Content uploaded by Terungwa Simon Yange
Author content
All content in this area was uploaded by Terungwa Simon Yange on Jul 27, 2020
Content may be subject to copyright.
12th Health Informatics in Africa Conference (HELINA 2019)
Peer-reviewed and selected under the responsibility of the Scientific Programme Committee
*Correspo nding author: Address: University of Agriculture, Makurdi, Nigeria. Email: lordesty2k7@ymail.com Tel: +2348064067803,
+2347056808523
© 2019 HELINA and JHIA. This is an Open Access article published online by JHIA and distributed under the terms of the Creative Commons
Attribution Non-Commer cial License. J Health Inform Afr. 2019;6(2):64-73. DOI: 10.12856/JHIA-2019-v6-i2-256
A Fraud Detection System for Health Insurance in Nigeria
Terungwa Simon Yangea*, Oluoha Onyekwareb and Hettie Abimbola Soriyanc
aDepartment of Mathematics/Statistics/Computer Science, University of Agriculture, Makurdi, Nigeria
bDepartment of Computer Science, University of Nigeria, Nsukka, Nigeria
cDepartment of Computer Science and Engineering, Obafemi Awolowo University, Ile-Ife, Nigeria
Background and Purpose: This research developed a Fraud Detection System for National Health
Insurance Scheme (NHIS) in Nigeria. This was with a view to addressing the fraudulent activities of
some stakeholders in NHIS; as many researches have proven that the lack of appropriate tools to do
this has negatively affected service providers as well as the beneficiaries of this Scheme.
Methods: In order to achieve the aim of this research, an inspection of organizational documents,
direct observation and collection of existing data from NHIS accredited health facilities and Health
Maintenance Organizations in Nigeria were carried out. The system was designed using Unified
Modelling Language (UML) tools. The implementation of the system was done using MongoDB as
the big data storage mechanism for the input, Comma Separated Values (CSV) files as a storage
facility for the intermediate results generated during processing and MySQL as the storage
mechanism for the final output, Apache MapReduce as the big data processing platform, Association
Rule Mining as the data analytics algorithm, and Java programming language as the implementation
technology.
Results: The system modules of comprised of four modules: user management, enrollment
processing, referral processing and claims processing. With this, it identified different types of
frauds in NHIS such as double billing, billing for services not provided, ghost patients,
identity theft, self-referral, collusion with providers and kickback schemes.
Conclusions: This paper developed a system for the detection of the fraudulent activities of the actors of NHIS.
This system employed data from the Nigerian NHIS which was categorised into: enrollment, referral and claim
data with different file formats: pdf, jpg, png, csv and excel.
Keywords: Fraud, Detection, Health, Insurance, Prevention, NHIS.
1 Introduction
Health insurance covers the entire or fragment of the risk of a patient incurring medical expenses,
spreading the risk over a large number of persons [1]. By estimating the total risk of healthcare and other
expenses in the healthcare system over the pool of risk, an insurer can develop a routine finance structure,
such as a monthly premium or payroll tax, to provide the money to pay for the healthcare benefits
specified in the insurance treaty. The benefit is administered by a central body such as a government
agency, private business or charity organisations [2]. Healthcare insurance has attracted substantial
interest in the past. It is a social security plan that promises the delivery of the needed healthcare services
to a person on the contribution of a token to provide economic protection to the participants. It is also, a
mechanism for protecting families against the unforeseen exorbitant costs of illness by sharing the risks
of future costs among healthy and sick populations in the form of regular predictable payments [3].
The National Health Insurance Scheme (NHIS) is a scheme put in place by the Federal Government of
Nigeria to provide full access to quality healthcare service in the Country [4]. The scheme covers civil
Yange et al. / A Fraud Detection System for Health Insurance in Nigeria
65
© 2019 HELINA and JHIA. This is an Open Access article published online by JHIA and distributed under the terms of the Creative Commons
Attribution Non-Commer cial Lic ense. J Health Inform Afr. 2019;6(2):64-73. DOI: 10.12856/JHIA-2019-v6-i2-256
servants, the armed forces, the police, the organized private sector, students in tertiary institutions, self-
employed, vulnerable persons, and the unemployed among others [5]. Social security is a human right, as
well as an economic and political necessity; it is an indispensable part of an efficient market economy.
Social security programmes are usually established as a means of improving the well-being of the poor,
reduce inequality within society and conciliate different social demands, thus avoiding the social and
political conflicts [6]. Adequate social security policies could be an important factor in the process of
socio-political development and economic growth of our dear Country, Nigeria.
