Article

Reducing Waiting Times to Improve Patient Satisfaction:

A Hybrid Strategy for Decision Support Management
Jenny Morales, Fabián Silva-Aravena * and Paula Saez

Facultad de Ciencias Sociales y Económicas, Universidad Católica del Maule, Avenida San Miguel 3605,
Talca 3460000, Chile; [email protected] (J.M.); [email protected] (P.S.)
* Correspondence: [email protected]

Abstract: Patient satisfaction and operational efficiency are critical in healthcare. Long waiting times
negatively affect patient experience and hospital performance. Addressing these issues requires
accurate system time predictions and actionable strategies. This paper presents a hybrid framework
combining predictive modeling and optimization to reduce system times and enhance satisfaction,
focusing on registration, vitals, and doctor consultation. We evaluated three predictive models:
multiple linear regression (MLR), log-transformed regression (LTMLR), and artificial neural networks
(ANN). The MLR model had the best performance, with an R2 of 0.93, an MAE of 7.29 min, and
an RMSE of 9.57 min. MLR was chosen for optimization due to its accuracy and efficiency, making
it ideal for implementation. The hybrid framework combines the MLR model with a simulation-
based optimization system to reduce waiting and processing times, considering resource constraints
like staff and patient load. Simulating various scenarios, the framework identifies key bottlenecks
and allocates resources effectively. Reducing registration and doctor consultation wait times were
identified as primary areas for improvement. Efficiency factors were applied to optimize waiting and
processing times. These factors include increasing staff during peak hours, improving workflows,
and automating tasks. As a result, registration wait time decreased by 15%, vitals by 20%, and
doctor consultation by 25%. Processing times improved by 10–15%, leading to an average reduction
of 22.5 min in total system time. This paper introduces a hybrid decision support system that
integrates predictive analytics with operational improvements. By combining the MLR model with
Citation: Morales, J.; Silva-Aravena, simulation, healthcare managers can predict patient times and test strategies in a risk-free, simulated
F.; Saez, P. Reducing Waiting Times to
environment. This approach allows real-time decision-making and scenario exploration without
Improve Patient Satisfaction: A
disrupting operations. This methodology highlights how reducing waiting times has a direct impact
Hybrid Strategy for Decision Support
on patient satisfaction and hospital operational efficiency, offering an applicable solution that does
Management. Mathematics 2024, 12,
not require significant structural changes. The results are practical and implementable in resource-
3743. https://doi.org/10.3390/
math12233743
constrained healthcare environments, allowing for optimized staff management and patient flow.

Academic Editor: Prasenjit

Keywords: patient satisfaction; healthcare operations efficiency; predictive modeling; simulation-
Chatterjee
based decision support; wait-time optimization
Received: 31 October 2024
Revised: 22 November 2024 MSC: 62J05; 90B50; 90C90
Accepted: 26 November 2024
Published: 28 November 2024

1. Introduction
According to [1,2] and other authors, patient satisfaction and operational efficiency
Copyright: © 2024 by the authors.
Licensee MDPI, Basel, Switzerland.
are two fundamental pillars in the management of modern healthcare systems. Prolonged
This article is an open access article
waiting times in hospitals and healthcare facilities are a critical factor that negatively
distributed under the terms and affects the perceived quality of care, patient satisfaction, and, ultimately, the operational
conditions of the Creative Commons performance of hospitals [3–5]. Several studies have shown that reducing waiting times can
Attribution (CC BY) license (https:// significantly improve the patient experience and optimize the use of limited resources, such
creativecommons.org/licenses/by/ as medical staff and hospital infrastructure [6–8]. In a context where healthcare systems
4.0/).

Mathematics 2024, 12, 3743. https://doi.org/10.3390/math12233743 https://www.mdpi.com/journal/mathematics

Mathematics 2024, 12, 3743 2 of 15

face increasing demands, optimizing service delivery times has become a priority for
hospital managers.
Predictive methodologies and simulation models have gained traction in recent years
as essential tools for improving decision-making in healthcare systems [9–11]. Accurate
prediction of system times, including waiting and processing times across different stages
of care, allows hospital managers to identify bottlenecks and allocate resources more effec-
tively [12,13]. However, the challenge lies not only in accurately predicting system times
but also in integrating these predictive insights into an optimization framework that can
enhance operations and reduce inefficiencies without compromising service quality [14,15].
In this context, regression models have been widely used to predict system times in
hospitals, including MLR and more advanced models such as ANN [16–18]. Each of these
models offers specific advantages, such as the simplicity and interpretability of MLR or the
ability to capture more complex nonlinear relationships through ANN and deep learning
(see, e.g., [18–20]). However, recent studies suggest that combining predictive models with
simulation and optimization can provide a more robust and effective solution for managing
patient flow in hospitals, especially in environments with limited resources and fluctuating
demand [21–23].
The aim of our paper is to build a methodology to reduce waiting times at critical
care points and improve patient satisfaction through a hybrid prediction and simulation
model. The novelty lies in the combination of an MLR model with simulation, which allows
real-time decision-making, effectively optimizing resources in hospital environments. To
achieve this, the hybrid approach combines predictive modeling with simulation-based
optimization techniques to reduce total system times in the hospital. Using an MLR
model as a basis, we simulate various operational scenarios, taking into account resource
constraints such as medical staff availability and patient demand during peak hours. By
simulating these scenarios, the proposed framework enables the identification of key
bottlenecks and generates specific recommendations for optimal resource allocation and
reduction of waiting times at critical stages such as registration, vital signs verification, and
medical consultation (see, for example, [24–26]).
Over the past decade, simulation has become a crucial tool for improving healthcare
system management without disrupting daily operations [27,28]. By integrating predictive
models within a simulation platform, healthcare managers can foresee the impact of
different operational decisions in real time, allowing for faster, data-driven decision-making.
This approach offers a crucial advantage by enabling the evaluation of multiple scenarios
without compromising service quality or causing disruptions to the workflow [29,30].
The structure of this paper is as follows: Section 2 provides a comprehensive review
of the literature on predictive models and simulation in the hospital setting; Section 3
details the methodology used, with a focus on multiple linear regression and the simulation
techniques applied. Section 4 presents the results obtained from the predictive models and
simulation framework. Section 5 offers a detailed discussion, and in Section 6, we present
the main conclusions of the paper.

