Hospital management society: A framework based on fuzzy logic and association rule mining towards well-being society

Abstract

Non-uniformity in medical procedures, expensive medical treatments, and the shortage of medicines in different areas are health care problems in our country. This paper aims to resolve that problem by developing a web-based-application called Hospital Management Society (HMS) based on a novel Dynamic Optimized Fuzzy C-mean Clustering and Association Rule Mining (DOFCCARM). The purpose of HMS is to enhance the hospitals (and clinics) by regulating, overseeing and accrediting them to bring uniformity in health care facilities, to make the medical treatment cost effective, to find common diseases in a particular age and area, and to help government in identifying the areas facing the shortage of licensed medicines. Therefore, HMS creates a single platform for both the doctors of central hospital (CH) and the doctors of member hospitals (MH). The CH provides clinical practice guidelines for various diseases. A team of doctors at CH evaluate the medical treatment provided by MH. If a hospital fails to maintain the standard then HMS blacklists such hospital. In our approach, we take a range of values to distinct successive partitions and generate a parallel membership function to make fuzzy sets of patients report, rather than single partitioning point. We determine the effectiveness of our approach through experiments on a dataset. The results revealed the most common age, symptoms and location for a particular disease and shortage of particular medicine in a specific area.

Keywords

Fuzzy C-mean association rule mining hospital management society intelligent system

1 Introduction

Health and education are among the basic needs of people anywhere in the world. Unluckily, in developing countries, people do not have availability of these facilities throughout their country. In the context of health facilities, doctors with minimal experience and no supervision, doctors lacking latest knowledge and researches related to medical procedures [13], substandard medicines, and expensive tests prescribed by the doctors are the major issues in health care. Hospital (or clinic) is an organization to provide surgical, treatment, and nursing care for wounded or sick people. Unfortunately, many hospitals in a developing country, especially small hospitals and clinics, are not providing accurate and worthy treatment to their patients [18]. In contrary, the treatment of a big hospital is comparatively more accurate and worthy, however, it is too expensive for general public to afford it. Moreover, non-uniformity in medical procedures and the shortage of medicines in different areas are also well established health care problems [15]. Therefore, the development of a system is inevitable to enhance the treatment services by regulating, overseeing, and accrediting the hospitals.

To overcome the difficulties of providing uniform standard health care treatments and services, we develop a framework to establish a platform called Hospital Management Society (HMS). Our solution involves a team of medical experts and specialists from a major hospital, i.e. central hospital (CH), who would develop guidelines for the society members, i.e. doctors elected to be members of the society. The experts provide guidelines for various diseases, epidemics, and viruses. It also considers varying age groups and other peculiarities as per the latest researches and practices in the field of medicine. The main objectives of HMS includes 1) To bring uniformity in health related treatment and services, 2) Cost effective medical treatment for everyone, 3) Finding out what medicines are short in a locality and what are the most common diseases in a particular age-group, weight and location, and 4) Efficient provision of medicines by the government.

To achieve these objectives, we have implemented Dynamic Optimized Fuzzy C-mean Clustering and Association Rule Mining (DOFCCARM) approach. Fuzzy C-mean is a clustering technique that permits a chunk of data to be a member of two or more clusters [12]. In our approach, we take a range of values, while previous techniques take only single partitioning point, to distinct successive partitions and optimize membership function to make fuzzy sets of patients report attributes. We have considered four diseases with related symptoms and medications to treat them. Age, weight, disease symptoms, medication, location and disease name are the attributes on which DOFCC is applied, where disease being the decision attribute. Then the concepts of ARM [9 , 21] are applied on these dynamically generated clusters. ARM technique is used to discover frequent patterns and hidden associations between the data [21].

The remainder of the paper is organized as follows: Section 2 describes the methodology of our research and various steps we adapted to do our research. Section 3 provides empirical results of our research. Section 4 discusses literature review on medical procedure, fuzzy clustering algorithms and association rule mining. In Section 5 we conclude this paper and suggest future work directions.

