Artificial Intelligence Enabled Healthcare Ecosystem Model:

AIEHEM Project

Luigi Lella

, Ignazio Licata

and Christian Pristipino

ISEM, Ins. For Scientific Methodology, PA, Italy

Interventional and Emergency Cardiology Unit, San Filippo Neri –ASL Roma 1, Rome, Italy

Keywords: Healthcare Management Systems, Data Mining and Data Analysis, Decision Support Systems.

Abstract: The AIEHEM project aims to analyze the data made available by the regional health system, using an

unorganized Turing machine model (A-Type) trained with a swarm-evolutionary hybrid algorithm. The goal

is to identify the main factors related to certain outcomes that the healthcare organization intends to achieve

(which can be economic, organizational, social or environmental). The chosen AI model is used to enhance,

not to replace the analytical capabilities of the healthcare system management. The insights of the AI model

are in fact used not only to identify the main objects of study to be taken into consideration, but also to

define the areas of intervention and consequently also the stakeholders to be involved in the organizational

change project to be carried out through the Theory of Change methodology. AI is therefore used to identify

the most suitable ecosystem for solving the considered problem.

1 INTRODUCTION

The development of decision support systems (DSS)

is a potentially gamechanger for executives and

managers of health systems and organizations

because they offer the possibility of managing an

elevate number of variables (Longaray et al., 2016;

Khademolqorani and Hamadani, 2015). These

particular DSSs are indeed characterized by a high

level of complexity that leads to the definition of

groups or hierarchies of variables to be taken into

account to solve certain critical issues and problems.

As health is a good of a fundamental and

irreplaceable nature (Diaby et al., 2013), there is a

high level of responsibility in the decision of

adopting IT solutions to support management. This

is particularly true if one considers that wrong

management decisions can directly endanger

patients, but it can also happen that the improvement

of the medical state of some patients can negatively

affect that of other patients (Marsh et al., 2014).

Unfortunately, such DSSs are often based on

machine learning and AI algorithms working as

"black boxes", where the assumptions of their

predictions and/or choices are concealed (Academy

of Medical Royal Colleges, 2019). This increases the

difficulty in assessing the degree of reliability of

such systems, making them particularly vulnerable

to bias and deliberately malicious attacks.

Among the main machine learning and AI

models used to implement DSSs are algorithms

based on decision trees, linear and logistic

regression, Bayesian inference and classification

(Bashir et al., 2014; Zandi, 2014; Roumani et al.,

2013).

Such systems often use Multicriteria Decision

Analysis methods (MCDA) and Multicriteria

Decision Making methods (MCDM) (Aghdaiea et al.

2014).

MCDAs are algorithms that allow to simplify

complex problems by bringing them back to a series

of elementary criteria to be considered in finding the

solution (Angelis et al., 2017). In recent decades,

MCDAs have been applied in various areas

including the management of health systems

(Longaray et al., 2016). Specifically, they have been

used in the clinical (Gasol et al., 2022; Berner, 2007)

and in the health management fields (Marsh et al.,

2014; Ju et al., 2012; Wu et al., 2007; Baltussen et

al, 2006).

MCDMs are algorithms that allow to find

solutions in the presence of multiple objectives (San

Cristobal, 2013) and also have been used been used,

among others, in the clinical (Bashir et al. 2014) and

232

Lella, L., Licata, I. and Pristipino, C.

Artiﬁcial Intelligence Enabled Healthcare Ecosystem Model: AIEHEM Project.

DOI: 10.5220/0011604300003414

In Proceedings of the 16th International Joint Conference on Biomedical Engineering Systems and Technologies (BIOSTEC 2023) - Volume 5: HEALTHINF, pages 232-238

ISBN: 978-989-758-631-6; ISSN: 2184-4305

 2023 by SCITEPRESS – Science and Technology Publications, Lda. Under CC license (CC BY-NC-ND 4.0)

in the health management fields (Narci et al., 2014;

Ishizaka and Nemery, 2013).