2. Literature Review
In the healthcare field, patient satisfaction and the effectiveness of healthcare systems
are closely related [31]. It has long been known that long waiting times have a negative
impact on patient satisfaction and overall perception [32,33]. According to several studies,
increasing operational efficiency, especially by reducing waiting times, is essential to
improve patient satisfaction and the quality of treatment provided (see, e.g., [34]). This
correlation between patient satisfaction and waiting time highlights the need to adopt
efficient approaches that can alleviate bottlenecks and improve resource allocation in
hospitals [4,35]. Therefore, healthcare services should prioritize optimizing system times
to ensure that patient needs are met in a timely manner, which has been shown to help
improve satisfaction levels [36–38].
Mathematics 2024, 12, 3743 3 of 15

Predictive modeling has emerged as a good tool in the patient care field to improve
decision-making, particularly in predicting patient flow and waiting times [4,39]. Hospital
managers can anticipate potential delays and adjust staffing and resources appropriately
by using predictive models such as multiple linear regression and more sophisticated
models such as artificial neural networks (ANN) that leverage past data [16,40]. According
to [9], predictive models increase the accuracy of predicting healthcare outcomes, which is
crucial for optimizing hospital operations. To fully realize their potential, predictive models
should be incorporated within a broader decision-support framework [41]. According
to [11], a hybrid technique combining simulation with predictive models provides a more
complete solution by simulating various situations and predicting outcomes to increase
system efficiency.
The application of simulation-based strategies in healthcare is well documented.
Simulation allows healthcare services to experiment with various operational strategies,
providing insights into how changes to one part of the system, such as staffing or resource
allocation, can impact overall performance [13,25]. Research by [21,22] shows that simu-
lation models can effectively reduce waiting times and improve resource management,
particularly in high-demand settings such as emergency departments. The flexibility of
simulation makes it a good tool for hospital management, allowing them to test different
strategies before implementation [29]. By combining simulation with predictive analytics,
healthcare services can achieve a holistic approach to managing patient flow, significantly
reducing system times without compromising the quality of care.
Even with advancements in simulation and prediction model construction, there is a
gap in their integration, particularly in frameworks for real-time decision-making [18,19].
The dynamic capabilities needed to react instantly to changes in healthcare environments
are frequently absent from predictive models. However, although useful for testing sce-
narios, simulation models do not always offer the degree of predictive accuracy needed
for long-term planning [23]. This disparity emphasizes the necessity of hybrid models
that can effectively combine simulation and predictive analytics to offer a strong decision
support system that can handle the pressing issues in healthcare [20]. Recent studies have
highlighted the potential of hybrid models in achieving this balance, offering solutions that
are both flexible and data-driven [42].
Authors such as [11,17] and others emphasize hybrid approaches of predictive models
with simulations for waiting time management and how such integration can contribute to
real-time operational optimization without disrupting hospital flow. Combining simulation
and predictive modeling has two benefits: it increases prediction accuracy and gives
hospital administrators a controlled setting in which to test out various operational tactics.
In complex systems like hospitals, where resource limitations and variations in patient
flow can have a major influence on system performance, this hybrid approach is especially
advantageous [24]. According to research by [27], simulation can be used to enhance
the general quality and safety of healthcare services in addition to optimizing patient
flow. Healthcare systems can decrease wait times at crucial points in the care process, like
registration and consultations, and increase patient satisfaction by incorporating simulation
into predictive models [30].
A hybrid decision support system that combines simulation and predictive analytics
is required to enhance healthcare delivery [28,43]. This strategy guarantees that medical
institutions are prepared to manage both expected and unanticipated changes in patient
demand. Authors such as [21] have shown that simulation-based optimization frameworks
can better reallocate resources and pinpoint major bottlenecks that cause delays, hence
reducing total system times. Additionally, the objectives of contemporary healthcare
administration, which center on raising patient happiness through operational effectiveness,
are in line with this hybrid approach. Incorporating machine learning methods into
these hybrid models will probably increase predicted accuracy and system adaptability as
research advances, providing healthcare institutions with a useful tool to satisfy demanding
patient care requirements [28].
Mathematics 2024, 12, 3743 4 of 15

3. Materials and Methods

In this section, we show the data preprocessing process and the steps to follow for
its preprocessing, including the handling of missing data and outliers, as well as the
formulation to obtain the total system time based on the waiting times and the service
process. The steps that allow us to prepare the data for the use of the methods used and the
optimization strategies are described below.

3.1. Data Preprocessing and Postprocessing Process

The public data correspond to 500 patients from a public hospital in Pakistan who
went through a healthcare system consisting of three key stages: registration, vital signs,
and medical consultation. The total time each patient spent in the system was recorded,
along with the waiting and processing times for each stage [44,45].
We implemented various preprocessing techniques, addressing the handling of miss-
ing data and outliers and preparing the data for use in predictive models. A brief descrip-
tion follows.
1. First, we inputted missing values in critical variables (such as times at registration,
vital signs, and medical consultation stages). We chose to eliminate incomplete records
to avoid bias.
2. Second, we applied the interquartile range (IQR) method to identify and handle
outliers since these outliers can affect the distribution of the data and the accuracy of
the predictive models.
3. Third, we calculated the waiting time at each stage (registration, vital signs, and
medical consultation) as the difference between the end time of the previous stage
and the start time of the next stage. We also calculated the processing time as the
difference between the start and end of each stage. We assumed that each patient
follows the same flow (registration → vital signs → medical consultation) without
interruptions, which allowed us to calculate the total time in the system sequentially.
4. Fourth, we checked the distribution of the post-processing data. By removing outliers,
we sought to ensure that the data followed a closer-to-normal distribution, especially
during waiting and processing times.
5. Fifth, to simplify the actual complexity of the hospital flow, we considered that the
waiting and processing times at each stage were independent of each other.
6. Sixth, we normalized the times between 0 and 1 to improve the convergence of the
models and reduce the bias caused by different time magnitudes at each stage.
Pre- and post-processing of the data allowed us to adequately prepare them for
predictive and optimization analysis. This strategy allowed us to ensure that the
methodology was built with high-quality data.