2 Methodology

We have developed a platform that involves a team of medical experts and specialists from CH, and doctors from private hospitals and clinics, hence forming a society called HMS as shown in Figure 1. The doctors from central hospital provide clinical practice guidelines for different diseases as applicable to different age groups and weight, as per the latest researches and practices in the field of medicine; and check the adherence of private doctors with these guidelines. Users of this system will be doctors, a team of IT experts and general public. The doctors are divided into three categories:

Doctors of central hospital: This is a team of doctors from CH. They expect to upload guidelines for a certain disease and evaluate the treatment procedures followed by doctors of member hospitals shown in Figure 1(a). These guidelines contain disease name, its symptoms and prescription. These doctors have the authority to allow any new hospital to become member of HMS.

Doctors of member hospitals: This is a team of doctors from member hospitals. They suppose to upload a medical report (audio file) of the patient’s medical examination and treatment that contains information such as age, weight, location of patient, symptoms, and prescription suggested by the doctors as depicted in Figure 1(b). It is permissible for MH doctors to view the guidelines as provided by the doctors of CH.

Evaluator: An expert from ministry of health to evaluate the guidelines and reports to add or remove any hospital from HMS as shown in Figure 1.

Fig. 1

Perspectives of Proposed System Architecture (a) Central Hospital, (b) Member Hospital.

Other users of our system are transcription team and general public. Transcription team operate through CH to receive audio files (reports) from society members and insert the data into the database. General public is able to view the hospital list, i.e. a set of hospitals having membership of the HMS.

2.1 Data management

It is an interesting task to discover various health related aspects such as: 1) which medicines are being prescribed by most of the doctors for the same disease, 2) what alternative medicines the society members are prescribing, and 3) which disease is common in a group of people with varying age, weight, and location etc. We have used Entity Relationship Model for representing database design and requirements gathering, as shown in Figure 2. It is a structural diagram of HMS where doctor, hospital, transcription team, report, account, guidelines, and evaluator are major entities in our problem domain. In this model, PK represents Primary Key of an entity. Every entity has its own PK to uniquely identify its records. The nature of relationships among entities are represented through cardinality such as one-to-one, one-to-many, and many-to-many. For instance, a doctor can upload many reports (i.e. one-to-many) and many reports are viewed by many doctors (many-to-many). On the other hand, a doctor can have one account, i.e. one-to-one.

Fig. 2

Entity Relationship Model of Hospital Management Society.

2.2 System behaviour

We provide an abstract illustration of the behavior of HMS through state machine diagram in Figure 3. This behavior is represented as a sequence of events that can happen in one or more probable states. The doctors of a central hospital have the following dynamic behavior.

On the event of system open, system transit from initial Idle state to Login state.

The account is then checked by the system.

On the event of valid email and password, system transit to the Open Account state.

if a user wants to view or update hospital list, system transits to the Hospitals state.

If a user wants to view reports, system transits to the Reports state.

If a user wants to view or update guidelines, system transits to the Guidelines state.

After the above operation is fulfilled, system transits to the Final state.

Fig. 3

State Machine Diagram for Doctors and Evaluator.

For the doctors of a member hospital, the following dynamic behavior is expected.

On the event of system open, if a user is already member of HMS then system transits from initial Idle state to Login state; otherwise system transits from initial Idle state to Create Account state.

The account is then checked by the system.

On the event of valid email and password, system transits to the Open Account state.

In case a user wants to view hospital list, system transits to the Hospitals state.

If a user wants to view or write reports, system transits to the Reports state.

If a user wants to view guidelines, system transits to the Guidelines state.

After the above operation is fulfilled, system transits to the Final state.

Figure 3 also represents state machine diagram for an evaluator. It has following dynamic behavior:

On the event of system open, system transits from initial Idle state to Login.

The account is then checked by the system.

On the event of valid email and password, system transits to the Open Account state.

When a user wants to view or update hospital list, system transits to the Hospitals state.

In case a user wants to view reports, system transits to the Comments state.

If a user wants to comment on hospital and guidelines, system transits to the Reports state.

When a user wants to view or update guidelines, system transits to the Guideline state.

After the above operation is fulfilled, system transits to the Final state.