Conceptually, NHIS is a welcomed innovation and development in the Nigeria health sector given its
objective. Although at present, only public servants in Federal establishments benefit from this scheme
since the Federal Government is the only public-sector that has implemented the programme. On
reduction of dependence on government for funding health services, [5] acknowledged that the scheme
has reduced the burden on the government and improved the funding of health service through its
contributory strategy. The 15 percent deduction from basic salaries of workers, which is remitted to the
NHIS and the co-payment system, has increased healthcare funding.
Currently, the implementation of NHIS has not been easy due to inadequate physical health facilities and
personnel, administrative and logistics bottlenecks [7]. The nation does not have enough healthcare
providing institutions with adequate medical facilities and personnel for effective implementation of the
Scheme. Besides, the administration of the Scheme has not been easy given the delays in processing
document of registered beneficiaries and remitted contributions to the NHIS, and Health Maintenance
Organisations (HMOs) and Heath Provides (HPs). Furthermore, the informal sector is very difficult to
organized for the Scheme. Even private hospital and clinic are becoming unwilling to embrace the
scheme. This has made NHIS prone to different types of fraudulent activities.
According to [8], healthcare fraud is defined as “an intentional deception or misrepresentation made by a
person or an entity, with the knowledge that the deception could result in some kinds of unauthorized
benefits to that person or entity”. The NHCAA projected conservatively that at least 3%, or more than 60
billion dollars, of the United State of America’s yearly healthcare expenses was lost to fraud in 2010.
This loss which was also estimated by other government and law enforcement agencies was as high as
10%. Besides the monetary loss, fraud has severely hampered the healthcare system from providing
quality care to authentic beneficiaries. Hence, the need for an effective fraud detection is imperative for
providing easy access to healthcare services, improving the quality and reducing the cost of healthcare
services. In any discussion concerning fraud, it is important to state clearly the distinction between fraud
prevention and fraud detection. Fraud prevention define procedures to stop fraud from taking place. Fraud
detection on the other hand, encompasses recognizing fraud as quickly as possible once it has been
committed. Several fraud detection cases involve huge datasets that are continuously changing. In the
nutshell, fraud detection surfaces once there is failure in fraud prevention [9].
Instead of prioritising quality healthcare delivery, fraudulent activities are prevalent among
professionals in NHIS. These fraudulent activities include unrealistic bills from hospitals, cooking up
cases in order to extort the Scheme by the hospitals, collusion of patient with providers or providers with
HMOs, ghost patients, managed care fraud, reverse false claim cases, lying about eligibility, scheduling
extra visits for patients, billing for services rendered by unqualified personnel, ganging, using the wrong
diagnosis, false negotiation cases, providing unnecessary care and maximizing care, billing for services
not provided, submitting double bills etc. [1][6][10]. The absence of a robust and functional health
information system and lack of adequate modern information technology infrastructure has hindered the
detection of fraud, sharing of information, prompt and timely data processing, creation of a database
between various stakeholders in the scheme [6][7].
This paper developed a system for the detection of fraudulent activities in the health insurance scheme in
Nigeria, NHIS. The system employed Association Rule Mining algorithm, MapReduce Framework,
MongoDB, MySQL and Java Programming Language.
1.1 Health Insurance Fraud
The substantial increase in medical expenditures required to satisfy the quest for high quality and high-
technology services has given birth to a greater demand for health insurance schemes. Most people now
bank on health insurance systems, which are either funded by the government or managed by the private
sector, to share the high cost of healthcare. With this intensive need for health insurance, fraudulent
Yange et al. / A Fraud Detection System for Health Insurance in Nigeria
66
© 2019 HELINA and JHIA. This is an Open Access article published online by JHIA and distributed under the terms of the Creative Commons
Attribution Non-Commer cial Lic ense. J Health Inform Afr. 2019;6(2):64-73. DOI: 10.12856/JHIA-2019-v6-i2-256
behaviours become a serious problem. For instance, [11] reported that 10% of United States annual
spending on healthcare is loss to fraud. The health insurance programmes of other countries are also faced
with similar challenges [12]. The other healthcare crimes (i.e., medical and drug) which involve surgeries,
invasive testing, certain drug therapies etc., even place their patients at significant physical risk and affect
patients’ health.