3.1.1. Handling Missing Data and Outliers

In the healthcare dataset, missing data are common, especially in recorded times for
various stages of the patient care process. We addressed this issue by the following:
1. Eliminating records with missing values in critical variables: registration times, vitals
times, doctor consultation times, and total system times. This decision was made to
avoid introducing bias or inaccuracies caused by inputting essential variables.
Let X be the dataset with variables X1 , X2 , . . . , Xn representing the waiting and
processing
 times at different stages.
2. If Xi ∈ Tw,i , Tp,j (wait time fo each stage i and processing time for each stage j),

record X is removed if ∃i ∈ 1, 2, 3  such as Xi = 0, where 0 represents missing values.
After this step, only complete X ∈ Tw,i , Tp,j records were retained, ensuring that all
relevant information for each patient was available.
3. The presence of extreme outliers in the waiting and processing times could skew
the results of our regression models. Therefore, outliers were identified using the
interquartile range (IQR) method for each of the variables, and extreme values were
Mathematics 2024, 12, 3743 5 of 15

removed to ensure normal distribution and prevent model distortion. For each
variable Xi , we calculated the interquartile range, and for the value of Xi that fell
outside the range, we considered outliers and removed them.
4. The final dataset contained 480 complete cases that were used in this article.

3.1.2. Calculation of Waiting Times and Total System Time

The dataset records timestamps at different stages of the patient care process. From
these timestamps, we calculate the waiting times and processing times as follows:
The waiting time for each stage is the difference between the end time of the previous
stage and the start time of the current stage. The waiting times for registration, vitals, and
doctor consultation are calculated as follows:
Tw,r = Ts,r − Te,r , where w is the wait, r is registration, s is the start, and e is the end;
Tw,v = Ts,v − Te,v , where v refers to vitals;
Tw,d = Ts,d − Te,d , where d is the doctor.
Therefore, Ts,r , Ts,v , and Ts,d are the timestamps for the start of each process. Te,r , Te,v ,
and Te,d are the timestamps for the end of each process.
The processing time for each stage is the time elapsed from the start to the end
of the stage. The processing times for registration, vitals, and doctor consultation are
calculated as follows:
Tp,r = Te,r − Ts,r , where p refers to the process;
Tp,v = Te,v − Ts,v ;
Tp,d = Te,d − Ts,d .
The total system time, Tt , is the sum of all the waiting and processing times across the
various stages of the patient care process: Tt = ∑3n=1 Tw,i + ∑3n=1 Tp,j , where Tt is the total
time a patient spends in the system, index i refers to the three stages of waiting (registration,
vitals, and doctor), and index j refers to the three stages of processing (registration, vitals,
and doctor).

3.2. Data Characterization

Figure 1 shows the distribution of waiting times for each of the stages. Although
waiting times at each stage are concentrated at the beginning, there are cases in which
patients wait a long time during all stages. Waiting times at the doctor stage are the longest,
with waits of up to 50 min.
For the bottleneck analysis, we considered the Tt for each patient as a variable, that is,
from their arrival until the end of their process with the doctor. In Table 1, we calculate the
durations of the times for each stage and present the contribution of each one to the Tt .
In Table 2, we present a correlation matrix to evaluate the relationships between each
stage with Tt . The analysis confirms that the highest correlation is between Tt and the
doctor’s times (0.469), confirming the bottleneck at that stage.

Table 1. Statistical summary of bottlenecks by stage (in minutes).

Summary Time Total System (Tt ) Registration Vitals Doctor

Count 126 126 126 126
Mean 59.31 2.212 6.10 14.45
Std 32.89 2.17 5.74 9.31
Min 14 0 0 0
25% 13 1 3 7
50% 54.5 1 4 12
75% 78.5 3 7 20
Mathematics 2024, 12, 3743 6 of 15

Table 2. Correlation matrix of times per stage (in minutes).

Correlation Matrix Total System (Tt ) Registration Vitals Doctor

Total System (Tt ) 1.000 0.093 0.221 0.469
Registration 0.093 1.000 −0.027 −0.150
Vitals 0.221 −0.027 1.000 0.062
Doctor 0.469 −0.150 0.062 1.000

Figure 1. Distribution of time spent in each stage.

Now that we have determined the bottlenecks in the care process, after determining Tt ,
we present three predictive regression models and a decision support strategy to optimize
system times in a patient care setting.

3.3. Multiple Linear Regression

MLR is a model that uses the waiting times and processing times of the three stages of
the system under study to predict Tt . This section provides a detailed description of the
formulation and implementation steps for this model.
The MLR model is expressed as a linear combination of several independent vari-
ables [40,46]. In this case, the independent variables are the waiting times and processing
times at the three stages of the healthcare system, as presented in Section 3.1.2:
1. Registration stage: Tw,r and Tp,r ;
2. Vitals stage: Tw,v and Tp,v ;
3. Doctor stage: Tw,d and Tp,d .
The general form of the MLR model is presented as follows:

3 3
Tt = β 0 + ∑ β i · Tw,i + ∑ β j+3 · Tp,j + ϵ, (1)
i =1 j =1

where Tt is the dependent variable, representing the total time a patient spends in the
healthcare system, β 0 is the intercept, representing the baseline time when all independent
variables are zero, β i are the regression coefficients for the waiting times, Tw,i , (where
Mathematics 2024, 12, 3743 7 of 15


i ∈ 1, 2, 3 corresponds to registration, vitals, and doctor), β j+3 are the regression coeffi-

cients for the processing times, Tp,j , (where j ∈ 1, 2, 3 corresponds to registration, vitals,
and doctor), and ϵ is the error term (or residual), capturing the variance in Tt not explained
by the independent variables.
It is important to mention that the first summation term, ∑3i=1 β i · Tw,i , represents
the combined effect of waiting times at each stage, and the second summation term,
∑3j=1 β j+3 · Tp,j , represents the combined effect of the processing times at each stage.
The MLR model operates under several key assumptions, which are essential to ensure
the validity of the results [40]. We present its methodology below.
1. The relationship between the dependent variable Tt and the independent variables
(waiting times and processing times) is linear.