2.3 Organization of system data and procedures

In this section, we explain a logical construction of our system by representing it through Warnier diagrams shown in Figure 4. It is a hierarchical flowchart of HMS to indicate the organization of procedures and data of our system. The procedures and data flow for central hospital doctors are represented in Figure 4(a). The functionalities available to CH doctors includes login to the system through email id and password, creation of an account through register where user needs to provide required information such as email id, name, password, hospital name, hospital & doctor registration number, and phone number. Viewing guidelines and checking reports are among the possible capabilities of a CH doctor. Similarly, we have illustrated the flow of information and set of capabilities of transcription team, doctors of member hospitals, and evaluators in Figures 4(b), 4(c), and 4(d) respectively.

Fig. 4

Warnier Diagrams for (a) CH’s Doctors, (b) Transcription Team, (c) Doctors of member hospitals, (d) Evaluator.

2.4 System implementation

The objectives are to bring uniformity in health care through identifying most common diseases in a particular age, weight and location; and to know the alternative medicines, which are absent in the guidelines, prescribed by the member doctors to their patients. This helps the government authorities to determine the shortage of medicines in a specific area. Hence, provision of medicines would become efficient because it is not feasible to provide each medicine in a large quantity to all areas.

To achieve these aforementioned objectives, we propose to use an existing fuzzy c-means algorithm (FCM) along with the association rule mining approach to extract useful insights from data. The flow-chart of system implementation is shown in Figure 5. FCM clustering algorithm makes fuzzy clusters of the data, where each data point has membership of multiple clusters. It is a type of soft clustering where each data entry has varying degree of membership for each cluster. Moreover, sum of all the memberships of a data entry is equals to one. In our implementation, it was necessary to use soft clustering approach because:

A doctor can prescribe one medicine for multiple diseases,

A patient can suffer from multiple diseases at one particular time, and

Same disease may exist in different age groups.

Fig. 5

An Overview of System Implementation.

Table 1

Algorithm 1 The Pseudocode of DOFCCARM
1: Input: Raw data $\vec{D}$
2: Output: Extracted Rules R
3: ε ← 0.2
4: Initialize membership matrix U with random values
5: do
6: determine cluster centers ${\vec{v}}_{i} \in \vec{V}$
7: update membership matrix U_i+1 ← U_i + Δ
8: while (\|\|U_i+1 - U_i\|\| > ε)
9: Generate k soft clusters C_k
10: for Each cluster c_i ∈ Cdo
11: Calculate items SPRT values SprtV_i
12: if (v_j ∈ SprtV_i) < σthen
13: remove item v_j
14: end if
15: Generate frequent itemset matrix FTM_i
16: Compute confidence CON (FTM_i) of frequent items
17: if (con_j ∈ CON_i) ≥ σthen
18: add corresponding rule to R
19: end if
20: end for

The key steps of our system implementation, see the algorithm 1, involves random initialization of membership matrix U, determining cluster centers $\vec{V}$ , update the membership matrix and repeat the former two steps until the termination criterion met with no significant changes in the membership matrix, i.e. ε = 0.2. We represent data point $\vec{d} \in \vec{D}$ as vector containing information about age, weight, disease symptoms, medication, location and disease name. We generate corresponding soft clusters C based on the membership matrix and feed U the output to the association rule mining component. The support SprtV of each itemset is calculated with the help of cluster information; consequently, we produce a frequent itemset matrix FTM. This matrix helps us to determine the confidence CONF of of frequent itemsets. For instance, given the location information, what kind of disease symptoms found or what kind of medications were used.

3 Results and discussion

Fuzzy c-means clustering algorithm is applied on the data set collected from four different hospitals and a clinic in Rawalpindi city of Pakistan and managed as per the entity model presented in Figure 2. The names of the hospitals and clinic involved in our experiments are as follows: 1) Jinnah Memorial Hospital, 2) Cantonment Board Hospital, 3) Hearts International Hospital, 4) Noor Hospital, and 5) Bangash Clinic. The data set consists of four diseases, symptoms, age and weight of patients, and medicines prescribed to cure the disease by the doctors of the private hospitals. Hence, FCM uses a set of attributes that includes age, weight, symptoms, medication and disease; where disease being the decision attribute. The application of FCM resulted in soft clusters based on the aforementioned attributes. Further, we apply association rule mining on resultant clusters to analyze the co-occurrence of a set of attributes. The results showed the most common age, weight and location for a particular disease. We can also observe the medicines prescribed by the society members that are not mentioned in the guidelines. Hence, our system indicated the shortage of a particular medicine in a particular area. By entering a disease name, the system shows the potential age, weight and locality where the disease is commonly happening. The degree membership of clusters for Anemia, Typhoid, Malaria, and Whooping-Cough diseases against the parameters such as age, weight, medicine, and location are presented in Figures 6, 7, 8, and 9 respectively. Based on such analysis presented in the aforementioned figures, our system produce meaningful results as depicted in Figure 10.

