International Journal of Information Science and Intelligent System, 3(3): 51-60, 2014 Modeling Temporal Data in Electronic Health Record Systems Chafiqa Radjai 1, Idir Rassoul², Vytautas Čyras 3 1,2 Mouloud Mammeri University, LARI laboratory, Tizi Ouzou, Algeria 3 Vilnius University, Vilnius, Lithuania Abstract Received: 12 April 2014; Accepted: 10 June 2014 One of the prevalent applications implicated temporal information is Electronic Health Record (EHR) systems. The representation of temporal aspects offers a particular challenge. We need to define a good set of features that are able to represent well the different temporal aspects required in EHR, where data are temporal in nature. This paper focuses on the identification and modeling of temporal aspects in EHR based on pre-existing model. Keywords: Temporal Aspects; Electronic Health Record; Modeling; Unification Martin Science Publishing. All Rights Reserved. 1. Introduction The issues of time and change are essential to reasoning in many applications. The key challenge to develop a temporal conceptual model is how to support various temporal aspects. There exist various temporal phenomena, such us diagnostic applications, planning, geographic information systems, computer-aided design, computer-aided manufacturing, and computer-aided construction. The automation of these areas means that we can represent different temporal aspects of the information. The temporal dimension in the medical field is increasingly important. EHR systems require both past and current data. Each record in the data, collected for a specific patient, consists of Chafiqa Radjai. E-mail address: radjai.chafika@gmail.com
52 Chafiqa R / International Journal of Information Science and Intelligent System (2014) laboratory test results, medication orders and physiological parameters. The record may also provide information about the patient s deceases, surgical intervention and their outcomes. These data sets can change over time and may be linked with other data sets to obtain further information about the patient. This dimension of time is of major importance in the documentation and analysis of health data. For example, changes to skin coloration can indicate a serious medical problem that requires various medical investigations. For example, yellowing of the skin and eyes can indicate pancreatic cancer. We need a detailed record of patient s health status; the treatments applied to him during a specific stay at the hospital and the locations of the hospital where the patient assigned for treatment? This is necessary for medical, billing, legal reasons, search tasks, liability, statistical purposes, medical decision-making and patient privacy protection. In order to access and use these temporal medical data, the dimension of time should be taken into account in the development of EHR architectures. We should be able to manage the temporal semantics of medical data. Medical applications (like clinicians) must detect different temporal aspects in patient data (e.g., at 15:30 January 21, 2005, a Patient accepted an episode of heart disease that lasted 2 weeks). Temporality is a rather complicated factor to measure in medical information dynamics. With few exceptions, patient databases models do not capture such temporal features. The important question is how to take into account and model the different temporal aspects of medical data and their evolution? The process of developing a conceptual temporal model for identifying temporal aspects that potentially characterize the medical field includes two phases: Develop a conceptual temporal model that supports several temporal aspects of information system. Use the model developed to detect any temporal aspects that appear in medical in EHR. This paper presents a new model to formalize a temporal aspect of a system. The temporal model is presented in previous work and this paper aims at applying it in an e-health case. The conceptual model proposed is based on an object oriented paradigm. The rest of the paper is organized as follows: Section 2 introduces the temporal primitives needed in EHR. We first present a motivating example, a real clinical record of patient. Next, we give an overview of our system including how medical ontology is integrated into our system. The overview section is followed by detailed description of temporal modeling approach. Finally, conclusion is given in section 5. 2. Time in E-health This section describes scenarios where health history is used and enables implementers to understand better the clinical workflow in which time is integrated for patient care. The objective of this section is to illustrate the way a patient's health record is used.
