An Interrelated Decision-making Model for an Intelligent Decision Support System in Healthcare

The nature of decision making in healthcare is complex and crucial. It is essential to have a tool that helps with accurate and correct decisions based on real-time data. Moreover, the healthcare process itself is complex, comprising various stages from primary to palliative, closely related to each other, and the process is different depending on the type of disease. Each stage has a crucial decision to be made relying on other stages decisions. Thus, an intelligent decision support system (IDSS) model based on a data mining approach becomes a prominent solution. However, the existing IDSS and Group Decision Support System (GDSS) applied a single-stage approach and primarily focused on development at a certain stage for specific outcomes. In contrast, the nature of healthcare decision-making in each stage is related to the previous stages, which change dynamically. Therefore, this paper proposes an interrelated decision-making model (IDM) for IDSS in healthcare that aims to have an effective decision by utilizing knowledge from previous and following treatment stages known as IDM-IDSS-healthcare. The experiment was conducted using simulated diabetes treatments data that were validated by the medical expert. Eight data sets with distinct sizes were constructed and classified into two types of decision-making categories. Each data sets consists of primary and secondary care stages with a range of 25 to 58 attributes and 300-11,000 instances. The experiment results show algorithms J48, Logistic, NaiveBayes Updateable, RandomTree, BayesNet and AdaBoostM1 obtained the best accuracy in sequence from 46% to 99%. The result also shows the improvement of decision-making efficiency with the prediction model accuracy has increased up to 56%. In addition, all respondents agreed in a focus-group discussion with medical and information technology (IT) experts that the proposed IDM-IDSS-healthcare is practical as a healthcare solution. Moreover, the solution for the development of IDM-IDSS-healthcare should use the multi-agent approach.

conjunction with the rapid evolution of healthcare [9]. However, to date, the development of the DSS has only focused on data management and knowledge in certain stages of healthcare [10], [11]. This means today's DSS development is mostly in single-stage approach where data flow occurs only at one stage or is a single solution in nature.
In this study, the term 'single-stage' refers to situations in which data flows at one stage or from several stages, with one or more decision making at one particular stage only and does not take into account the everchanging state of the data from other stages. This situation means that today's healthcare decisionmaking process is not based on the overall context of the respective healthcare stage. In contrast, a decision in healthcare should consider the interconnectedness between each healthcare stage holistically [12]. The lack of holistic DSS development for healthcare has an inevitable impact on the accuracy of results [13]. Holistic decision making in healthcare allows observation and evaluation of the entire situation to produce more accurate results than the previous approach.
As a result, research is needed on a decision-making flow model in healthcare based on the stages of care, with this model able to be used as a foundation for DSS development in general. The realities of the best decision-making process in healthcare require the sharing of information among key decision-makers. Despite the existence of the group decision support system (GDSS) which allows many decision-makers to be involved in decision making, the architecture and analysis of the data obtained are still independent because the GDSS architecture is independent. Even though the GDSS has now been improved with intelligent techniques such as data mining, to date, it remains a single-stage or single solution [14], [15].
The use of information and communications technology (ICT) now allows access to medical information from one setting to another, thus facilitating the sharing of medical information. The use of various technological devices in healthcare helps physicians with more and more diagnoses. However, none of these technologies can currently handle the overall flow of information in the healthcare sector, amid the various technological advances currently available in this sector. Thus, another important factor here is how to use the knowledge generated from this information [16]. This study asserts that DSS development in healthcare must be based on a holistic decision-making model in healthcare to tackle these limitations.
As a result, the key contributions of this study are summarized as follows: 1. An interrelated decision-making model for an intelligent decision support system in healthcare, known as IDM-IDSS-healthcare, is proposed, with this model to be used to obtain accurate decision making in the field of healthcare. 2. A unique feature of IDM-IDSS-healthcare is a decisionmaking flow model that can include the interrelated relationships between each stage of care in healthcare. 3. IDM-IDSS-healthcare consists of: (i) healthcare stages based on the area of the problem, (ii) decision flow occurring between healthcare stages, (iii) data attributes for the respective healthcare stages, and (iv) the amount of data flowing and not flowing from a respective healthcare stage to another respective healthcare stage and vice versa. The precision of decision making was found to be influenced by large volumes of data. All contents of IDM-IDSS-healthcare will be analyzed, operated, and used to make effective decisions in the respective healthcare stages. The remainder of this paper is structured as follows. Section II presents an overview of the healthcare system as well as a review of DSS in healthcare. Section III presents the methodology applied in this research and the details of the proposed decisionmaking model. Section IV presents the experiments and Section V presents the results. Finally, Section VI presents conclusions about the proposed decision-making model as well as future research directions.

