Explainable process trace classification: An application to stroke.

Medical process trace classification exploits the activity sequences logged by an healthcare organization to classify traces themselves on the basis of some performance properties; this information can be used for quality assessment. State-of-the-art process trace classification resorts to deep learning, a very powerful technique which however suffers from the lack of explainability. In this paper we aim at addressing this issue, motivated by a relevant application, i.e., the classification of process traces for quality assessment in stroke management. To this end we introduce the novel concept of trace saliency maps, an instrument able to highlight what trace activities are particularly significant for the classification task. Through trace saliency maps we justify the output of the deep learning architecture, and make it more easily interpretable to medical users. The good results in our use case have shown the feasibility of the approach, and let us make the hypothesis that it might be translated to other application settings and to other black box learners as well.

Leveraging semantic labels for multi-level abstraction in medical process mining and trace comparison.

Many medical information systems record data about the executed process instances in the form of an event log. In this paper, we present a framework, able to convert actions in the event log into higher level concepts, at different levels of abstraction, on the basis of domain knowledge. Abstracted traces are then provided as an input to trace comparison and semantic process discovery. Our abstraction mechanism is able to manage non trivial situations, such as interleaved actions or delays between two actions that abstract to the same concept. Trace comparison resorts to a similarity metric able to take into account abstraction phase penalties, and to deal with quantitative and qualitative temporal constraints in abstracted traces. As for process discovery, we rely on classical algorithms embedded in the framework ProM, made semantic by the capability of abstracting the actions on the basis of their conceptual meaning. The approach has been tested in stroke care, where we adopted abstraction and trace comparison to cluster event logs of different stroke units, to highlight (in)correct behavior, abstracting from details. We also provide process discovery results, showing how the abstraction mechanism allows to obtain stroke process models more easily interpretable by neurologists.

Managing proposals and evaluations of updates to medical knowledge: theory and applications.

The process of keeping up-to-date the medical knowledge stored in relational databases is of paramount importance. Since quality and reliability of medical knowledge are essential, in many cases physicians' proposals of updates must undergo experts' evaluation before possibly becoming effective. However, until now no theoretical framework has been provided in order to cope with this phenomenon in a principled and non-ad hoc way. Indeed, such a framework is important not only in the medical domain, but in all Wikipedia-like contexts in which evaluation of update proposals is required. In this paper we propose GPVM (General Proposal Vetting Model), a general model to cope with update proposal⧹evaluation in relational databases. GPVM extends the current theory of temporal relational databases and, in particular, BCDM - Bitemporal Conceptual Data Model - "consensus" model, providing a new data model, new operations to propose and accept⧹reject updates, and new algebraic operators to query proposals. The properties of GPVM are also studied. In particular, GPVM is a consistent extension of BCDM and it is reducible to it. These properties ensure consistency with most relational temporal database frameworks, facilitating implementation on top of current frameworks and interoperability with previous approaches.

NRTS: A Client-Server Architecture for Supporting Education in a Neonatal Resuscitation Simulation Scenario.

In this paper, we describe Neonatal Resuscitation Training Simulator (NRTS), an Android mobile app designed to support medical experts to input, transmit and record data during a High-Fidelity Simulation course for neonatal resuscitation. This mobile app allows one to automatically send all the recorded data from the Neonatal Intensive Care Unit (NICU) of Casale Monferrato Children's Hospital, (Italy) to a server in the cloud managed by the University of Piemonte Orientale (Italy). The medical instructor can then view statistics on simulation exercises, that may be used during the debriefing phase for the evaluation of multidisciplinary teams involved in the simulation scenarios.

High-Fidelity simulation-based program improves flow state scale in the perinatal team.

BACKGROUND: We aimed to evaluate the degree of realism and involvement, stress management and awareness of performance improvement in practitioners taking part in high fidelity simulation (HFS) training program for delivery room (DR) management, by means of a self-report test such as flow state scale (FSS). METHODS: This is an observational pretest-test study. Between March 2016 and May 2019, fourty-three practitioners (physicians, midwives, nurses) grouped in multidisciplinary teams were admitted to our training High Fidelity Simulation center. In a time-period of 1 month, practitioners attended two HFS courses (model 1, 2) focusing on DR management and resuscitation maneuvers. FSS test was administred at the end of M1 and M2 course, respectively. RESULTS: FSS scale items such as unambiguous feed-back, loss of self consciousness and loss of time reality, merging of action and awareness significantly improved (P < 0.05, for all) between M1 and M2. CONCLUSIONS: The present results showing the high level of practitioner involvement during DR management-based HFS courses support the usefulness of HFS as a trustworthy tool for improving the awareness of practitioner performances and feed-back. The data open the way to the usefulness of FSS as a trustworthy tool for the evaluation of the efficacy of training programs in a multidisciplinary team.

