Predicting Hospital Readmissions: Laura Hamilton Interview (Additive Analytics CEO)


Data Science Weekly Interview with Laura Hamilton – Founder/CEO of Additive Analytics - who is using predictive analytics and machine learning to lower hospital readmission rates

We recently caught up with Laura Hamilton, Founder and CEO of Additive Analytics. We were keen to learn more about the evolution of the health analytics space, how data science / machine learning is helping, and what she is building at Additive Analytics ...

Hi Laura, firstly thank you for the interview. Let's start with your background...

Q - What is your 30 second bio?
A - I graduated from the University of Chicago with a mathematics degree. From there I joined Enova International, where I launched 3 businesses within a 3 year span; 2 of those were profitable within 18 months of launch. After Enova, I joined ecommerce startup BayRu as Head of Analytics. At BayRu, I built a proprietary analytics engine that compares the company's performance to benchmarks. In September 2013, I launched a healthtech startup called Additive Analytics. We provide analytics for hospitals.

Q - How did you get interested in working with data? Was there a specific "aha" moment when you realized the power of data?
A - It was during my econometrics class at the University of Chicago. For the final project, we did a linear regression on some labor data using Stata. I just liked the idea that I could get real, actionable, objective results with a computer program and a few commands. Then at Enova, I brought on an additional data vendor, which improved our underwriting and reduced our default losses. It's often more effective (and easier) to go out and obtain additional data sources than it is to keep trying to make your algorithm more sophisticated!


Laura, interesting background and context - thank you for sharing! Next, let's first talk about Healthcare Analytics..

Q - What have been some of the main advances that have fueled the rise of Healthcare Analytics in recent years?
A - We've seen a dramatic increase in the number of providers using Electronic Health Records (EHRs) in recent years as a result of government incentives. There's all this clinical data sitting in machine-readable form right now. It used to be all paper charts in boxes in the basement. Now that there's all this data, people are really excited to harness the data and use it to provide better care at a lower cost. In the future, I think that we will see a lot more analytics geared towards patient engagement. I am really excited about the potential of Blue Button+, which is an initiative by the federal government to enable patients to view their own personal health data online or via mobile devices.

Q - What are the main types of problems now being addressed in the Healthcare Analytics space?
A - One of the key priorities of the Centers for Medicare and Medicaid Services (CMS) is moving from a fee-for-service payment model to a pay-per-episode-of-care model. Currently, 60 hospitals are participating in CMS' advanced bundled payments model. These hospitals need to start taking full financial risk by October 2014. Under the new model, a participating hospital will receive a flat payment up front (depending on the patient's condition). The hospital will not receive any additional payments if the patient is readmitted within 30 days of discharge. That type of financial risk is new to hospitals. Hospitals are used to fee-for-service. As a result, there's a lot of demand for analytics solutions that will reduce 30-day readmissions. For example, the Henry Ford Innovation Institute has issued a challenge to find innovative, technology-driven solutions to reduce 30-day hospital readmissions.

Q - Who are the big thought leaders?
A - There are several...

  1. Healint is using data from cell phone sensors to identify neurological conditions such as stroke.
  2. AliveCor built a $199 ECG sensor that attaches to a cell phone.
  3. Google is using search data to uncover flu trends/li>
  4. 23andme analyzes patients' DNA to identify genetic disease risks.
  5. IBM researchers have found a way to extract heart failure diagnostic criteria from free-text physicians notes.
  6. Additive Analytics predicts which patients are likely to get readmitted to the hospital within 30 days of discharge and gives providers actionable steps to take to reduce readmissions.
  7. AllScripts is building an ecosystem of technology and analytics apps that integrate with its electronic medical record system.

Q - What excites you most about bringing Healthcare and Machine learning together?
A - Healthcare in the United States is so broken right now. The United States spends $2.8 trillion per year on healthcare. That's $8,915 per person. The United States spends 17.2% of its GDP on healthcare - far more than its peers spend. Take a look at this graph:

Healthcare Spending as % of GDP
Image Credit: Additive Analytics

By using the right analytics, we can reduce costs while increasing the quality of care.

Q - What are the biggest areas of opportunity / questions you would like to tackle?
A - One of the key areas of focus for us is reducing 30-day hospital readmissions. Sixty hospitals have joined CMS' most advanced payments model. They've agreed to receive a single payment covering the whole episode of care, including all hospital readmissions up to 30 days after discharge. We're offering a solution to enable hospitals to understand their data and reduce their readmissions.


Definitely sounds like an exciting time to be developing technology in this space! On that note, let's talk more about Additive Analytics...

Q - How did you come to found Additive Analytics?
A - My background is in technology and analytics, and I wanted a way to leverage that. And I think now is the right time to be working on technology and analytics for healthcare. We are just now getting electronic access to patients' medical records. Also, there are lots of payment changes coming in the near future as a result of the Affordable Care Act. With those two changes happening right now, I think there is a ton of opportunity in the healthtech and healthcare analytics space.

Q - Got it. So what specific problem does Additive Analytics solve? How would you describe it to someone not familiar with it?
A - We provide analytics for hospitals. In the past few years, most healthcare providers have moved from paper charts to electronic charts. Now there is a huge amount of clinical data. Additive Analytics takes that clinical data and generates useful insights from it. For example, our model can identify what patients are likely to get readmitted to the hospital within 30 days of discharge. We suggest actionable steps that providers can take to reduce 30-day hospital readmissions. By reducing readmissions, we can save money and save lives.