3 3
E( Tt | Tw,i , Tp,j ) = β 0 + ∑ β i · Tw,i + ∑ β j+3 · Tp,j . (2)
i =1 j =1

2. The residuals from one observation should not be correlated with the residuals from
another observation. In other words, there should be no autocorrelation among the
residuals. One way to check for autocorrelation in a regression model is by using the
Durbin–Watson (DW) statistic.
3. The variance of the error term ϵ is constant across all values of the independent
variables. This is known as homoscedasticity and is expressed as follows:

Var (ϵ | Tw,i , Tp,j ) = σ2 . (3)

4. The error term ϵ follows a normal distribution, ϵ ∼ N (0, σ2 ).

5. The independent variables Tw,i and Tp,j are not highly correlated with each other.
Multicollinearity could distort the estimation of the regression coefficients.
Now, we estimate the coefficients β i and β j+3 using the ordinary least squares method
(OLS) [47,48]. This method minimizes the sum of the squared residuals, which is the
difference between the observed and predicted values of Tt .
n
β̂ 0 , β̂ 1 , . . . , β̂ 6 = arg min
β 0 ,β 1 ,...,β 6
∑ (Tt,k − T̂t,k )2 , (4)
k =1

where Tt,k is the actual total system time for the k-th patient, and T̂t,k is the predicted total
system time for the k-th patient based on the estimated regression coefficients.

3.4. Log-Transformed Multiple Linear Regression Model

The log-transformed multiple linear regression model (LTMLR) is derived from the
original model, MLR (see Equation (1)). This type of strategy has also been used in
health studies, such as [49,50]. We transform it using the natural logarithm, resulting in
the following:

3 3
log( Tt ) = β 0 + ∑ β i · Tw,i + ∑ β j+3 · Tp,j + ϵ, (5)
i =1 j =1

where the components of the function are the same as in Equation (1), except that now
log( Tt ) is the natural logarithm of the Tt .
We fit the model to revert the predictions from the log-transformed scale back to the
original scale and apply the exponential function to the predicted values:
 
Ŷ β 0 +∑3i=1 β i Tw,i +∑3j=1 β j+3 Tp,j
Tt = e = e , (6)
Mathematics 2024, 12, 3743 8 of 15

where Ŷ = log (Tt ) is the predicted log-transformed system time, and the exponential
transformation eŶ brings the prediction back to the original time scale.

3.5. Artificial Neural Network Model

We have now developed an ANN model to predict Tt based on several input features,
such as the waiting times and processing times at different stages of the patient care flow.
For this ANN, we follow a multilayer perception (MLP) architecture with one hidden layer
and the sigmoid activation function [17,51].
The input layer of the ANN receives the following variables, mentioned in
Sections 3.1.2 and 3.3 (i.e., Tw,r , Tp,r , Tw,v , Tp,v , Tw,d and Tp,d ), and which represent the key
factors that contribute to the Tt .

Neural Network Architecture

For the ANN model, we consider the following components: (1) input layer, i.e., it
receives the input features (waiting and processing times and the number of neurons in this
layer is six [it is equal to the number of input variables]); (2) hidden layer, i.e., it contains an
experiment between 10 and 20 neurons, each neuron applies a sigmoid activation function
to its input. The output of the hidden layer is calculated as follows:

(h) 1
zk = (h) (h)
, (7)
−(∑in=1 wk,i xi +bk )
1+e
(h)
where wk,i are the weights connecting the input variables to the neurons in the hidden
(h)
layer, and bk is the bias term for each neuron in the hidden layer.
In our model, the final predicted value of Tt is produced by the output layer, which
combines the outputs of the hidden layer using the following linear activation function:
m
∑ wk
(0) ( h )
T̂t = zk + b (0) , (8)
k =1

(0)
where wk are the weights connecting the hidden layer to the output layer, and b(0) is the
bias term in the output layer.
We use the sigmoid function as the activation function in the hidden layer to introduce
nonlinearity into the model. The sigmoid function is defined as follows:

1
f (z) = . (9)
1 + e−z
This activation function ensures that the output of each neuron in the hidden layer
is between 0 and 1, allowing our ANN model to capture complex relationships between
input variables.
For the loss function, we use MSE, which measures the squared difference between
the predicted T̂t and the Tt .

3.6. Statistical Metrics to Validate Regression Models

1. We measure the proportion of the variance of the dependent variable, Tt , that is
explained by the independent variables of the model [52,53]. To do so, we calculate
R2 as follows:

∑nk=1 ( Tt,k − T̂t,k )2

R2 = 1 − (10)
∑nk=1 ( Tt,k − T̄t,k )2

where Tt,k is the actual observed value of the dependent variable for the k-th obser-
vation (in this case, the total time in the system, Tt ); T̂t,k is the predicted value of Tt
Mathematics 2024, 12, 3743 9 of 15

from the regression model for the k-th observation; and T̄t is the mean of all observed
Tt values.
2. We also obtained MAE as a measure of error to determine the absolute difference
between the observed and predicted values of Tt .
n
1
MAE =
n ∑ |Tt,k − T̂t,k |. (11)
k =1

3. Finally, we calculate RMSE as another error measure, which allows us to mea-

sure the square root of the average squared differences between the observed and
predicted values.
s
n
1
RMSE =
n ∑ (Tt,k − T̂t,k )2 . (12)
k =1

3.7. Strategy for Reducing Patient Wait Times: A Decision Support Management
We propose a decision support strategy to minimize Tt in patient care by leveraging
optimization techniques and selecting the best regression model. We include multiple
simulations to explore the impact of various constraints and resource allocations, mak-
ing the strategy more comprehensive and adaptable. The details of its components are
presented below:
1. The objective is to minimize Tt , which is influenced by waiting times and processing
times at the three main stages of patient care: registration, vital signs, and medical
consultation. In this way, we define the objective function as follows:

3 3
MinTt = β 0 + ∑ β i · Tw,i + ∑ β j+3 · Tp,j . (13)
i =1 j =1

2. The key constraint in this optimization problem is the availability of resources such
as staff, technological infrastructure, and time. These resources directly impact the
possible reductions in waiting and processing times. Let Ri be the resources available
min and T min be the minimum
for each stage i (registration, vitals, and doctor). Let Tw,i p,j
feasible times for waiting and processing given available resources.
3. min and (2) T ≥ T min .
The constraints are (1) Tw,i ≥ Tw,i p,j p,j
4. Then, to optimize Tt , we introduce efficiency factors ηw,i and η p,j , where η ∈ (0, 1],
representing how much waiting and processing times can be reduced based on the
resources available. In this way, the optimized waiting and processing times are
opt opt
Tw,i = ηw,i · Tw,i ; and Tp,j = η p,j · Tp,j .
5. Finally, the optimized Tt of the system becomes the following:

3 3
= β 0 + ∑ β i · ηw,i · Tw,i + ∑ β j+3 · η p,j · Tp,j .
opt
Tt (14)
i =1 j =1

4. Results
The results of our research show an evaluation of the performance of the developed
regression models in terms of goodness of fit and prediction accuracy. We also present a
simulation of the decision support strategy that links the model with a time optimization
strategy to improve patient satisfaction.