Supporting human interaction and human resources coordination in distributed clinical guidelines.

Clinical guidelines (GL) play an important role in medical practice: the one of optimizing the quality of patient care on the basis of the best and most recent evidence based medicine. In order to achieve this goal, the interaction between different actors, who cooperate in the execution of the same GL, is a crucial issue. As a matter of fact, in many cases (e.g. in chronic disease treatment) the GL execution requires that patient treatment is not performed/completed in the hospital, but is continued in different contexts (e.g. at home, or in the general practitioner's ambulatory), under the responsibility of different actors. In this situation, the correct interaction and communication between the actors themselves is critical for the quality of care, and human resources coordination is a key issue to be addressed by the managers of the involved healthcare service. In this paper we describe how computerized GL management can be extended in order to support such needs, and we illustrate our approach by means of a practical case study.

Artificial Intelligence in Clinical Decision Support: a Focused Literature Survey.

OBJECTIVES: This survey analyses the latest literature contributions to clinical decision support systems (DSSs) on a two-year period (2017-2018), focusing on the approaches that adopt Artificial Intelligence (AI) techniques in a broad sense. The goal is to analyse the distribution of data-driven AI approaches with respect to "classical" knowledge-based ones, and to consider the issues raised and their possible solutions. METHODS: We included PubMed and Web of ScienceTM publications, focusing on contributions describing clinical DSSs that adopted one or more AI methodologies. RESULTS: We selected 75 papers, 49 of which describe approaches in the data-driven AI area, 20 present purely knowledge-based DSSs, and 6 adopt hybrid approaches relying on both formalized knowledge and data. CONCLUSIONS: Recent studies in the clinical DSS area demonstrate a prevalence of data-driven AI, which can be adopted autonomously in purely data-driven systems, or in cooperation with domain knowledge in hybrid systems. Such hybrid approaches, able to conjugate all available knowledge sources through proper knowledge integration steps, represent an interesting example of synergy between the two AI categories. This synergy can lead to the resolution of some existing issues, such as the need for transparency and explainability, nowadays recognized as central themes to be addressed by both AI and medical informatics research.

Planning: supporting and optimizing clinical guidelines execution.

A crucial feature of computerized clinical guidelines (CGs) lies in the fact that they may be used not only as conventional documents (as if they were just free text) describing general procedures that users have to follow. In fact, thanks to a description of their actions and control flow in some semiformal representation language, CGs can also take advantage of Computer Science methods and Information Technology infrastructures and techniques, to become executable documents, in the sense that they may support clinical decision making and clinical procedures execution. In order to reach this goal, some advanced planning techniques, originally developed within the Artificial Intelligence (AI) community, may be (at least partially) resorted too, after a proper adaptation to the specific CG needs has been carried out.

Applying artificial intelligence to clinical guidelines: the GLARE approach.

We present GLARE, a domain-independent system for acquiring, representing and executing clinical guidelines (GL). GLARE is characterized by the adoption of Artificial Intelligence (AI) techniques in the definition and implementation of the system. First of all, a high-level and user-friendly knowledge representation language has been designed. Second, a user-friendly acquisition tool, which provides expert physicians with various forms of help, has been implemented. Third, a tool for executing GL on a specific patient has been made available. At all the levels above, advanced AI techniques have been exploited, in order to enhance flexibility and user-friendliness and to provide decision support. Specifically, this chapter focuses on the methods we have developed in order to cope with (i) automatic resource-based adaptation of GL, (ii) representation and reasoning about temporal constraints in GL, (iii) decision making support, and (iv) model-based verification. We stress that, although we have devised such techniques within the GLARE project, they are mostly system-independent, so that they might be applied to other guideline management systems.

Supporting therapy selection in computerized clinical guidelines by means of decision theory.

Supporting therapy selection is a fundamental task for a system for the computerized management of clinical guidelines (GL). The goal is particularly critical when no alternative is really better than the others, from a strictly clinical viewpoint. In these cases, decision theory appears to be a very suitable means to provide advice. In this paper, we describe how algorithms for calculating utility, and for evaluating the optimal policy, can be exploited to fit the GL management context.

Automatic treatment of temporal issues in clinical guidelines in the GLARE system.