Fig. 6

Degree of Membership for Clusters of Anemia (a) Age, (b) Weight, (c) Medicine, and (d) Location.

Fig. 7

Degree of Membership for Clusters of Typhoid (a) Age, (b) Weight, (c) Medicine, and (d) Location.

Fig. 8

Degree of Membership for Clusters of Malaria (a) Age, (b) Weight, (c) Medicine, and (d) Location.

Fig. 9

Degree of Membership for Clusters of Whooping Cough (a) Age, (b) Weight, (c) Medicine, and (d) Location.

Fig. 10

System Outcome for (a) Anemia, (b) Typhoid, (c) Malaria, and (d) Whooping Cough.

4 Literature review

In this section, we highlight existing studies from literature in relevance to our work in terms of clinical practice guidelines, treatment and diagnosis evaluation tools and techniques, evidence-based medicine, data driven approaches to assist practitioners in clinical decision making.

4.1 Clinical practice guidelines (CPG)

The clinical practice guidelines (CPG) is an active area of research, which describes the conclusions and recommendations related to proper treatment based on scientific evidence [13]. CPGs help the practitioner in decision making regarding diagnostic, treatment etc. These guidelines also make the whole process cost effective by focusing only on necessary examinations and drugs. By keeping in view their importance, it is proposed to develop many tools and languages which will help in computerizing (or automating) CPGs. A CPGs validation application is proposed in [13]. This proposed application suppose to detect, in the light of specified CPGs, any inconsistencies present in medical diagnosis and treatment. A computer-based tool [10] is created to help doctors in decision making and providing quality and safe treatment to their patients. Rough sets are helpful in diagnosis as well as in treatment, as the authors in [10] have developed a rough sets based software and tested it on real life data. This software is helpful for the doctors in automated decision support. Applying international guidelines and standards in medical treatment is a guarantee of safe treatment [10]. But it is observed that sometimes the guidelines are not followed as per the requirements. One of the possible reasons is the busy schedules of doctors and they do not find time to read the guidelines.

CPGs have inconsistencies associated with them because they are complex and are written in natural language. Therefore, a formal modeling language is required to produce or represent these guidelines, in order to eliminate factor of inconsistency. Medical domain has its own domain specific language and various Ontologies; therefore the modeling language must cover those Ontologies [6, 19]. Checking the satisfiability of the CPGs is another area of research, where researchers have introduced decidable metric interval-based description logic and a tableau-based algorithm for this purpose [1 , 15]. The guidelines satisfiability can be checked through their tableau algorithm.

4.2 Treatment and diagnosis evaluation tools and techniques

A system of databases has been established by Dutch Society of Cardiology as a standard tool for evaluation of procedures used for the diagnosis and treatment of cardiology diseases [18]. The data stored in these databases comprises of prevalence of disease, incidence details, and number & results of the procedures. These databases are implemented and managed by a separate foundation named as National Cardiovascular Data Registry (NCDR) [18]. The count and results of cardiovascular procedures are registered in these databases. From the contents of these databases, it is possible to compare these results to some distinctive hospitals or cardiologists. It is expected to improve the quality of care [18]. NCDR databases are supported by the government and participation to these databases indicates the performance of all hospitals. It will help in improving the performance of individual clinicians or hospitals.

4.3 Evidence-based medicine

Evidence-based medicine is to use best evidence in decision making about patient’s care. The practice of evidence-based medicine is through integration of individual clinical expertise with the best available external clinical evidence from systematic research [14]. Individual clinical expertise means skill and power of decision they gain by clinical practicing. The more skills they have in diagnosing a disease and the more efficient they are in making clinical decisions; they are regarded as clinical experts. The research related to medicines and clinical examination such as diagnosis is known as best available external clinical evidence. External clinical evidence replaces the old treatment procedures and guidelines with latest and more effective procedures. It is best to use the combination of both, i.e. individual clinical expertise and the best available external evidence. They are both dependent on each other. External evidence cannot be applied without the help of clinical expertise. Furthermore, not using the current best external evidence means individual doctors are following the out-dated procedures.