Both MCDAs and MCDMs work by grouping or

ranking possible alternative solutions to

administrative problems. For this reason they also

behave like decision-making black boxes that hide

the criteria that guided the choice of the suggested

solution, as well as the biases and all the other

factors that influenced the selection. Therefore, it is

crucial instead to develop algorithms that may help

in explaining the causal links between the variables

considered in the identification of the solution. This

will allow a posteriori to verify whether the

automated decision-making process has been based

on a sufficient number of data, whether these are

characterized by a sufficient level of quality and

consistency, whether or not their selection has been

influenced by cognitive biases. Indeed, a simplified

logical representation of the causal links between the

key variables selected by the DSSs, can help human

decision-makers in assessing the reliability of the

system and in making the most appropriate

decisions.

All these considerations led to the definition of

the project called "Artificial Intelligence Enabled

Healthcare Ecosystem Model" (AIEHEM), a DSS

based on an innovative AI model capable of

providing not only accurate predictions on the

outcome of certain operational or management

activities, but also to identify the main critical

factors correlated with the achievement of certain

organizational objectives.

The example shown in this article is related to

the optimization of the management of patients at

high risk of death. In this preliminary phase of

analysis we decided to use the exceeding the

regional expenditure threshold linked to patient

treatment as a proxy variable correlated with this

organizational outcome.

The used AI model is able not only to

autonomously identify the independent variables

most closely related to the chosen dependent

variable, but it is also able to tie them together

within a single logical formula. This knowledge

representation can be considered as a kind of

guideline used to evaluate the possibility of

incurring an overrun of care costs.

The advantage of defining a rule expressed

through a logical formalism is that it allows a group

of human experts to evaluate subsequently the

appropriateness of the reasoning followed by the AI

model. Just examining the set of independent

variables taken into consideration by the system in

making its predictions, it is possible to understand if

it has left out important factors of analysis and if its

reasoning was influenced by the cognitive biases of

the human experts who selected the data to be

analyzed. Furthermore, once the appropriateness of

the logical formula proposed by the AI model is

verified, it is possible to identify the stakeholders of

the healthcare ecosystem to be involved in order to

find a solution to the analysed organizational

problem. The next processing of the AI model on the

updated database can allow to understand whether or

not the healthcare ecosystem has managed to

intervene effectively on the critical factors identified

by the system and if other critical factors to be taken

into consideration have emerged over time. In this

way it is possible to establish a Deming cycle (plan-

do-check-act) to progressively improve the results

achieved by the healthcare ecosystem (Taylor et al.,

2014).

The AIEHEM project aims to assess the

multidimensional key factors which are related to

the strategic goals of a regional health system, by

using a specific artificial intelligence model.

Strategic goals can be: economic (e.g.: optimize the

consumption of system resources in chronicity, etc.),

organizational (e.g.: reduce waiting lists, increase

the quality of services provided, etc.), social (e.g.:

highlight the risk conditions for a pro-active risk

management, etc.) and environmental (e.g.: reduce

the environmental impact of health facilities, etc.).

The innovation of this project is based on the

specific methods, as well as on its global, systemic

approach. For the first time, in fact, an AI model will

support the identification of the actors, the facilities,

the activities, the knowledge, the opportunities, and

the most suitable ecosystem to achieve the the

regional health system goals, involving the relevant

end users.

The AIEHEM project is in line with the "One

Health" approach described by the 6

Mission of the

National Recovery and Resilience Italian Plan, as it

aims to promote interdisciplinarity and inter-

professionalism through the enhancement of

available health system data and information. The

purpose is to enhance the efficacy of planning and

decision-making of social and health services

through the use of an AI model built to simplify the

volume and complexity of social-health interactions.

The collection, refinement and processing of the

available data will be in compliance with the code

for the protection of personal data (GDPR) and with

the ethical values and fundamental principles of

health promotion and public health, through the

involvement of the data protection officer and the

regional ethics committee.

Artiﬁcial Intelligence Enabled Healthcare Ecosystem Model: AIEHEM Project

233

2 DSS DESCRIPTION

The recent pandemic crisis has highlighted the

problems that can arise from poor healthcare

management or unoptimized health system

resources. All these problems derive essentially from

the lack of effective cooperation strategies between

the various regional stakeholders resulting in an

insufficient integration of territory services,

hospitals and social services in facing health

challenges.

The AIEHEM project is aimed at supporting the

regional health management with an AI-based DSS

to identify the relevant areas of intervention,

stakeholders, and resources associated to an

effective achievement of their strategic objectives.

