Enhancing Healthcare in Emergency Department Through Patient and

External Conditions Proﬁling: A Cluster Analysis

Mariana Carvalho

and Ana Borges

CIICESI, Escola Superior de Tecnologia e Gest

ao, Polit

ecnico do Porto,

Rua do Curral, Casa do Curral, Margaride, Felgueiras, 4610-156, Portugal

Keywords:

Patient Proﬁling, Data Mining, Clustering, Healthcare, Emergency Departments.

Abstract:

Improving healthcare delivery in emergency departments (EDs) is of paramount importance to ensure efﬁcient

and effective patient care. This study aims to enhance healthcare in the ED by employing cluster analysis

techniques to proﬁle patients and external conditions. Through a comprehensive analysis of patient data and

factors associated with the ED environment, we seek to identify patterns, optimize resource allocation, and

tailor interventions for improved outcomes. The identiﬁcation of distinct patient proﬁles and understanding of

the impact of external factors allows to understand the complex dynamics of the ED. Additionally, it enables

healthcare professionals to better understand patient populations, anticipate healthcare needs, and tailor treat-

ment plans accordingly. Therefore, in this paper, we apply a clustering technique to obtain three clusters with

different characteristics, both at the patient level and at the level of external factors, with different emergency

room inﬂows.

1 INTRODUCTION

The Emergency Department (ED) plays a crucial role

in providing immediate and life-saving care to pa-

tients in urgent medical situations. As the demand for

emergency healthcare services continues to rise, there

is a pressing need to optimize ED operations and en-

hance the quality of care delivered. To achieve these

goals, it is essential to understand the complex inter-

play between patient characteristics, external condi-

tions, and healthcare outcomes within the ED envi-

ronment.

In the dynamic environment of an ED, the ability

to understand patient characteristics and effectively

allocate resources is of paramount importance. Proﬁl-

ing patients in the ED provides healthcare providers

with invaluable insights into patient demographics,

arrival patterns, and accompanying factors, ultimately

leading to improved patient care and optimized re-

source allocation. Moreover, understanding the in-

ﬂuence of external conditions helps optimize ED op-

erations, adapt stafﬁng levels, and improve resource

utilization to accommodate varying demands. This

paper aims to delve into the signiﬁcance of patient

proﬁling in the ED by conducting a comprehensive

https://orcid.org/0000-0003-2190-4319

https://orcid.org/0000-0003-4244-5393

cluster analysis.

Proﬁling patients in the ED not only provides a

deeper understanding of patient characteristics but

also sheds light on their arrival patterns and associ-

ated factors. This information plays a pivotal role in

enhancing patient care and addressing pressing chal-

lenges such as overcrowding and sub-optimal patient

ﬂow. By uncovering unique patient proﬁles, health-

care providers can tailor their strategies to meet the

speciﬁc needs of different patient groups, leading to

more efﬁcient and effective care delivery.

External conditions, encompassing factors beyond

individual patient attributes, signiﬁcantly inﬂuence

the functioning and efﬁciency of the ED. These con-

ditions may include the time of arrival, weather con-

ditions and calendar variables, among others. Un-

derstanding how these external factors interact with

patient proﬁles can inform strategies for optimizing

resource allocation, adapting stafﬁng levels, and im-

proving overall ED performance.

The comprehensive cluster analysis approach em-

ployed in this paper seeks to contribute to the existing

body of knowledge on enhancing healthcare delivery

in ED. For that, we explore various factors such as pa-

tient’s demographic information, consultation charac-

teristics, calendar information and weather informa-

tion, aiming to provide novel insights into managing

Carvalho, M. and Borges, A.

Enhancing Healthcare in Emergency Department Through Patient and External Conditions Proﬁling: A Cluster Analysis.

DOI: 10.5220/0012173300003598

In Proceedings of the 15th International Joint Conference on Knowledge Discovery, Knowledge Engineering and Knowledge Management (IC3K 2023) - Volume 1: KDIR, pages 257-264

ISBN: 978-989-758-671-2; ISSN: 2184-3228

257

patient proﬁles, optimizing resource allocation, and

ultimately improving patient outcomes.

