GE HealthCare has launched CareIntellect™ for Operations, a cloud-first, Software-as-a-Service application designed to help health systems expand patient access and enhance the patient experience while helping optimize staffing, capacity and throughput.
Health systems often have to adjust their operations to balance patient demand, staffing, bed capacity, and the availability of clinical services. Because many hospitals operate close to capacity, a relatively small constraint can cascade into broader disruption. For example, a delay in imaging can postpone a clinical decision, hold up a discharge, leave an inpatient bed occupied longer than expected, and delay a patient’s progression through the care pathway.
CareIntellect™ for Operations leverages the power of artificial intelligence to help optimize hospital operations and harness the insights embedded in hundreds of patient-specific and operational data points. The application predicts where bottlenecks are likely to develop and recommends actions that can help teams respond earlier.
CareIntellect™ for Operations fits with day-to-day hospital operations by leveraging data in real time through integration with existing systems, including the electronic medical record and resource-management systems, and providing an hour-by-hour forecast of expected patient flow bottlenecks up to 72 hours into the future.
The application predicts multiple operational factors associated with the availability of and demand for beds, staff and resources, along with wait times across departments and supporting services.
CareIntellect™ for Operations is powered by machine learning techniques with two proprietary AI models front and center. The Pressure Forecast model predicts where operational strain may emerge across the hospital. The Estimated Day of Discharge model predicts when individual patients are likely to be discharged. Together, they connect hospital-wide capacity planning with patient-level care coordination.
Pressure forecast
Hospitals rarely experience operational pressure in isolation. A surge in emergency admissions can create boarding in the emergency department, increase demand for inpatient beds and produce longer queues for imaging or procedural recovery. These effects may develop over several hours, often before leaders have assembled a complete view of the problem. Operational teams may spend substantial time collecting data from separate systems and deciding which constraint requires attention first.
What the model predicts
The Pressure Forecast model predicts factors that can indicate emerging strain at the hospital, department and unit levels. It covers inpatient and observation units, the emergency department, imaging, procedural recovery, physical and occupational therapies, and incoming transfers.
The factors that vary by department can include patient census, available beds, staffing ratios, emergency department boarding, imaging turnaround times and therapy consultation delays. In an imaging department, a related factor might be the time between an order and completion of a scan. In an inpatient unit, relevant measures might include census and the gap between required and available nursing resources.
The model forecasts these measures up to 72 hours in advance. When a factor is expected to move beyond its normal operating range, it can serve as an early indication that a department or unit may face strain, allowing hospital leaders to get a handle on potential operational pressure before it happens.
How the forecast works
To keep the predictions rooted in the day-to-day realities of the hospital, CareIntellect™ for Operations compares each pressure factor with the hospital’s recent operating history. It then converts different measurements into a common pressure scale and weighs them according to their operational importance.
The model uses N-BEATS, a deep-learning architecture developed for time-series forecasting. A time series is a sequence of measurements collected at regular intervals, such as the number of occupied beds recorded every 15 minutes. The model learns recurring patterns in those measurements and uses them to estimate how each factor is likely to change.
For every pressure factor, the model studies 14 days of historical data captured at 15-minute intervals and generates predictions for the following 72 hours at the same level of detail.
CareIntellect™ for Operations uses the forecast to identify the factors behind current or expected pressure, and connect them with rules-based recommendations. For example, a surge in patients admitted to the emergency department will prompt the application to identify discharge candidates, surface their barriers and support reprioritization of service queues accordingly. Recommendations are updated as admissions, procedures, staffing and patient movement change, enabling hospital personnel to review the suggestions before acting.
However, hospital operations can’t be viewed exclusively at a holistic and abstract level. Individual patients are the heartbeat of hospital operations. A hospital-wide forecast still depends on what happens to individual patients. A bed’s “open” status is dependent on whether the patient occupying it can complete the final steps of discharge.
Estimated day of discharge
A patient discharge may depend on laboratory results, pending consultations, a final imaging examination, whether transportation is available to take them home, home health arrangements or an appropriate placement in another care setting. When teams lack a reliable estimate, case managers may struggle to decide which patients need priority discharge planning, beds may remain occupied longer than necessary and admitted patients may continue waiting in the emergency department.
Here’s where the Estimated Day of Discharge model comes in.
What the model predicts
The Estimated Day of Discharge model estimates when each inpatient or observation patient is likely to be discharged. It calculates the probability of discharge across several future periods, in the coming days or over the longer term. Predictions are updated hourly as new patient information becomes available.
This forecast is designed to help operations leaders, care managers, discharge planners, nursing leaders and patient-flow coordinators identify which patients are likely to be ready for discharge soon, and the planning that needs to be prioritized.
How the prediction works
The model analyzes a patient’s longitudinal medical record, meaning the sequence of clinical events recorded over the course of care. It uses structured information including age, admission class and source, prior encounters, laboratory results, procedures, diagnoses, imaging and other orders, medications, transfers and previous discharge information. These events are arranged over time so the model can assess both recent activity and the broader course of the hospital stay.
The underlying model is adapted from BERT, a language-model architecture developed to understand relationships between words within a sequence. In this case, the sequence consists of clinical events rather than words. The model examines how orders, procedures, test results, transfers and medications unfold over time and identifies patterns associated with when similar patients were discharged.
To accommodate patients with lengthy medical records, the model uses a sliding-window approach. This gives greater weight to recent events while retaining enough earlier information to preserve the overall trajectory of care.
The Pressure Forecast and Estimated Day of Discharge models approach hospital capacity from different directions. One tracks the demands placed on hospital resources, while the other estimates how eligible patients may be moved along the care journey. CareIntellect™ for Operations brings those forecasts into the same operational view, alongside current data and recommended actions.
For health systems operating near their limits, these predictions are targeted to give leaders time to align people, beds and services before a local constraint becomes a hospital-wide disruption. If hospitals can get a clearer view of what is coming next, their teams can be better equipped to act earlier and help keep care progressing through an increasingly complex system.
