AI in Length of Stay and Reimbursement

A hospital’s goal is not to minimize Length of Stay (LOS), but to optimize it at the intersection of quality and cost. By knowing the expected LOS up front, hospitals can better plan care, coordinate with partners, and line up discharge plans.
Michelle Wyatt, Senior Director of Clinical Best Practice, Xsolis

Diagnostic Related Groups (DRGs) are part of the financial fabric of hospital reimbursement. 

DRGs can determine how much hospitals will get reimbursed for patient care and set expectations for the appropriate length of stay (LOS) for each patient. DRGs are, and likely always will be, deeply connected to health system and hospital length of stay management.

How is the DRG determined? Unfortunately, final DRGs are calculated after the patient has left the hospital. Waiting until after the patient’s discharge disposition is set and reached creates two fundamental issues:

  • Geometric Mean Length of Stay (GMLOS) is not determined until after the patient is discharged.
  • Missing data that validates the DRG or associated comorbidities may not be detected until the patient has left the hospital.

What would it be like if clinicians had access to that information earlier in the stay? Could it help to improve discharge planning and decrease and manage length of stay?

Predicting a DRG

Predictive DRG (pDRG) is a machine learning model within Xsolis’ CORTEX platform, now known as Dragonfly, that combines management of LOS and AI for powerful results. It allows hospitals to receive and utilize the most probable DRG, and the corresponding length of stay, for a patient within 24 hours of admission. 

That means valuable insight to inform care and length of stay planning. The pDRG model moves discharge planning from a reactive workflow to a much more proactive one.

The pDRG engine suggests the three most probable DRGs and associated GMLOS for each patient, updating continuously and in real time. Machine learning can take new information into account and provide new insights for clinicians.

This isn’t only a real-time prediction of inpatient length of stay for discharge prioritization. Comparing and planning case expectations with DRG classifications also allows clinicians and care teams to ensure patient care is optimal, necessary, and timely. 

With pDRG, Xsolis offers a concurrent feedback loop for optimizing length of stay. That means better care for patients and more appropriate and timely discharges.

Prioritizing Near-Term Discharges 

As a new and added feature to the pDRG model, Xsolis has developed the Discharge Prioritization Scoring (DPS) Report. This tool serves as a method of prioritizing discharge planning by predicting the likelihood of near-term discharge. 

The DPS brings forth data and insights targeted toward helping prioritize and inform discharge planning workflows and managing length of stay. It moves past simple but useful concepts, like discharge planning checklists, to offer predictive insight that helps providers operate more effectively.

The DPS Report identifies patients who are likely to discharge within 24 hours using the p24 model. This model generates predictions about a patient’s likelihood of discharge and prioritizes the census based on that probability. 

Having this key information on a daily basis allows hospitals to: 

  • Prioritize discharge planning.
  • Focus on key data points to support decision-making.
  • Explore new and unique ways to utilize clinical information in Dragonfly to increase workflow efficiency.

The pDRG model empowers hospitals and health systems to achieve a number of positive objectives. Having an accurate prediction of LOS early on makes it easier to plan and provide care. This information also facilitates coordination with partners and effective discharge plans.

Want to learn more about the pDRG model and the DPS Report available only through Dragonfly? Read the fact sheet.

To explore how these models can help transform utilization management at your organization, schedule a demo today.