Q - Could you tell us a little more about the technology? Firstly, how does it work?
A - Our software integrates directly with hospitals Electronic Medical Record (EMR) systems. Our software takes clinical data from the EMR and runs various analyses on it. From there, we generate intuitive graphs that give hospitals insights into their performance, both on clinical measures as well as financial measures. We also provide concrete steps that hospitals can take in order to improve their quality measures. For example, if a hospital wants to reduce 30-day hospital readmissions, we can provide specific steps on a per-patient basis that the hospital can take to prevent excess readmissions.

Q - What tools / applications are you using?
A - I like to use Octave, Python, and Vowpal-Wabbit. Sometimes I find it's helpful to do some initial summary and graphing with Excel. The Additive Analytics web application is built with Ruby on Rails. It sits on top of a Postgres database. For data visualization, I like D3 and DataTables. If I need a quick chart for the Additive Analytics blog, sometimes I will use Infogr.am.

Q - How is Machine Learning helping?
A - Machine learning helps us make sense of the huge amounts of clinical data in hospitals' EMRs. For example, we can use Natural Language Processing to extract meaning from free-text physicians' notes. Also, we can use techniques such as logistic regression and neural networks to predict which patients are likely to get readmitted to the hospital within 30 days.

Q - What is the most surprising insight you have found?
A - Simply giving patients a phone call after they are discharged from the hospital reduces the risk of 60-day readmission by 22%. It's so simple, but it's so powerful.

Q - What is your favorite example of how Additive Analytics is having real-world impact?
A - Last week we launched an online tool that expectant parents can use to compare maternity wards at different hospitals. It was written up in TechCrunch. Our goal is to give patients tools to evaluate providers based on objective, quantitative quality metrics. We hope to provide increased transparency to hospitals' performance. Currently, you can go on Yelp and find the best restaurant or you can go on Angie's List and find the best plumber. You can go to US News & World Report to find the best college. We think you should be able to go on the internet and find the best healthcare provider, too.

Q - What advances could your approach / technology enable going forward?
A - We can run analytics on clinical EMR data to figure out which treatments are working better than others. For example, we could analyze the outcomes of patients who were treated with proton therapy via a $150 million cyclotron. We could compare how those patients fared versus patients treated with traditional (much cheaper) methods. Perhaps we would find that proton therapy didn't improve outcomes at all; that could provide significant cost savings to hospitals as well as payers.


Very interesting - look forward to hearing more about Additive Analytics going forward! Finally, it is advice time!...

Q - What does the future of Healthcare & Machine Learning look like?
A - A few thoughts ...

  1. Researchers have found a way to extract Framingham heart failure diagnosis criteria from free-text physicians' notes using Natural Language Processing. In the future, I think we'll see many more applications of Natural Language Processing for diagnosis, for anomaly detection, and for billing.
  2. I think that we're going to see a much tighter integration of the clinical side of things with the financial side of things in the future. It will be much easier for physicians and hospital administrators to understand which treatments are the most cost effective.
  3. We have a huge number of data sources now. I have a Withings scale. My husband has a Fitbit. We've got all these new sources of data from wearables and even from our cell phones. In the future, I think a lot more of the data will be connected. Your physician will be able to see a chart of your Withings data, your Fitbit data, your cell phone data.
  4. I think that we're going to get better at finding adverse drug events. Now that we have electronic medical records for millions of patients, we can mine that data to find drug interactions and problematic side effects—even ones that only affect a small subset of patients. Problems such as those with Vioxx and Thalidomide will be found more quickly, and fewer patients will be affected.
  5. We're going to have a better understanding of disease transmission. Already, we can use Google search terms to understand flu trends. If we combine social media data with electronic medical record data and perform aggregate analyses, we can predict epidemics and take steps to halt disease transmission in its tracks.

Q - Any words of wisdom for Machine Learning students or practitioners starting out?
A - 5 things:

  1. Take Dr. Andrew Ng's Machine Learning course on Coursera.
  2. Take Dr. Abu-Mostafa's Learning from Data course on edX
  3. Get as many features as you can. Think about where you can get additional data, and add as many new data sources as you can.
  4. Data visualization is as important as the model. You can have the most sophisticated model in the world, but if you don't present it in a way that's intuitive to the user it will be useless. All analyses should be actionable
  5. Beware overfitting!


Laura - Thank you so much for your time! Really enjoyed learning more about the evolving Health-tech landscape and what you are building at Additive Analytics. Additive Analytics can be found online at http://www.additiveanalytics.com and Laura on twitter @LauraDHamilton.

Readers, thanks for joining us!


P.S.If you enjoyed this interview and want to learn more about
  • what it takes to become a data scientist
  • what skills do I need
  • what type of work is currently being done in the field
then check out Data Scientists at Work - a collection of 16 interviews with some the world's most influential and innovative data scientists, who each address all the above and more! :)

You might also enjoy these interviews because you are awesome:

Back

Sign up to receive the Data Science Weekly Newsletter every Thursday

Easy to unsubscribe. No spam — we keep your email safe and do not share it.