4.1. Performance of Regression Models

We present, in Table 3, the performance metrics for our predictive models used to
estimate the Tt in a healthcare setting. The model is evaluated using several statistical
Mathematics 2024, 12, 3743 10 of 15

measures, which indicate how well the model fits the data and how accurate its predictions
are. The results confirm the MLR model presents the best performance.

Table 3. Comparative table of regression models (error metrics in minutes).

Model R2 MAE MSE RMSE

MLR 0.93 7.29 91.67 9.57
LTMLR 0.85 7.62 97.02 9.85
NN 0.86 7.05 190.04 13.79

4.2. Statistical Assumptions Results

We evaluate whether the MLR model meets the statistical assumptions. To do so, we
check the aspects discussed in Section 3.3.
From Figures 2 and 3 and the information provided by Tables 4 and 5, we confirm
that all these results indicate that the MLR model meets the key assumptions for linear
regression, making it a robust and valid model for predicting Tt .

Figure 2. Residuals vs. fitted values (linearity and homoscedasticity).

Figure 3. Q–Q plot of residuals (normality).

Mathematics 2024, 12, 3743 11 of 15

Table 4. Variance inflation factor (VIF).

Feature VIF
Registration wait time 1.40
Vitals wait time 1.64
Doctor wait time 1.60
Registration time 1.78
Vitals time 1.83
Doctor time 2.68

Table 5. Summary of statistical test results.

Test Value Interpretation

Durbin–Watson Close to 2 No autocorrelation in residuals (independence)
Breusch–Pagan Test (p -value) >0.05 Residuals have constant variance (homoscedasticity)
Shapiro–Wilk Test (p -value) >0.05 Residuals are normally distributed
Variance Inflation Factor (VIF) All < 10 No multicollinearity

4.3. Decision-Making Framework

According to the strategy defined in Section 3.7, we propose an example to evaluate
the reduction of patient total times in the system. Let us assume that we simulate resource
constraints based on staff availability at each stage:
min = 20 min
Stage 1: Registration has a staff of two, allowing for a minimum wait time of Tw,r
and a processing time of Tp,r min = 8 min.

Stage 2: Vitals has a staff of three, reducing waiting time to Tw,v min = 15 min and

processing time to Tp,v min = 4 min.

Stage 3: One doctor is available, limiting waiting time to Tw,dmin = 30 min and processing
min = 10 min.
time to Tp,d
Now, we simulate a scenario where (1) the registration process achieves an efficiency
of ηw,r = 0.85 (15% reduction in wait time) and η p,r = 0.9 (10% reduction in processing
time); (2) for vitals, ηw,v = 0.8 (20% reduction in wait time) and η p,v = 0.85 (15% reduction in
processing time); (3) for doctor consultation, ηw,d = 0.75 (25% reduction in wait time) and
η p,d = 0.85 (15% reduction in processing time).
We generate multiple scenarios with different levels of efficiency factors and resource
availability. For each scenario, we calculate Tt using the MLR model. For each scenario, we
opt
calculate Tt and compare it with Tt .
We assume the baseline times for a patient are (1) Tw,r = 30 min; (2) Tp,r = 10 min;
(3) Tw,v = 20 min; (4) Tp,v = 5 min; (5) Tw,d = 40 min; (6) Tp,d = 15 min. We use the
efficiency factors and the information above to determine the optimized times, which are
presented below:
opt opt
Registration: Tw,r = 0.85 · 30 = 25.5 min, Tp,r = 0.9 · 10 = 9 min.
opt opt
Vitals: Tw,v = 0.8 · 20 = 16 min, Tp,v = 0.85 · 5 = 4.25 min.
opt opt
Doctor: Tw,d = 0.75 · 40 = 30 min, Tp,d = 0.85 · 15 = 12.75 min.
opt
The result of calculating the optimized time is Tt = 25.5 + 9 + 16 + 4.25 + 30 + 12.75 =
97.5 min. Base Tt was Tt = 30 + 10 + 20 + 5 + 40 + 15 = 120 min. Therefore, the total system
opt
time reduction is ∆Tt = Tt − Tt = 120 − 97.5 = 22.5 min.

5. Discussion
In this work, we introduce a novel decision support strategy that integrates multiple
regression models with simulation techniques to optimize total system times in health-
care. The proposed strategy offers a significant contribution to the field of healthcare
operations management by providing a data-driven approach to reducing waiting times
Mathematics 2024, 12, 3743 12 of 15

