Abstract

Have we MedB4 ? Estimating future medical services demand using a machine learning heuristic model

Author: Adam Strier MD/Bnai Zion Surgery department, Haifa Israel
Ronen Tal-Bozer PhD/ Sense/ Medical Bioinformatics-Bar-Ilan University, Israel
Ibrahim Mattar MD/ Head of Bnai Zion Surgery department, Haifa Israel
Ron Unger PhD/ Associate Professor, Head of the Biomedical Informatics Program at Bar-Ilan

Background
As medical personnel, we seldom have a good estimate for the next day’s requirement. Biology dictates large fluctuations in sickness and health patterns over time and space, with pandemics hitting record high occupancy throughout emergency departments in winter, in complete contrast to the silent wards on holiday eve.
We present a novel concept that
A. Assists medical managements in predicting medical needs for medical services providers. Estimates of needs are made for medical supplies, on-duty and on-call medical and paramedical personnel - on a daily, weekly, monthly or yearly basis.
B. Performs online pattern recognition of clusters of patient complaints/symptoms/diagnoses, and alerts when deviations from expected patient loads are identified, allowing recognitions of epidemics, natural disasters, air pollution events, massive food poisoning, etc.
C. Future expansion can be forecasted - in patient income and expenses, training needs for future medical teams, planning of temporary and permanent staff re-allocations, and making estimates of projected efficiency and cost-effectiveness of medical teams and services
Methods:
The system uses supervised learning to establish presets for an Artificial Neural-Network algorithm as the system’s core function – future estimation from several sources of data.
As a modular, scalable, and easily-integrated system, MedB4 servers interconnect with the clinic or hospital medical systems, including Patient visits to the ER and admissions to medical and surgical wards. The system is inter-connected to the patient database holding past and present patients’ data, labs, imaging, OR reports, as well as to Human Resource management servers.
The system receives input from the weather forecast, the weekly, monthly and yearly calendar, including data from same day last week, month, and year, data from air pollution sources, and from live news, identifying local news affecting medical requirements (e.g. mass casualty events).
Offline time intervals will be used for last estimate correction and re-calibration using back-propagation and unsupervised learning.