Page 239 - Data Science Algorithms in a Week

Managing Overcrowding in Healthcare using Fuzzy Logic 223

increased accordingly. In other observations such as 8 and 9, when factor scores such as
workload increased while another factor such as boarding decreased, the resulting
crowding score exhibited no change. In observation 21 when other scores exhibited
minimal change, a sharp increase in the demand score can be attributed to the sharp
increase in crowding, demonstrating the significance of the role of demand in crowding.
The agreement between GIEDOC and expert assessment is analyzed in Table 11,
where assessments are documented according to the “low”, “medium”, and “high”
classes (2, 3, and 4) from Table 11. The GIEDOC issued 4 assessments for “low” scores,
15 for “medium”, and 5 for “high”, while the expert provided 3 “low” assessments, 13
“medium”, and 8 “high”. For the low class, the GIEDOC and the expert issued the same-
assessment agreements twice, while they agreed eleven times for the medium class, and
five times for the high class. When measured against the expert assessments, the
GIEDOC overestimated once for the low class, (providing a score of “medium” where
the expert provided a score of “low”), and underestimated the medium class twice
(providing “low” while the expert provided “medium”), while underestimating the high
class three times. It should be noted that the insignificant and extreme classes could not
be predicted, as the ED during this study was neither empty nor extremely overcrowded
according to both scores from the expert and the GIEDOC. Most activity regarding the
major operation factors occurred in the third level or “medium” class according to their
scores.
The Kappa value found for the system was 0.562, 95% CI [0.45, 0.66], which
indicates moderate agreement between the objective and subjective scores of GIEDOC
and the expert.

CONCLUSIONS AND FUTURE WORK

This study proposed a framework for quantifying overcrowding within different
healthcare contexts, seeking to overcome the shortcomings of previous indices by
founding the framework upon the perspective of multiple experts and stakeholders. With
a method for quantifying overcrowding in qualitative and quantitative terms provided by
a variety of experts, and identifying and reducing bias, this study strives for
reproducibility of results in other settings.
With regard to the design of the fuzzy system, future research could focus on either
increasing the number of inputs to the system, or identifying more crowding
determinants. Other design improvements could include an expansion of the hierarchical
fuzzy system, in which more subsystems could be implemented in association with other
identified inputs or determinants of crowding. In designing the knowledge base, further
research could attempt to integrate other quantitative tools into the fuzzy system to
process some inputs independently, such as patient demand. Methods such as simple