The ﬁndings of this study have practical impli-

cations for healthcare professionals, administrators,

and policymakers involved in emergency care. Un-

derstanding patient proﬁles and their relationship with

external conditions can guide the development of tai-

lored interventions, resource allocation strategies, and

stafﬁng optimization approaches.

This paper is structured as follows: Firstly, we

provide a comprehensive background, outlining the

importance of patient proﬁling in the ED and its im-

pact on healthcare delivery. This is followed by the

presentation of hospital data, where we describe the

data preprocessing methods employed and present the

ﬁndings of exploratory data analysis.

Subsequently, we delve into the methodology sec-

tion, where we detail the application of the k-means

clustering algorithm for patient proﬁling in the ED.

We explain the steps involved in the clustering pro-

cess, including the selection of appropriate features

and the determination of the optimal number of clus-

ters.

Moving forward, we present the results section,

which encompasses the characterization of the ob-

tained clusters. We analyze the distinct cluster

proﬁles examining any patterns or trends identiﬁed

within the clusters, providing further insights into pa-

tient demographics, arrival patterns, and associated

factors.

Finally, we conclude the paper by summariz-

ing the key ﬁndings of our analysis and discussing

their signiﬁcance for enhancing healthcare delivery in

ED. We highlight the implications of patient proﬁl-

ing through cluster analysis for addressing challenges

such as overcrowding, improving patient ﬂow, and

optimizing resource allocation. Moreover, we iden-

tify areas for future research and emphasize the im-

portance of continued efforts to enhance the effective-

ness and efﬁciency of ED operations.

2 BACKGROUND

Studies conducted by (Yatoo et al., 2021), (Guivarc’h

et al., 2020), (Wardrop et al., 2021), and (Yeniocak

and Topacoglu, 2018) have demonstrated the bene-

ﬁts of patient proﬁling in the ED. (Yatoo et al., 2021)

discovered that a majority of patients in their study

were over 60 years old and tended to present dur-

ing evening hours, highlighting the importance of tai-

loring resources to accommodate this demographic

and time-speciﬁc demand. (Guivarc’h et al., 2020)

identiﬁed two distinct patient proﬁles in a dental ED:

younger patients seeking acute pain relief and elderly

patients seeking follow-up care. This ﬁnding empha-

sizes the need for tailored approaches to meet the di-

verse needs of different patient groups.

(Wardrop et al., 2021) emphasized the value of un-

derstanding how patients arrive at the ED in planning

healthcare services’ response to ﬂuctuating demand.

By analyzing arrival patterns and associated factors,

healthcare providers can better anticipate and respond

to changing needs, ensuring effective resource alloca-

tion. Furthermore, (Yeniocak and Topacoglu, 2018)

found that the number of people accompanying pa-

tients in the ED varied based on sociocultural factors

such as gender, age, literacy, and education level. Pro-

ﬁling patients allows healthcare providers to account

for these factors and ensure that the necessary support

and resources are available.

Various approaches to clustering ED patients have

been proposed in the literature that highlight the po-

tential of clustering methods in managing overcrowd-

ing, optimizing patient ﬂow, and improving overall

ED operations. (Feretzakis et al., 2022) compared

a clustering-related technique for ED patients using

the k-means algorithm and its impact on admission

output. (Albarakati and Obradovic, 2019) introduced

a multi-domain and multi-view networks model to

cluster hospital admissions originating from the ED.

(Valipoor et al., 2021) employed data-driven design

strategies to address issues of crowding and board-

ing in the ED. (Wartelle et al., 2022) conducted an

analysis to evaluate the effects of opening new on-

demand care services based on variations in patient

ﬂow within a large hospital’s ED.

3 HOSPITAL DATA

The hospital data was collected from the electronic

platform of a hospital located in the north of Portugal.

The following subsection describes the data in detail

as also the created and derived variables. All data was

pre-processed and analyzed using R (Ihaka and Gen-

tleman, 1996) and RStudio (R Core Team, 2023).