As an example, we present here the use of the

AIEHEM methodology to optimize the management

of system resources for patients at high risk of death

in Marche region (Italy). Particularly, we focus on

the need of enhancing sustainability by reducing the

costs of inappropriate use of resources in the last

quarter of life of the patients, which also lead to an

unjustified and dangerous delay of the diagnostic

services for patients with the right indications.

The strategic aim was to encourage palliative

care by avoiding clinical investigations that do not

help in improving the prognosis of patients. In order

to identify the factors related to an above-average in-

hospital consumption of resources in patients in the

last trimester of life, the AI model analysed the

2019 hospital discharge records (12344 records).

Data relating to the pandemic period were excluded

as the allocation of resources for the management of

the pandemic could influence the outcome of the

analysis.

The variables taken into account by the AI

model are the following: gender, age class, type of

hospitalization, method of discharge from the

previous structure, main diagnosis category,

residency, the exceeding the regional threshold for

assistance expenditure. The sex variable assumed

two values (male or female), the age was divided

into 5 strata (0-60, 61-70, 71-80, 81-90, 90+), the

type of hospitalization (reqType) was cathegorised

in 4 values (scheduled not urgent, urgent,

compulsory treatment, scheduled with

prehospitalization), the method of discharge

(disMode) encompassed 8 values (without the

proposal of the family doctor, with sending of the

family doctor, scheduled discharge, discharge from

public hospital, discharge from accredited private

structure, discharge from non-accredited private

structure, discharge from other hospitalization

regime, discharge from emergency urgency

department). In addition, 25 major diagnostic

cathegories (MDC) values were taken into account

and the regional threshold for assistance expenditure

was set by calculating the average diagnosis related

group (DRG) value, equal to 4579.71 euros

(Mistichelli, 1984).

To identify the analysis variables most correlated

with the problem that we intend to investigate, we

have chosen to use a single-state type A model of

Unorganized Turing Machine (UTM) (Turing,

1948), consisting of a combinatorial network of

NAND gates whose optimal configuration is

selected by the evolution of a population of

individuals each of which represents the encoding of

a UTM configuration. In this way the UTM is

generated "in an unsystematic and random way"

from a set of two-input NAND gates. Turing chose a

NAND gate because any other logical operation can

be performed by a set of NAND gates. An

Unorganized Turing Type A Machine can be

considered "a kind of Boolean neural network

without a layered structure, since recurrent

connections are allowed without constraints"

(Teuscher and Sanchez, 2000).

Every possible configuration of the NAND gates

that make up the UTM was coded with a binary

vector and to identify the optimal configuration, that

is the vector that would allow to maximize the

predictive accuracy of the model, a swarm-

evolutionary hybrid algorithm was used, which we

have called the Evolutionary Bait Balls Model

(EBBM), in which NAND gate configuration

vectors are considered as individual members of a

swarm. Each of them is able to perform only three

elementary operations (repulsion from others,

attraction to another particularly performing

individual, orientation towards others). The

evolution of this population leads to the appearance

of emerging behaviors (the state in which a sort of

bubble is formed in which most individuals tend to

orient themselves with respect to others),

manifesting a kind of collective intelligence (Lella et

al., 2022).

The original evolutionary model of the bait ball,

which inspired our EBBM, was developed by

researchers who found that within the group of fish

trying to escape predators a spontaneously generated

core constitutes what they called "selfish herd"

(Roberts, 2021; Yang, 2018). This denomination

comes from the selfish theory of the pack according

to which individuals within the population attempt to

reduce the risk of predation by placing other

conspecifics between them and predators. Returning

HEALTHINF 2023 - 16th International Conference on Health Informatics

234

to the bait ball model, it is precisely this "selfish"

behavior adopted by individuals that leads to the

formation of the optimal collective configuration.

The EBBM algorithm used as an optimization

algorithm can be described as follows:

Input: Array of individuals I to be

updated

Output: The position vectors (binary

vectors) of each individual in I will be

changed.

