|
(This work has been reported in conjunction with
Andrzej Ceglowski (Faculty of IT), and Jeff
Wassertheil (Faculty of Medicine/Peninsula
Health))
Healthcare systems the world over are
experiencing difficulty in providing accessible,
timely and cost-effective treatment of patients.
Hospital emergency departments, a vital link in
the healthcare system, are experiencing problems
symptomatic of the whole. Patients arrive at
emergency departments at any time of the day
and expect to be treated quickly and effectively.
Any blockage in flow through the emergency
department directly affects patient treatment and
is quickly propagated upstream in the supply
chain, impacting upon ambulance services and
other hospitals in the region. Since a large TimesNewRomanproportion of hospital admissions arise from
emergency department presentations, the flow of
patients through an emergency department has
significant downstream impact on hospital
resources.
Operational Research seeks to play a part in
solving the problems that beset hospital
emergency departments. While some progress
has been made in modelling emergency
departments and valuable changes implemented
as a result of insights gained from these models,
understanding of resource utilisation in
emergency departments remains limited.
Existing models of emergency departments tend
to focus on non-core aspects of emergency
department operations such as patient flow or
length of stay, contributing little to understanding
of the root causes of emergency department
blockage.
A change of context for emergency department
modelling from the pervasive „maximising
patient throughput¾ to „effectiveness of patient
treatment¾ provides a radical revision of
hypotheses. One hypothesis derived from
„effectiveness of patient treatment¾ would be, „It
is possible to identify similar treatment
activities¾. Surely patients who receive similar
treatment will have similar needs, follow similar
paths through the emergency department and
require similar resources? Such a hypothesis
elicits a model context that works towards
understanding and segmentation of patient
treatment, not necessarily from a purely clinical
perspective, but rather from a resource
consumption process perspective. If such a
model could be built it would contribute to real
understanding of emergency department
operation and, ultimately, to improved
operational decision making.
This talk describes the beginnings of such a
model. It outlines how the use of data mining as
a knowledge discovery tool supported model TimesNewRomandevelopment. It describes how existing data was
explored and ultimately yielded a feasible
segmentation of the hugely diverse patient
treatment data into a manageable set of
¾averaged¾ treatments. This segmentation of
treatment has been repeatedly tested and found to
be sound. Expert medical opinion has been that
they are clinically defensible. Patient presenting
problem and treatment cluster have aligned in
numerous analyses. The logic inherent in the
segmentation has been confirmed by a search for
rules within patient treatment and linkages have
been revealed between patient demographics and
treatment.
This non parametric segmentation, while yet
imperfect, has provided a large number of
insights into the workings of the emergency
department. These insights have built knowledge
about emergency department operations that
provided a unique starting point for a range of
applications, including:
- Guidance for future data gathering
exercises;
- Treatment based Monte Carlo cost
models of the emergency department;
- Validation of both data entry and data
interpretation within and across multiple
emergency departments;
- Assistance to both qualitative and
quantitative process modelling methods;
- A discrete event simulation of the
emergency department that is showing real
promise in providing a resource optimisation
capability;
- Predictive decision support that can raise
emergency department information systems from
a reactive work sequencing role to a proactive
one.
A flavour will be imparted with the intention of
emphasising that freely available data should not
be dismissed out of hand because of concerns
about „data-led¾ modelling. While there are
dangers inherent in data-led modelling, existing
data can support model development appropriate
for MS/OR.
| 