3.1 Data Preprocessing Methods

The collected data (the original dataset) is composed

of: 739401 observations and 18 variables. The set of

variables is described as followed:

• id Appointment: Identiﬁer of the appointment.

• id Patient: Identiﬁer of the patient.

• Age: Patient’s age in years.

• Gender: Patient’s gender.

KDIR 2023 - 15th International Conference on Knowledge Discovery and Information Retrieval

258

• Marital status: Patient’s marital status.

• Occupation: Patient’s occupation.

• District: Patient’s address district.

• Municipality: Patient’s address municipality.

• id Specialty: Identiﬁer of the Specialty.

• Specialty: Specialty description.

• Appointment type: Type of the appointment.

• First Appointment ﬂag: First Appointment indi-

cator.

• Emergency ﬂag: Emergency indicator.

• Appointment Type ﬂag: Appointment type indi-

cator.

• Appointment Date: Date of the Appointment.

• Appointment State ﬂag: State of the Appoint-

ment indicator.

• Appointment State: State of the Appointment.

• Appointment Schedule Date: Date of the Sched-

ule of the Appointment.

The data preprocessing phase started by correcting ty-

pos and errors in the dataset. As the data contained

in the dataset is derived from collecting information

from software, there is always the problem that some

of the entered data is not validated, leading to errors.

So, several measures were taken to address this con-

cern: the observations that contain the Appointment

Schedule Date (date of schedule of the appointment)

later than the Appointment Date were erased. Ad-

ditionally, observations containing patients with an

age above 100 were eliminated. Furthermore, all the

empty strings values, i.e., ””, were declared as ’Not

Available’ (NA) values.

The variable Occupation was deleted since all the

observations were missing. And, since we are try-

ing to characterize patients that reach the ED, we are

only considering observations in which the Specialty

variable contains the value ”Emergency Department”.

Also, the variable, Date Month Appointment, Week

Day Appointment and Season were created using the

appointment date.

The ﬁnal dataset was composed of 56150 obser-

vations and 10 variables: Age, Gender, Marital Sta-

tus, Type of Appointment, Date of Appointment, Date

Month Appointment, Week day Appointment, Season,

Precipitation and Temperature.

3.2 Exploratory Data Analysis

In the present section we will brieﬂy describe the

dataset contents to get a baseline on the interpretation

of the results obtained with the application of cluster-

ing technique. As previously mentioned, the dataset

used in this study exclusively includes records catego-

rized as urgent appointments within the hospital. The

dataset encompasses patient information and appoint-

ment data spanning from 2018 1

Nov and 2020 30

Nov. The age distribution of patients is depicted in

Figure 1. The graph illustrates a signiﬁcant surge of

patients aged 0 to 5 arriving at the ED, followed by

another peak in the age range of 40 to 45.

Figure 1: Distribution of Patients’ Age.

Regarding the gender distribution of patients ac-

cessing the ED of this hospital: more than half of the

patients are Female (51.3%), while the remaining are

Male (48,7%).

When analysing the marital status of the patients,

the highest percentage belongs to: Single individuals,

accounting for 65.8% of the total, followed by Mar-

ried patients with as percentage of 30.3%. Widowed

patients constitute 1.8%, Divorced patients account

for 1.5%, co-habiting patients a percentage of 0.5%

and ﬁnally Other with the remaining 0.1%.

The distribution of appointment types, i.e., the

nature of the appointments performed in the Emer-

gency Department (ED), reveals that: approximately

48.9% of the occurrences were Emergency Acts, fol-

lowed by Pediatric Consultations with approximately

37.6%. Followed by Nursing Acts with a percentage

of 6.4% of the cases, and ﬁnally, Incidents represent

approximately 6.4% of the occurrences. The remain-

ing 0.4% are characterized by Accident Act - Continu-

ation, which means that, in this kind of appointments,

there is a follow-up of a previous accident act.

