Every day, patients in healthcare facilities around the world find themselves sicker than they were when they were admitted, due to hospital-acquired infections (HAIs).
In the United States alone, around one in 31 hospital patients have at least one HAI, according to the Centre for Disease Control. These infections, also referred to as healthcare-associated or nosocomial infections, can have devastating physical, emotional and financial consequences for patients. In some cases, they can be deadly.
That makes infection prevention and control (IPC) measures to fight HAIs a top priority for healthcare providers everywhere. In Atlanta, Georgia, Piedmont Healthcare is now several years into its 10-year journey to create a ‘zero harm’ environment for patients, by eliminating all HAIs by 2026.
An important part of this programme at Piedmont, which operates 11 hospitals in the greater Atlanta area, has been the building of an infection prevention dashboard, based on a data warehouse from Exasol and front-end reporting from Tableau.
This enables staff to analyse lab results regarding infections to figure out the root cause, enabling them to determine whether the infection was acquired in hospital, and if so, how. That in turn puts them in a better position to swap out equipment or make changes to caregiving protocols that might be putting patients at risk.
Previously, this data was poorly structured and it took intensive processing to extract and analyse the data, which meant that the previous day’s results were not ready for hospital staff until 1.30pm the following day. Today, the dashboard shows them that information at 8.30am.
This is just one example of the kind of reporting that has been made easier by the decision to move to the Exasol data warehouse just over two years ago, says Mark Jackson, Piedmont Healthcare’s Head of Business Intelligence:
A large part of our success in reducing HAIs across our healthcare system is because we’ve been able to put data in front of people so that they can see, every day, whether there are new infections from monitoring microbiology results that have come in and do a huge amount of data parsing to extract from notes-based text what the actual organisms are and then use that data to drive our improvement processes.
Since FY2016, the latest information I have for us is that we’re down 68% on HAIs and that’s despite adding some five new hospitals to the system during that period. So that’s pretty impressive.
While the organization was already very happy with the reporting capabilities provided by Tableau (which it’s been using for around 8 years), he says, it was clear that more processing power was required at the back end to deal with the huge (and growing) volume of data the organisation holds and the need for faster reporting.
More specifically, the preference was for a data warehouse based on massively parallel-processing (MPP) technology. This, it was felt, would speed up data analysis work exponentially, by partitioning the warehouse across multiple servers, with each one having memory and processors to process data locally and concurrently.
While keeping Tableau’s front end for all its visualizations, Piedmont swapped out an elderly Microsoft SQL Server system for Exasol as the data warehouse, integrating this with EPSi, a healthcare financial decision support system and Epic, an electronic healthcare record system. Piedmont also looked at HP Vertica and Microsoft SQL Server Parallel Data Warehouse, before ultimately settling on Exasol.
Prognosis: machine learning
Today, in addition to the infection dashboard, this implementation provides the basis for use cases in almost every aspect of Piedmont Healthcare’s work. It hosts around one trillion data points, says Jackson - not just patient information, but also financial data, insurance claims processing information, patient satisfaction data and more. He and his team are still in the process of migrating other data sources to Exasol to support new use cases, but despite this, he’s already looking ahead:
There’s still a huge amount of descriptive analytics that we need to do and a lot of people in our hospitals that need information, but once we’ve tackled these use cases, I’m excited to start exploring how we might apply machine learning algorithms to some of the data held in Exasol.
On the clinical side, for example, he thinks more sophisticated analytics could be applied to healthcare records, with algorithms processed by Exasol used to identify if a particular patient is at risk of deteriorating, enabling hospital staff to take action immediately. The arrival of more connected healthcare devices, he says, will add hugely to the data load, but also open up exciting new avenues of analysis.
And, on the business side of things, Piedmont could make use the geospatial query capabilities in Exasol to plan its expansion strategy, he says. By feeding third-party data into Exasol, the organization could figure out where it should establish new healthcare offices for primary and specialty care, for example, and what local demand for these might look like:
We could figure out what healthcare provision already exists in a given area, what particular healthcare concerns exist in that area, and compute drive-time distances to offices. That would be something really cool to do with predictive analytics and I’m excited to put this kind of thinking to work for the benefit of our patients.