II. RELATED WORK
Various efforts have been done toward having an effective healthcare management system. However, one of the most important factors in determining one's quality of life is access to high-quality health care [17], [18]. As a result, the goal of the entire field of healthcare is to improve the health of the population as a whole [19]. Taylor, et. al. (2003) stated that a healthcare system is defined as a system that consists of interconnected elements and stages [20]. On the other hand, Kovacic, et. al. (2008) defines a healthcare system in more holistic as it is not just limited to healthcare organization but includes political, economic and cultural, technical and organizational factors, relationships, processes and elements, in which individuals, groups and communities are interconnected, have a goal to meet their health needs [21]. The health care system, in general, is made up of three interconnected components: (1) healthcare service consumer groups, (2) healthcare provider groups such as doctors, and (3) organizations that manage healthcare services [22], [23]. However, the focus of this work is on decision making in healthcare according to the structure of the healthcare system and the interrelationships between the stages of healthcare itself. Only a few prior studies in the literature have explored the stages of healthcare, with few researchers having conducted work on these stages. Literature reviews have indicated that healthcare stages are traditionally shown horizontally, with only a sequential relationship between these stages [24], [25]. With the passing of time, the stages of healthcare services have also changed. Today, the stages of healthcare have undergone a transformation from a sequence in linear form to a cycle that can be iterative.
Besides, currently, healthcare has progressed from the traditional one stage to the current five stages. According to the standard practice of the Health Care Services System, the stages of health care consist of five stages of care, (1) Primary Care (PC) is a basic health care service to identify early symptoms, (2) Secondary Care (SC), has been identified as diseased but focus on identifying disease in the early stages, (3) Tertiary Care (TC), contains treatment to reduce the effects of the disease, (4) Quaternary Care (QC), contains treatment with a very high medical level and (5) Palliative Care (Pal-C), supportive care services for patients with the disease seriously [26]- [28]. These five stages of care are interrelated with each other. In medical, most stages of care have a sequential relationship with the term 'refer to' and have a returned relationship with the term 'continue follow-up' among these stages. Thus, the interrelationship among these five stages of healthcare was identified as recursive and iterative [29], [30].