and improving patient satisfaction. In contrast to traditional models, our approach lever-
ages multiple predictive models and applies them in a simulation framework, allowing
healthcare managers to evaluate and implement strategies in a controlled environment.
The results of the study demonstrate that the MLR model outperformed other models
in terms of predictive accuracy, with an R2 of 0.93, an MAE of 7.29 min, and an RMSE
of 9.57 min. These findings are consistent with prior studies that have highlighted the
robustness of MLR in healthcare settings (see, e.g., [42,54]). Additionally, the simulation
results show that the proposed strategy effectively reduced total system times by an average
of 22.5 min, with the most significant reductions observed in the registration and doctor
consultation stages.
MLR is useful for hospital decision-making by providing coefficients that show the
influence of each variable on total system time. However, in complex environments, wait
times are not always linear, especially during peak hours or when resources are scarce.
Additionally, factors like case severity also impact wait times, which MLR does not capture
in a simple linear relationship.
We chose thrre types of models to meet the different needs of our study: MLR for its
simplicity and interpretability, LTMLR to handle moderate nonlinear relationships and
reduce distributional biases, and ANN to capture more complex relationships in the data.
However, the limitations of each underscore that each model has areas where it could be
ineffective, which is important to consider in the interpretation of the results and in the
practical application of the study.
One of the limitations of this study is the reliance on a single dataset from a public
hospital. While the models performed well within this context, additional research is
needed to generalize these findings across different healthcare systems. Moreover, the
simulation framework could be further enhanced by incorporating more complex variables,
such as patient acuity and staff availability, to provide a more comprehensive analysis.
Additional limitations of our study include that it does not include detailed informa-
tion on patients’ clinical factors and their variability in system perception, which may affect
patient satisfaction. Furthermore, the study does not consider external factors that may
influence waiting times, such as disease outbreaks, health crises, or seasonal changes in
patient volume.
Despite these limitations, the methodology presented in this paper offers a robust
framework for reducing system inefficiencies and improving patient satisfaction. By
integrating predictive modeling with simulation, healthcare managers can test various
strategies and implement the most effective solutions without disrupting ongoing oper-
ations. Future research should explore the use of machine learning techniques to fur-
ther enhance predictive accuracy and expand the applicability of the model to different
healthcare environments.

6. Conclusions
In this work, we present an integrated decision support framework designed to
minimize total system times in healthcare by leveraging predictive modeling and simulation
techniques. The MLR model was identified as the most effective predictive model, achieving
an R2 of 0.93, an MAE of 7.29 min, and an RMSE of 9.57 min. The optimization framework
built around this model demonstrated significant reductions in system times in 22.5 min,
contributing to improved patient satisfaction and operational efficiency.
The contributions of this study are two-fold. First, we provide a comparative analysis
of multiple regression models to determine the most accurate predictor of system times.
Second, we integrate these models into a simulation-based optimization framework that
enables healthcare managers to implement strategies for reducing waiting times and
improving resource allocation.
Summarizing, the decision support strategy proposed in this paper offers a practical
and effective solution for optimizing system times in healthcare. By reducing waiting times
and improving resource allocation, the framework enhances both operational efficiency and
Mathematics 2024, 12, 3743 13 of 15

patient satisfaction, making it a valuable tool for healthcare managers aiming to improve
service delivery.
Future research should investigate the application of advanced machine learning
techniques to enhance predictive accuracy and broaden the model’s applicability across
diverse healthcare settings. For instance, approaches such as fuzzy multiple linear regres-
sion analysis could offer valuable insights. Furthermore, validation of the model across
different healthcare systems and its adaptation to more intricate patient care workflows
should be considered.
Incorporating real-time data into the simulation framework represents a critical avenue
for improving the model’s responsiveness to evolving healthcare dynamics. Additional
directions for future work include the following: integrating patient demographic data
to refine predictive accuracy; optimizing resource allocation based on patient profiles;
conducting longitudinal studies to evaluate the impact of wait time management on patient
satisfaction; developing a user-friendly interface that enables hospitals to intuitively utilize
the simulation and predictive model in real-time.
Moreover, adapting the model to specialized areas within the hospital, such as emer-
gency departments, intensive care units, or surgical units, could assess its versatility in
varied clinical contexts. Finally, integrating alert systems responsive to epidemiological or
climatic events could equip hospitals to anticipate and manage significant fluctuations in
patient volume effectively.

Author Contributions: Conceptualization, F.S.-A. and J.M.; data curation, F.S.-A. and P.S.; formal
analysis, F.S.-A. and J.M.; funding acquisition, J.M. and P.S.; investigation, F.S.-A. and J.M.; methodol-
ogy, F.S.-A., J.M. and P.S.; project administration, J.M.; supervision, J.M. and P.S.; writing—original
draft, F.S.-A., J.M. and P.S.; writing—review and editing, F.S.-A., J.M. and P.S. All authors have read
and agreed to the published version of the manuscript.
Funding: This research was funded by the “ANID Fondecyt Iniciacion a la Investigación 2024
N° 11240214”.
Data Availability Statement: The original contributions presented in the study are included in the
article, further inquiries can be directed to the corresponding author.
Conflicts of Interest: The authors declare no conflicts of interest.

References
1. Parente, D.H.; Pinto, M.B.; Barber, J.C. A pre-post comparison of service operational efficiency and patient satisfaction under
open access scheduling. Health Care Manag. Rev. 2005, 30, 220–228. [CrossRef] [PubMed]
2. Boland, G.W.; Duszak, R.; McGinty, G.; Allen, B. Delivery of appropriateness, quality, safety, efficiency and patient satisfaction. J.
Am. Coll. Radiol. 2014, 11, 7–11. [CrossRef] [PubMed]
3. Martinez, D.A.; Zhang, H.; Bastias, M.; Feijoo, F.; Hinson, J.; Martinez, R.; Dunstan, J.; Levin, S.; Prieto, D. Prolonged wait time is
associated with increased mortality for Chilean waiting list patients with non-prioritized conditions. BMC Public Health 2019,
19, 233. [CrossRef] [PubMed]
4. Silva-Aravena, F.; Álvarez-Miranda, E.; Astudillo, C.A.; González-Martínez, L.; Ledezma, J.G. Patients’ prioritization on surgical
waiting lists: A decision support system. Mathematics 2021, 9, 1097. [CrossRef]
5. Silva-Aravena, F.; Gutiérrez-Bahamondes, J.H.; Núñez Delafuente, H.; Toledo-Molina, R.M. An intelligent system for patients’
well-being: A multi-criteria decision-making approach. Mathematics 2022, 10, 3956. [CrossRef]
6. Khani, R.A. Improving Waiting Time and Patients’ Experience in a Medical Retina Clinic. Ph.D. Thesis, Royal College of Surgeons
in Ireland, Dublin, Ireland, 2015.
7. Godley, M.; Jenkins, J.B. Decreasing wait times and increasing patient satisfaction: A lean six sigma approach. J. Nurs. Care Qual.
2019, 34, 61–65. [CrossRef]
8. Pushadapu, N. Optimization of Resources in a Hospital System: Leveraging Data Analytics and Machine Learning for Efficient
Resource Management. J. Sci. Technol. 2020, 1, 280–337.
9. Toma, M.; Wei, O.C. Predictive modeling in medicine. Encyclopedia 2023, 3, 590–601. [CrossRef]
10. Nwaimo, C.S.; Adegbola, A.E.; Adegbola, M.D. Transforming healthcare with data analytics: Predictive models for patient
outcomes. GSC Biol. Pharm. Sci. 2024, 27, 25–35. [CrossRef]
11. Rackauckas, C.; Ma, Y.; Noack, A.; Dixit, V.; Mogensen, P.K.; Byrne, S.; Maddhashiya, S.; Santiago Calderón, J.B.; Nyberg, J.;
Gobburu, J.V.; et al. Accelerated predictive healthcare analytics with pumas, a high performance pharmaceutical modeling and
simulation platform. BioRxiv 2020 . [CrossRef]
Mathematics 2024, 12, 3743 14 of 15