Examining the months with the highest number of

patients visits to de ED: December (12) leads with

a total of 7700 visits (13.7% of the total visits), fol-

lowed by January (1) with 7622 visits to the ED

(13.6% of the total visits), and February (2) with 7073

visits (12.6% of the total visits). Conversely, August

(8) records the fewest patient visits to the ED with a

total of 2498 visits (4.4% of the total visits), followed

by September (9) with 2557 visits (4.6% of the total

visits), and October (10) with 3024 visits (5.4% of the

total visits).

Concerning the number of appointments by week-

Enhancing Healthcare in Emergency Department Through Patient and External Conditions Proﬁling: A Cluster Analysis

259

day: Monday (1) is the day with the highest number

of ED visits, totaling 9249 visits (16.5% of the total

visits). It is followed by Sunday (0) with 8322 visits

(14.8% of the total visits), and Saturday (6) with 7919

patient visits to the ED (14.1% of the total visits). On

the other hand, the days of the week with the fewest

visits are Friday (5), Thursday (4), and Tuesday (2)

with 7628 (13.6% of the total visits), 7641 (13.6% of

the total visits), and 7673 visits (13.7% of the total

visits), respectively.

When analyzing appointments by season, Winter

stands out with the highest number of patient visits

to the ED, accounting for 37.8% of the total visits.

Spring follows with 28.8%, Summer with 19.9%, and

ﬁnally, Fall has the fewest patient visits to the ED,

representing only 14.4% of the total number of visits.

Following the data pre-processing stage and a con-

cise analysis of the dataset’s content, we will proceed

to the modeling phase.

4 METHODOLOGY

The k-means algorithm (Hartigan and Wong, 1979)

is a widely utilized clustering technique employed to

partition datasets into distinct clusters. It is an it-

erative algorithm that aims to minimize the within-

cluster sum of squares: maximizing intra-clustering

similarity and minimizing inter-clustering similarity.

The algorithm proceeds as follows: Firstly, the

number of clusters (k) to be formed is determined,

and k initial cluster centroids are randomly assigned.

Each data point is then assigned to the nearest cen-

troid, forming k initial clusters based on the Euclidean

distance. The centroids of the clusters are recalcu-

lated in each iteration until convergence is achieved

or a maximum number of iterations is reached.

An important consideration in applying k-means

is the need to specify the number of clusters before-

hand. Determining the best value for k can be chal-

lenging, and the results obtained can vary depending

on the initialization.

4.1 Determining the K Value

Finding the best value of k, the number of clusters,

is of utmost importance before applying the k-means

algorithm. Selecting the appropriate value of k can

signiﬁcantly impact the quality and interpretability of

the clustering results. If the value of k is too low,

the algorithm may merge distinct clusters together, re-

sulting in a loss of meaningful insights. Conversely,

if the value of k is too high, it may lead to over-

segmentation, generating clusters that are too small

and lacking sufﬁcient distinctiveness.

To ﬁnd out the best number of k clusters to per-

form the k-means algorithm, we used the available

package NbClust (Charrad et al., 2014) in RStudio in

a sample of 10% of all data. The NbClust package

provides 30 indices - although we only performed 26

indices - for determining the number of clusters and

proposes the best clustering scheme from the differ-

ent results obtained by varying all combinations of the

number of clusters, distance measures, and clustering

methods. The output of NbClust is as follows and it

is shown in detail in the Figure 2.

* Among all indices:

* 5 proposed 2 as the best number of clusters

* 6 proposed 3 as the best number of clusters

* 2 proposed 4 as the best number of clusters

* 3 proposed 5 as the best number of clusters

* 1 proposed 6 as the best number of clusters

* 1 proposed 8 as the best number of clusters

* 5 proposed 10 as the best number of clusters

***** Conclusion *****

* According to the majority rule,

the best number of clusters is 3

Figure 2: Suggested k number of cluster by Index.

5 RESULTS: CLUSTERS

CHARACTERIZATION

In this section, we present a detailed characterization

of each cluster obtained through the application of the

k-means clustering algorithm. Our aim is to provide

a comprehensive understanding of the distinct patient

proﬁles identiﬁed within the clusters. To achieve this,

we analyze the frequency and cross-information of

each variable considered during the clustering pro-

cess.