A. DECISION SUPPORT SYSTEM IN HEALTHCARE
The results of the literature review showed that different decisions produced different risk stages [31]. This highlights the need to identify a decision-making process that takes the healthcare stage into account to make accurate decisions in the healthcare sector. It is recognizable from the differences in these risk stages that they form the basis for the healthcare system's division into several stages of care. Thus, the focus of this study is on the complex theory of healthcare based on the decision-making aspects in various stages of healthcare. Previously, computer experts have investigated and manufactured DSSs to help people make better choices in different aspects of their lives.
Thus, based on Figure 1, we can see the evolution of DSS began with Personal DSS, which emerged from the evolution of technologies that began with computer-based information systems, operations research or management science, and behavioural decision theory [32]. Following the existence of Artificial Intelligence (AI) and the development of Social Psychology, DSS evolved into IDSS and GDSS. During this period, the first DSS for healthcare, MYCIN, was developed in 1975, followed by INTERNIST-I (1982) and DXplain (1987) [33]- [35]. Since then, DSS in healthcare has evolved in tandem with technological advancements.
These developments have led to the existence of a Group Decision Support system (GDSS), Executive Information System (EIS) and Intelligent Decision Support System (IDSS). As far as GDSS technology is concerned, even though a GDSS includes multiple decision-makers and has been enhanced with smart techniques, the GDSS design and data analysis still comprise a single solution. The same applies to EIS and IDSS. Technology is constantly evolving. IDSS has evolved into knowledge management-based IDSS, interactive and integrated IDSS, and natural language-based IDSS. Meanwhile, EIS and NSS evolved into analytics-based IDSS and cloud-based IDSS [36].
Further, the usage of DSSs in the healthcare sector is growing and increasing rapidly in line with the rapid evolution of the healthcare sector itself. The computer-based Healthcare Information Management System, also known as the Electronic Health Record System, is broadly classified into three types: administrative information management system, clinical information management system, and financial information management system [37]- [39]. This study, however, only looks at the Clinical Decision Support System (CDSS), which is a subcategory of the clinical information management systems. DSS has evolved till recently, with all technologies attempting to be incorporated, resulting in Multiagent-based IDSS. The same applies to CDSS. Then, CDSS has evolved to Intelligent CDSS (ICDSS) until recently to Multiagent ICDSS [40], [41].
However, the use of ICDSS technologies nowadays were still rising as shown in Table 1. Table 1 showed, the first aspect is whether the source of the data analysis for the DSS in healthcare is obtained from different stages of care. The second aspect is whether the decision model of each DSS in Table I is dynamic. The third aspect is whether the DSS in Table I applies artificial intelligence (AI) techniques, while the fourth aspect considers the DSS from the perspective of the type of intelligence techniques used. For the first aspect, the analysis results found that none of the DSSs in Table I originated from more than one healthcare stage. This shows that these DSSs have all been developed specifically for only certain stages of care within the healthcare sector. Therefore, most existing DSSs have been developed separately based on a specific healthcare stage.
the DSSs in Table 1 have been developed for secondary care, except for the Intelligent system for anticipating intradialytic hypotension in chronic hemodialysis [56] and the Early warning system [57] which was developed for tertiary care. Therefore, the data collected for analysis in the DSS for each of these studies were only from one stage, with that stage being independent, with no relationship to either the next stage or the previous stage of healthcare. In reality, decision making in the healthcare system is interrelated at every stage. Therefore, to improve the quality of decision making, research into the development of the DSS must take into account the interrelationships between the different stages of healthcare. In terms of the second aspect, as shown in Table I, most of the existing DSSs have a static decision-making model, with none using a dynamic decision-making model. For example, H2RM is a DSS that uses patient charts and online guidelines to generate a decision model once in the system.
Aside from that, most DSSs showed in Table 1 use a variety of data sources, such as the ADDIS data model, which was created through the assisted extraction of trial registries, manual extraction of abstract databases, text mining in industrial databases, and extraction of ontology-based eXtensible (OBX) and ontology of clinical research (OCR) mapping rules [58]. Apart from ADDIS and H2RM, existing DSSs with a static decision-making model include the Smart healthcare monitoring system, Early warning system, and MCDSS for COVID-19.
Even though these data models are built from various data sources, the generation of the decision model occurs at one time only. This again confirms that a dynamic decision-making model is needed to improve the effective of decision making. Through a dynamic decision-making model, the DSS will always have the Fuzzy-neuro decision support system for back pain diagnosis (FNDSB [42] No latest decision model from which to generate decision support. The third aspect is whether the DSS applies artificial intelligence (AI) techniques. As shown in Table I, almost all existing DSSs use AI techniques, except for the DSS called HeartMan and the DSS for the early detection of system inflammation response syndrome (SIRS) [48], [59]. The AI techniques are now essential for producing accurate predictive models for effective decision-making models in DSSs. Furthermore, the AI methods used are diverse. The fourth aspect concerns the type of AI techniques used, with data mining currently one of the most popular techniques used in the existing DSSs, as shown in Table 1. The process of data mining is interactive and iterative. The data mining approach can also be used to identify important data attributes that have a high influence on the process of obtaining accurate results. Furthermore, the data mining approach can be used to identify critical data attributes that have a high influence on the process of obtaining accurate results [60]- [62]. Consequently, the form and status of the data collected for the data mining process are vital. Therefore, data mining research is required. The findings are considered in the data mining process to ensure quality results by using data mining methodology. [63], [64], [76].