Chafiqa R / International Journal of Information Science and Intelligent System (2014) 53 When a patient is admitted for treatment the patient s health habits and past illnesses and treatments will be taken to detect (find) and diagnose the disease. Patient will be assigned to nurses, locations, and consulting doctors. These assignments may change and need to be recorded during treatment as the patient is relocated. The patient s record includes complete information about problems and symptoms over time and the related tests and treatments done by the entire medical employees in the different hospital locations. Patients may return to the hospital and we need to relate new treatments to past treatments. A patient may discontinue treatment at any time for any reason this information will be on record. 3. Related Work Time is significant in EHR data modeling. Representing, maintaining, querying, and reasoning about time-oriented clinical data are critical success factor. This section gives an overview of some important approaches. A middle layer framework [2] is proposed to allow the representation and the storage of different temporal objects and temporal relationships among them. The framework uses object versioning to manage different versions of patient medical data. Each temporal object version is assigned to a timestamp. The authors explained how to add metadata elements to the representation of sub-mappings in object versioning. Such metadata include information about why the mapping is needed (the cause of data changes) and how the mapping is (was) generated. This metadata can help in interpretation of data. A different approach for incorporating time dimension in EHR is presented in [3]. The authors investigated three different EHR approaches in terms of temporality: CEN, HL7 and openehr. Furthermore, they analyzed how temporal structures such as time series can be developed from the basic data types. They conclude that all the three EHR can have temporality components incorporated in them. The temporality incorporation can be achieved using either generic components or predefined specific structures. HMAP [4] is a temporal data model managing intervals with different granularities and indeterminacy from natural language sentences. In HMAP, absolute intervals are explicitly represented by their start, end, and duration: in this way, we can represent valid times as in December 2011 for two days, from May 1999, for three weeks, from March 2003 to October 15, 2005, between 9 and 11:30 p.m.. HMAP is based on a three-valued logic, for managing uncertainty in temporal relationships. Formulas involving different temporal relationships between intervals, instants, and durations can be defined, allowing one to query the database with different granularities, not necessarily related to that of data. HMAP fails in representing sentences referring to the relative time and cannot represent different semantics of temporal information. The work proposed in [5] focus on the visualization of typical patient data at different temporal granularities on Mobile. The system uses a temporal representation at multiple granularities levels (minutes, hours, days, weeks, months). The visualizations provide overviews of four types of considered data (temperature, blood pressure, medicines or events) in
54 Chafiqa R / International Journal of Information Science and Intelligent System (2014) a single screen to give the user the possibility of visually relating them. However, the system needs to add additional visualization that could be useful. In addition, more evaluation of the system strengths and weaknesses in clinical and homecare contexts is needed. The model NLTM (Temporal Model of Natural Language) [6] is proposed for handling uncertain temporal medical information. The timestamp types for representing NLTM are Time point, Uncertain time point, Interval, Uncertain interval, Durations, Uncertain temporal element. Moreover, NLTM divided temporal attributes into two parts: date elements (year, month and the day) and time-of-day (hour, minute, second and so on). These temporal attributes can be certain or uncertain. The model defines several notations for representing different temporal expressions. Besides, it establishes the predicates and functions of temporal primitives to deal with all kinds of queries. EHR contains temporal data as patient history. However, it is not always possible to describe all the semantics of temporal medical data in precisely since the observation and capturing of some data are not perfect; hence its modeling and representation are deficient. 4. Temporal Medical Information in EHR In our special case, we deal with information related to patient and his medical care team. Time-oriented patient data include symptoms, pathologies, measured parameters and therapies. In addition to these medical data, the patient care team is an important factor involved in handling this information. 4.1 Dynamic life Span of Patient Care Team The temporal dimension becomes another concern that adds to determine patient care team members. The patient care team consists of nurses, doctors, technicians and other professionals may change when new progress, new orders or new situations happen. Each care team member has its own functions and responsibilities. This group participates in the care of a defined group of patients. The duration of patient care teams changes severely depending on each patient situation. Identifying the dynamic lifespan of teams and teams members is essential to protect, control patient privacy and for legal reasons uses. This scenario here is a typical patient treatment drawn from field observations conducted in the ED [7]. When a patient walks into the hospital, a receptionist or nurse asks for the chief compliant and checks the patient s vital signs. After, a triage nurse assigns the patient to a specific room according to the severity of the situation and doctor s expertise. The room nurse keeps an eye on the patient situation. Then, she receives doctor s orders and works on them. Doctor sees patients in their rooms, and prescribes medications or lab orders to nurses and/or technicians. Lastly, after the patient s situation is stabilized in the ED, a case manager will start preparing the transfer or discharge paperwork for the patient. 4.2 Patient Medical Information Record
