DATA PRIVACY PRESERVING MODEL FOR HEALTH INFORMATION SYSTEM

ABSTRACT
Information systems are employed by organizations for the collection, filtering, processing, creation and distribution of data. In healthcare delivery, patients are required to share information with certain categories of health personnel to facilitate correct diagnosis and to determine appropriate treatment. There have been cases of unauthorized access to patient information by health personnel. Some of these personnel eventually cause great harm to the patient by divulging sensitive information. The existing Data Privacy Preservation (DPP) models are designed for Clinical Decision Support Systems with inadequate information available for DPP in Health Information Systems (HIS) in Nigeria. This research, therefore focused on the development of a model for Data Privacy Preservation (DPP) in HIS to address this inadequacy.
A model for DPP in HIS was developed using the iterative design technique. The model developed comprises a local database that contains the health information of patients, the Random Forest Decision Tree (RFDT) algorithm, an attribute blocking module that employs the RFDT algorithm, an attribute unblocking module which also uses the RFDT algorithm and a module for the computation of time elapsed in unblocking attributes. Mandatory Role-based Access Control was used to restrict the access health professionals have to patient data; each category of health worker can only view the attribute(s) needed for them to provide the service required to fulfill their role. An application based on the RFDT algorithm, was developed to instantiate the model following the Waterfall Software Development Life Cycle. Netbeans Integrated Development Environment, MySQL server, Java Development Kit 8, Scenebuilder 2.0, and Navicat 8 query editor constitute the programming environment. The application was evaluated against the machine learning approach to DPP that employed the classification technique, by comparing its efficiency with the Waikato Environment for Knowledge Analysis (WEKA) version 3.8 software in ensuring DPP using the RFDT algorithm.
 
The model developed in this study provides a generic framework for DPP in HIS that reveals the necessary components. This model provides a template that could be adapted for use in studies on DPP in HIS. The application provides the health personnel with Graphical User Interfaces that depict the professional’s access to the patient database while restricting access to attributes not allowed for such category of health workers. The use of the RFDT algorithm in WEKA for DPP gave an efficiency of 73.77% while the approach that employed the application gave an efficiency of 78.32%.
The model presented in this study wouldhelp preserve sensitive patient data from being accessed by health workers who are not authorized to do so. The study showed that the application is more efficient than the WEKA software in ensuring DPP using the RFDT algorithm.The DPP model proposed in this study could also be employed in other domains outside the health sector to curb the challenges resulting from weak DPP.

INTRODUCTION

1.1 Background to the Study

Health Information Systems(HIS) provide the bedrock for decision-making and has four key
functions: data generation, compilation, analysis and synthesis, and communication and use. The HIS gathers data from the health sector and other relevant sectors, analyzes the data and ensures their overall relevance, quality, and timeliness, and converts data into information for health-related decision-making.In addition to being essential for monitoring and evaluation, the information system also providesearly warning capability,supports patient and health facility management, facilitate planning, supports and stimulatesresearch, permits health situation and trends analysis, supports global reporting, andunderpins communication of health challenges to diverse users (WHO, 2009).
To improve the quality of medical care around the globe,efforts are being made to increase the practice of evidence-based medicinethrough the use of an HIS called Clinical Decision Support Systems (CDSS). Clinical Decision Support provides clinicians, patients, or caregivers with clinical knowledge and patient-specific information to help them reach decisions that enhance patient care (Osheroff, Teich & Middleton, 2011). The patient’s information is matched to a clinical knowledge base, and patient-specific appraisals are then communicated effectively at appropriate times during patient care. Some CDSS include forms and templates for entering and documenting patient information, and alerts, reminders, and order sets for providing suggestions and other support. The use of CDSS comes with many potential benefits. Importantly, CDSS can increase adherence to evidence-based medical knowledge and can reduce unnecessary variation in clinical practice. CDSS can also assist with information management to support the physicians’ decision making abilities, reduce their mental workload, and improve clinical workflows (Karsh et al., 2010). When well designed and implemented, CDSS have prospects that can improve health care quality, and also to increase efficiency and reduce health care costs (Berner, 2010).
 