4.4 Data driven approaches

Guidelines are designed to help practitioners assimilate, evaluate and implement the ever-increasing amount of evidence and opinion on best current practice [17]. These guidelines help the practitioner in making decisions on how to endow their patients with quality medical care. To further assist this process, fuzzy based solutions were adopted in this domain. Fuzzy C-mean (FCM) is most widely used clustering algorithm to extract knowledge from the data sets in which data points have partial relation to the cluster [4 , 21]. The data points of the data sets have partial membership to the already defined cluster centers. These cluster centers are outside the data set and are chosen randomly. New cluster centers are updated through an iterative process, hence, both cluster centers and membership values of the data points are changed in each iteration. A data point can have the membership of all clusters with the corresponding membership value. Fuzzy clustering approach tackles the noisy data and outliers, and it also has an ability to deal with the data having various types of variables [4 , 16].

The examination of financial, sensor, medical and sentimental streaming data, also referred as serial data, is hard to use due to a problem called sharp boundary problem. It is tough to elect the boundary values. This data is often transformed into discrete data such as categorical data and boolean data for effective utilization. Sharp boundaries problem is addressed in [8 , 21]. It is achieved by developing a novel fuzzy association rule mining method [21]. This method can be applied in a wide range of classifying problems, such as the classification of sentiment strength. In this technique, the authors use original data sets of physical and emotional diseases to make fuzzy sets. In this dual compromise scheme, the first trade-off balances well the performance of out-putting association rules and more extensively appropriate fuzzy membership function, although the second trade-off decreases the time parameter [2, 20]. In DOFARM technique [21], instead of only single partitioning points, the authors use a series of values to efficiently separate two successive partitions and develop a corresponding membership function to make fuzzy sets for original data sets of physical and emotional diseases.

5 Conclusion

In this work, we developed a system called hospital management society to address some health care related issues in our country. CPGs and associate tools/databases are only effective in validation of underlying clinical practices and improving the performance of individual hospitals. On the other hand, evidence-based and data-driven approaches can help the practitioners to follow procedures based on best external evidences and guidelines. However, our solution brings uniformity in the health care facilities, to make the medical treatments cost effective, to find out most common diseases in a particular age & area, and to help government in identifying the areas facing shortage of licensed medicines. Data mining techniques i.e. fuzzy c-means clustering algorithm and concepts of association rule mining were used to obtain the aforementioned functionality. Application of FCM resulted in the identification of common diseases in a particular age group and location, shortage of a medicine in an area; which may assist the government in the efficient provision of those medicines in selected areas. Therefore, the overall system brings uniformity in health care, where everyone have access to the same quality of medical treatment and medicines, and the procedure of medical treatment becomes cost effective by following best available clinical practices as per the guidelines. We are planning to extend our proposed system by considering other attributes, e.g. weather, for better analysis.

Footnotes

Acknowledgment

This work is supported by the Deanship of research at Islamic University of Madinah(IUM), Saudi Arabia (Takamul-11, No. 208). We give special thanks to the administration of IUM for their support in every aspect of this work. We would like to thank and Acknowledge the work done by all the stakeholders of this project, especially the hospitals for providing patients data for our experiments. The content is solely the responsibility of the authors and does not necessarily represent the official views of the Islamic University of Madinah, KSA or Fatima Jinnah Women University, Pakistan. We are also thankful to the anonymous reviewers for their constructive comments and suggestions.

References

Arcos-Aviles

, Pascual

, Marroyo

, San-Chis

and Guinjoan

, Fuzzy logic-based energy management system design for residential grid-connected microgrids, IEEE Transactions on Smart Grid 9 (2016), 530–543.

Borah

and Nath

, Identifying risk factors for adverse diseases using dynamic rare association rule mining, Expert Systems with Applications 113 (2018), 233–263.

