Formalization and acquisition of temporal knowledge for decision support in medical processes
Introduction
Time is intrinsic to medical practice. Therefore, time must be part of any representation of medical practice knowledge or only instantaneous clinical decisions could be made. Long-term interventions and planning in medicine are common and they demand a description of time constraints about the clinical actions involved. These actions are correct not only if they are done, but also if they are done at and during the right time. The health conditions of patients start and hold during a set time or over a time period. Clinical interventions occur at one or more points in time or over periods of time [42]. However, this temporal dimension is not always available, or is just partially considered, in Clinical Practice Guidelines (CPGs). CPGs are conceived to support decision-making processes in health care, covering health maintenance, prevention, diagnosis, treatment, patient self-care, and education. They describe medical procedures as a narrative set of recommendations for the management of patients who have a particular disease. CPGs dealing with chronic disorders require the patient’s treatment to be continued over time, often for the remainder of the patient’s life [6]. Some CPGs can include certain time-related parameters, such as the recommended time between two consecutive encounters during dosage titration. Nevertheless, very often, times are unspecified and, consequently, physicians are supposed to fill these information gaps with their own knowledge and experience.
Several studies have shown that the quality of health care is directly related to the experience of the physicians involved [8], [24], [25], [32]. The argued rationale is that physicians with experience can reason better than less experienced or inexperienced physicians because they have a greater ability to combine different sorts of knowledge acquired from several sources and from their own professional education, training, and experience. Consequently, they use to make wiser decisions. However, even senior physicians may have serious difficulties, or they might be reluctant to ascertain general time constraints for some diseases or clinical actions. To our knowledge, currently, there are no mechanisms to help physicians obtain evidence-based knowledge about the time constraints that should be included in CPGs. However, health care centers use medical records and information systems to register data about clinical processes and patients, including information about the times in which the encounters took place, the prescriptions that were made, or when other clinical actions happened [3], [5], [25], [27], [29], [30], [47], [48]. Information overload is becoming an issue as patients’ data and medical information are being accumulated. We are heading towards an era of big data, where the issue is not only the amount of data available, but also the velocity of accumulation and sharing, as well as its variety. Many issues appear related to this data explosion, such as the complexity and heterogeneity of clinical data. The interest in discovering hidden information is emerging [28]. This information is not necessarily evidence-based, but gathered from the experience of daily medical practice. It can be recognized and transformed into medical knowledge with the help of computer intelligent data-analysis techniques and tools to support decision-making [3], [4], [5], [27], [29], [30], [47], [48], [49]. Consequently, medical records and health care information systems are promising sources of information to detect medical knowledge about time in medicine. When modelled, this experience-based knowledge about time can complement the evidence-based knowledge about clinical procedures contained in the CPGs, thus allowing Timed Medical Decision Support Systems’ (TMDSS) development and exploitation.
In [21], Extended Timed Transition Diagrams (eTTDs) were proposed as an intuitive, easy, and efficient mechanism to formalize knowledge about medical procedures. eTTDs are state-transition diagrams in which states describe clinical situations in terms of a set of state variables with a clinical sense. Transitions between pairs of eTTD states are possible. They are represented as state-to-state edges that are labelled with (1) Clinical conditions that the patients crossing the edge satisfy, (2) Medical activities or actions describing the treatment provided to these patients, and (3) A time interval constraint representing the time for them to evolve to the next state. An eTTD can have transitions showing clinical situations in which a patient’s condition remains the same following a clinical activity and time interval. Clinical actions in the transitions that correspond to pharmacological interventions can include a dose (e.g., 200 mg) and an intake frequency (e.g., every 8 hours). Both the dose and the intake frequency define the dosage of the pharmacological action.
In [21], eTTDs were used to represent knowledge about medical practice for the management of chronic patients at three levels of detail: Therapy strategy, dosage titration, and drug intolerance management. The modelling of these three levels of clinical knowledge was applied to offer help in decision-making processes for long-term treatments. The main limitation of that work was the lack of formalized representation for time constraints.