B. ISSUES AND CHALLENGES IN A COMPLETE HEALTHCARE DECISION SUPPORT SYSTEM
Various methods are used to produce high-quality DSSs for healthcare. Still, data mining is widely used in the development of DSS research. It shows that the data mining method used in various types of DSS has proved effective in improving the accuracy of results [52], [53], [55]. Despite the significant achievements and success of IDSSs, acceptance and use of many types of DSS in healthcare have not yet been widely achieved [74]. The ineffective use of ICDSS is one of the identified problems. This issue arises due to several factors, including a lack of understanding of the ICDSS workflow's consequences. Furthermore, DSS health researchers argue that systems need to be studied to regulate complex workflows [10]. In addition, DSS health researchers argue that systems need to be developed to regulate complex workflows [75].
According to the findings of prior ICDSS observations, as set out in Table 1, ICDSS development is mostly based on specific stages of healthcare. The lack of development of decision-making support systems for overall healthcare affects the quality and effectiveness of decision making. Comprehensive decision making allows all aspects to be observed and assessed. This situation produces more accurate results than options considering only one stage or one aspect. It is required to explore a decision-making model based on all healthcare stages to use as the cornerstone for the development of ICDSS. [48], [67], [70]. Figure 2 illustrates the current form of ICDSS data flow and a single decision-making concept in most of the current existing ICDSS architecture. Despite the fact that certain studies take into consideration the flow of data from multiple sources, however, the decision-making is still performed once and at a specific stage of healthcare only [78], [79]. Besides, even though several studies produce results to various stages of healthcare, however, decision-making is also still achieved once and at a certain stage in healthcare only [80], [81].
The reality is that the best decision-making process in healthcare requires information to be shared between key decision-makers. Despite the existence of GDSSs that allow many decision-makers to be involved in decision making, the fact remains that the architecture and data analysis remains independent. The reason is that the GDSS architecture is itself independent. The GDSS continues to be improved, even today, with intelligent techniques such as data mining. However, the concept of the GDSS remains independent of other stages as a single solution [22], [59]. Figure 3 display the form of the current GDSS data flow and multiple decisionmaking concepts in GDSS architecture. Meanwhile Figure 4 illustrates the reality of data flows and iterative decisionmaking concept in ICDSS architecture to be proposed.   Figure 5 shows the research methodologies for this study that consists of three stages: formulate IDM-IDSS model for healthcare, experiment, and evaluation. This research applies a mix-method qualitative, quantitative, and experiment. There is an extensive literature review, analysis of healthcare documents used locally and internationally, and past research on healthcare decision-making scenarios to formulate the IDM_IDSS model.  [25], [26], [29].

III. METHODOLOGY
Two dedicated doctors from Advanced Medical and Dental Institutes (AMDI), University Science Malaysia (USM), are involved in the model formulation processes. One is the diabetes specialist doctor, and the second is the children specialist doctor, the Head of Regenerative Medicine Cluster, who advised and validated the proposed model. They also validated that the Diabetes IDM model includes the data preparation. The Diabetes IDM model covers the five stages of iterative decision-making. However, the experiment only focuses on preparing the data set for  diabetic primary care and secondary care stages only. The data sets have two types of decision-making: diagnosis on a diabetic level and stage of care, either remain at the current stage, previous stage, or next stage. Hence, eight data sets are prepared. Each data set consists of two data as a primary care and secondary care stage but has four different sizes of instances and attributes, ending up with 192 experiments as each data set is experimented with using six machine learning algorithms. The performance of IDM is evaluated based on the changes in accuracy prediction models towards the size. Lastly, the proposed IDM Model for Healthcare, the Diabetes IDM model, and the experiment results are shown to twelve doctor practitioners for quantitative feedback evaluation using a focus group discussion method.

A. AN INTERRELATED DECISION-MAKING MODEL FOR IDSS IN HEALTHCARE
The model, IDM-IDSS-healthcare, is characterized as a set of elements (stages) that require decision making following the type of disease and stage supported by the healthcare organization or centre. ABC Hospital (a pseudonym) has, for example, only assisted the management of heart disease from Stages 1 to 3. Meanwhile, Stage 4 management should be applied at Hospital DEF (also a pseudonym). The model, IDM-IDSS-healthcare, thus comprises an IDSS collection consisting of care and data flow stages for each intelligent decision support system (IDSS). The treatment stage is indicated by P, while the data flow between care levels, represented by A, consists of outflow and data entry. The IDM-IDSS-healthcare set consists of several stages depending on the area or organization of the problem. Thus, in summary, the P element in the IDM-IDSS-healthcare set can be represented by the following equation: The stages of care P comprise the following components:

P = {Decision Maker, Goal, Task, Data}
Therefore, P can be represented by representatives as follows: Table 3 then describes the elements in P and presents their definitions.