1: call function to alter the positions

of each individual

2: for all i ϵ I do

3: perform ZOR, ZOA, ZOO sets

calculations

4: if individual detected in ZOR then

5: perform repulsion (R)

6: else if individual detected in ZOO

then

7: perform orientation (O)

8: else if individual detected in ZOA

then

9: perform attraction (A)

10: end if

11: end for

Where ZOR is the Repulsion Zone: one

individual cannot occupy the position of another,

that is, it cannot be represented by the same binary

vector. In this case it assumes another random

position (every single bit of the individual is

modified with probability RepulsionRate). ZOA is

the Zone of Attraction: an individual tends to

approach individuals characterized by a greater

fitness (with a probability equals to the attraction

rate, every single bit of the individual can assume

the same value of the bit in the same position of the

best performing individual in the ZOA set). ZOO is

the Orientation Zone: an individual tends to orient

itself, among the individuals close to it, towards the

most performing one (with a probability equals to

the orientation rate, every single bit of the individual

can assume the same value of the bit in the same

position of the best performing individual in the

ZOO set). To define the sets ZOR, ZOA, ZOO, the

parameters ZORrange, ZOArange and ZOOrange

were introduced, representing the maximum number

of different bits between the vector of the individual

considered and that of the individual belonging

respectively to the ZOR, ZOA and ZOO zone

(attraction rate=0.05, orientation rate=0.3, repulsion

rate=0.5, ZOA range=70, ZOO range= 5, ZOR range

= 0). The fitness function of the individual is set as

the prediction accuracy of the corresponding UTM.

With this swarm-evolutionary hybrid algorithm,

better results are obtained in terms of predictive

accuracy than other classical evolutionary models

such as the genetic algorithm, as demonstrated in

(Lella et al., 2022) where the EBBM model was

used to implement an expert system capable of

diagnosing with a fair level of accuracy the risk of

incurring type II diabetes mellitus.

Every possible UTM configuration, which

corresponds to a given binary vector, has been coded

as follows. The first 59 bits represent the values that

can take all the classes of variables that can be

selected for the first input of the NAND gates of the

UTM model. The following 59 bits represent the

values that can take all selectable variable classes for

the second input of the NAND gates of the UTM

model. The remaining 54 bits were used to encode

the architecture of the 18 available NAND gates.

The values of these variables were encoded in binary

format using a single bit for the sex variables and the

exceeding of the regional spending threshold and a

"one-hot" encoding for all the others, that is, using n

bits for all the n possible values of the variable and

valuing to 1 only the bit whose position is associated

with the corresponding category. The variable of the

exceeding of the regional expenditure threshold is

considered as class variable, all the other ones are

considered non-class variables.

Each NAND gate has been encoded with three

bits. If the value of the first bit is 1 the first input of

the NAND gate considered is a first class of input

variable, otherwise the first input is connected to the

output of the next NAND gate. If the value of the

second bit is 1 the second input of the considered

NAND gate is a second class of input variable,

otherwise the first input is connected with the output

of another NAND gate. If the value of the third bit is

1 it means that the inputs of the NAND gates

considered are short-circuited and only the first input

should be considered. In this way each individual,

which represents a possible NAND network

configuration, is represented by 59+59+54=172 bits.

To represent a combinatorial NAND, when a

individual is tested for suitability, all the first classes

of input variables, all the second classes of input

variables, and all 18 available NAND gates are

selected sequentially once. The first NAND gate

(NAND#1) of the 54-bit sequence is the network

output gate. If the first bit of its code is 1, NAND#1

input 1 is the first input class variable that can be

selected. If the first bit of its code is 0, input 1 of

NAND#1 is the output of NAND#2, the code of

which is represented by the following three bits of

the 54-bit sequence. If the second bit of NAND#1 is

1, NAND#1 input 2 is the first input class variable

that can be selected. If the second bit of NAND#1 is

0, input 2 of NAND#1 is the output of NAND#3, the

Artiﬁcial Intelligence Enabled Healthcare Ecosystem Model: AIEHEM Project

235

code of which is represented by the third bit triplet

within the 54-bit sequence.

3 EBBM UTM PERFORMANCE

The performance of the UTM model trained by

EBBM (EBBM UTM) was compared with other AI

and machine learning models that allow to model an

explicit representation of the causal links identified

between the study variables considered. All the

models were trained using 60% of the available data

and tested with the remaining 40% of the data. Table

1 shows the predictive accuracies of the tested

models together with precision, recall and F1-score

measures. Table 2 reports the number of decision

nodes that make up the model and the number of

non-class variables taken into account by the models

to make their predictions.

The ZeroR model (Witten et al., 2011) was used

as a benchmark to verify that all other algorithms

used have been configured and used correctly.