Despite the promise of CDSS, there are several barriers that can hinder their development and implementation. Till date, Medical knowledge base is incomplete in part because of insufficient clinical evidence (Englander & Carraccio, 2014). Moreover, methodologies are still being designed to convert the knowledge base into computable code, and interventions for conveying the knowledge to clinicians in a way they can easily usein practice are in the nascent stages of development. Low clinician demand for Clinical Decision Support is another encumbrance to broader CDSS adoption. Clinicians’ lack of motivation to use CDSS appears to be related to usability issues with the Clinical Decision Support intervention, its lack of integration into the clinical workflow, concerns about autonomy, and the legal and ethical implications of adhering to or overriding recommendations made by the CDSS (Berner, 2010). In addition, in many cases, acceptance and use of CDSS are hinged uponthe adoption of electronic medical records (EMRs), because EMRs can include Clinical Decision Support applications as part of Computerized Provider Order Entry (CPOE) and electronic prescribing systems.
One of the five recommendations made for CDSS in connection with the practice of Evidence-based Medicine was to “develop maintainable technical and methodological foundations for computer-based decision support” (Sim, Gorman & Greenes, 2011). Also, the medical domain is “characterized by much judgmental knowledge”. Consequently, a CDSS that can provide suggestive knowledge representations based on data sets with patient attributes that are synonymous with the attributes of the patient in context is valuable to a medical practitioner. Invariably, there are situations where the number of local samples to draw conclusions from, is none or few. Several current challenges have not been sufficiently addressed during the development of CDSS. From latest research, the lists of challenges include: improvement of the human-computer interface, dissemination of best practices in CDSS design, development, and implementation, creation of an architecture for sharing executable CDSS modules and services, combination of recommendations for patients with co-morbidities,summary of patient-level information, prioritization and filtering of recommendations to the user, prioritization of CDSS content development and implementation, creation of Internet-accessible clinical decision support repositories, usage of free text information to drive clinical decision support, and mining of huge clinical databases to create new CDSS (Kumar&Prabha, 2016).
Psychiatry is one branch of medicine that urgently needs HIS owing to the fact that there are relatively few specialists in that area of medicine (Saha,Chant, Welham, & McGrath, 2015). According to the National Alliance on Mental Illness, mental illnesses are medical conditions that disrupt a person’s clear thinking, feeling, mood, ability to relate to others, decision making ability and daily functioning (NAMI, 2011). Mental illnesses include schizophrenia, depression, bipolar disorder, obsessive-compulsive disorder (OCD), posttraumatic stress disorder (PTSD), borderline personality disorder, anxiety disorder and others. However, schizophrenia involves a relatively higher display of psychotic symptoms than most other mental illnesses (Amin, Agarwal & Beg, 2013).
Schizophrenia is a chronic and debilitating illness characterized by perturbations in cognition, affect and behavior, all of which have a bizarre aspect (Lehman et al., 2010). Due to the fact that schizophrenia is a stigmatized illness it is important for schizophrenic patients’ data to be kept with a high degree of secrecy so as to avoid sensitive patient data being divulged. It is therefore expedient that in Clinical Decision Support Systems that contain data of Schizophrenic patients, access to patient data by the healthcare givers be restricted based on their roles in the hospital. This can be achieved by employing access control. The Mandatory Role-Based Access Control is a type of access control and can be employed for such a study as this. To boost the security of a Health Information System (HIS) through data privacy preservation, this study proposes a model for implementing data privacy preservation in a HIS. This model would help boost the security of the HIS in question through the restriction of access of users to its database.
This study proposesa Data Privacy Preservation (DPP) model for HIS. In order to guarantee the secrecy of sensitive patient data domiciled in a HIS, the study involved the development of an application named Schizoapp which was used to instantiate the proposed DPP model and effected data privacy by blocking attributes on a patient database based on the MandatoryRole-Based Access Control (MAC) model which is used to assign access rights to different categories of health professionals based on their role in the hospital. The study also compared the use of the application (Schizoapp) developed in this study for data privacy preservation with the machine learning approach to data privacy preservation which employed the Random Forest Decision Tree algorithm embedded in the WEKA software.

1.2 Statement of the Problem

In healthcare delivery, patients are required to share information with certain categories of health personnel to facilitate correct diagnosis and to determine appropriate treatment. However, patients would most of the time prefer their sensitive information to be kept secret particularly from persons that need not have access to such information especially in cases of health problems such as schizophrenia as the disclosure of such private information may lead to social stigma and discrimination. There have been cases where health personnel who by virtue of their role ought not to have access to certain patient information gained access to such information. Some of these health personnel cause harm to the patient in question by divulging such details to other individuals thereby jeopardizing the patient’s health. Hence, the healthcare system becomes the worse for it as a number of patients may relapse to worse states they already improved from and the retrogression in the patients’ health status will in the long run take a toll on the healthcare system.
The existing Data Privacy Preservation (DPP) models are designed for Clinical Decision Support Systems with inadequate information available for DPP in Health Information Systems (HIS) in Nigeria. This research, therefore focused on the development of a model for Data Privacy Preservation (DPP) in HIS to address this inadequacy.