Fraud in the healthcare insurance involve three parties [12][13]: the healthcare service provider (i.e.,
the physician, pharmacist, laboratory scientist, health centre, pharmacy, laboratory, and even ambulance
companies) which render healthcare services; the healthcare service consumer or beneficiary or insurance
subscriber (i.e., patient) which receive healthcare service from the provider; and the healthcare insurance
carrier which collect steady premiums from subscribers and make the commitment to pay healthcare cost
on their behalves. These parties exchange information amongst them in the course of care delivery. This
is basically in the form of service requested by the subscriber (patient visit) to the provider, explanation of
benefits which contain the detail services rendered by the provider to the subscriber, claim/bill which is
sent to carrier for the services rendered to the subscriber by the provider, and the payment to the provider
based on the claim submitted to the carrier [11]. As the number of beneficiaries (patients) of this scheme
increases, high volume of data is generated by both the providers and the carriers; and consequently,
some fraudulent activities (such as billing services that were never rendered, performing medically
unnecessary services, misrepresenting non-covered treatments as medically necessary covered treatments,
and misrepresenting applications for obtaining lower premium rate) are carried by these actors
(beneficiary, provider and insurer) which give rise to the need to investigate such acts in an attempt to
identify perpetrators, and this requires a proper analysis tool for the purpose [13][14].
The reasons why health insurance fraud has become a prevalent practice is that majority of those
involved find it beneficial in diverse ways. Several surgeons see it as necessary to provide quality care for
their patients [15]. Most patients, although disapproving of the idea of fraud, are occasionally more eager
to admit it when it affects their own medical care. Programme Administrators are often compassionate on
the issue relating to healthcare insurance fraud as they seem to take full advantage of the services of their
providers [14]. A summary of healthcare frauds is highlighted below.
Kickback schemes: One of the widespread and discussed form of healthcare fraud is kickbacks [11].
These occur in many forms. For instance, pharmacists can fill a prescription with a specific brand of
drugs instead of the other that yields a bonus from the pharmaceutical company. Aside, financial
implications, these drugs might also be harmful to the patient’s health. Physicians themselves can
fraudulently write prescriptions for money, essentially a kickback from the downstream illegal sale of
these drugs.
Self-referral: This refers to the transfer of a patient to a facility with which the referring healthcare
personnel has a financial relationship [12]. This might involve a kickback scheme if the facility where the
patient is referred to, pays a certain amount of money back to the physician, but other financial
relationships are conceivable [16].
Identity fraud: This kind of fraud occurs when an uninsured individual assumes the identity of an
insured person so as to benefit from the services packaged in the scheme or to hide a particular illness.
The healthcare services eventually provided to the person ‘lending’ their identity could be negatively
affected, since this will be at par with their actual health records. This kind of fraud can also be
perpetrated without the consent of the actual owner [14] [16].
Double bills: Many care providers submit the same claim multiple times, in order to get paid different
times for performing the same action (i.e., submitting a particular claim multiple time for the same
service) [10] [17]. This is known as double billing and it is also a fraud.
Billing for services not provided: Here, claims are generated and submitted to insurance companies
for healthcare services that are not provided or for drugs that have not been delivered to the patient. This
is known as phantom billing [9] [13].
Ghost Patients: The submission of a claim for healthcare services provided to a patient who either
does not exist or who never received the service or item billed in the claim or patients that are dead or
have changed their provider [13].
Collusion with providers: Both provider and the member collude to submit false claims where the
physician receives the benefits from the false claims [18].
Yange et al. / A Fraud Detection System for Health Insurance in Nigeria
67
© 2019 HELINA and JHIA. This is an Open Access article published online by JHIA and distributed under the terms of the Creative Commons
Attribution Non-Commer cial Lic ense. J Health Inform Afr. 2019;6(2):64-73. DOI: 10.12856/JHIA-2019-v6-i2-256
Among these frauds, the ones committed often by health service providers accounts for the greatest
proportion of the total healthcare fraud. The reason for this is that the historically prevailing attitude in the
medical profession is one of “fidelity to patients”. Although the vast majority of service providers are
honest and ethical, the few dishonest ones may have various possible ways to commit fraud on a very
broad scale, thus posing great damage to the healthcare system. Some service providers’ fraud, such as
that involving medical transportation, surgeries, invasive testing, and certain drug therapies, even places
patients at a high physical risk.
1.2 Investigation of Healthcare Insurance Fraud
[19] developed a fraud detection system using data mining techniques. In all the categories of
insurance including health insurance, fraud is a major issue. According to the researchers, fraud in health
insurance is perpetrated via the intentional deception or misrepresentation of facts for gaining some shady
benefit in the form of healthcare expenses. Data mining tools and techniques are used to identify
fraudulent activities in large insurance claim datasets. Based on some identified cases of fraud from a
sample dataset, the anomaly detection technique calculates the likelihood or probability of each record to
be fraudulent by analysing the past insurance claims. The analysts can then have a closer investigation for
the cases that have been marked by data mining software.
