Unlock the full potential of bed capacity with AI

Every day, hospital staff do their best to navigate the daily chaos of inpatient care by making educated guesses about what’s going to happen during the day. Relying on the team throughout the day, staff look at Excel or paper spreadsheets to predict how many beds will open and when. They are trying to estimate the demand for these beds at the time of day, unsure of when to deploy “surge capacity.” Some days this method works well. But more often than not, staff’s best efforts result in long patient wait times, unwanted staff overtime and, ultimately, lower access to care.

The problem with traditional bed care is that the usual approach of using spreadsheets to get a periodic reading of patient flow, then trying to unlock capacity by discharging patients more quickly, simply doesn’t work. It requires sophisticated algorithms and real-time predictive and prescriptive analytics to shape demand, successfully match bed availability, place the right patient in the right bed at the right time, and identify and address discharge barriers.

The return on improving patient flow

Beds are a significant financial investment – a single bed can be worth $10,000. Keeping a steady flow of patients in and out of beds is a difficult but very important element in the overall management and efficiency of a hospital. It is also the core of providing a positive patient experience.

Historically, health systems have invested extensive resources to improve patient flow and reduce length of stay. Avoidable days, or the number of days a patient remains an inpatient even though he/she/they is medically ready for discharge, can cost a hospital thousands of dollars each month. Avoidable days usually occur due to an avoidable delay, such as failing to secure necessary durable medical equipment (DME) for a patient after discharge or failing to secure a room in a post-acute care facility (ie, skilled nursing facility (SNF) or rehabilitation) for the patient to come in when he/she/they are no longer hospitalized. Addressing these barriers as early as possible in the patient stay, so that the patient can be discharged to the next step in the care journey at the right time, is essential to avoid long and costly delays and turn over new beds.

If hospitals invest in the right user-friendly tools, supported by a predictive and prescriptive analytics engine combined with the use of Natural Language Processing (NLP), they can proactively identify and address discharge barriers earlier, streamline patient flow and ultimately improve patient outcomes and the bottom line. Here are four opportunities to improve bedside treatment with AI and NLP-based analyses:

1) Using sophisticated supply and demand models to allocate patient beds

The best way to place patients is to accurately predict and match supply and demand – on a unit-by-unit, minute-by-minute, day-by-day level – every day. Similar to how apps like Waze take baseline predictions from traffic speed for each section of road for every minute of every day of the week, solutions are now available that model current and future bed availability in each device. Supply and demand must be approached in different, yet compatible, ways.

Supply: Model the availability and timing of beds that will become available in each unit. By using historical data to mathematically create a “fingerprint” (a model for each unit that predicts the likely number of patients who will be discharged), hospital staff can make specific placement decisions about individual patients. Since the predictions are augmented with real-time feeds, these decisions will be more accurate and less speculative.

Demand: Similar to the supply side, create specially tailored models for “upcoming demand signals” at any time of the day for each element of demand. These elements can include various factors, such as incoming volumes from surgical and emergency departments, as well as external transfers. Models can be updated by real-time feeds that capture any delays or cancellations of operations to ensure up-to-date accuracy.

Side-by-side supply and demand models can then be elevated to patient placement managers, giving them visibility into upcoming demand and supply for beds. This leads to dramatically better results than a system based only on reaction.

2) Make data-driven internal transfer decisions

Internal bed transfer requests are often seen as an additional burden, pushed to the side to be carried out only when convenient. However, transfers can actually serve as a strategic lever since they can free up a bed that will be needed in the near future. Using the predictive modeling tool described above, plus moving the right patients to the appropriate open beds, placement teams can open up the right openings to meet the expected demand for high-value beds.

3) Forecasted demand with surgical equalization

On any given day, 20-25% of the bed requirement is the flow of patients from the operating theater to beds. This often results in peaks in the patient count. Surprisingly, this flow is actually more “controllable” than the ED count contribution, as optimizing the elective surgical plan with respect to recovery time can produce a flatter admitted count. The practice of “surgical leveling” can be done by predicting the volume and case mix of surgeries, using AI-based tools to develop templates for planning.

4) Use predictive discharge planning to focus case teams and social services

The most common discharge delays occur towards the end of a patient’s stay – typically around insurance, transport, follow-up of outpatient or home care, or if necessary availability at skilled nursing facilities (SNF) or extended care facilities. This last discharge barrier has become particularly challenging in recent years, as SNFs suffer from severe post-pandemic operational and staffing constraints.

Many discharge delays can be avoided if case managers were notified of the problem earlier in the patient’s stay. Historical data on avoidable discharge delays can be collected and a machine learning model can be used to identify key case attributes that indicate possible discharge delays early on. The most powerful solution also uses NLP, which makes this information actionable for all members of the care team and unit staff, so they can proactively remove barriers and optimize patient flow. This helps to reduce costly avoidable days and gives patients a smoother treatment journey.

Each of these pillars of bed management is critical when it comes to improving processes and resource utilization. While each health system will have different criteria and contributing factors for proper bedside care and patient outcomes, and while new tools require learning and patience, the investment is definitely worth it.

About Sanjeev Agrawal

Sanjeev serves as President and Chief Operating Officer of LeanTaaS, the leading AI/ML analytics company in healthcare. LeanTaaS’s predictive analytics software powers over 130 health systems and 500 hospitals to improve access and reduce costs.

Sanjeev is also the co-author of the book “Better Healthcare Through Math”. Prior to LeanTaaS, Sanjeev was Google’s first Chief Product Marketing Officer and led three successful startups – CEO of Aloqa (acquired by Motorola), VP of Product and Marketing at TellMe Networks (acquired by Microsoft) and Founder and CEO of Collegefeed (acquired by AfterCollege) .