as an expert in advanced business analytics, what additional benefits did you
envision for the business side of the equation – the advantages your model presents
in helping hospitals better manage beds, ICU resources and other practical allocations?
And how does this more realistic insight into eventual treatment needs and associated
costs benefit insurers?
|
| Pankush: |
For the hospital administrator, it's very easy to use the model. They simply plug
in the history of their patients and some basic demographics and they'll get the numbers
they need based on the data we’ve compiled for millions of other patients in similar
circumstances.
For insurers, knowing the treatments a covered patient will eventually need at the
beginning of their hospital stay rather than later on provides a valuable, realistic
perspective that they can use to project the financial liabilities associated with
each patient. Armed with the results that come from employing this new and efficient
predictive model, insurers will be able to better plan for and accommodate the real-world
costs of each patient admission.
|
| HCOB: |
What do hospital administrators and insurance providers each need to do to realize
the benefits of your model? |
| Pankush: |
For hospital administrators, a lot is already being done in terms of leveraging the
value of the electronic medical records systems (EMRs) required to use the model.
In many instances, administrators have, or can easily acquire, the medical records
of incoming patients in electronic form, according to leaders of some of the largest
EMR providers in the world. So, they have much of what they need for many patients
right now.
As for insurers, we made a point to speak with a few of the largest healthcare insurance
companies in the country, who also welcomed and helped structure the research we conducted.
From a business perspective, we would like to reach out to the practitioners to share
the results our model generates to help them realize the full benefits of this transparency
to their bottom line.
|
Degrees & Programs
Faculty & Staff
Career Development
Recruiters & Industry
The research, entitled “When will I get out of the hospital? Modeling Length of Stay
using Comorbidity Networks,” was co-authored by Pankush Kalgotra, Assistant Professor,
Business Analytics in the Department of Systems and Technology at Auburn University’s
Harbert College of Business, and Ramesh Sharda, Vice Dean, Watson Graduate School
of Management and Regents Professor of Management Science and Information Systems
in the Spears School of Business at Oklahoma State University. The peer-reviewed study
was recently accepted for publication in the Journal of Management Information Systems,
a top journal in the information systems discipline.
A key determinant of length of stay (LOS) when a patient enters a hospital is the
level of information available to admissions staff at the time of intake – it defines
the patient’s initial treatment plan as well as any insurance coverage originally
authorized. But current intake processes typically consider only a small portion of
critical patient information presented at admissions – current patient symptoms, their
medical history (where available) and almost no information regarding how patients
with similar symptoms end up progressing during their stays. This can be severely
limiting, given that patients often develop new diseases during their stay in the
hospital that expand the treatment needed and increase insurance coverage extensions
required.