Here, in order to complement that work and to identify common time constraints in medical practice, we analyzed diagnostic and therapeutic medical procedures for 50 randomly chosen patients registered in the primary care information system of the SAGESSA Health Care Group1 (Spain) [41]. We also analyzed several CPGs about chronic diseases in order to detect the sorts of time constraints contained, and studied the time constructs of the standard data models EHRcom [11] and CDA HL7 [19], as well as several languages to computerize CPGs, to detect the sort of time constraints that they can represent. Based on this quadruple analysis, we synthesized two representation formalisms of time constraints in medicine: micro- and macro-temporality. Here, we also provide algorithms to generalize time constraints in the form of micro- and macro-temporalities from hospital datasets. When detected, constraints can be used to (1) complement evidence-based knowledge from CPGs with experience-based temporal knowledge obtained from data and to (2) study the adherence of physicians to the time constraints suggested in CPGs.
The rest of this paper is organized as follows. Section 2 deals with our proposal of time modelling for medical procedures. It begins with a classification of the time constraints for medical processes that we found after our analyses were conducted. It continues with a proposal on modelling the temporal dimension of medical procedures, and provides validation of the proposed temporal model. In Section 3, we propose computer algorithms for automatic generation of time constraints derived from data on medical processes. It also provides two types of tests and a discussion on the obtained results considering their medical correctness and adherence. Discussion and conclusions are presented in Section 4 and Section 5, respectively.
Section snippets
Time modelling of medical procedures
In order to formalize medical time constraints, we analyzed the sort of time indications that can be found in the medical processes registered in the medical records of several patients who were attended in the health care centers of the SAGESSA group. Moreover, we reviewed six clinical practice guidelines for three of the most prevalent cardiovascular diseases according to the World Health Organization (WHO) [50]. Finally, we also studied two of the most used standards for health care data
Detecting time constraints of medical procedures from medical records
Physicians use their medical knowledge to make diagnostic and therapeutic decisions. Medical procedures followed in health care centers are registered in databases. In 2016, Madkour et al. [28] argued that approximately 80% of saved data was unstructured and found in so-called clinical narratives. This constitutes a major challenge for information extraction where natural language processing emerges as a possible solution [28], [52], in combination with intelligent data analysis algorithms [4],
Discussion
In medicine, procedural knowledge permits a physician to determine the best course of action possible. This procedural knowledge can be extended with a temporal component and therefore become time-dependent. Time plays a major role in the clinical domain, as it helps us to understand the chronological sequence of medical procedures [28]. Physicians use their medical knowledge to make diagnostic and therapeutic decisions. Decision-making processes depend on whether clinicians can recall, find,
Conclusion
Our work showed that the lack of evidence about temporal indications in CPGs can be complemented with the use of computer technology for the intelligent analysis of the data about the clinical treatments registered in health care information systems. All this knowledge that comes from clinical experiences can be captured with two types of structures (micro- and macro-temporality) that have been shown not only able to describe all kind of temporal constraints in the management of cardiovascular
Conflict of interest statement
The authors declare that they have NO affiliations with or involvement in any organization or entity with any financial or non-financial interest in the subject matter or materials discussed in this manuscript.
Acknowledgments
The authors acknowledge financial support from the Slovenian Research Agency (Research Core Funding No. P2-0057), and the RETOS P-BreasTreat project (DPI2016-77415-R) of the Spanish Ministerio de Economia y Competitividad.
The authors acknowledge Suzana Kert, MD and Antoni Collado, MD, for providing the medical support that was essential for this research.
The authors appreciate reviewers’ valuable comments which improved the clarity and quality of the paper.
Aida Kamišalić is a Teaching Assistant and Researcher at the Faculty of Electrical Engineering and Computer Science at the University of Maribor. She has held a Ph.D. in Computer Science since 2014. Her research interests include database technologies, medical procedure modelling for chronic patients and time knowledge acquisition for medical procedures.
References (52)
- et al.
Temporal detection and analysis of guideline interactions
Artif. Intell. Med.
(2017) - et al.
Different medical data mining approached based prediction of ischemic stroke
Comput. Methods Prog. Biomed.
(2016) - et al.
Automatic generation of clinical algorithms within the state-decision-action model
Expert Syst. Appl.
