Page 237 - Data Science Algorithms in a Week

Managing Overcrowding in Healthcare using Fuzzy Logic 221

FRAMEWORK IMPLEMENTATION AND VALIDATION

This section details the process for implementing and testing the accuracy of the
proposed fuzzy model framework, which will be described as the Global Index for
Emergency Department Overcrowding, or GIEDOC. One of the main goals of the
GIEDOC is to produce reliable results which can be reproducible in EDs of other
healthcare systems. The design of the GIEDOC accounts for this in the knowledge base,
as ten healthcare experts from a nation in question may provide data to be fed into the
knowledge base, allowing the fuzzy system to produce results. This is why the design of
GIEDOC is unlike other developed indices, which when tested outside their countries of
origin, do not show adequate reproducibility when implemented. In order to accurately
assess the GIEDOC, it must be implemented in real ED environments to measure the
level of crowding, and at the same time, an expert assessment of a native expert must be
made of the same environment to compare the results from the GIEDOC.
For the purposes of measuring the accuracy of the GIEDOC, five classes within the
GIEDOC were defined by five equal intervals on a scale from 0 to 100, so that the classes
could be compared to the subjective assessment of experts. These five classes for
assessing ED crowding on five subjective levels were: 1 for “insignificant”, 2 for “low”,
3 for “medium”, 4 for “high”, and 5 for “extreme”. In other words, this was done to
compare the agreement of the index to experts, by determining if this scale reflects the
expert perspective for crowding. The GIEDOC was implemented for three days in a
public Saudi Arabian hospital in Jeddah, which sees more than one hundred thousand
patients in its emergency department on a yearly basis, possessing more than 400
inpatient beds and 42 emergency beds. During the validation, twenty-four observations
were made to collect data which focused on factors including the capacity of the
emergency department, the number of patients in the waiting area, ER, and boarding
areas, the number of present physicians and nurses, the average patient complexity in
both the waiting area and the ER, and finally a healthcare expert’s subjective assessment
of crowding. These results are detailed in Table 11, where the ED crowding level scale
can be compared to class number assigned by experts Kappa analysis was used to test the
agreement between the computed GIEDOC scores and the subjective assessment of the
healthcare experts. These statistics allow for the comparison of the accuracy of the results
from GIEDOC to those of other indices when assessing ED crowding.
Table 11 provides the data obtained from the twenty-four observations conducted for
validation of the GIEDOC, resulting in calculated scores for the major operational
factors. The demand scores ranged from values of 8 to 61.4 according to the demand
indicator of the GIEDOC, while staffing scores ranged from 50 to 85.1, and ED workload
ranged from 33.33 to 89.2. It should be noted that the majority of staffing scores obtained
their maximum values, indicating that over the three days of validation, the selected ED
almost always maintained adequate staffing. There was higher variation in the range of