Phoenix Children’s data lake helps improve patient care
When Phoenix Children’s embarked on a big data initiative a few years ago, the health system’s leaders first agreed that any piece of information, large or small, could be of value. Then, says CIO David Higginson, it was an exercise in peeling back the layers of its data onion.
“We started with our primary systems—the EHR, lab, surgery, etc.—and there are pretty standard ways to connect those to an enterprise warehouse and pool that data,” he says. “But then we kept going, and found it more and more challenging to connect systems and data that had never been done before.”
The solution? The healthcare system, one of the largest pediatric medical systems in the country, slightly more than five years ago deployed a “data lake,” a repository that serves as a collection point for vast amounts of information.
The project went from the drawing board to being an operational repository in three months.
“One reason we were so aggressive in building our data lake is that we knew that, at Phoenix Children’s and the wider industry, we are years behind when it comes to combining data from different information systems to give real insights,” Higginson says. “Every step you take in a hospital shows you five things that could be made better if you had real decision support. Any so many of those problems are connected, as is the data about them. You just have to make the connections.”
Today, Phoenix Children’s has 65 different systems, including enterprise EHR, radiology, time and attendance and door-scanner data, streaming into the data lake, according to Higginson.
Having virtually every scrap of data available for its analytics efforts has enabled the health system to create new safeguards around medication dosing that have virtually eliminated medication errors. It’s also developed real-time analytics that monitor kidney function for every patient in the hospital, which has helped caregivers head off kidney injuries in pediatric patients, an all-too-common problem when treating patients who are on multiple medications for a variety of ailments.
“When you combine all the data from those secondary systems with information from your core systems, you provide context to every action you’re taking, and that provides a new level of insight into how to improve your clinical and financial operations,” Higginson says. “When we determine what kind of problems need to be addressed, we approach it with an open mind and don’t assume we have the answer. Having all our information in the data lake helps us get to the root of the problem.”
Phoenix Children’s data lake is built on a stack from Microsoft. The base platform is Microsoft SQL server, with business intelligence capabilities, and the company’s SSIS ETL tools are used to move data from the source systems into the data lake. Microsoft’s SQL server package is integrated with the R language to enable users to develop statistical models and predictive algorithms for analytics projects. And Microsoft’s Reporting Services is used to create reports on demand.
“The efficiencies really made this a no-brainer for us,” Higginson says. “The Microsoft products work so well together that we could normalize the baseline for extracting data from all our source systems and didn’t have to spend the time or money incorporating other software into the data lake. And the BI tools cost us nothing since we had already purchased the Microsoft SQL server platform. We could have spent $100,000 on some other BI tools and perhaps have slightly better capabilities, but that didn’t make any sense to us.”
One of Phoenix Children’s first data lake initiatives was tackling medication dosing errors. There are copious amounts of medical literature about standard medication dosages, says Phoenix Children’s chief medical information officer Vinay Vaidya, but it doesn’t truly address the issues in a pediatric hospital.
“We are treating premature babies as well as 300-pound 15-year-old football players, and the available literature doesn’t provide a sense of what are the soft and hard stops for dosage levels for children of vastly different age and size. And there’s no off-the-shelf computerized physician order entry [CPOE] system that has a module available with that data,” Vaidya says.
Phoenix Children’s went to its data lake to find a solution. The health system analyzed more than 1 million medication orders made over 11 years at its main hospital to determine safe dosage ranges for children based on age and weight. The analytics pieced together disparate bits of information from the EHR, lab, pharmacy and other information systems to provide a coherent view of the medications provided to patients.
And its data infrastructure enabled it to go big. “The analytics are typically done one-by-one for each drug, but we were able to take all 1,200 drugs we prescribe and use a single table and programming code for all of them at once,” Vaidya says. The result is that the CPOE system now generates hard and soft stops when the amount of a prescription is suspect.
To understand the importance of developing hard and soft stops on medication dosages, Vaidya points to its effect on prescriptions of potassium. The drug is used to treat a variety of health concerns, but potassium can also be lethal if taken in large amounts. In the five years since Phoenix Children’s developed its hard and soft stops for potassium based on its analytics, its CPOE system electronically flagged and stopped 108 orders of potentially harmful doses.
In most cases, Vaidya says, a wrong keystroke caused an unsafe dose to be entered, but in other cases in was unfamiliarity with the appropriate dosage levels for that specific patient. When hard stops occur, the CPOE system is programmed to not generate the order, eliminating the possibility that a dangerous medication dose could flow downstream to the pharmacy. Soft stops are flagged and then reviewed by both the ordering physician and pharmacist to ensure the clinical conditions warrant such a high dose.
“When you’re dealing with a pediatric population, small changes in medication dosages can be the difference between a soft stop for a dose level that’s a bit high, to a hard stop because the dosage is potentially lethal.”
Another example of how the data lake is improving patient care is Phoenix Children’s recently launched effort to head off kidney injuries.
Vaidya spoke with a nephrologist at Phoenix Children’s who was concerned that the health system didn’t have an effective way to monitor the kidney functions of the 300 patients in the hospital. Many children are being treated for serious conditions such as cancer, asthma and infections that require multiple medications. But those medications combined can negatively impact kidney function.
“There are all kinds of disparate pieces of data in the EHR about each patient’s condition and medications, and it’s basically impossible for a physician to cognitively calculate all the possible side effects and interactions from the different drugs available,” Vaidya says. “So what we needed to do was look at the data and provide physicians and pharmacists with very early warnings about potential kidney damage.”
The kidney function monitoring system taps into the data lake in real-time and pulls together data from myriad data streams to analyze creatinine tests, which measure the functions of the kidney, along with information such as diagnoses and medications to gauge kidney function trends for every patient at the hospital. Most of those patients, Vaidya notes, were not admitted for kidney problems and hadn’t been monitored consistently for how their various treatments were having an impact on kidney function.
Now that physicians and pharmacists have real-time information about trends, they can quickly identify any changes in kidney functions and address those changes before any damage occurs.
The speed and success of the effort is an example of how the technological and structural efficiencies help Phoenix Children’s big data efforts deliver, Higginson says.
The organization is structured very flat, he says, and instead of committee meetings there are one-on-one interactions with clinical and IT leaders. So instead of running the kidney monitoring system though a number of committees, Higginson and Vaidya met with a few colleagues to sketch out the data requirements and design for the system.
The prototype was put together in a week and reviewed by clinicians, who said it could make a huge safety impact. Within hours, the prototype was available to clinicians.
“We are now on version 15 or so of this kidney functioning monitoring system, but instead of waiting months to perfect the system we put it out and it had an immediate impact, and now we’re consistently working to improve it based on the feedback from end users,” Higginson says. “We’ve tried very hard to knock down the technological barriers as well as the organizational barriers to using our data, every bit of it, to improve our operations.”