The power of prediction: how AI can help hospitals forecast and manage patient flow

For hospital leaders tasked with managing unexpected surges in patient demand, the ability to anticipate and adapt to rapidly changing circumstances has become more essential than ever. What if we could predict potential bottlenecks in patient flow in real time – and prevent them before they occur?

While the pandemic has put critical care capacity under the spotlight like never before, hospitals around the world have long faced challenges with bed and staffing shortages to meet demand for acute care. Emergency departments (EDs) in many countries struggle with overcrowding even under regular circumstances. Intensive care units (ICUs) may be operating at or near capacity. All too often, waits and delays are the result – causing frustration, anxiety, and potentially harmful outcomes in patients, while adding to the pressure for staff [1].

It can be tempting to think that the solution lies in adding more beds or more staff. But typically, the problem is not merely one of resources. It’s also about better managing the beds you have. The real challenge is often one of patient flow: anticipating and knowing when to transition a patient from one care setting to the next.

It’s a highly complex and dynamic orchestration challenge, with many moving parts. Which patient waiting in the ED should get the next ICU bed? Which patient in the ICU can I safely move to a step-down unit to free up a bed? And who is ready to be discharged for home monitoring?

Managing patient flow requires an enterprise-wide view across different parts of the hospital or hospital network. However, that’s often precisely what’s missing today. With clinical and operational data dispersed across disparate systems, care teams lack wider situational awareness beyond their unit or department. It’s this lack of readily available and actionable information that can hamper patient prioritization, slow down patient transitions, and lead to unforeseen bottlenecks in patient flow.

The COVID-19 crisis has exposed and exacerbated many of these challenges. But it has also given rise to smart ways of tackling them. Healthcare providers have embraced centralized care collaboration models, sharing data in real time to visualize untapped capacity and facilitate patient transfers. And they’re not just relying on that data to get an overview of what’s happening from moment to moment. They’re also using it to forecast and prepare for future demand. For example, hospitals have successfully used predictive models to estimate the number of beds, equipment and staff needed for COVID-19 patients in the ICU and other hospital wards [2,3].

As we begin to think beyond the pandemic, there’s a unique opportunity to embed these data-driven practices into the everyday management of patient flow – from hospital admission all the way to hospital discharge and, ultimately, monitoring in the home. Using the power of AI and predictive modelling, we can extract relevant patterns and insights in patient flow and patient care needs from vast amounts of real-time and historical hospital data. After initial validation, the resulting algorithms can be updated on a regular basis to take recent trends and circumstances into account, thereby further optimizing predictive value. This enables hospital leaders and patient flow coordinators to orchestrate care more effectively across settings and rapidly adapt to changing circumstances.

Here’s what that may look like for one patient’s journey.

Once Rosa has received critical care in the ICU to help stabilize her condition, Jennifer can already start planning ahead to facilitate Rosa’s care journey. Smart algorithms support Jennifer in estimating when Rosa will be ready to be transferred to a lower-acuity care setting in the hospital for telemetry monitoring. Based on a transition review list that is updated dynamically, Jennifer can support physicians in prioritizing clinical evaluation of patients who may be ready for transfer. At the same time, she has an up-to-date overview of available telemetry monitoring. She can also see how many patients in the ED are waiting for an inpatient bed. This helps Jennifer identify potential bottlenecks early and manage patient flow accordingly.

Upon admission to a step-down unit where she is placed on telemetry monitoring, Rosa remains under the caring eye of clinical staff – with predictive algorithms again helping Jennifer orchestrate next steps proactively. Based on an analysis of physiologic deviations and alarm trends over the last 12 hours, Jennifer is able to evaluate when Rosa’s condition is stable enough for her to be considered for transfer to the medical-surgical unit.

We now come to the end of Rosa’s in-hospital care journey. As she’s being monitored in the medical-surgical unit – her last stop on the way to hospital discharge – Rosa’s discharge readiness score and readmission risk indicate that she is physiologically stable and well on her way to recovery. After the responsible physician has reviewed her condition, Rosa receives the reassuring message she has been waiting for. She is ready to go home.

Continuous adaptation is key here. Unlike some other industries such as manufacturing, where processes follow a fixed and predetermined sequence, healthcare is a human endeavor where unexpected events happen. For example, a patient like Rosa may not respond to her treatment plan as anticipated and need to be readmitted to the ICU for further intensive care. Predictive algorithms can help to pick up early signs of deterioration, but in-the-moment clinical judgment remains essential in assessing a patient’s condition and deciding what to do next.

That’s why central command centers in healthcare are not about top-down management. Rather, they should support collaboration among physicians and care administrators across the hospital or hospital network. The future of care will be all about predictive insights informing networked decision-making, with physicians and staff remaining in charge of clinical decisions, supported by a central hub that oversees the bigger operational picture to proactively manage patient flow.

References
[1] Rutherford PA, Anderson A, Kotagal UR, Luther K, Provost LP, Ryckman FC, Taylor J. Achieving Hospital-wide Patient Flow (Second Edition). IHI White Paper. Boston, Massachusetts: Institute for Healthcare Improvement; 2020. http://www.ihi.org/resources/Pages/IHIWhitePapers/Achieving-Hospital-wide-Patient-Flow.aspx
[2] Mayo Clinic COVID-19 Predictive Analytics Task Force, Pollock BD, Carter RE, et al. Deployment of an Interdisciplinary Predictive Analytics Task Force to Inform Hospital Operational Decision-Making During the COVID-19 Pandemic. Mayo Clin Proc. 2021;96(3):690-698. https://www.mayoclinicproceedings.org/article/S0025-6196(20)31482-8/fulltext
[3] Weissman GE, Crane-Droesch A, Chivers C, et al. Locally Informed Simulation to Predict Hospital Capacity Needs During the COVID-19 Pandemic. Annals of Internal Medicine. Volume 173, Issue 1: 21-28. https://www.ncbi.nlm.nih.gov/pubmed/32259197
[4] Foley M, Kifaieh N, Mallon W. Financial Impact of Emergency Department Crowding. Western Journal of Emergency Medicine, Volume XII, no. 2: May 2011, 192-197. https://pubmed.ncbi.nlm.nih.gov/30193738/
[5] Reducing hospital readmissions with integrated COPD care. Philips case study. [Results are specific to the institution where they were obtained and may not reflect the results achievable at other institutions.] https://www.philips.com/a-w/about/news/archive/case-studies/20191001-reducing-hospital-readmissions-with-integrated-copd-care.html