(2012) - et al.
Supporting a distributed execution of clinical guidelines
Comput. Methods Prog. Biomed.
(2013) - et al.
GLIF3: a representation format for sharable computer-interpretable clinical practice guidelines
J. Biomed. Inf.
(2004) - et al.
Disseminating medical knowledge: the PROforma approach
Artif. Intell. Med.
(1998) - et al.
Rationale for the Arden syntax
Comput. Biomed. Res.
(1994) - et al.
Bayesian networks in biomedicine and health-care
Artif. Intell. Med.
(2004) - et al.
Temporal data representation, normalization, extraction, and reasoning: a review from clinical domain
Comput. Methods Prog. Biomed.
(2016) - et al.
Learning causal and predictive clinical practice guidelines from data
Stud. Health Technol. Inform.
(2007)
The PRODIGY project–the iterative development of the release one model
AMIA Symposium Proceedings
Computer technologies to integrate medical treatments to manage multimorbidity
J. Biomed. Inform.
The SDA model: a set theory approach
Proceedings of the Twentieth IEEE International Symposium on Computer-Based Medical Systems
Workflow mining: a survey of issues and approaches
Data Knowl. Eng.
Biomedical time series clustering based on non-negative sparse coding and probabilistic topic model
Comput. Methods Prog. Biomed.
Estimating a reasonable patient panel size for primary care physicians with team-based task delegation
Ann. Fam. Med.
Family medicine panel size with care teams: impact on quality
J. Am. Board Fam. Med.
Effectiveness and Efficiency: Random Reflections on Health Services
GASTON: an architecture for the acquisition and execution of clinical guideline-application tasks
Med. Inf. Internet Med.
ESC Guidelines for the diagnosis and treatment of acute and chronic heart failure 2008
Eur. Heart J.
ESC Guidelines for the diagnosis and treatment of acute and chronic heart failure 2012
Eur. Heart J.
ESC Guidelines on the management of stable angina pectoris
Eur. Heart J.
ESC Guidelines on the management of stable coronary artery disease
Eur. Heart J.
Cited by (8)
Temporal reasoning and query answering with preferences and probabilities for medical decision support
2022, Expert Systems with ApplicationsCitation Excerpt :Therefore, they are explicitly recorded in many CPGs and CIGs. Notably, Kamisalic et al. have proposed an interesting approach to mine CPG temporal constraints from medical data, and to represent them (Kamisalic et al., 2018). This contribution is very important, since it may be used as a starting point to derive temporal constraints from experience, for those CPGs which do not natively contain them.
Multi-level medical knowledge formalization to support medical practice for chronic diseases
2019, Data and Knowledge EngineeringCitation Excerpt :One of the main drawbacks of the obtained AH models is that they offer limited support for temporal information. Although eTTDs are capable of representing temporal knowledge in a natural way [26], we were unable to find this sort of information for AH. Not considering the temporal information implies that all the clinical decisions have to be interpreted instantaneously.
AS-SIM: An Approach to Action-State Process Model Discovery
2022, Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Aida Kamišalić is a Teaching Assistant and Researcher at the Faculty of Electrical Engineering and Computer Science at the University of Maribor. She has held a Ph.D. in Computer Science since 2014. Her research interests include database technologies, medical procedure modelling for chronic patients and time knowledge acquisition for medical procedures.
David Riaño is an Associate Professor at the Department of Computer Science and Mathematics at Universitat Rovira i Virgili. He is the head of the Research Group on Artificial Intelligence at that university. He received his Ph.D. in Artificial Intelligence in 1997. His research interests include the knowledge-based modelling of medical procedures for chronic and multimorbid patients, the development of machine learning technologies to discover and represent clinical procedures from raw data, and the simulation of patient evolutions with knowledge-based systems.
Tatjana Welzer is a Full Professor at the University of Maribor, Faculty of Electrical Engineering and Computer Science. She is the Head of the Database Technologies Laboratory of the Institute of Informatics. She received her Ph.D. in Computer Science in 1995. Her research work covers areas of database technologies, cross-cultural communication and problems in media communications.