Chafiqa R / International Journal of Information Science and Intelligent System (2014) 55 Patient s problems denote a change in state. These problems occur at some time points (e.g., blood in the urine in July 2009) and can hold during a period of time (e.g., headache, fever). These problems require diagnosis. The diagnosis requires medical history of the patient and different measurements (physical examination, laboratory testing) to identify the cause of the problem, and finally to generate a treatment. The different steps of treatment must be applied in a precise order, with a given frequency and for a certain span of time in order to be effective. Let us consider the following examples: The patient had a severe headache on February 10, 1993, from 2 2.30 p.m. to 6 6.15 p.m., lasting between 3 hours 35 minutes and 3 hours 50 minutes ; On December 14, 2004, between 2 and 2.15 p.m., the physician measured the bloodpressure of the patient: it was 120/80 mmhg ; In 1993, an anticoagulation-therapy was administered to the patient for 60 65 days ; The patient had aphasia from 6.30 p.m. to 9 p.m., May 13,1991 ; We need to represent these temporal semantics of medical data. The following sections discuss issues for defining temporal features in a conceptual model, the abstract data types supporting time description, representing and reasoning about the temporal nature of the clinical environments. 5. Demonstration In this section, we present the temporal conceptual data model we propose for the representation of temporal aspects. Our proposed model follows the object-oriented programming standard. Our demonstration will use this model to create a temporal conceptual model for a hypotetical medical application. We will show how to quickly create the medical application diagram using UML and represent different temporal aspects. In the demonstration, we will discuss the modeling process. 5.1 Modeling Temporal Aspects The proposed model [1] is designed to meet the requirements of modeling time. The methodology adopted to develop the model is as followed: 1) Proposing a model of time; 2) İdentifying and modeling the temporal object that will be used as basis for modeling IS evolution; 3) Presenting a conceptual schema that will be usefull for modeling versions of Information System on the axis of time.
56 Chafiqa R / International Journal of Information Science and Intelligent System (2014) Its most important features are the following: 1) It is a conceptual data model, focusing on taking into account the needs of modeling temporal aspects of real-world applications ; 2) The model is a result of a combination of several models that provide a good representation of time and the evolution of information system. 3 ) The model is able to cover both the validity time and the transaction time in a flexible manner. It provides the necessary primitives: time instants, time intervals and time elements. Each time primitive can be either absolute or relative, on the one hand and determinate or indeterminate, on the other hand. The model supports the definition of specific additional application calendars with their respective granularities. The models supports all notions of time that can be considered relevant to the current practical applications. (1) Modeling time. The model of time includes multiple temporal dimensions. Figure 1 shows that a class Time may have different types: instant (class Instant), interval (class Interval), set of instants (class Set_Of_Instants), set of intervals (class Set_Of_Intervals) or period (class Period). An instant can be relative (class Instant_R) or absolute (class Instant_A). Instant and Period classes are expressed in a given calendar (Class Calendar) at a precise granularity. Class Instant is fully described by the five attributes: granularity, position, duration, uncertainty and isshifted. These attributes are fully detailed in [8]. Figure 1. Model of time
Chafiqa R / International Journal of Information Science and Intelligent System (2014) 57 (2)Modeling temporal objects and associations. For modeling temporal aspects of objects and associations, the stereotype called <<Temporal>> is proposed. It contains the tag LifeSpan that has boolean type. It is used to denote the different temporal classification of a class (Active, Suspended, scheduled, Disabled). If this aspect should be captured, we attribute to its value True. Temporal attributes are marked by the stereotype <<Timestamp>>. It contains two tags: TransactionT and ValidTime which have Boolean type. If the database designer decides to capture the time of validity of an attribute, the ValidTime tag has a value True. If the transaction time is captured, the TransactionT tag has a value True. If both are captured, the two tags have True values. In addition, we attach a constraint to the stereotype that ensures for valid time, if it is a set of intervals, they cannot overlap while the case if it is a set of instants, they must be distinct and as there is no future transaction time. Figure 2 shows the stereotypes proposed for modeling temporal objects and associations. Figure 2. The stereotypes <<Temporal >> and <<Timestamp>>. To store the history of a class, we must register all temporal classification of each instance of the class. When a given instance is declassified from a class we must update its timestamp lifespan attribute instead of physically deleting it. 5.2 Modeling Patient Care Team The aim of this section is to represent time aspects of health care organization. In this section, we describe the architecture of the medical care system that we are developing for efficient exploitation of our unified temporal data model. A patient care team is assigned to a patient through temporal associations. Figure 3 shows associations between temporal classes: Doctor, Nurse, Technician and class Patient that describes the corresponding assignments. Nurse specialty can be room nurse, case manager or receptionist. Each medical employee often participates in multiple care teams. Time information is crucial. Almost every team care member must contain information of time because without time it is impossible that which person serve the patient for which specific interval. Thus, patient care team generally does not have the same doctor, nurse or technician.