Subcomponent Definition Decision Maker (PK)
People who have the right to use the system

Goal (M)
A goal or something to accomplish on a task at a certain level

Task (T)
Activities are undertaken to achieve the goal Data (D) Record with a list of components and data attributes at the care level The element 'data' D comprises the following elements based on Table 4: Meanwhile, components kp is composed of several attribute elements a with the last component, kpn = ks, as follows. Table 4 above presents the data D, together with its elements or members. Subsequently, component A is formed by the relationships between the stages of care P. Three types of relationships, R, K, and S, are formed between P, as shown in Figure 6.
Definition of this flow A is subject to IDM-IDSS-healthcare principles as set out in Table 5: In addition, Figure 7 shows the general IDM-IDSShealthcare model with the maximum number of care stages (n = 3). Based on Figure 5, flow K is always going to a higher stage, while flow S is always going to be lower than the current stage. Meanwhile, the flow R is always at the same level. In addition, the flow K was not only sequentially increased but also increased by more than one level. At the same time, S-flows may occur sequentially with lowering stages or may occur with a decrease of more than one level at a time. Following are the steps for the development of IDM-IDSS-healthcare: 1) Identify the stages involved and the maximum number of stages involved. The identification needs to be carried out by the field specialist at this point.
2) Identify the flow between the different stages involved. At this point, the identification needs to be carried out with the approval of the specialist in the field.
3) Identify the data attributes list for each stage. Again, identification must be done with the approval of a specialist in the field at this point.

B. AN INTERRELATED DECISION-MAKING MODEL FOR DIABETES
The data flow value of IDM-IDSS-healthcare for the diabetes case study shows that the concept of IDM-IDSShealthcare has been able to deliver significantly better results. Based on the principles of data mining, more information will produce more accurate results. The concept of IDM-IDSS-healthcare forecasting can therefore help in the process of producing more accurate decision making. Decision making in healthcare by its nature is very interlinked between stages. Preserving information on treatment and moving among the stages leads to accurate healthcare decision making. This situation follows the basic principle of data mining in that the more data, the more accurate the decision-making process.
Therefore, this section describes the results of the development of IDM-IDSS-healthcare for the diabetes case study. The result showed how well IDM-IDSS-healthcare could be used to increase the accuracy of healthcare decision making. Following are the steps in the development of the IDM-IDSS-healthcare case study on diabetes: • Identify the stages The first step is to identify the stages of healthcare and the maximum number of stages of healthcare involved in diabetes. In the case of diabetes, this may involve all stages of healthcare. The maximum number of medical stages for the diabetes case study is therefore five. Subsequent stages of healthcare were identified as primary, secondary, tertiary, quaternary, and tertiary. • Identify the flow between each level Existing streams between each level are identified, following the identification of the existing levels of care for cases of diabetes. The identification of these emerging streams between the different levels is integrated after the information has been collected through interviews and discussions with healthcare professionals on diabetes diagnosis and the determination of healthcare levels. Figure 6 shows the data streams identified that are either at a specific stage of care or between one stage of care and another stage of care. These data streams are represented as a, b, c, d, e, f, g, h, i, j, k, l, m, n, and o, as shown in Figure 6. • Identify the list of data attributes A list of data attributes for each stage of care was collected after identifying the current stages of healthcare and trends for the diabetes case study. Next, the list of data attributes was evaluated and validated by healthcare professionals through interviews and discussions. • Calculate the total data This step calculates the sum of the data that flow and the sum of the data that do not flow. The calculation is made based on the number of data records at each level of healthcare involved. The example of total data records for each stage of healthcare in the diabetes case study is shown in Table 7 and Table 8. The amount of data is based on the list of data attributes for each stage of healthcare (primary, secondary, tertiary, quaternary, and palliative) and the list of data attributes involved in the data flow between the stages of healthcare • Complete the development of IDM-IDSS-healthcare for diabetes The final step in the development of IDM-IDSS-healthcare is now discussed with its use in the diabetes case study producing the data flow results illustrated in Figure 8 and Figure 9 which shows the IDM-IDSS-healthcare data flow values. The values, assuming that all stages of care are provided, are derived from approximately 300 treatment records. Further, in order to support the evaluation in this study, focus groups discussion (FGD) were chosen as one of the approaches under the formative evaluation approach, which is one of the CDSS evaluation categories [79].

