The following app is a proof of concept using the patient flow machine-learning model to predict root cause for the 810 patient test data set. It is a working prototype running on a iPhone with surprising accuracy.
If you are part of an organization that is experiencing long delays in patient flow and are interested in learning more about this machine-learning model please contact us.
Patient flow in a hospital is an extremely complex and intertrelated process with many critical factors impacting it. Identifying root cause for delays requires a multi-discipline team of physicians, nurses, care management, transport, housekeeping, registration, finance and information technology. Analyzing and documenting delays is a time consuming and manual task.
However, the investment in time can be leveraged by a machine-learning model to predict future bottlenecks that will improve key performance indicators such as left without being seen (LWBS) in the ED, lower the length of stay (LOS) for inpatients and increased patient satisfaction. Improving patient satisfaction is hard to measure but invaluable. Improving LWBS and LOS translates to better patient care and increased revenue.
The process begins when a patient presents in the emergency department and ends when they are checked into an inpatient bed. The figure on the right is a list of the steps in the process.
The DayOfWeek field is a calculated value based on the date the patient presented. This value is used to determine if there is a correlation based on the day of the week. All other fields are captured timestamps from clinical applications.
The machine-learning model was trained to predict where the bottleneck is in the 20-step patient flow process outlined above.
The dataset contained approximately 1155 records, 345 were used to train the model and 810 are used to test the model.
The model was built using the logistic regression algorithm. Logistic regression is a supervised learning algorithm that learns to associate an input with and output. The output is difficult to collect automatically and usually requires humans (or “supervisors”) to provide the output.
The accuracy of the model was improved from 77.1% to 99.71% by standardizing the 20 variable inputs to the ML model as double precision values and increasing the maximum iterations from 10 to 50. Going beyond 50 iterations yielded no additional improvement.
The model was built in under a minute, is 131K bytes in size and will run on an iPhone with sub-second response.
One of the key lessons learned from this project is the relationship between the amount of data needed to train the model and the accuracy. More training data improves model accuracy with a higher probability that the prediction is correct.