The Challenge
The Solution
Case Study Highlights
- IEMS, the largest provider of emergency medical services in Indiana, needed to correlate and extract insights out of millions of medical and operational data points across various government agencies
- They tried using Cognos but were slowed down by manual data cleansing and batch-oriented processing, taking weeks to generate a single report
- IEMS turned to Elastic partner Perceivant, the company behind Data Dojo: a HIPAA-compliant, cloud-based data analytics platform, powered by Elasticsearch, to create real-time, interactive dashboards to get immediate insights out of their data
- They've not only improved their operations, but are now able to incorporate unstructured data from doctor’s notes into their analyses, helping them proactively respond to the threat of Ebola in 2014
Perceivant’s Journey with the Elastic Stack
The Challenge
Indianapolis Emergency Medical Services (IEMS) is the largest provider of emergency prehospital medical care in Indiana, responding to nearly 100,000 911 calls each year. The organization not only serves the nearly 1 million residents of Marion County, Indiana through emergency care, but also through outreach and educational programs. In order to deliver the best level of care, as well as optimize operations as part of its duty to the city's taxpayers, IEMS turned to data to help guide decisions around how they operate.
This required pulling together millions of data points across various government agencies, from clinical and patient data from local hospitals, to response and logistical data from 911 calls and ambulance dispatches, and census data like demographics and crime statistics. IEMS tried using business intelligence and performance management software, Cognos, but found it too cumbersome due to its batch-oriented processing. Between data cleansing and manual correlation in extensive pivotal tables, getting insights out of their data would take weeks, making it nearly impossible to make decisions that could have an immediate impact on protocol and resource allocation.
The Solution
To help them gather and analyze data more efficiently, IEMS turned to Perceivant, which makes a HIPAA-compliant, cloud-based data analytics platform powered by Elasticsearch called Data Dojo.
In working with Perceivant, IEMS exported data from various medical systems into CSV, XML, HL7, and JSON files. These files were then securely imported to Data Dojo, where the data was indexed in Elasticsearch. From there, the data is directly output to an OLAP multi-tenant reporting service where users can run real-time queries and build interactive dashboards, eliminating a lot of the manual data processing and correlation that IEMS used to perform. The NoSQL nature of Elasticsearch and Perceivant's flexible analytics structure allowed the addition of new structured columns on-demand to further analyze the data, eliminating the need to make any coding changes to the medical systems the data was being pulled from. And, thanks to the full-text search and analytics capabilities of Elasticsearch, Data Dojo also let IEMS incorporate unstructured data from doctor's notes, helping create new dashboards that let them respond to unforeseen challenges and extend their use case in ways they couldn't before imagine.
One such example is the 2014 Ebola outbreak epidemic. IEMS knew they needed to get real-time insights from their data to get ahead of the virus, so they engaged Perceivant to help them build dashboards in Data Dojo to identify changes in normal influenza patterns. Elasticsearch was essential to searches of the unstructured doctors notes to identify symptoms like ‘fever', ‘fatigue', or ‘headache'. By comparing historical influenza data to real-time reported symptoms, IEMS could easily spot a spike or trend that they could immediately further investigate.
The Results
Since implementing Data Dojo, IEMS has been able to significantly improve their operations. Reports that used to take weeks to generate now exist in real-time, interactive dashboards. IEMS data scientists can explore new ways to correlate data through the platform's interactive dashboards, and, with Data Dojo being based in the cloud, IEMS can grant anyone access to it, empowering doctors and nurses to track specific patients' health.
IEMS has been able to utilize data to make sure they equip ambulances with the proper equipment and relevant medics based on the types of emergency calls likely to occur in certain geographic locations, analyze the most costly types of treatments and how they impact the long-term success of a patient, and measure the impact certain equipment has on the outcome of a patient. IEMS was also able to track Ebola symptoms daily as opposed to their prior norms of running monthly reports. And, in addition to tracking the flu, Ebola, and other public health epidemics with greater ease, IEMS also monitors incidents of product overdose, gunshot victims, and weather patterns that are correlated with certain disease outbreaks. If they observe a trend, the IEMS can immediately Tweet and broadcast the information out to the affected community.
The How
Data Dojo is built on top of AWS' HIPAA-compliant cloud. Data can be manually imported to the Data Dojo platform either via SFTP or Amazon S3 bucket using an updatable CSV file, or via Perceivant's API for real-time updates. Once the data is on the platform, it is indexed in Elasticsearch, and from there, output to an OLAP multi-tenant reporting service where users can build Pivot Tables with drag and drop functionality, create graphs based on Pivot Table results, as well as drill down on cross-sections of data in the Pivot Table and perform queries against it. With Elasticsearch's open APIs, Perceivant also built Apache Spark and R connectors for customers that have data scientists that want to do more complex statistical computing and graph building.
In its quest for a text search and analytics solution, Perceivant evaluated Elasticsearch, Hadoop and MongoDB. Hadoop queries were too slow, taking about 30 minutes to return results vs. the 10 seconds response times they achieved with Elasticsearch. They also found the Hadoop and MongoDB user interfaces too hard to use unless you were a Hadoop or Mongo coder. The fact that Elasticsearch can power text search and analytics at blazing fast response times was what made Perceivant eventually choose Elasticsearch to build its platform on top of.
Perceivant’s Clusters
- Nodes27
- Indexes1,347
- Documents667 million
- Query Rate~12 per hour
- Daily ingest rate15 million
- Replicas1
- Time-based Indices1 per day
- Node SpecificationsCompute units on High Frequency Intel Xeon E5-2670 v2 (Ivy Bridge) Processors 7.5 GB of RAM (50% allocated to ES) SSD