1.3 Objective of the Study

The main objective of this study is to propose and implement a DPP model for HIS.
The specific objectives are to:

1. propose a model for DPPin HIS;
2. develop a DPP application to instantiate the model for DPPin HIS;
3. implement DPP in a HISusing the application developed in ii and
4. evaluate the prototype application developed for its efficiency

1.4 Methodology Overview

1. Major existing DPP models were reviewed and grouped into three clusters from which the most recent model in each cluster was selected. The three models chosen from the clusters are (A Cloud-Based eHealth Model for Privacy Preserving Data Integration by Dubovitskaya, Urovi, Vasirani, Aberer and Schumacher(2015); A Data Privacy Preserving Model for a Clinical Decision Support System by Deshmukh, Tijare and Sawalkar(2016) and A Privacy Preserving Data Classification Model by Desale and Javheri(2016)). In the course of reviewing the models, the flaws in each of the models were highlighted. Taking into consideration the flaws identified in the models, a model for data privacy preservation in an HIS was proposed. The proposed model consists of:
2. a local database that contains the health information of patients
3. the Random Forest Decision Tree (RFDT) algorithm

• an attribute blocking module that employs the RFDT algorithm

1. an attribute unblocking module which also uses the RFDT algorithm and
2. a module for the computation of time elapsed in unblocking attributes.
3. An application for data privacy preserving in a Health Information System named Schizoapp was built using the Waterfall Software Development Life Cycle Model and the following tools were employed:
4. Netbeans Integrated Development Environment (IDE)
5. MySQL server

• Java Development Kit (JDK) 8

2. Scenebuilder 2.0
3. Navicat 8 query editor
4. JavaFX
5. Using the Mandatory Role-Based Access Control, access to patient data as regards the three sensitive attributes of the eleven attributes in the dataset by the four categories of healthcare professionals considered in this study (doctors, nurses, psychologists and social workers) is restricted such that each category is only allowed to view the attribute(s) needed for them to provide the service needed by the patient.

The attribute(s) of the three which each category is not allowed to see is blocked in the database so that the health worker in that category can only see the other attributes in order to ensure the preservation of patient data privacy. Graphical User interfaces were generated to depict the view of each healthcare professional to patient data.

4. The machine learning approach to data privacy that involved the use of the Random Forest Decision Tree algorithm in the WEKA softwarefor DPP was compared with the application based approach which employed the proposed DPP application. Both approaches were evaluated for efficiency based on the quantum of time taken to unblock the attributes. Hence, the better approach for data privacy preservation is the one which took a longer time for the blocked attributes to be unblocked.

1.5 Justification forthe Study

This study will bring to the fore, the need for Psychiatric hospitals in Nigeria to adopt Electronic Health Records for patient data rather than the present method used by most of them which employs paper records for patient data. The study when implemented by the Psychiatric hospitals in Nigeria will help mitigate the intrusion of patient data privacy by restricting access to patient data only to the persons that are eligible to view such information by virtue of their role as healthcare professionals needed to keep the patient in a state of good health. By implementing the data privacy preserving model that was proposed in this study, the menace of schizophrenic patients in Nigeria being stigmatized based on their schizophrenic status would be mitigated to a reasonable degree.

1.6 Scope of the Study

The study focused on preserving the privacy of data belonging to some schizophrenic patients and some of the people that have been interrogated by psychiatrists at one time or the other to ascertain if they were schizophrenic or not. For the purpose of this study, two Psychiatric hospitals were visited to gather the data required for the study. The two hospitals were Federal Neuropsychiatric Hospital, Yaba, Lagos and Neuropsychiatric Hospital, Aro, Abeokuta. Two hundred and sixty three anonymous records of persons that have earlier visited Federal Neuropsychiatric Hospital, Yaba on account of showing symptoms suggestive of schizophrenia and Two hundred and forty eight anonymous records of persons that have visited Neuropsychiatric Hospital, Aro, Abeokuta earlier on account of being linked with schizophrenic symptom(s) were gotten, giving a total of five hundred and eleven records, which were used for the research. The study used five hundred and eleven records due to the fact that this was the number of records both Psychiatric Hospitals used in this study were willing to release.

1.7 Operational Definition of Terms

Health Information System:Thisrefers to any system that captures, stores, manages or transmits information related to the health of individuals or the activities of organizations that work within the health sector.
Model: This is a representation of an idea, an object or even a process or a system that is used to describe and explain phenomena that cannot be experienced directly.
Data Privacy: This deals with the ability an organization has to determine what data in a Health Information Systemcan be shared with health personnel.
Mandatory Access Control: Thisrefers to a type of access control by which the operating system constrains the ability of a subject or initiator to access or generally perform some sort of operation on an object or target.
Role Based Access Control: Thisis a policy neutral access control mechanism defined around roles and privileges used to restrict system access to authorized users.
Decision Tree is a flow-chart-like structure, where each internal (non-leaf) node denotes a test on an attribute, each branch represents the outcome of a test, and each leaf (or terminal) node holds a class label.
Random Forests are an ensemble learning method for classification, regression and other tasks, that operate by constructing a multitude of decision trees at training time and outputting the class that is the mode of the classes (classification) or mean prediction (regression) of theindividual trees.
Machine Learning is a type of artificial intelligence (AI) that provides computers with the ability to learn without being explicitly programmed.