[10] developed a data mining approach for the investigation of fraud in health insurance scheme using
knee-point k-means algorithm. NHIS was considered as the case study for the work. The research focuses
on the use of some computer-based methods that could help to properly target investment in the
healthcare sector and also drastically reduce fraud in health insurance by healthcare providers. To this
effect, they applied the knee-point k-means clustering method, which was capable of detecting fraudulent
claims by health service providers. Cluster-based outliers were examined. Health providers’ claims
submitted to HMO were grouped into clusters. Claims with similar characteristics were grouped together.
The claims were grouped into two clusters: fraudulent and non-fraudulent.
In a survey of hybrid methods for the uncovering fraud in healthcare insurance by [20], acts carried out
with the aim of obtaining a fraudulent outcome from health insurance processes were carefully examined.
According to the researchers, when providers tries to enjoys some benefits or advantages to which they
are not entitled then that attempt is considered as insurance fraud and it has become a major concern for
health insurance companies. They proposed a hybrid framework that applied some data mining
techniques to detect frauds. This framework considered the analysis of the characteristics of healthcare
insurance data, some preliminary knowledge of healthcare system and the fraudulent behaviours. The
framework harnessed the advantages of both the supervised, semi-supervised and unsupervised learning
methods to identify fraudulent claims.
[21] investigated the benefits of applying big data techniques in the detection of fraud in public health
insurance system in Romania. They outlined the benefits of using big data technology in combating crime
in the healthcare industry and came up with the following conclusions:
i. that big data technology and its distributed processing power has taken fraud detection in
healthcare insurance to a higher level. Few years ago, insurance fraud detection was not
considered cost-effective as it takes time and was also too expensive, and hence, so many
organisations prefer to pay claims without proper investigation.
ii. that using big data analytics approaches can culminate into speedy identification of fraudulent
claims, and also generates a new set of tests to automatically reduce the section that was
potentially fraudulent or detect new patterns of fraud, previously not known.
iii. that the processing of complex data using sophisticated tools reveals its huge potential, and also
shows that orthodox tools for data analysis cannot handle them. The analytics tools applied in the
field of healthcare insurance were briefly described, each of them being effective for a particular
type of fraud or a particular stage of the fraud detection process. All these culminate into the
deduction that the best solution for detecting fraud in the health insurance system is, at present, a
hybrid solution, both in terms of technologies and in terms of models of analysis.
iv. that in healthcare insurance frauds are country-specific, usually based on gaps or weaknesses in
the country’s rules and regulations. Models are continuously changing fraud as dubious
individuals are always seeking for new ways to evade the law.
Yange et al. / A Fraud Detection System for Health Insurance in Nigeria
68
© 2019 HELINA and JHIA. This is an Open Access article published online by JHIA and distributed under the terms of the Creative Commons
Attribution Non-Commer cial Lic ense. J Health Inform Afr. 2019;6(2):64-73. DOI: 10.12856/JHIA-2019-v6-i2-256
Accordingly, approaches for detecting and preventing fraud must always be adjusted and ready to
rediscover the fraudulent actions [22][23][24]. To add to the lapses in the country’s constitution, each
country has unique economic, political, social, and institutional opportunities for and barriers which
makes fraud examination different amongst countries. A crucial and peculiar issue in the NHIS is the high
level of corruption in the sector, lack of accountability and clear sense of irresponsibility [5][22].
[25] proposed a model using big data in investigating real time crime in the health insurance in the
cloud. This approach utilizes fraud management solution to detect potential frauds in the cloud. The
solution was based on massive amount of historical data, predictive statistical models and social network
analytics. The model renders its services via client components like apps and web-services. Just like [21]
and [20][25][26] did not implement any working system for their research. Also, as opined by [21],
healthcare crimes are country specific and Nigeria has not adopted the cloud services and there are no
healthcare laws relating to data on the cloud, and therefore, this model cannot be used to investigate crime
in our healthcare system.
[27] developed a data analytics framework for Health Insurance data using Association Rule Mining.
This was in a bid to identify fraudulent claims as deliberate cheating by concealing and omitting facts
while claiming from health insurance providers has become in the health insurance domain which has led
to significant amount of monetary loss to the providers. In view of the above, careful scanning of the
submitted claim documents need to be conducted by the insurance companies in order to spot any
discrepancy that indicates fraud. For this purpose, manual detection was neither easy nor practical as the
claim documents received keep increasing and for diverse medical treatments. The researcher was able to
detect fraudulent health insurance claims by identifying correlation or association between some of the
attributes on the claim documents. With the application of a data mining techniques of evolving clustering
method, association rule mining and support machine, this study was able to successfully determined
correlated attributes to address the discrepancies of data in fraudulent claims and thus reduce fraud in
health insurance. However, the study was used for structured data which made it unfit to be applied in big
data which is highly unstructured. With the numerous data mining techniques implemented traditionally,
it would consume more resources when applied to big data.