58 Chafiqa R / International Journal of Information Science and Intelligent System (2014) Figure 3. Modeling patient care team Figure 3 describes also the hierarchical arrangement of a Medical Group. Medical group includes hospitals. Each hospital includes some Buildings. Each treatment Location is part of a building. Class Location assigned to an inpatient, nurses or technicians include the patient s room, laboratories or other places for specific treatments. 5.3 Modeling Patient Medical Information Medical treatment of a patient should contain information about patient (diagnosis, planned treatment, condition, knowledge, medication, experiences, examination results and interval). Figure 4 describes the modeling process of patient medical information. At a given instant, a patient may have a single problem for example a headache or fever. However, when we store the whole history, a patient may have had several problems along his life span. These problems require a diagnosis.
Chafiqa R / International Journal of Information Science and Intelligent System (2014) 59 Figure 4. Modeling patient medical history A diagnosis includes of several measurements. For example, a blood pressure measurement is described using the class Measurement. A measurement instance may have had a single value in an instant t but it may have had different values along his lifespan. All instances of class Measurement are associated with temporal dimension. The treatment of patient s problem involves two approaches: (1) medications (such as insulin therapy, pharmacotherapy, and chemotherapy), (2) procedures (such as surgery, physical therapy, massage therapy, radiotherapy). We use the stereotype <<Timestamp>> temporal attributes e.g. the doze of medications is temporal and may change according to the patient status. The treatment of patient is temporal. It may be changed, paused or ended. The patient may be treated with different treatments. For example, in case of breast cancer, a patient starts to be treated by a chemotherapy treatment for a certain period of time after that the patient will be treated by radiotherapy treatment. 6. Conclusion In this paper, we presented our formalism which could provide powerful temporal modeling as needed in the electronic health records systems. It facilitates the consistent representation of temporal aspects in those data, thus allowing performing temporal reasoning and temporal queries on these data.
60 Chafiqa R / International Journal of Information Science and Intelligent System (2014) The model permits to define and control the temporal dimension of patient care teams and their members. In addition, we can refer to any past state of patient s problems, diagnosis and treatments. We have very favorable evaluation results, that our approach truly meets the desired requirements. Although the research has made a progress, much work still to be done. Efficient implementations will be carried out in further research. Acknowledgments This work has been done during my doctoral mobility stay at the faculty of Mathematics and Informatics of Vilnius University under European Commission ERASMUS MUNDUS program. References [1] C. Radjai, and I. Rassoul, Towards the Unification of Modeling Temporal Aspects of Information Systems, Journal of Information Technology Review, Vol. 4, no. 1, pp. 37-45, 2013. [2] T.Mallaug, and K. Bratbergsengen, K, Integrated Electronic Health Record Access by Object Object Versioning and Metadata, ADBIS Research Communications, 2006. [3] W. Gall, G. Duftschmid, and W. Dorda, Temporal Components in Architectures of Electronic Health Records, Tagungsband der, Vol. 49. [4] C. Combi, and G. Pozzi, HMAP A temporal data model managing intervals with different granularities and indeterminacy from natural language sentences, The VLDB Journal, Vol. 9, no 4, pp. 294-311, 2001. [5] L. Chittaro, Visualization of patient data at different temporal granularities on mobile devices, Proceedings of the working conference on advanced visual interfaces, ACM, pp. 484-487, 2006. [6] X. ZHANG, A Temporal Data Model for Handling Uncertain Temporal Medical information, Journal of Computational Information Systems, Vol. 8, no. 10, pp. 3971-3978, 2012. [7] Y. Chen, and H. Xu, Privacy management in dynamic groups: understanding information privacy in medical practices, Proceedings of the 2013 conference on Computer supported cooperative work, ACM, pp. 541-552, 2013. [8] J. Benzler, and S. J. Clark, Toward a Unified Timestamp with explicit precision, Demographic research, Vol. 12, no. 6, pp. 107, 2005.