'The Ferrari of data science': 7 hospital execs share how they use predictive analytics

  • Small
  • Medium
  • Large

Artificial intelligence-powered predictive analytics tools can improve hospital operations by optimizing capacity demands, alerting care teams of patients who may be at risk for adverse events and anticipating supply shortages, among other uses.

Here, seven hospital executives share the most important way their health system uses the technology.

Editor's note: Responses have been lightly edited for clarity and style. 

Atefeh Riazi. CIO at Memorial Sloan Kettering Cancer Center (New York City): MSK is using predictive analytics in so many areas. One is to identify patients who are at high risk for certain clinical events — including a malignant airway obstruction, a pelvic fracture from bone metastasis, or a cancer-associated thrombosis — so as to suggest prophylactic interventions. MSK is also applying it in digital pathology to identify and quantify various cancer biomarkers for targeted therapy. Finally, in the radiology space, we are integrating AI models into clinical workflows, such as lymphoscintigraphy examinations or the diagnosing of neuroendocrine tumors.

Allen Hsiao, MD. Chief Medical Information Officer at Yale New Haven (Conn.) Health: Predictive analytics is one of the most enticing promises of the digitization of healthcare data. At Yale New Haven Health and the Yale School of Medicine, we have implemented numerous algorithms to predict readmissions, emergency department volumes, sepsis, acute kidney injury, congestive heart failure, deterioration and one-year mortality as well as modeling COVID-19 patient volumes, length of stay and discharge readiness. Some are homegrown, others are open source or commercially available and customized to our population.

We are also studying the impact of predictive models to determine whether they actually work to change outcomes.  

Zafar Chaudry, MD. Senior Vice President & CIO at Seattle Children's: Our chief data officer has led our organization to HIMSS Adoption Model for Analytics Maturity Stage 7. In the predictive analytics space, our team of data scientists has developed models to predict census and the capacity needed to support it, such as tools to eliminate surgical cancellations, shift from diverts to directed placement for those kiddos that are diverted are lower acuity, improve capacity for the operating room, and eliminate clinical processes like surgical huddles. 

Another model is used to predict whether demand in the ED is going to be busier than planned over the next couple hours; this allows the ED to staff up ahead of this unexpected demand. On the clinical side, we have developed our own models to predict sepsis and deterioration, and models for pediatric sequential organ failure assessment.

Mike Seim, MD. Chief Quality Officer at WellSpan Health (York, Pa.): At WellSpan Health, we know the application of predictive analytics, if done right, can be a critical tool in identifying patients at high risk for a rapid decompensation in their condition and who may need to be transferred to a higher level of care. To improve safety, our clinical teams engaged with technical and informatics experts to review that data and recommendations, resulting in the development of evidence-based treatment/care pathways and interventions embedded within the EHR for adoption and spread across our hospitals. 

We had great success implementing this strategy in identifying patients with sepsis resulting in a significant decrease in mortality. We have also implemented a similar strategy to measure SF ratios in our COVID-19 patients to identify patients at highest risk for developing acute respiratory distress syndrome. When patients are identified as having a decreasing SF ratio, we are able to begin to prone patients earlier in their disease course. We've seen improved outcomes and a decrease in emergency transfers to the intensive care unit directly related to this sort of predictive analytics and that's why we're leaning into it at WellSpan.

​​Sunil Dadlani. CIO at Atlantic Health System (Morristown, N.J.): Predictive analytics aren't just an integral part of the strategy for IT at Atlantic Health System — they are critical to the way healthcare is delivered today and going forward. At Atlantic Health System, we are using cutting-edge AI and machine learning-enabled predictive and prescriptive analytics across the organization to help us better tailor care throughout our specialties and service lines.

This helps us on all levels, from individual patients — predicting which patients are likely to develop adverse events such as congestive heart failure or be readmitted, or forecasting appointment no-shows — to broader, population health-level considerations, like predicting which counties and ZIP codes will likely see surges in COVID-19 patient volumes, or risk-scoring for chronic diseases.

These analytics can also help us predict supply chain and logistics bottlenecks, so that we can prepare and create contingency plans, ensuring that the quality of our care is never interrupted.

Deb Muro. CIO of El Camino Health (Mountain View, Calif.): As a healthcare industry, we have focused for years upon the creation of EHR platforms in support of interoperability. The outcome of automating vast amounts of patient information has resulted in expansive and at times unwieldy "data lakes" fraught with challenges to bring value or drive actionable results. It is important now, more than ever, to realize the benefit of the tremendous investments in data automation to harness the power of information to predict and forecast aspects of the patient experience and care.

Our first foray into predictive analytics has involved the use of algorithms capable of ingesting real-time patient data to forecast expected changes in patient condition and status. If a patient's condition is expected to degrade in a defined time frame, a rapid response team is notified to intervene appropriately, which has resulted in a decrease in adverse patient events. We have only scratched the surface with these capabilities as the use cases are endless for utilizing data to predict what will happen within one hour, four hours or 24 hours regarding patient events, volumes, staffing and supply chain needs. The time is now to disrupt the healthcare experience in positive and impactful ways.  

Scott LaRosa. Executive Director of Enterprise Analytics and IT at Southcoast Health (New Bedford, Mass.): Predictive analytics is ultimately the payoff of clean workflows and accurate/aligned data alongside an advanced, data-literate culture that is prepared to understand and respond to such information. While predictive analytics may seem to be the Ferrari of data science, we are choosing to start with fundamentals of data governance, data stewardship and overall data use education. 

We have seen predictive analytic tools fail due to a canned approach to a regionalized issue, such as readmissions. The logic must match the need. The way this happens is by including key C-suite executives for support of the logic behind the build prior to launching the tool itself for end users to make decisions from. 

Where we have seen value, once we have this level of support, is when we mix predictive analytics with real-time reporting and day-to-day workflows, such as by mixing a predictive no-show metric on an existing ambulatory schedule that shows today's appointments. By doing this, we have the ability to target which patients should have a more aggressive call/contact approach to be sure they remember their visit that day, and the busy provider schedule stays on track.  

Culture, executive support, workflow and education are all key aspects that should come before implementing a predictive tool at any organization.

Copyright © 2021 Becker's Healthcare. All Rights Reserved. Privacy Policy. Cookie Policy. Linking and Reprinting Policy.


Featured Whitepapers

Featured Webinars