2 Materials and Methods
2.1 Data Collection
The data was collected from NHIS. Data collection was one of the most difficult task in this work as most
of the data was collected manually in paper form by NHIS through HMOs and stored in file cabinets.
This was done via document examination and observation which in either case, the data was collected
from journals and NHIS databases. This data was classified into four (4) categories: enrollment, update,
referral and claims. The formats of the data were: CSV, PDF, text, excel and images. The features of the
data include: Enrollment: Name, Address, Date of birth, Sex, Next of Kin, Email address, Mobile,
Telephone no. fixed, National ID no, Employer NHIS no., Date of NHIS registration, Nationality,
Location of Posting, Photograph, Blood group, Genotype, Allergies, Relationship (Principal, Spouse,
Child, Extra-dependant), Expiry date, Primary provider. Claim: Name, NHIS No. of patient, Name and
NHIS No. of patient’s primary provider, Name and NHIS No. of Secondary Provider, drug prescription
sheet, Diagnosis/disease code, Treatment given, Date of treatment, Amount billed, Co-payment received
(when applicable).
2.2 Design of the System
The design was done using Unified Modelling Language (UML). This was used to specified the
developed model into a representative model to aid software system development. Two (2) different types
of diagrams in UML were used to model the system: use case and sequence diagrams.
2.2.1 Use Case Diagram
The use case diagram for the system is shown in Figure 1. It has four (4) actors (i.e., finance manager,
data manager, contributory manager and the admin officer), and eight (8) use case (login /logout, upload
data, analyse claims, analyse enrollment, analyse referral, analyse update, preprocess data and report).
These are discussed as follows.
Yange et al. / A Fraud Detection System for Health Insurance in Nigeria
69
© 2019 HELINA and JHIA. This is an Open Access article published online by JHIA and distributed under the terms of the Creative Commons
Attribution Non-Commer cial Lic ense. J Health Inform Afr. 2019;6(2):64-73. DOI: 10.12856/JHIA-2019-v6-i2-256
Login/Logout: The model is designed to be implemented in a web-based environment and hence,
security is important. With this mindset, all users must possess valid user credentials before accessing
the system. Before accessing the system, the user must provide these credentials and the system must
acknowledge that these credentials are valid.
Upload Data: Uploading data needed for the system come in two folds. First, the existing data are
digitized and uploaded to the system, and this is done by the data manager. The second method, the data
is uploaded via the automated systems.
Analyse Claims: This is the processing of claims submitted for reimbursement. This produced two
results: fraudulent and non-fraudulent.
Figure 1: Use Diagram for the System
Analyse Enrollment: This is the processing of enrollment data submitted for new registration. This
produced two results: fraudulent and non-fraudulent.
Analyse Update: This is the processing of update data submitted for addition of a dependant, change of
primary facility or HMO etc. This produced two results: fraudulent and non-fraudulent.
Analyse Referral: This is the processing of referral requests made for referral of patients to higher
facilities. This produced two results: fraudulent and non-fraudulent.
Preprocess Data: The cleansing of the data submitted for the processing.
Reports: These are insights generated after the processing of data.
2.2.2 Sequence Diagram
Sequence diagram shows object interactions arranged in time sequence. It depicts the objects and
classes involved in the scenario and the sequence of messages exchanged between the objects needed to
carry out the functionality of the scenario. Sequence diagrams are typically associated with use case
realizations in the logical view of the system under development. Sequence diagrams are sometimes
called event diagrams or event scenarios. Figure 2 shows the sequence diagram depicting the static
features of the proposed basic flow of the model. This shows the different interactions by components of
the model, the analytics and the results to be generated and their associated relationships. In the Figure 2,
the data in the real world is captured and loaded into the big data warehouse (MongoDB) using the
loadData() function. The data is extracted using the extractData() function and it is pre- processed and the
attributes of interest are defined and captured using the dataPreprocessing() function. Also, if the data is
structured, it is retrieved from the big data warehouse and preprocess directly with the aid of the
dataPreprocessing() function. After preprocessing, the data is loaded into the map phase of the
MapReduce where the different attributes are collated together using the collateAttributes() and rules are
generated from the collated attributes using the generateRules() function. The generated rules are then
pruned so as to drop the weak ones and pick the strong ones using the values of their confidence and
support as computed by the pruneRules() function. These rules are used to generate insights using the
generateReport() function. These insights are applied by the stakeholders using the apply() function, and
Finance Manager
Login/Logout
Upload Da ta
Analyse C laims
Preprocess
Reports
Analyse Enrollment
Analyse Referral
Analyse Up date
Admin Of ficer
Contributory Manager
Data Manager
Yange et al. / A Fraud Detection System for Health Insurance in Nigeria
70
© 2019 HELINA and JHIA. This is an Open Access article published online by JHIA and distributed under the terms of the Creative Commons
Attribution Non-Commer cial Lic ense. J Health Inform Afr. 2019;6(2):64-73. DOI: 10.12856/JHIA-2019-v6-i2-256
are also stored in the relational database (MySQL) for future reference using the storeOutput() function.
