This is the second part of a series written by Charles Boicey from UC Irvine Health (part 1 is here). The series will demonstrate a real case study for Apache Hadoop in healthcare and also journal the architecture and technical considerations presented during implementation.
It has been 232 days since the last post. Much has transpired including a rebranding of the organization from UCI Medical Center to UC Irvine Health. I am happy to report we have a production Saritor environment up and running on the Hortonworks Data Platform.
Here are some highlights from the past 232 days:
In collaboration with our medical device integration partner, iSirona, we are developing a system to acquire home monitoring data and transmit it to Saritor. Our first deployed device will be a scale. This may sound simple, but in-home monitoring of the daily weights of Congestive Heart Failure patients is essential for the prevention of those patients readmitting to the hospital.
Home monitoring data will not be transmitted directly to the Electronic Medical Record (EMR), for a very specific reason. Home device data from thousands of patients transmitted directly to the EMR would be a nightmare for clinicians to manage. It would be too much data. By sending the data to Saritor first, an algorithm can determine which changes in weight indicate risk of re-admittance and then notify clinicians about those cases. All home monitoring data will be viewable in the EMR via an API to Saritor.
We are working on a pilot to enhance patient monitoring in the hospital. In California, nurses typically have up to five patients to care for, and it can be challenging to be with a patient at the bedside and also keep a close eye on all the small changes in vitals across all patients.
Soon, hospitals will be able to provide each new inpatient with a wearable disposable patch that monitors vital signs such as heart rate, temperature, pulse oximetry and wirelessly transmit that data every minute to Saritor. An algorithm can “watch” that data for patterns that the nursing team might not be able to catch. Because nurses cannot watch a monitor for every minute of their shift, Saritor has “got their back”. Nurses can go about the business of caring for patients and Saritor will notify them when there is a disturbing pattern in a patient’s vitals. A data warehouse might be able to run a similar algorithm, but with 24-hour latency. That’s too much latency for a nurse to respond quickly to an emergent situation.
With the increasing numbers of patients joining the “Quantified Self” movement we see Saritor as the ideal environment to help receive more health data generated by the patients themselves. We want to store and make use of patient-generated data from personal health records and home monitoring. Sites such as Fitbit, 23 and Me and others could also feed in data. With open APIs to a patient’s personal health record this data can be ingested into Saritor and then be made available to clinicians via the EMR. Score cards from the EMR data in Saritor can also be pushed back out to the patients.
Other Lessons We’ve Learned
Hadoop Plays Well with Others
One awesome discovery we made was that the Hadoop Ecosystem plays well with other systems. We were able to start ingesting data into Hadoop, without having to change anything within the current IT environment. For example, all of the healthcare data ingested into Saritor goes into HDFS. For the monitoring of inpatients, Map Reduce jobs run against HDFS and then push that data into MongoDB. Algorithms in Mahout run against the data in MongoDB and can push notifications to the EMR via an event engine.
For graph analysis of healthcare data MapReduce jobs run against HDFS and then output in graph form for input into Neo4j.
Legacy Healthcare Data Is Valuable
We ended up with 9 million patient records spanning 22 years and 1.2 million patients. Our original estimate was 3 million records. We are using this data to build our surveillance algorithms.
Social Media Is an Important New Source of Information
Saritor is capable of storing social media data related to UC Irvine Health and a UCI student project is underway to develop a sentiment analysis dashboard to better understand the social media environment external to UC Irvine Health. As part of the patient experience feedback loop we will be able to reach out and connect with patients to better understand their concerns so that we can enhance the patient experience.
Others in the Healthcare Community Are Interested in Adopting Hadoop
I’ve spoken with many other healthcare providers that are trying to solve the same type of problems, all are eager to exchange Hadoop best practices.
In the next installment, I’ll give an update on the results of our monitoring pilots, describe our progress on surveillance algorithms, and tell you more about our collaboration with other hospitals and clinics.
If you’re considering your own Hadoop implementation, then click here to learn more about Hortonworks Data Platform.