Additionally, we go beyond the traditional vari-

ables used in clustering and extend our analysis to

incorporate the number of emergency consultations.

This allows us to identify speciﬁc proﬁles and risk

factors associated with frequent attendance at emer-

gency rooms.

KDIR 2023 - 15th International Conference on Knowledge Discovery and Information Retrieval

260

The K-means clustering analysis resulted in the

formation of three clusters which are represented in

Figure 3. The sizes of these clusters are as follows:

Cluster 1 with 24,119 observations (43.0% of the to-

tal observations), Cluster 2 with 18,495 observations

(32.9% of the total observations), and Cluster 3 with

13,536 observations (24.1% of the total observations).

These sizes represent the number of data points as-

signed to each cluster.

Figure 3: Representation of the clusters.

The clusters’ centroids are represented in the Fig-

ure 4 which displays the average values of different

variables for each cluster.

Figure 4: Representation of the clusters’ centroids.

In an initial analysis of the centroid representation

image and the cluster representation image, it is evi-

dent that all clusters exhibit distinguishing character-

istics, leading us to believe that we can differentiate

patient proﬁles and other external characteristics.

In the following subsections, we analyse and char-

acterize the clusters obtained.

5.1 Cluster1

Cluster 1 has the largest number of observations,

which means that it will be the one associated with

higher afﬂuence in the ED. And as mentioned ear-

lier, accounting for approximately 43% of all obser-

vations.

Concerning the nature of visits to the EDin cluster

1 there is a clear dominance of Emergency Act with a

total of 88.7%, followed by Nursing Acts with a total

of 11.3%, Accident Act - Continuation with approx-

imately 1%, and ﬁnally Pediatric Consultation has a

percentage of less than 1%.

Regarding the patients’ age, there is a clear peak

in the age range between 35 and 50 years, while the

lowest range of values is observed between 5 and 15

years.

In terms of patients’ gender distribution: 57.3% of

the patients were Female, and the remaining 42.7%

were Male.

Regarding the marital status of the patients, clus-

ter 1 is characterized by a signiﬁcant proportion of

Married patients, as shown in Figure 5, accounting

for approximately 82.6% of patients with Emergency

Act consultations.

Figure 5: Distribution of Appointment Type by Patients’

Marital Status in cluster 1.

Figure 6 allows to analyze the nature of the ap-

pointment and its relation to the seasons. And as illus-

trated in this ﬁgure, although a balanced distribution

is observed, there is a higher inﬂux during Spring for

Emergency Acts (accounting for 36.8% of total visits),

contrasting with the percentages for Summer (with

27.3%), Fall (with 19%), and Winter (with 16.9%).

Figure 6: Distribution of Appointment Type by Season in

cluster 1.

It is possible to detail even more the analysed in-

formation above by adding information about the in-

ﬂux each month. The visits to the ED are almost

equally distributed every month, but there is a clear

higher inﬂux in January and February, representing

16.7% and 13.7% of all Emergency Act consultations,

respectively.

Enhancing Healthcare in Emergency Department Through Patient and External Conditions Proﬁling: A Cluster Analysis

261

Regarding the distribution of visits on each day of

the week, visits are distributed as follows: Sunday (0)

has the highest inﬂux with 16.0% of visits, followed

by Monday (1) with 15.9%, and Saturday (6) with

14.9%. The days with the lowest inﬂux are Thursday

(4), Friday (5), and Tuesday (2) with 12.9%, 13.1%,

and 13.5%, respectively.

5.2 Cluster2

Cluster 2 contains 32.9% of the total observations and

is the second largest cluster.

Analyzing the nature of ED visits in Cluster 2,

there is a clear trend with a signiﬁcant inﬂux of Pe-

diatric Consultation visits, accounting for 81.5% of

the total visits allocated to this cluster. This is fol-

lowed by Incidents visits, representing 15.9% of the

total visits, and Emergency Act visits, accounting for

only 1.7% of the total visits. Visits categorized as In-

surance Claims - Nursing Act, Nursing Act, and Pedi-

atric Telephone Consultation - Trace-COVID have a

percentage lower than 1%.