12. Alshareef, I.M.A.; Al Shaman, H.M.H.; hadi Al Mansour, I. The Role of Data Analytics In Medical Administration: Leveraging
Information For Decision-Making. J. Namib. Stud. Hist. Politics Cult. 2023, 36, 12–23.
13. AlBakary, W.A.; Sedky, A.A.H.; Abdelmoez, W. Interactive Bottleneck Detection in Data Driven Business Process Simulation
in Healthcare: Egyptian Case Study. In Proceedings of the 2022 32nd International Conference on Computer Theory and
Applications (ICCTA), Alexandria, Egypt, 17–19 December 2022; pp. 234–240.
14. Varanasi, S.; Malathi, K. Self-improved COOT optimization-based LSTM for patient waiting time prediction. Multimed. Tools
Appl. 2024, 83, 39315–39333. [CrossRef]
15. Yanamala, A.K.Y. Cost-Sensitive Deep Learning for Predicting Hospital Readmission: Enhancing Patient Care and Resource
Allocation. Int. J. Adv. Eng. Technol. Innov. 2022, 1, 56–81.
16. Pak, A.; Gannon, B.; Staib, A. Predicting waiting time to treatment for emergency department patients. Int. J. Med. Inform. 2021,
145, 104303. [CrossRef] [PubMed]
17. Kuo, Y.H.; Chan, N.B.; Leung, J.M.; Meng, H.; So, A.M.C.; Tsoi, K.K.; Graham, C.A. An integrated approach of machine learning
and systems thinking for waiting time prediction in an emergency department. Int. J. Med. Inform. 2020, 139, 104143. [CrossRef]
18. Srinivas, S. A machine learning-based approach for predicting patient punctuality in ambulatory care centers. Int. J. Environ. Res.
Public Health 2020, 17, 3703. [CrossRef]
19. Yousefi, M.; Yousefi, M.; Fathi, M.; Fogliatto, F.S. Patient visit forecasting in an emergency department using a deep neural
network approach. Kybernetes 2020, 49, 2335–2348. [CrossRef]
20. Sudarshan, V.K.; Brabrand, M.; Range, T.M.; Wiil, U.K. Performance evaluation of Emergency Department patient arrivals
forecasting models by including meteorological and calendar information: A comparative study. Comput. Biol. Med. 2021,
135, 104541. [CrossRef]
21. Zhang, H.; Best, T.J.; Chivu, A.; Meltzer, D.O. Simulation-based optimization to improve hospital patient assignment to physicians
and clinical units. Health Care Manag. Sci. 2020, 23, 117–141. [CrossRef]
22. Tavakoli, M.; Tavakkoli-Moghaddam, R.; Mesbahi, R.; Ghanavati-Nejad, M.; Tajally, A. Simulation of the COVID-19 patient flow
and investigation of the future patient arrival using a time-series prediction model: A real-case study. Med Biol. Eng. Comput.
2022, 60, 969–990. [CrossRef]
23. Pianykh, O.S.; Guitron, S.; Parke, D.; Zhang, C.; Pandharipande, P.; Brink, J.; Rosenthal, D. Improving healthcare operations
management with machine learning. Nat. Mach. Intell. 2020, 2, 266–273. [CrossRef]
24. Amissah, M.; Lahiri, S. Modelling granular process flow information to reduce bottlenecks in the emergency department.
Healthcare 2022, 10, 942. [CrossRef] [PubMed]
25. Safdar, K.A.; Emrouznejad, A.; Dey, P.K. An optimized queue management system to improve patient flow in the absence of
appointment system. Int. J. Health Care Qual. Assur. 2020, 33, 477–494. [CrossRef] [PubMed]
26. Ansah, J.P.; Ahmad, S.; Lee, L.H.; Shen, Y.; Ong, M.E.H.; Matchar, D.B.; Schoenenberger, L. Modeling Emergency Department
crowding: Restoring the balance between demand for and supply of emergency medicine. PLoS ONE 2021, 16, e0244097.
[CrossRef]
27. Lamé, G.; Dixon-Woods, M. Using clinical simulation to study how to improve quality and safety in healthcare. BMJ Simul.
Technol. Enhanc. Learn. 2020, 6, 87. [CrossRef]
28. Ordu, M.; Demir, E.; Tofallis, C.; Gunal, M.M. A novel healthcare resource allocation decision support tool: A forecasting-
simulation-optimization approach. J. Oper. Res. Soc. 2021, 72, 485–500. [CrossRef]
29. Jahangiri, S.; Abolghasemian, M.; Ghasemi, P.; Chobar, A.P. Simulation-based optimisation: Analysis of the emergency department
resources under COVID-19 conditions. Int. J. Ind. Syst. Eng. 2023, 43, 1–19. [CrossRef]
30. Faeghi, S.; Lennerts, K.; Nickel, S. A system dynamics model application to operating room planning and management. J. Simul.
2023, 17, 58–75. [CrossRef]
31. Manzoor, F.; Wei, L.; Hussain, A.; Asif, M.; Shah, S.I.A. Patient satisfaction with health care services; An application of physician’s
behavior as a moderator. Int. J. Environ. Res. Public Health 2019, 16, 3318. [CrossRef]
32. Abidova, A.; Alcântara da Silva, P.; Moreira, S. Predictors of patient satisfaction and the perceived quality of healthcare in an
emergency department in Portugal. West. J. Emerg. Med. Integr. Emerg. Care Popul. Health 2020, 21, 391.
33. Sonis, J.D.; Aaronson, E.L.; Lee, R.Y.; Philpotts, L.L.; White, B.A. Emergency department patient experience: A systematic review
of the literature. J. Patient Exp. 2018, 5, 101–106. [CrossRef] [PubMed]
34. De Simone, S.; Planta, A.; Cicotto, G. The role of job satisfaction, work engagement, self-efficacy and agentic capacities on nurses’
turnover intention and patient satisfaction. Appl. Nurs. Res. 2018, 39, 130–140. [CrossRef] [PubMed]
35. Bronick, A.K. Reducing Patient Wait Times in a Primary Care Setting to Improve Patient Satisfaction. Ph.D. Thesis, Georgetown
University, Washington, DC, USA, 2024.
36. Ala, A.; Alsaadi, F.E.; Ahmadi, M.; Mirjalili, S. Optimization of an appointment scheduling problem for healthcare systems
based on the quality of fairness service using whale optimization algorithm and NSGA-II. Sci. Rep. 2021, 11, 19816. [CrossRef]
[PubMed]
37. Adhicandra, I.; Nurhidayati, S.; Fauzan, T.R. Optimization of Hospital Queue Management Using Priority Queue Algorithm and
Reinforcement Learning for Emergency Service Prioritization. Int. J. Softw. Eng. Comput. Sci. (IJSECS) 2024, 4, 512–522. [CrossRef]
Mathematics 2024, 12, 3743 15 of 15

