Chronic Care Disease Management and Reducing Hospital Re-admissions: Using streaming data and AI to help get the job done

Related Topics: Data & Analytics, Healthcare, Intelligent Innovation

In a recent eBook, Dr. Rob Nelson shared how to achieve business at the speed of now by driving business results with intelligent analytics and decision-making using real-time streaming data.

There are many sources for data appropriate to streaming processing and advanced analytics. Let’s dive into a specialized topic and look closely at chronic care disease management and reducing hospital re-admissions. This use case can be quite relevant and transferable to other types of data and circumstances. Plus, you can see how bringing different data sets into the mix with AI is very powerful.

Intelligent Streaming Analytics

Real-time or streaming analysis allows healthcare organizations and businesses to access data within seconds or minutes of ingestion to encourage faster and better decision making. Unlike batch analysis, data points are fresh and findings remain topical. Your prospects, customers, users and workers can respond to the latest insight without delay. Add the use of advanced analytics, AI and machine learning models and your real-time analytics get smart.

While we feel the need for speed in the age of “now”, speed isn’t the only advantage of real-time analytics. A high impact solution is equipped to handle high volumes of complex data and also deliver actionable insight at blistering speeds.

In the world of healthcare, In the era of VALUE-BASED CARE amid the adoption of AI, connected devices, 5G and advancements in personalized medicine and genomics there are many opportunities. In fat, with healthcare, even the speed of now isn’t fast enough. One such area is the problem of hospital re-admissions and chronic disease management.

The persistent and complex problem of re-admissions

A hospital readmission, as most healthcare workers know, is when a patient comes back to hospital within 30 days after being discharged from an earlier hospital stay. Reducing hospital re-admissions—especially those that result from poor inpatient or outpatient care—has long been both an important goal because it represents an opportunity to lower health care costs, improve quality, and increase patient satisfaction at once.

Further, the Centers for Medicare and Medicaid Services, create benchmarks for the 30-day readmission rate, and they financially penalize hospitals that have higher readmission rates. As described by one leading physician and expert, Robert Berenson, MD at the Urban Instituted, improving readmission rates is a “..win, win, win”.

This is a complex issue prime for AI, machine learning, big data and real-time data streaming, prediction and intervention.

Now, consider this time also, of a pandemic where healthcare resources are so scarce yet the ability to prioritize the right patients for hospital admission while keeping others at home is even more critical.

The problem is big, costly, and lives are at stake

Let’s consider the rate at which patients return to the hospital.

The numbers and the cost of those re-admissions are staggering by most accounts. The readmission costs the United States Federal Government over $20 billion a year, and about one fourth of these 30-day re-admissions are considered preventable. While figures vary depending on the age and health status of the patients, studies show that 17.6 percent of Medicare patients were readmitted to hospital within 30 days of discharge, accounting for $15 billion in spending (1). Of those re-admissions, the Agency for Healthcare Research and Quality study of patients admitted to hospital with preventable admissions found nearly 20% had at least one preventable readmission within six months. The cost of those admissions was $729 million, or $7,400 per readmission.

The numbers are not much better in the commercial population. When one commercial health system reviewed discharge data for its enrollees, it found readmission rates at hospitals ranged to as high as 44 percent, with an average around 10 percent. So, it is not surprising that this is a top priority all players in the healthcare system with a special emphasis on those with chronic disease and comorbidity.

The national average of hospitals penalized for avoidable readmission rates is 50%. But, there are states, such as New Jersey, where as many as 97% of hospitals have been penalized for hospital readmission rate that exceed the average. Most importantly, patients are at far greater risk if they reach a point that requires another hospital admission – especially one that could have been avoided through better prevention, intervention and care.

Many variables necessitate AI models to find a solution

Enter in the applicability and power of both real-time streaming data and AI and machine learning in healthcare to prevent unnecessary re-admissions through appropriate care interventions – yet prioritizing patients who should be readmitted.

The re-admissions problem is complex and presents a number of input variables related to the patient including their demographics, the diagnoses, the comorbidity index and more. Categories of data such as healthcare utilization, social determinants of health, diagnosis history and other data make is possible to build a machine learning model to predict and reduce readmission. Then, we can lever real-time patient data off of monitoring devices, self-reported data including weight, vitals, blood pressure, temperature, O2 levels and more.

Not surprisingly – and as we learned in the world of the coronavirus pandemic – people who tend to have multiple conditions and multiple illnesses tend to be greater candidates for readmission. Other variables include socio-economic status and provider information and characteristics like facility type.