ZeroR always predicted the most frequent class

variable in the presence of any combination of input

variables. Given its simplicity it typically had a

much lower level of predictive accuracy than the

other algorithms that have been tested. Alternatively,

the result found may be due to a bad selection and

encoding of the input data with which the models

were trained and tested or to a bad configuration of

the models used.

The OneR, which stands for "one Rule" (Holte,

1993), is nothing more than a one-level decision

tree. In various areas and predictive tasks this model

has proved to be much more performing than other

more complex models, and it is always appropriate

to verify whether the problem under consideration

can be effectively treated using this model that uses

a reduced amount of resources.

The J48 (Witten et al., 2011) is a decision tree

based on the "divide and conquer" strategy used

recursively. At each training step the node

characterized by the highest amount of information

is selected and split into a series of nodes

corresponding to some possible values that the

original node can assume. The process ends when all

instances considered reference the same value as the

class attribute.

The Bayesian network (Ben-Gal, 2007) is a

probabilistic graphic model that represents a set of

stochastic variables with their conditional

dependencies through the use of a direct acyclic

graph.

A random forest is an aggregate classifier

obtained by bagging aggregation of decision trees.

The name comes from the random decision forests

that were first proposed by Kam Ho (1995).

Table 1: Prediction accuracy of the AI tested models.

Model

Prediction

Accuracy

(%)

Precision Recall

F1-

Score

Zero R 72.11 0 0 0

One R 80.57 0.731 0.48 0.58

J 48 81.29 0.792 0.446 0.571

Bayes

Network

80.70 0.745 0.469 0.575

Random

Fores

75.92 0.578 0.508 0.541

EBBM

UTM

81.14 0.796 0.435 0.563

Compared to the other considered machine

learning and AI models, the UTM EBBM allowed to

create a predictive model characterized by a limited

number of nodes and which also takes into account

only a subset of the non-class variables presented in

making its predictions.

Table 2: Complexity level of the final knowledge

representations.

Model

Number

of nodes

Considered variables

Zero R 0 -

One R 1 MDC

J 48 31

MDC, type of

hospitalization, residency,

mode of discharge, age,

gende

Bayes Network 26

MDC, type of

hospitalization, residency,

mode of discharge, age,

gende

Random Forest 30

MDC, type of

hospitalization, residency,

mode of discharge, age,

gende

EBBM UTM 17

MDC, type of

hospitalization, mode of

discharge

This model allows human experts to evaluate the

criteria used by the model, explaining possible biases

in the selection of the data used for its training.

Because of its simplicity, the model obtained through

the UTM EBBM, could also be used, after a proper

validation by a team of human experts, to write

decision-making guidelines to be adopted by

managers or clinicians (figure 1).

HEALTHINF 2023 - 16th International Conference on Health Informatics

236

Figure 1: UTM forecast model selected through EBBM.

Only the UTM EBBM model was able to

establish that sex, age and residency were negligible

variables. Consequently, this finding allows to

exclude the referents of the territorial districts as

stakeholders to be involved. Also no sex-based or

age-based inequalities were observed, therefore the

involvement of associations of patients who deal

with them may not be necessary. The same AI

model built in this first analysis suggested to involve

psychiatric and hospital services in the management

of certain types of scheduled hospitalizations.

However, to be noted that the choice to use the

exceeding of the regional spending threshold as a

class variable may have introduced cognitive bias.

We will repeat the analysis of the UTM EBBM

model using different proxies to assess the

appropriateness of the assistance that has been

provided to patients close to death. Alternative

proxies could be given by the number of accesses to

the emergency room or the cost of drug therapies

that have been administered to patients close to

death. These new simulations will allow to validate

or not the predictive model that has been defined.

ACKNOWLEDGEMENTS

Special thanks to Dr. Remo Appignanesi, Dr. Maria

Rita Mazzoccanti, Dr. Antea Maria Pia Mangano,

Dr. Elena Di Tondo, Dr. Marco Morbidoni, Dr.

Pietro Serafini, Dr. Mariaflavia Spagna, Dr. Cristina

Omenetti and Dr. Cristiana Sisti for their kind and

valuable support.

4 CONCLUSIONS

The AIEHEM project has the ambitious goal of

suggesting to the organizational management the

most suitable strategy to undertake in achieving a

certain long-term outcome, using the support of an

AI model.

