Skip to:
Personnel:
Harish Manchukonda (Principal Investigator)
Dr. Shahram Rahimi
Alexander Sommers
The goal of this work is the optimizing of the “flow” of patients through hospital systems. This is accomplished by quantifying the characteristics of when those patients show up, what their behavior is when they are admitted, and what changes can be taken to anticipate negative patient management outcomes. Our current focus is on the emergency department.
We have completed the following:
1. A Study of Emergency Department Patient Admittance Predictors.
In this work we introduced and compared two prediction systems on the task of replicating the human decisions regarding patient admittance in typical American emergency departments. For this purpose we used a dataset describing the patient trajectories in a 60,000 patient per-year emergency department in the United States. The data described the patients along many dimensions but those of prime importance were their severity of condition and the time they had been waiting to be admitted from the waiting room to the department proper. A recurrent neural network (RNN) was developed to learn the task of selecting the next patient, from the waiting-room/queue, to be admitted for treatment. This trained model was then compared to a heuristic-based selection algorithm currently used in industry for hospital simulation applications. We demonstrate achievable accuracies of 75.29% and 84.97% using the RNN, depending on the type of data preprocessing used. These accuracies are only potentially and theoretically achievable, respectively. The former’s validity hinges on whether or not certain “anomalous cases” are outliers or not, the second is achieved with the assumed existence of a method which flags these same cases as anomalous prior to becoming input for the RNN, which may or may not be achievable, pending further consultation with industry experts. What our results suggest hinges on whether or not such cases are outliers, though in either case a more sophisticated dataset is desired. If they are not outliers a superior dataset is likely necessary to apply machine learning, or at least our methods, meaningfully to this prediction problem for use in simulation or in real-world hospitals.
2. Predictive Analytics for “Left Without Treatment” in the emergency department.
When patients arrive in a hospital’s Emergency Department (ED) they are assigned an Emergency Severity Index (ESI) score that indicates the severity of their condition and the type of resources they may need. Patients with lower ESI scores are usually sicker, require more immediate service, and are more likely to require multiple resources. Patients who perceive their wait time as unacceptable may choose to leave the department before they are treated. Leaving the emergency department without receiving complete care is both a risk to the patient and a quality of care issue. This work seeks to produce a regression model that can predict the likelihood that a given patient will leave after having waited a specific amount of time in the emergency department. Such a model could be used to optimize the patient’s queue of an ED in an effort to minimize the likelihood of patients leaving without receiving care.
3. Predicting the patient arrival count based on weather parameters, calendar parameters, and Holiday Information.
We found that patient arrivals in the emergency room correlate meaningfully with a number of calendar and meteorological events, so much so that they can be used to anticipate approximate patient load and be used to recommend a redistribution of staff work hours to insure more personnel are on hand when needed, and less when the load is lighter.