Improving population health with a secure, next-generation research tool
Do you know how many of your organisation’s patients are likely to be readmitted within the next four weeks? Knowing this number can help your clinical teams improve the care they provide, while making better use of your health economy’s limited resources.
What if you could predict which specific patients in a population group of hundreds of thousands – if not millions – was likely to be readmitted over the same period?
An internationally renowned Californian health system is using Cerner’s intelligence tools to develop advanced data models which are able to do exactly this – with a stunning degree of accuracy.
Dr William Feaster, chief health information officer at the Children’s Hospital of Orange County (CHOC), explains: “We can now predict a patient’s risk of readmission pretty accurately by combining historical data about patients with the current data on a new patient, and running that through our readmission algorithm. If the algorithm designates a patient as ‘high risk’ at the time of discharge, it turns out that over half of those patients are actually readmitted within seven days.
“For the last few years we have been doing a lot of work in data science. We're applying sophisticated techniques, statistical inference and machine learning algorithms to our data.
By combining and utilising different types of clinical, patient, social and open source data, the team at CHOC, including a PhD data scientist, can build intelligent models which anticipate patient outcomes based on a multitude of medical and social factors.
The models are then continuously fed back into the provider’s electronic health record system, to inform clinical workflows and patient pathways, and deliver vital insights to care professionals on the frontline.
Dr Feaster explains how this tool helps clinical decision-making: “If a doctor is getting ready to discharge a patient but they see that they're flagged as high risk, the doctor, nurse and care management staff will be able to see the patient’s risk classification for readmission.
“They will also see the top six reasons why the patient was determined to be high risk and they will be able to determine whether the patient is ready for discharge. The goal is not to unnecessarily delay discharge, but drive attention to certain details such as education about medications, or addressing transportation issues to follow-up visits in order to prevent readmissions. Then, the whole care team can start working on it and make sure the patient is ready to go before they leave the hospital.”
While this capability has great potential to improve the treatment that individual patients receive, Dr Feaster’s key focus is the impact such innovations can have on population health to prevent citizens becoming patients in the first place.
“One of the goals of population health is to take people that have chronic disease and make sure that they're as well as they can be. We have all sorts of tools that help do that.
“But now we are taking a look at those individual tools and those individual patients to see if applying artificial intelligence and data science-type techniques can help us manage that population even better.”
Alongside the predictive readmission algorithm, this data-rich approach to problem solving has informed the team’s work in developing a multitude of tools, allowing CHOC to quietly become one of the world’s most forward-thinking healthcare providers.
As you can imagine, research of this depth is incredibly time-consuming and resource intensive. Until last autumn, the team had been gathering, collating, storing and analysing the data required to write these models themselves.
Dr Feaster describes this process: “We now have a vast amount of data in our HealtheIntent® population health database. It's everything we have done for the past five years in both inpatient and outpatient, including all the financial data for the 150,000 children that we are fully at risk for [through a capitated payment system]. Utilising this data for our work was not an easy process.
“The tough part is identifying the data elements you need and determining where that data is located, importing it into a secure database on premise, and running predictive models on servers in your data centre utilising the ‘R’ programming language and environment.” All time-consuming steps with a slow turnaround.
However, this all changed in November 2017 when Cerner invited CHOC to work with them to build a toolbox designed to allow healthcare organisations, research organisations and academic bodies to accommodate this kind of big data research.
The outcome of this collaboration is HealtheDataLab™, a ground-breaking tool that leverages the capabilities of HealtheIntent, Cerner’s market-leading data intelligence and population health platform, to provide a protected cloud-based environment in which unlimited amounts of data can be interrogated, at scale and with speed.
To do this, the hospital moved from using its own data centre to an Amazon Web Services’ cloud platform – where HealtheDataLab resides. Cerner then provided its technical capabilities and owned the process of the data onboarding and migrating the provider’s patient records alongside a succession of other data sets required for patient care.
Once data is brought into this environment it is consistently maintained by Cerner so that current, relevant data is always at hand.
The chief health information officer likens the new research possibilities that HealtheDataLab offers to his data scientists to being “a kid in a candy store”.
“We now have all that data in an analytic environment that is pretty easy to use. We can now use all the sophisticated tools of data science [such as Python, R and Jupyter], artificial intelligence and deep learning, creating neural networks - all the stuff you hear people talk about. It’s pretty amazing,” he explains.
“The other thing is that we can load other data sets in alongside the data that Cerner put in there, so if we want to load in genomic data, we can do that.”
HealtheDataLab is now helping Dr Feaster’s team improve their predictive data model. “We put that model into HealtheDataLab and we're now refining it. Instead of 98 variables, we now have hundreds of variables that we're tracking. Instead of ten categories of medications, we now have 120 categories at our disposal.
“With larger data sets come potentially improved accuracy of models that you develop.”
Separately, the CHOC team is now using HealtheDataLab to develop a ‘clinical deterioration model’ - an algorithm to detect patient deterioration throughout their care pathway.
Dr Feaster outlines its applications: “We're picking all those patients who during their hospitalisation were transferred to an intensive care unit.
“Now, with the ability to marshal thousands of variables for the model, we have a pretty good suspicion that we are going to accurately determine the likelihood of another patient needing a transfer some time during their hospitalisation and calculate that on an ongoing basis.”
This harnessing of datasets into practical tools is not simply a process designed to help those providing care. Dr Feaster explains that in the near future, citizens and their families will be able to access tools created with the help of HealtheDataLab to help prevent hospitalisation.
One fertile area for the technology is the ability for those living with asthma to monitor air quality in their community, as Dr Feaster explains: “We're going to be feeding analytic data back to the patient, saying 'There's a lot of pollution today - your child is sensitive to pollution, so take a peak flow test this morning [to indicate if their airways have narrowed]'.
“They could get an alert on their smartphone – and message through Cerner’s HealtheLifeSM portal... Ideally, they should see that in conjunction with their asthma action plan. It will give them instructions on what to do.”
Dr Feaster describes this form of patient activation as “medical rocket science” – something that has yet to be done in mainstream medicine to this degree. “This is very advanced, so we’re still developing it,” he adds.
Given his experience of using patient data as a tool to quantify risk, predict outcomes, and better care for large patient cohorts, how does Dr Feaster judge the impact these capabilities could have on the provider sector in the UK?
“We need to make sure that healthcare has good access to data science, which up until now it really hasn't - because people haven't had a tool to use. If you have data sitting in some database somewhere, it's not accessible and there's no tools to analyse it, it doesn't have much utility.
“My advice to the NHS is that you have plenty of reasons to start doing this now as a way of improving the health of your population and reducing unnecessary utilisation and cost.
“There's a wealth of data already there in primary care systems that could be imported into this.
“Don't waste any time - begin to adopt analytic tools that can help you analyse your population and then provide tools back to the care givers to improve the care they provide.”