Regarding the analysis of patients’ age who vis-

ited the ED, a peak can be identiﬁed between 0 and 8

years of age.

In terms of the distribution of patients’ gender,

56.6% of the patients were Male, and the remaining

43.4% were Female.

Regarding the distribution of patients’ marital sta-

tus, a clear pattern can be identiﬁed, as seen in Figure

7. The majority of patients visiting the ED are Single

- 100% of all patients with a Pediatric Consultation

visit and 97.3% of all observations in Cluster 2.

Figure 7: Distribution of Appointment Type by Patients’

Marital Status in Cluster 2.

Concerning the analysis of visits by season, Clus-

ter 2 shows a high inﬂux during Spring, as depicted

in Figure 8. Pediatric Consultation visits in Spring

account for 38.6% of all pediatric consultations, con-

trasting with the percentages for Summer (21.9%),

Winter (21.3%), and Fall (18.2%).

Providing further details on the distribution of vis-

its by season, we can include information on visits in

each month. There is a clear higher inﬂux in January

Figure 8: Distribution of Appointment Type by Season in

Cluster 2.

and February, representing 21.3% and 20.1% of all

Pediatric Consultations, respectively.

Concerning the number of visits on each day of

the week, they are distributed as follows: Monday (1)

has the highest inﬂux with 17.9% of visits, followed

by Thursday (4) with 14.5%, and Tuesday (2) with

14.3%. The days with the lowest inﬂux are Saturday

(6), Sunday (0), and Friday (5) with 11.9%, 13.3%,

and 13.9%, respectively.

5.3 Cluster3

Cluster 3 contains 24.1% of the total observations and

it is the smallest cluster with the fewest number of

observations.

Analyzing the information obtained from Cluster

3, speciﬁcally the nature of the visits to the Emer-

gency Department (ED), the most frequent types of

visits are Pediatric Consultation and Emergency Act,

with similar percentage values, accounting for 44.8%

and 43.9% of the total observations allocated to the

cluster, respectively. This is followed by Nursing Act

and Incidents, which account for 6.2% and 4.6% of

the total visits, respectively. The types of visits la-

beled as Accident Act - Continuation and Insurance

Claims - Nursing Act have a percentage lower than

1%.

Regarding the age distribution of patients visiting

the ED, there is a clear peak between 0 and 10 years

of age, as well as a secondary peak between 34 and

50 years of age.

In terms of the analysis of the distribution of pa-

tients’ gender, 51.5% of the patients were Female, and

the remaining 48.5% were Male.

With regard to the analysis of the graph in Figure

9: 100% of all patients with a Pediatric Consultation

are Single, which corresponds to 55.8% of the total;

and concerning the Emergency Act type consultations,

64.3% of patients are Married patients.

Concerning the distribution of visits by season, we

found an interesting pattern in this analysis. The ma-

jority of the observations were realized in Winter, rep-

resenting 99.9% of all the visits to the ER.

KDIR 2023 - 15th International Conference on Knowledge Discovery and Information Retrieval

262

Figure 9: Distribution of Appointment Type by Patients’

Marital Status in Cluster 3.

To conﬁrm this information and detail in which

months of winter there are more consultations, we

performed an analysis of the distribution of consul-

tations in each of the months of the year. The greatest

inﬂux of consultations is in November (11) and De-

cember (12). Regarding Pediatric Consultation and

Emergency, a percentage of 58.2% and 57.5% of total

consultations in each type of consultation were held in

December, respectively. Compared to November, in

the same type of consultations a percentage of 41.8%

and 42.3%, respectively, were performed.

Figure 10: Distribution of Appointment Type by Month in

Cluster 3.

The number of visits on each day of the week is

described as follows: Monday (1) has the highest in-

ﬂux with 16.4% of visits, followed by Sunday (0) with

15.6%, and Saturday (6) with 15.4%. The days with

the lowest inﬂux are Wednesday (3), Tuesday (2), and

Thursday (4) with 12.7%, 12.7%, and 13.4%, respec-

tively.