IV. EXPERIMENT
The evaluation was carried out to demonstrate the effectiveness of the proposed IDM-IDSS-healthcare, therefore the data preparation is the crucial part. The source of data used in the experiment was based on data from the diabetes case study as illustrated in Figure 8 and Figure 9. However, because of the constraints to obtain complete real data, the data for this experiment were generated simulation data based on analysis documents healthcare and analysis decision making past researches.

A. Data Set
As stated in methodology, each stage has two types of decision making: diagnosis diabetes level and predict the stage of care to go. The data sets only prepared data set at primary and secondary stage only. Furthermore, each data set is divided into two types of data changing the size of the data set: a data set with a relationship with R only and a data set with relationships between R, K, and S (refer to Figure  6).
The first type of data set consists of all data but only at one healthcare stage either at primary or secondary only. This type of data set has four different sizes of instances which is 300, 600, 2000 and 5000 for both diabetes diagnosis and predict the stage of care as shown in Table 7 and Table 8. Table 8 shows the second type of data set which have an R, K and S relationship. It involves a combination of data between stages which is data from previous and following treatment stages. Four different sizes of data sets have been prepared for this research. The number of attributes and instances were different based on the stage of care.

B. EXPERIMENT SETTING
The experimental conducted for each data set using WEKA (version 3.8) data mining tool. The WEKA was applied because the IDM-IDSS-healthcare will be implemented based on multi-agent development in the Java Agent Development Framework (JADE) environment in future. The JADE environment will be applied later to become future multi-agent ICDSSs in healthcare. The experiments were conducted using CPU intel 334Mhz with memory 16 Mb RAM. Table 7 and Table 8 show that for the two categories of diagnosis class and stage of care class as 8 data sets. Each data set has primary and secondary care data. Therefore, the experiment conducted 16 sets of data using six types of algorithms to obtain the accuracy model at primary and secondary stages which end with 192 experiments times. The six algorithms are J48, Logistic, NaiveBayes Updateable, RandomTree, BayesNet and AdaBoostM1. The experiment uses a 10-cross validation method and selects the best accuracy for each data set. Figure 8 and Figure 9 show a recursive flow of data from one stage to another in the case of the R-relationship (see Figure 6). Figure 8 shows a recursive flow of data for prediction on diagnosis model while Figure 9 shows a recursive flow of data for prediction on the stage of care model. The R-relationship is a common relationship within existing decision support systems (DSSs). The K-relationship, on the other hand, indicates the data flow from one stage of healthcare to another. Meanwhile, relationship S shows the flow of incoming data from one healthcare stage to another. In the current DSSs, the relationship between K and S does not consider. Therefore, this study has proposed the IDM-IDSShealthcare model in order to enable the K-S relationship in today's decision support system (DSS). Thus, the DSS can produce more accurate results with both K-and Srelationships. This situation is consistent with the principle of data mining, which indicates that when more information is provided, more knowledge is generated, and decision making is more accurate.

V. RESULT
The findings showed IDM-IDSS-healthcare was workable and improved the decision-making accuracy. This was supported by the findings of expert evaluation via the focus group method. Thus, the following section discussed detail the experiment results in section A and evaluation expert in section B.