The chosen model (EBBM-based UTM) is able

not only to identify the critical factors related to the

achievement of a certain strategic objective, but through

the extrapolation of a rule that binds them it can help to

better understand the phenomenon that underlies the

achievement of the objective, suggesting which

stakeholders to involve in defining an adequate

intervention strategy.

The main beneficiary of the outcomes deriving from

the adoption of the AIEHEM methodology is the patient.

In the given example aimed at optimizing the system

resources in the care of patients at risk of death, the

medium-long term advantages are mainly the

improvement in the quality of palliative care and the

reduction of the waiting lists for specialty visits and

instrumental examinations. As a matter of fact, our

analysis revealed the use of unnecessary investigations,

perhaps deriving from defensive medicine practices.

Therefore, the insights obtained with the AIEHEM

methodology may support the staff operating in the

territorial and hospital structures to better coordinate their

activities and optimize the use of their resources. The

enhanced quality of services will therefore affect not only

the patients, but also their family members and caregivers

who would be given more precise instructions on how to

manage these patients.

It should be underlined that the AI model taken into

consideration, however, must serve to extend, complement

and support, not to replace, the analytical skills of human

experts who must in any case prepare the best lines of

intervention to achieve the specific management goals. In

a wide perspective, the EBBM UTM model may give its

best as a support tool in structured strategies of

management of change. The "Theory of Change", as used

by UNICEF and other UN organizations, has proved

particularly effective in finding solutions to contingent

problems that require political, administrative and

organizational interventions (ToC Center, 2021) and may

be an ideal framework to enhance AI-driven decision

support tools.

An estimate of the potential savings by optimizing the

system resources for patients in their fourth quarter of life

with the AIEHEM methodology is 16 million euros per

year.

REFERENCES

Academy of Medical Royal Colleges (2019), Artificial

Intelligence in Healthcare, retrieved on June 29th,

2022 from https://www.aomrc.org.uk/reports-

guidance/artificial-intelligence-in-healthcare/ .

Aghdaiea, M.H., Zolfanic, S.H., Zavadskas, E.K.,

Synergies of Data Mining and Multiple Attribute

Decision Making, Procedia – Social and Behavioral

Sciences, vol.73, pp. 388-395.

Angelis, A., Kanavos, P. (2017), Multiple Criteria

Decision Analysis (MCDA) for evaluating new

medicines in health technology assessment and

Artiﬁcial Intelligence Enabled Healthcare Ecosystem Model: AIEHEM Project

237

beyond: The advance value framework. Soc. Sci.

Med., vol. 188, pp. 137-156.

Baltussen, R., Stolk, E., Chisholm, D., Aikins, M. (2006),

Towards a multi-criteria apporach for priority setting: an

application to Ghana. Health Econ, vol 15, pp. 689-696.

Bashir, S., Qamar, U., Khan, F.H. (2014), Heterogeneous

classifiers fusion for dynamic breast cancer diagnosis

using weighted vote based ensemble. Quality &

Quantity, DOI: 10.1007/s11135-014-0090-z.

Ben-Gal, I. (2007). Bayesian Networks. In Ruggeri F,

Kennett RS, Faltin FW (eds.). Support-

Page. Encyclopedia of Statistics in Quality and

Reliability. John Wiley & Sons.

Berner, E.S. (2007), Clinical Decision Support Systems,

Theory and Practice. Second Edition Springer, 2007.

Diaby, V., Campbell, K., Goeree, R. (2013), Multi-criteria

decision analysis (MCDA) in health care. A

bibliometric analysis. Oper Res Health Care, vol.2,

PP.20-24.

Gasol, M., Bosch, J.A.; Pontes, C., Obach, N. (2022),

Early Access to Medicines: Use of Multicriteria

Decision Analysis (MCDA) as a Decision Tool in

Catalonia (Spain). J. Clin. Med., vol.11, p.1353.

Holte, R.C. (1993). Very simple classification rules

perform well on most commonly used datasets.

Machine Learning.

Ishizaka, P., Nemery, P. (2013), Multi-Criteria Decision

Analysis: Methods and Software. Wiley, 2013

Ju, Y., Wang, A., Liu, X. (2012). Evaluating emergency

response capacity by fuzzy AHP and 2-tuple fuzzy

linguistic approach. Expert. Syst. Appl., vol 39,

pp.6972-6981.