38. Vázquez-Serrano, J.I.; Cárdenas-Barrón, L.E.; Vicencio-Ortiz, J.C.; Matis, T.; Gaitán-Mercado, C.M.; Peimbert-García, R.E. Hybrid
optimization and discrete-event simulation model to reduce waiting times in a primary health center. Expert Syst. Appl. 2024,
238, 121920. [CrossRef]
39. Silva-Aravena, F.; Delafuente, H.N.; Astudillo, C.A. A novel strategy to classify chronic patients at risk: A hybrid machine
learning approach. Mathematics 2022, 10, 3053. [CrossRef]
40. Silva-Aravena, F.; Morales, J. Dynamic surgical waiting list methodology: A networking approach. Mathematics 2022, 10, 2307.
[CrossRef]
41. Loftus, T.J.; Altieri, M.S.; Balch, J.A.; Abbott, K.L.; Choi, J.; Marwaha, J.S.; Hashimoto, D.A.; Brat, G.A.; Raftopoulos, Y.; Evans,
H.L.; et al. Artificial Intelligence–enabled Decision Support in Surgery: State-of-the-art and Future Directions. Ann. Surg. 2023,
278, 51–58. [CrossRef]
42. Boonprasope, A.; Tippayawong, K.Y. Predicting Healthcare Mutual Fund Performance Using Deep Learning and Linear
Regression. Int. J. Financ. Stud. 2024, 12, 23. [CrossRef]
43. Malik, M.; Abdallah, S.; Ala’raj, M. Data mining and predictive analytics applications for the delivery of healthcare services: A
systematic literature review. Ann. Oper. Res. 2018, 270, 287–312. [CrossRef]
44. Kaiser, J. Dealing with Missing Values in Data. J. Syst. Integr. 2014, 5, 42–51. [CrossRef]
45. Ibrahim, E.; Shouman, M.A.; Torkey, H.; El-Sayed, A. Handling missing and outliers values by enhanced algorithms for an
accurate diabetic classification system. Multimed. Tools Appl. 2021, 80, 20125–20147. [CrossRef]
46. Singh, M.; Dixit, S.; Ranjan, R.; Sharma, R.; Singla, R.; Goyal, P. Correlation between Waiting Time and Satisfaction Among Those
Attending Outpatient Departments of a Tertiary Care Social Security Hospital, Faridabad, Haryana. Med. J. Dr. D Y Patil Univ.
2020, 13, 480–485. [CrossRef]
47. Safe, M.S.; Barouni, M.; Saif, S.M. Health impact on Economy by Artificial Neural Network and Dynamic Ordinary Least Squares.
Health Manag. Inf. Sci. 2017, 4, 107–113.
48. Kaya Samut, P.; Cafrı, R. Analysis of the efficiency determinants of health systems in OECD countries by DEA and panel tobit.
Soc. Indic. Res. 2016, 129, 113–132. [CrossRef]
49. Wang, H.; Sambamoorthi, N.; Robinson, R.D.; Knowles, H.; Kirby, J.J.; Ho, A.F.; Takami, T.; Sambamoorthi, U. What explains dif-
ferences in average wait time in the emergency department among different racial and ethnic populations: A linear decomposition
approach. J. Am. Coll. Emerg. Physicians Open 2024, 5, e13293. [CrossRef]
50. Pereira, V.; Oyekunle, T.; Janes, J.; Amling, C.J.; Aronson, W.J.; Cooperberg, M.R.; Kane, C.J.; Terris, M.K.; Klaassen, Z.; Freedland,
S.J.; et al. Time from biopsy to radical prostatectomy by race in an equal-access healthcare system: Results from the SEARCH
cohort. Prostate 2023, 83, 1011–1019. [CrossRef]
51. Choudhury, A.; Urena, E. Forecasting hourly emergency department arrival using time series analysis. Br. J. Healthc. Manag.
2020, 26, 34–43. [CrossRef]
52. Al-Mousa, A.; Al-Zubaidi, H.; Al-Dweik, M. A machine learning-based approach for wait-time estimation in healthcare facilities
with multi-stage queues. IET Smart Cities 2024. [CrossRef]
53. Hijry, H.M. Crowding Reduction and Waiting Time Analysis in Health-Care System Using Machine Learning. Ph.D. Thesis,
Oakland University, Rochester, MI, USA, 2022.
54. Li, Y.; Cao, L.; Zhang, J.; Jiang, Y.; Han, Y.; Wei, J. Energy benchmarking in healthcare facilities: A comparative study. J. Constr.
Eng. Manag. 2021, 147, 04021159. [CrossRef]

Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual
author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to
people or property resulting from any ideas, methods, instructions or products referred to in the content.