To make reference to these insights, a request is sent the stakeholders using the request() function and the
response to the request is sent back to the stakeholders from the database using the response() function.
3 Results
The implementation of the system was developed using Java Enterprise Edition technology. It is web-
based and can be used in a distributed setting. The front-end is built using Java Server Pages, the business
logic is implemented using Enterprise Java Beans. and the back-end is implemented in two folds: input
storage was implemented using mongoDB and the output storage was implemented using Structured
Query Language (SQL-MySQL). The input storage (MongoDB) accepts data in different formats (e.g.,
pdf, jpeg, png, gif, csv and excel) as inputs and process them to produce report for the stakeholders which
is stored in MySQL. The large input in the different formats is first divided into smaller units using the
Association Rule Mining Algorithm, which is implemented in the MapReduce Framework to ease the
processing of this data. The system extracts the data stored in these formats via an OCR component built
in it.
This implement the parallel distributed processing model of the MapReduce using the Apriori algorithm
of the association rules mining. This classes contains constants, variables and methods
Figure 2: Sequence Diagram for the System
that are used in fraud analytics. The business logic module has a sub module known as the fraud analytics
module which implement the MapReduce framework and the Apriori Association Rule Mining algorithm.
Whenever a task is forwarded to this sub module, details about the task is sent to the JobTracker which
coordinate, allocates and control all the activities within the analytics sub module. When these details get
to the JobTracker, it assigns the TaskTrackers the job which in turns load the task unto the mapper
classes. The TaskTracker also feed the JobTracker with every detail of what happens with the task, both
in the mapper and the reducer classes. The relationship between the JobTracker and the TaskTracker is a
master-slave one, the JobTracker is the master while the TaskTracker is the slave. The mapper and
reducer classes are the kernel of the analytics process. While the mapper carryout the dirty task of the
analysis, the reducer summarises whatever is produced as the outcome from the mapper. This output is
sent to the relational database, MySQL, through the entity bean. It is through this same entity bean that
request for retrieval of that from the database and responses to the JSPs follows.
Stakeholders M ongoDB
loadData()
Extraction
dataPreprocessi ng()
Preprocessing
extractData()
dataPreprocessing ()
Mapper/
Reducer
loadData()
Generate
Rules
generateRules()
Report/Insight
request()
response()
MySQL
storeOutput()
apply()
CollateAttribute s()
generat eReport ()
pruneRules()
Yange et al. / A Fraud Detection System for Health Insurance in Nigeria
71
© 2019 HELINA and JHIA. This is an Open Access article published online by JHIA and distributed under the terms of the Creative Commons
Attribution Non-Commer cial Lic ense. J Health Inform Afr. 2019;6(2):64-73. DOI: 10.12856/JHIA-2019-v6-i2-256
The backend is designed using two different databases: MongoDB and MySQL, and temporary csv
files. MySQL is a relational database management system, and it is used for storing the output of the
analytics. The logic about the MySQL component of the back-end is housed in the entity bean. This
component stores only processed data. The MongoDB is a NoSQL database which is used for storing big
data as it is captured from the various health facilities. This is where data captured by other applications
too is channelled to.
Figure 3 shows the capturing of the data into the system. Figure 4 and Figure 5 show the detection of
fraud from claims and enrollment data.
4 Discussions
The modules of the application are discussed in the sections that follow below and it comprised of four
modules: user management, enrollment processing, referral processing and claims processing.
User Management: This system uses single-sign-on which implies that all users must login through a
single interface and are authenticated before authorizing them to access the system resources. The module
is responsible for adding new staff and assigning login credentials (username and password) to them.