A. EXPERIMENT RESULT
Even though during experiments has collected various types of measurement. However, Table 7 and Table 8 only shows the accuracy result for each experiment to show the impact of accuracy for the proposed interrelated decision-making. The table shows the accuracy of eight data sets for primary and secondary stages for both diabetes diagnosis and stage of care decision. The table shows that algorithm J48 consistently shows as the best model for all data sets experiment includes both data sets for diabetes diagnosis and stage of care with accuracy from 99.3% to 100% as marked with bold. The table also shows the second-best algorithm is Logistic, followed by NaiveBayes, Random Tree, BayesNet, and AdaBoostM1. On the other hand, the AdaBoostM1 algorithm shows the lowest average percentage accuracy as marked as italic, with accuracy between 46.8% to 92%. The algorithm is less stable compared to other algorithms.
Then, based on symbol representations shown in Table 9, Figure 10 and Figure Table 10 shows the accuracy increment ratio for both types of relationships. The table shows BayesNet shows the highest accuracy increment ratio, follow by AdaBoost, NaiveBayes, Random Tree, Logistic, and lastly, J48. It can be concluded        that increasing the instances has increased the accuracy except for J48. It is because the J48 has obtained high accuracy, and additional instances do not change the accuracy. Furthermore, the table also shows that most of the accuracy of the increment ratio, is either the same or reduced, while increasing the number of attributes and instances. However, the J48 showed a slight increment for the state of care model but remained similar for the diagnosis model.

B. EXPERT EVALUATION
The expert evaluation was performed using the FGD approach because FGD is one of the CDSS evaluation categories. The expert evaluations also took place at the same time as the system implementation evaluations, however, the statements presented here were exclusively concerning IDM. Table 11 shows three evaluation statements presented to twelve healthcare professionals in different specialities from AMDI, USM as respondents using the FGD method. The 12 respondents consisted of three paediatricians, two medical lecturers, a public health physician, a radiologist, a family physician, a surgeon, a transfusion specialist and two AMDI information technology officers. The three evaluation statements address three main aspects of IDM-IDSS-healthcare acceptability. The three main aspects were the acceptance of the IDM concept in general healthcare, the acceptance of Diabetes IDM, and the acceptance of IDM-IDSS-healthcare to aid healthcare decision making. The evaluation was done using five scales as shown in Table 12. As a result, 75% of respondents strongly agreed; 25% agreed to accept the IDM concept of healthcare in general, according to Figure 11. Following, 83% of respondents strongly agreed; 17% agree to accept the Diabetes IDM. Conclusively, all of the respondents strongly agree to accept the IDM-IDSS-healthcare to aid healthcare decision making in the healthcare system. These results showed that IDM-IDSS-healthcare is practically used in healthcare solutions.  Number Evaluation Statement 1.
The IDM concept of healthcare in general 2.
IDM based on diabetes data can provide valuable information to help with healthcare decisions.

3.
Based on the result of the experiments, IDM-IDSS-healthcare could adapt the real flow of healthcare decision making in the healthcare system

VI. CONCLUSION
This study has proposed an interrelated decision-making model for intelligent decision-making in healthcare based on the multi-agent solution. The proposed model shown is workable and proven by applying it to a diabetes case study. A diabetes IDM-IDSS consists of five interrelated stages of care: primary care, secondary care, tertiary care, quaternary care, and palliative care. The Diabetes IDM-IDSS model also successfully demonstrates the existence of in-and-out data flows at each stage. Furthermore, the case study has shown that each stage can have various types of decision-making. The experiment has shown two types of decision-making: diabetes diagnostic and stage of care in primary and secondary care. Even though the experiment only focuses on two stages, the result shows how iterative decision-making works and how accuracy of prediction model changes due to additional instances toward time. This research also has found the impact on the effectiveness of decision making: type of decision making, the type of algorithm uses in modelling, the number of instances, and the number of attributes. The experiments result consistently show that adding some of the instances have increased the accuracy model. However, if the accuracy is already nearly or reaches 100%, the additional data does not change. While additional attributes cause has to reduce the accuracy of the decision-making model. It is because the model is more complex and needs more instances to reach high accuracy. In summary, it concludes that the proposed interrelated decision-making approach is applicable uses for future IDSS-healthcare solutions as it supports the dynamic change of data that influence the decision-making accuracy in any healthcare stage. Furthermore, the expert evaluation also agreed that the proposed IDM IDSS-healthcare model is an acceptable solution for healthcare decision-making.
Therefore, we conclude that the IDM-IDSS-healthcare model is acceptable and that it is recognized as an aid in the decision-making process, particularly in the healthcare sector. However, before uses it in the real-world needs more robust experiments using real diabetes treatment data using more recent machine learning algorithms. It can be concluded that, this model can be applied in other fields characterized by stages, such as various stage of decision making in business, etc.