Kam Ho, T. (1995). Random Decision Forests, Proceedings

of the 3rd International Conference on Document

Analysis and Recognition, Montreal, QC, pp. 278–282

Khademolqorani, S., Hamadani, A.Z. (2015).

Development of a Decision Support System for

Handling Health Insurance Deduction. Int. J. of Adv.

Comp. Sci. And App., vol 6, no.2

Lella, L., Licata, I., Pristipino, C. (2022). Pima Indians

Diabetes Database Processing through EBBM-

Optimized UTM Model. In Nathalie Bier, Ana L. N.

Fred, Hugo Gamboa, editors, Proceedings of the 15th

International Joint Conference on Biomedical

Engineering Systems and Technologies, BIOSTEC

2022, vol 5: HEALTHINF, Online Streaming, February

9-11, 2022. pages 384-389, SCITEPRESS, 2022.

Longaray, A.A., Ensslin, L., Munhoz, P., Tondolo, V.,

Quadro, R., Dutra, A., Ensslin, S. (2016). A systematic

literature review regarding the use of multicriteria

methods towards development of decision support

systems in health management, Procedia Computer

Science, vol., 100, pp.701-710.

Marsh, K., Lanitis, T., Neasham, D., Orpanos, P., Caros, J.

(2014). Assessing the value of health care

interventions using multi-criteria decision analysis: A

review of the literature. PharmacoEconomics, vol. 32,

pp. 345-365.

Mistichelli, J.A. (1984), Diagnosis Related Groups

(DRGs) and the Prospective Payment System:

Forecasting Social Implications. Kennedy Institute of

Ethics, Center for Bioethics Library (January 1, 1984).

Narci, H.O., Ozcan, Y.A., Sahin, I. (2014), An

examination of competition and efficiency for hospital

industry in Turkey. Health Care Management Science,

DOI: 10.1007/s10729-014-9315-x.

Roberts T.J.: Dynamical and computational structures

under the sea: modelling of fish motion,

http://studentnet.cs.manchester.ac.uk/resources/library/

3rd-year-projects/2016/timothy.roberts-2.pdf , last

accessed 2021/09/30.

Roumani, Y.F., May J.H., Strum, D.P., Vargas, L.G.

(2013). Classifying higly imbalanced ICU data, Health

Care Management Science, vol.16, no.2, pp.119-128.

San Cristobal, J.R. (2013). Critical Path Definition Using

Multicriteria Decision Making: PROMETHEE

Method, J. Manag. Eng., vol 29.

Taylor, M.J., McNicholas, C., Nicolay, C. (2014).

Systematic review of the application of the plan–do–

study–act method to improve quality in healthcareBMJ

Quality & Safety 2014, vol.23, pp.290-298.

Teuscher C., Sanchez E. (2000). A Revival of Turing’s

Forgotten Connectionist Ideas: Exploring Unorganized

Machines. In Proceedings of the 6th Neural

Computation and Psychology Workshop, NCPW6,

University of Lige.

ToC Center, (2021). What is Theory of Change, retrieved

on June 29th, 2022 from https://www.theoryofchange.

org/what-is-theory-of-change/

Turing A. (1948). Intelligent Machinery. In Collected

Works of A.M. Turing: Mechanical Intelligence. Edited

by D.C. Ince. Elsevier Science Publishers, 1992.

Witten, I. H., Frank,E., Hall,M.A. (2011). Data Mining

Practical Machine Learning Tools and Techniques.

Morgan Kaufmann Publishers.

Wu, C.R., Lin, C.T., Chen, H.C. (2007). Optimal selection

of location for Taiwanese hospitals to ensure a

competitive advantage by using the analytics hierarchy

process and sensitivity analysis. Build. Environ.,

vol.42, pp.1431-1444.

Yang W. (2018). When the Selfish Herd is Unsafe in the

Middle. In: The 22nd Asia Pacific Symposium on

Intelligent and Evolutionary Systems

Zandi F. (2014). A bi-level interactive decision support

framework to identify data mining-oriented electronic

health record architectures, Applied Soft Computing,

vol.18, pp.136-145.

HEALTHINF 2023 - 16th International Conference on Health Informatics

238