Claim Processing: This module handles the processing of all the claims submitted by providers for fee-
for-service payment. The module verifies every item on the claim by comparing it with NHIS existing
rate and anyone that fall short is either recalculated or at worse rejected. In the course of processing, all
unprocessed claims in the repository are picked and process. This is to ensure that no claim spend up to
fourteen days before been processed. The system was able to detect double billing, upcoding, billing for
services not provided and identity theft.
Figure 3: Data Capturing
Figure. 4: Summary View of Processed Principals’ Enrollment Data
Yange et al. / A Fraud Detection System for Health Insurance in Nigeria
72
© 2019 HELINA and JHIA. This is an Open Access article published online by JHIA and distributed under the terms of the Creative Commons
Attribution Non-Commer cial Lic ense. J Health Inform Afr. 2019;6(2):64-73. DOI: 10.12856/JHIA-2019-v6-i2-256
Figure. 5 Summary View of Processed Claims Data
Enrollment Processing: This module is responsible for adding or updating enrollees’ information. These
include personal information, next of kin information, educational information, employment information,
primary providers etc. During enrollment, enrollees are checked to ensure that no enrollee registers twice,
the number of biological children for an enrollee is not more than four. Also, that a child that is more than
eighteen years is either de-register or not allowed to register. The enrollee is only allowed to register one
spouse at a time. Also, there is provision to add one extra dependant which can only be allowed when the
enrollee accept to be paying five thousand naira annually in addition to the normal monthly contribution.
Again, enrollees who are not staying together with their dependants are allowed to have different primary
facilities, otherwise, the system only allowed them to have one facility in accordance with NHIS
guideline. The module was able to detect double registration of enrollees, registration more than four
biological children, registration of children that are more than eighteen years and ghost enrollees.
Referral Processing: The module is responsible for the handling cases of referrals. A beneficiary is
referred to another NHIS accredited hospital when the medical condition of the beneficiary is more than
what the primary provider can handled. The principle of referral is that a beneficiary should be referred to
the next closest NHIS accredited facility for a more specialized care. When referrals are made, the system
evaluates the diagnoses in the referral request and compares it with what allowed by NHIS to be handled
by the next level of healthcare. It also checks the distance between the facility they are referring the
patient to and other facilities around the same location with the primary facility that is making the referral
on behalf of the enrollee. In an event where the distance is more than the other facilities, such referrals are
queried. Also, if the medical condition of the enrollee is not more than what should be handled at primary
facility, they are also queried. Again, if the diagnoses do not tally with what the next hospital is registered
to carry out, it is also queried. Otherwise, the NHIS guideline stipulates that referrals should not last for
more than forty-eight hours with the HMOs. The module was able to detect self-referral, and collusion
with providers.
5 Conclusion
In conclusion, this paper developed a system for the detection of the fraudulent activities of the actors of
NHIS. This system employed data from the Nigerian NHIS which was categorised into: enrollment,
referral and claim data with different file formats: pdf, jpg, png, csv and excel. The system was designed
using UML and implemented using Association Rule Mining, MapReduce Framework, MongoDB,
MySQL and Java Programming Language. The system was able to detect several fraudulent activities of
providers, insurers and beneficiaries. These ranges from self-referral, collusion with providers, double
registration of enrollees, registration more than four biological children, registration of children that are
more than eighteen years, ghost enrollees, double billing, upcoding, billing for services not provided to
identity theft. This system when adopted by NHIS, it will sanitise the entire Scheme.
Yange et al. / A Fraud Detection System for Health Insurance in Nigeria
73
© 2019 HELINA and JHIA. This is an Open Access article published online by JHIA and distributed under the terms of the Creative Commons
Attribution Non-Commer cial Lic ense. J Health Inform Afr. 2019;6(2):64-73. DOI: 10.12856/JHIA-2019-v6-i2-256
References
[1] Olaniyan AO. Assessment of the implementation of national health insurance scheme (NHIS) in south-western nigeria.
Unpublished PhD Thesis submitted to the Department of Public Administration, Ob afemi Awolowo University, Ile-
Ife, Nigeria, 2017.
[2] Etobe EI, Etobe UE. The national health insurance scheme and its implication for elderly care in Nigeria. IJSR. 2015; 4(2):
128-132.
[3] Eteng FO, Ijim-Agbor U. Understanding the challenges and prospects of administering the national health insurance scheme in
Nigeria. IJHSSR. 2016; 2(8): 43-48.
[4] NHIS (National Health Insurance Scheme) (2013). National health insurance scheme operational guidelines. Available from:
http://www.nhis.gov.ng/images/stories/hmoregister/NHIS_OPERATIONAL_GUI DELINES.pdf.