Castillo

and Atanassov

, Comments on fuzzy sets, interval type-2 fuzzy sets, general type-2 fuzzy sets and intuitionistic fuzzy sets, in Recent Advances in Intuitionistic Fuzzy Logic Systems, Springer, (2019), 35–43.

Chattopadhyay

, Pratihar

D.K.

and Sarkar

S.C.D.

, A comparative study of fuzzy c-means algorithm and entropy-based fuzzy clustering algorithms, Computing and Informatics 30 (2012), 701–720.

De Silva

C.W.

, Intelligent control: fuzzy logic applications, CRC press, (2018).

El-Sappagh

, Kwak

, Ali

and Kwak

K.-S.

, Dmto: a realistic ontology for standard diabetes mellitus treatment, Journal of Biomedical Semantics 9 (2018), 8.

Kurniasih

, Jasmi

K.A.

, Basiron

, Huda

and Maseleno

, The uses of fuzzy logic method for finding agriculture and livestock value of potential village, International Journal of Engineering & Technology 7 (2018), 1091–1095.

Lei

, Jia

, Zhang

, He

, Meng

and Nandi

A.K.

, Significantly fast and robust fuzzy c-means clustering algorithm based on morphological reconstruction and membership filtering, IEEE Transactions on Fuzzy Systems 26 (2018), 3027–3041.

Luna

J.M.

, Cano

, Pechenizkiy

and Ven-Tura

, Speeding-up association rule mining with inverted index compression, IEEE Transactions on Cybernetics 46 (2016), 3059–3072.

10.

Mlynarski

, Ilczuk

, Pilat

, Wakulicz-Deja

and Kargul

, Automated decision support and guideline verification in clinical practice, in Computers in Cardiology, 2005, IEEE, (2005), 375–378.

11.

Nayak

and Devulapalli

, A fuzzy logic-based clustering algorithm for wsn to extend the network lifetime, IEEE Sensors Journal 16 (2015), 137–144.

12.

Parker

J.K.

and Hall

L.O.

, Accelerating fuzzy-c means using an estimated subsample size, IEEE Transactions on Fuzzy Systems 22 (2013), 1229–1244.

13.

Pech-May

, Lopez-Arevalo

and Sosa-Sosa

, Toward the validation of patient data for clinical practice guidelines, in 2009 6th International Conference on Electrical Engineering, Computing Science and Automatic Control (CCE), IEEE, (2009), 1–6.

14.

Sackett

D.L.

, Evidence-based medicine, in, Seminars in Perinatology 21, Elsevier, (1997), 3–5.

15.

Sanati

M.Y.

, Maccaull

, Maibaum

and Majzoobi

M.M.

, Consistency checking of clinical practice guidelines with mitdl, in 2016 Eighth International Conference on Information and Knowledge Technology (IKT), IEEE, (2016), 281–287.

16.

Sheng

, Hou

, Jiang

and Chen

, A novel association rule mining method of big data for power transformers state parameters based on probabilistic graph model, IEEE Transactions on Smart Grid 9 (2016), 695–702.

17.

Twaddle

, Clinical practice guidelines, Singapore Medical Journal 46 (2005), 681.

18.

Van Der Velde

, Brinkhuis

, Kloosterman

, Van Der Putten

, Dijk

, Hoekema

, Dassen

, Brand

, Van Der Veen

, Boorsma

, et al., Implementation of a national database infrastructure for registration of clinical procedures and as tool for national benchmarking, in 2008 Computers in Cardiology, IEEE, (2008), 261–264.

19.

Viani

, Larizza

, Tibollo

, Napolitano

, Priori

S.G.

, Bellazzi

and Sacchi

, Information extraction from italian medical reports: An ontology-driven approach, International Journal of Medical Informatics 111 (2018), 140–148.

20.

Wang

, Li

, Duić

, Mi

, Hodge

B.-M.

, Shafie-Khah

and Catalão

J.P.

, Association rule mining based quantitative analysis approach of household characteristics impacts on residential electricity consumption patterns, Energy Conversion and Management 171 (2018), 839–854.

21.

Zheng

, He

, Huang

, Zhang

and Wang

, Dynamic optimisation based fuzzy association rule mining method, International Journal of Machine Learning and Cybernetics 10 (2019), 2187–2198.