6 DISCUSSION AND

CONCLUSION

EDs serve as critical hubs for providing immediate

medical care to patients in urgent need. The demand

for ED services continues to rise, leading to chal-

lenges such as overcrowding, long wait times, and re-

source limitations. To address these issues effectively,

it is crucial to have a comprehensive understanding of

patient characteristics, arrival patterns, and associated

factors.

In this study, we employed the k-means algo-

rithm, using NbClust using R in RStudio to determine

the optimal number of clusters, to analyze real data

from a hospital. Through this approach, we identiﬁed

and delineated three distinct clusters, each encapsulat-

ing unique patient characteristics and external factors.

Through the application of a comprehensive cluster

analysis, this paper aims to expand the understanding

of healthcare delivery in ED, ultimately driving im-

provements in patient outcomes.

Cluster 1, which represents 43.0% of the total ob-

servations, stands out as the largest cluster. It is char-

acterized by a high proportion of Emergency Act visits

(62.8%) and Incidents visits (23.3%), indicating a sig-

niﬁcant number of urgent and non-urgent cases. The

age distribution shows a relatively even distribution

across different age groups. In terms of gender, there

is a balanced representation between males and fe-

males. Moreover, the distribution of patients’ marital

status reveals a relatively even distribution across var-

ious categories. This cluster exhibits a consistent pat-

tern of visits throughout the seasons, with no speciﬁc

season dominating. Overall, Cluster 1 can be named

the ”Mixed Acuity Cluster” as it encompasses a mix

of emergency and non-urgent visits, and it represents

a diverse range of patients in terms of age, gender, and

marital status.

Cluster 2, accounting for 32.9% of the total ob-

servations, is the second largest cluster. The deﬁn-

ing characteristic of this cluster is the overwhelm-

ing dominance of Pediatric Consultation visits, rep-

resenting 81.5% of the visits within the cluster. There

is also a notable presence of Incidents visits (15.9%)

and a very small proportion of Emergency Act visits

(1.7%). The age distribution highlights a peak in the

0-8 years age group. Gender-wise, a majority (56.6%)

of the patients in this cluster are male. Furthermore,

all patients with Pediatric Consultation visits are pri-

marily single. The cluster exhibits a higher inﬂux

during the spring season compared to other seasons.

Given these characteristics, Cluster 2 can be named

the ”Pediatric Consultation Dominant Cluster,” as it

primarily consists of pediatric patients seeking con-

sultations and exhibits distinctive age, gender, and

seasonal patterns.

Cluster 3, representing 24.1% of the total observa-

tions, is the smallest cluster. It shows a relatively bal-

anced distribution of Pediatric Consultation (44.8%)

and Emergency Act (43.9%) visits. The age distribu-

tion reveals two peaks, one in the 0-10 years age range

and another between 34 and 50 years. Gender-wise,

there is a slight majority of female patients (51.5%).

Marital status analysis indicates that all patients with

Pediatric Consultation visits are single (as expected),

Enhancing Healthcare in Emergency Department Through Patient and External Conditions Proﬁling: A Cluster Analysis

263

while a signiﬁcant proportion of Emergency Act vis-

its come from married patients. Notably, the clus-

ter exhibits a clear preference for visits during the

winter season, particularly in November and Decem-

ber. Considering these characteristics, Cluster 3 can

be named the ”Mixed Acuity with Seasonal Prefer-

ence Cluster,” as it encompasses a mix of pediatric

and emergency visits, demonstrates distinct age and

seasonal patterns, and showcases variations in marital

status.

Regarding external factors like precipitation and

temperature, no discernible patterns were discovered

that had an impact on the utilization of the emergency

room.

In conclusion, this analysis contributes to the iden-

tiﬁcation of distinct groups with unique needs, facil-

itating the development of tailored approaches to op-

timize resource allocation, improve patient care, and

enhance the overall efﬁciency of the emergency de-

partment. The comprehensive characterization of the

clusters, including their underlying variables and the

impact of emergency consultations, enhances the un-

derstanding of the diverse patient proﬁles within the

emergency department.