[5] Agba MO, Ushie EM, Osuchukwu NC. National Health Insurance Scheme (NHIS) and employees’ access to healthcare
services in Cross River State, Nigeria. GJHSS. 2010; 10(7): 9-16.
[6] Oyegoke TO, Ikono RN, Soriyan HA. An integrated health management system for national health insurance scheme in
Nigeria. JETCIS. 2017; 8(1): 30-40.
[7] Alimi OM, Binuyo OG, Gambo IP, Jimoh K. Realtime national health insurance scheme (RNHIS): Means to achieve health
for all. IJCSEA. 2016; 6(2): 1-8.
[8] NHCAA (National Health Care Anti-Fraud Association) (2012). “What is health care fraud? Available from:
[www.nhcaa.org/resources/health-care-anti-fraud-resources/consumer-info- action.aspx]
[9] Thornton D, Brinkhuis M, Amrit C, Aly R. Categorizing and describing the types of fraud in healthcare. PCS. 20 15; 64: 713 –
720.
[10] Fasho to SG, Owolabi O, Sadiku J, Gbadeyan JA. Application of data mining technique for fraud detection in health insurance
scheme using knee-point and k-means algorithm. AJBAS. 2013; 7(8): 140-144 .
[11] Bagde PR, Chaudhari MS. Analysis of fraud detection mechanism in health insurance using statistical data mining techniques.
IJCSIT. 2016; 7(2): 925-927.
[12] Li J, Huang K-Y, Jin J, Shi J. A survey on statistical methods for healthcare fraud detection. HCMS. 2008; 11: 275-287.
[13] Dora P, Sek haran GH. Healthcare insurance fraud detection leveraging big data analytics. IJSR. 2015; 4(4): 2073-2076.
[14] Ekin T, Ieva F, Ruggeri F, Soyer R. Applications of bayesian methods in detection of healthcare frauds. CET. 2013; 33: 151-
156.
[15] Musal R. Two models to investigate medicare fraud within unsupervised databases. ESA. 2010; 37(12): 8628-8633.
[16] Liu Q, Vasarhelyi M. Healthcare fraud detection: A survey and a clustering model incorporatin g geo-location information. In
the Proceedings of the 29th World Continuous Auditing a nd Reporting Symposiu m November 21-22, 2013, Brisbane,
Australia.
[17] Jacqulin MJ, Shrijina S. Implementation of data mining in medical fraud detection. IJCA. 20 13; 69(5): 1-4.
[18] Travaille P, Thornton D, Müller MR, Hillegersberg J. Electronic fraud detection in the u.s. medicaid healthcare program:
Lessons learned fro m other industries. In the Proceedings of the Seventeenth Americas Conference on Information Systems,
Detroit, Michigan August 4th-7th 2011.
[19] Kirlidog M, Asuk C. A fraud detection approach with data mining in health insurance. SBS. 2012; 62: 989 – 994.
[20] Bagul PD, Bojewar S, Sanghavi A. Su rvey on h ybrid approach for fraud detection in health insurance. IJIRCCR. 2016; 4(4):
6918-6922.
[21] Bologa A, Bologa R, Florea A. Big data and specific analysis methods for insurance fraud detection. DSJ. 2010; 1(1): 30-39.
[22] Arodiogbu IL. Introducing social health insurance to solve problems of poor health sector financing in Nigeria. Unpublished
MSc Thesis Submitted to the Department of Health Management, Planning and Policy, University of Leeds, 2005.
[23] Dutta A, Hongoro C. Scaling up national health insurance in nigeria: Learning from case studies of india, colombia, and
Thailand. Washington, DC: Futures Group, Health Policy Project, 2013.
[24] Yunusa U, Irinoye O, Suberu A, Garba AG, Timothy G, Dalhatu A, Ahmed S. Trends and challenges of public health care
financing system in nigeria: The way forward. IOSR. 2014; 4(3): 28-34.
[25] Konasani V, Biswas M, Koleth PK. Healthcare fraud management using big data analytics. An Unpublished Report by
Trendwise Analytics, Bangalore, India, 2012.
[26] Oyegoke TO. Development of an integrated health management system for national health insurance scheme. An Unpublished
M.Sc. Thesis Submitted to the Department of Computer Science and Engineering, Obafemi Awolowo University, Ile-
Ife, Nigeria, 2015.
[27] Kareem S, R. B. Ahmad RB, Sarlan AB. Framework for the identification of fraudulent health insurance claims using
association rule mining. 2017 IEEE Conference on Big Data and Analytics (ICBDA). 2015; 99-104.