Perhaps a long-term care facility has other instances of issues like hospital acquired infections etc. Perhaps the patient stays at home but lacks good home care. The model can also look at claims data, inpatient visits, pharmacy visits. All of these in variables are input into a model to predict readmission and chronic disease issues such as with Diabetes or Heart Disease (CHF).

Use modeling to prioritize care resources and improve outcomes

Modeling can classify each patient into one of ten deciles from lowest probability to the highest probability of readmission. By prioritizing a top tier of patients in a managed population visits, like 10% of patient, inpatient visits identified by the model, the model may identify the most – at risk population that is contributing the most to re-admissions or chronic disease interventions. As few as 10% of the patients could drive a large percentage of re-admissions. The modeling enables tiered intervention programs. Strategies for real-time monitoring and intervention are possible such as the home visit such as pre-emptive paramedic intervention or telephonic follow-up.

Use real-time data for real-time interventions

Machine learning models suggests the best intervention strategy for each patient based on their decile, and based on the, the different types of intervention approaches.

For those patients that need more real-time monitoring, data can stream as it feeds a model that will trigger alerts and specific interventions by dispatching a team or alerting a care team that makes a decision on the right action and intervention for a patient. Further, with geo-location, care teams can be optimized on how they are deployed and staffed for a service area.

With ever evolving health monitoring capabilities (including personal health monitoring devices) and prescription drug administration monitoring to feed real-time data streams, including geo-location, from patient devices and check-ins, the machine learning model and run real-time to alert specific action and interventions by a care team before an emergency unfolds. So, this approach is use big data, AI and machine learning to proactively predict and classify patients at most risk of readmission and fire—off real-time alerts and workstreams to intervene…including some automated actions like alerting specific care teams.

In-home Care Example

For example, a high-risk diabetes and chronic heart disease patient is in home care environment on monitoring devices.
Streaming data has alerted that several key conditions exist that warrant intervention.

  • The model recognized that the patient missed a dose of medicine, their weight has increased, O2 levels have decreased and the EKG is showing abnormal readings.
  • The actual location of the patient is known and verified in the monitoring real-time via GPS tracking.

The health system employees a proactive paramedic team who automatically dispatched to check on the patient all while a care team telemedicine intervention is initiated. They are able to prioritize their visit routes based on an optimization model for their care route. The prediction model indicates a re-admission will be necessary without intervention.

Fortunately, the care team is able to intervene and stabilize the situation.

  • This data also enables performance from current industry standards, in addition to the prediction of which patients are at risk of being readmitted and dates of highest risk.
  • Near real-time predictions can be automated, easy-to-understand, cross-continuum tool.
  • Recommended actions in the best interest of the patient.
  • Action can be taken.

It helps prioritize precious resources

  • Which patients do we focus on?
  • What do we do?
  • When do we do it?

Intelligent, Real-time, Actionable Analytics

Analytics has become vital to improve patient experience, care delivery, quality of care and patient safety. It is also key to optimize budget spend and use of scare resources, enhance business processes, and find and eliminate anomalies. All of these eventually translate to improved capacity and outcomes for the populations healthcare provider and payer organizations serve.

Healthcare delivery organizations can particularly benefit from analytics as it enables them to communicate, monitor and coordinate care more effectively in an era where patients and their caregivers are more informed and respond better to personalized clinical care.

Developments with 5G, new devices, genomics and personalized medicine will only accelerate the opportunity and need for AI and streaming-real-time data for healthcare at the speed of now— but truly at the speed of TOMORROW and the NEXT YEAR and so on.

We can work with healthcare delivery organizations, payers and vendors help provide better care as part of your efforts to achieve better population management, value-based care, personalized care amid very thin margins and strained resources. This re-admissions example is just one of endless opportunities that we can help you uncover and address.

To discover how you can apply streaming data technologies with machine learning and AI algorithms to create a powerful foundation for driving your real-time business decisions and improved healthcare and business outcomes, contact us.

References

https://www.commonwealthfund.org/publications/newsletter-article/focus-preventing-unnecessary-hospital-readmissions#

J. Benbassat and M. Taragin (2000) Hospital Readmissions as a Measure of Quality of Health Care. Archives of Internal Medicine 160, 1074–1081.

B. Friedman and J. Basu (2004) The Rate and Cost of Hospital Readmissions for Preventable Conditions. Medical Care Research and Review 61, 225–240.

Subscribe to get the Latest Updates

Enter your email address below to get the latest news and updates from RCG.