This information serves as a powerful tool for im-

proving patient care, enhancing resource allocation,

and addressing challenges such as overcrowding and

sub-optimal patient ﬂow.

Future work in this context could explore alter-

native clustering algorithms or techniques to validate

and compare the results obtained using the k-means

algorithm. Such as hierarchical clustering, density-

based clustering, or model-based clustering to assess

their effectiveness in capturing the underlying pat-

terns in the data.

ACKNOWLEDGEMENTS

This work has been supported by national funds

through FCT – Fundac¸

ao para a Ci

encia e Tecnolo-

gia through project UIDB/04728/2020. The authors

thank the hospital for providing the real data used in

this study.

REFERENCES

Albarakati, N. and Obradovic, Z. (2019). Multi-domain

and multi-view networks model for clustering hospital

admissions from the emergency department. Interna-

tional Journal of Data Science and Analytics, 8:385–

403.

Charrad, M., Ghazzali, N., Boiteau, V., and Niknafs, A.

(2014). Nbclust: an r package for determining the

relevant number of clusters in a data set. Journal of

statistical software, 61:1–36.

Feretzakis, G., Sakagianni, A., Kalles, D., Loupelis, E.,

Tzelves, L., Panteris, V., Chatzikyriakou, R., Trakas,

N., Kolokytha, S., Batiani, P., Rakopoulou, Z., Tika,

A., Petropoulou, S., Dalainas, I., and Kaldis, V.

(2022). Exploratory Clustering for Emergency De-

partment Patients. IOS Press.

Guivarc’h, M., Saliba-Serre, B., Le Coz, P., and Bukiet, F.

(2020). A cross-sectional analysis of patient care path-

ways and proﬁles in a dental emergency department.

International Dental Journal, 70(1):21–28.

Hartigan, J. A. and Wong, M. A. (1979). Algorithm as

136: A k-means clustering algorithm. Journal of the

royal statistical society. series c (applied statistics),

28(1):100–108.

Ihaka, R. and Gentleman, R. (1996). R: a language for data

analysis and graphics. Journal of computational and

graphical statistics, 5(3):299–314.

R Core Team (2023). R: A Language and Environment for

Statistical Computing. R Foundation for Statistical

Computing, Vienna, Austria.

Valipoor, S., Hatami, M., Hakimjavadi, H., Akc¸alı, E.,

Swan, W. A., and De Portu, G. (2021). Data-driven

design strategies to address crowding and boarding in

an emergency department: A discrete-event simula-

tion study. HERD: Health Environments Research &

Design Journal, 14(2):161–177.

Wardrop, R., Ranse, J., Chaboyer, W., and Crilly, J. (2021).

Proﬁle and outcomes of emergency department pre-

sentations based on mode of arrival: A statewide retro-

spective cohort study. Emergency Medicine Australa-

sia, 34(4):519–527.

Wartelle, A., Mourad-Chehade, F., Yalaoui, F., Questiaux,

H., Monneret, T., Soliveau, G., Chrusciel, J., Duclos,

A., Laplanche, D., and Sanchez, S. (2022). Multimor-

bidity clustering of the emergency department patient

ﬂow: Impact analysis of new unscheduled care clinics.

Plos one, 17(1):e0262914.

Yatoo, G. H., Mufti, S., Jabeen, U., Malik, A., et al. (2021).

To study the proﬁle of the patients attending the emer-

gency medicine department of a tertiary care teaching

hospital of north india. Biomedical Journal of Scien-

tiﬁc & Technical Research, 34(3):26758–26762.

Yeniocak, S. and Topacoglu, H. (2018). A proﬁle of individ-

uals accompanying patients in the emergency depart-

ment: An analysis of 5046 cases. Nigerian Journal of

Clinical Practice, 21(10):1260–1264.

KDIR 2023 - 15th International Conference on Knowledge Discovery and Information Retrieval

264