Temporal constraints play a fundamental role in clinical guidelines. For example, temporal indeterminacy, constraints about duration, delays between actions and periodic repetitions of actions are essential in order to cope with clinical therapies. This paper proposes a computer-based approach in order to deal with temporal constraints in clinical guidelines. Specifically, it provides the possibility to represent such constraints and reason with them (i.e., perform inferences in the form of constraint propagation). We first propose a temporal representation formalism and two constraint propagation algorithms operating on it, and then we show how they can be exploited in order to provide clinical guideline systems with different temporal facilities. Our approach offers several advantages: for example, during the guideline acquisition phase, it enables to represent temporal constraints and to check their consistency; during the execution phase, it allows the physician to check the consistency between action execution-times and the constraints in the guidelines, and to provide query answering and temporal simulation facilities (e.g., when choosing among alternative paths in a guideline).

Automatic checking of the correctness of clinical guidelines in GLARE.

Representing clinical guidelines is a very complex knowledge-representation task, requiring a lot of expertise and efforts. Nevertheless, guideline representations often contain several kinds of errors. Therefore, checking the well-formedness and correctness of a guideline representation is an important task, which can be drastically improved with the adoption of computer programs. In this paper, we discuss the advanced facilities provided by the GLARE system to assist physicians to produce correct representations of clinical guidelines.

Towards a comprehensive treatment of repetitions, periodicity and temporal constraints in clinical guidelines.

OBJECTIVE: In this paper, we define a principled approach to represent temporal constraints in clinical guidelines and to reason (i.e., perform inferences in the form of constraint propagation) on them. We consider different types of constraints, including composite and repeated actions, and propose different types of temporal functionalities (e.g., temporal consistency checking). BACKGROUND: Constraints about actions, durations, delays and periodic repetitions of actions are an intrinsic part of most clinical guidelines. Although several approaches provide expressive temporal formalisms, only few of them deal with the related temporal reasoning issues. METHODOLOGY: We first propose a temporal representation formalism and two temporal reasoning algorithms. Then, we consider the trade-off between the expressiveness of the formalism and the computational complexity of the algorithms, in order to devise a correct, complete and tractable approach. Finally, we show how the algorithms can be exploited to provide clinical guideline systems with different types of temporal facilities. RESULTS: Our approach offers several advantages. During the guideline acquisition phase, it enables to represent temporal constraints, and to check their consistency. In the execution phase, it checks the consistency between the execution times of the actions and the constraints in the guidelines, and provides query answering and simulation facilities.

Case-based retrieval to support the treatment of end stage renal failure patients.

OBJECTIVE: In the present paper, we describe an application of case-based retrieval to the domain of end stage renal failure patients, treated with hemodialysis. MATERIALS AND METHODS: Defining a dialysis session as a case, retrieval of past similar cases has to operate both on static and on dynamic features, since most of the monitoring variables of a dialysis session are time series. Retrieval is then articulated as a two-step procedure: (1) classification, based on static features and (2) intra-class retrieval, in which dynamic features are considered. As regards step (2), we concentrate on a classical dimensionality reduction technique for time series allowing for efficient indexing, namely discrete Fourier transform (DFT). Thanks to specific index structures (i.e. k -d trees), range queries (on local feature similarity) can be efficiently performed on our case base, allowing the physician to examine the most similar stored dialysis sessions with respect to the current one. RESULTS: The retrieval tool has been positively tested on real patients' data, coming from the nephrology and dialysis unit of the Vigevano hospital, in Italy. CONCLUSIONS: The overall system can be seen as a means for supporting quality assessment of the hemodialysis service, providing a useful input from the knowledge management perspective.

Integrating model-based decision support in a multi-modal reasoning system for managing type 1 diabetic patients.

We present a multi-modal reasoning (MMR) methodology that integrates case-based reasoning (CBR), rule-based reasoning (RBR) and model-based reasoning (MBR), meant to provide physicians with a reliable decision support tool in the context of type 1 diabetes mellitus management. In particular, we have implemented a decision support system that is able to jointly exploit a probabilistic model of the glucose-insulin system at the steady state, a RBR system for suggestion generation and a CBR system for patient's profiling. The integration of the CBR, RBR and MBR paradigms allows for an optimized exploitation of all the available information, and for the definition of a therapy properly tailored to the patient's needs, overcoming the single approaches limitations. The system has been tested both on simulated and on real patients' data.

Supporting decisions in medical applications: the knowledge management perspective.

In the medical domain, different knowledge types are typically available. Operative knowledge, collected during every day practice, and reporting expert's skills, is stored in the hospital information system (HIS). On the other hand, well-assessed, formalised medical knowledge is reported in textbooks and clinical guidelines. We claim that all this heterogeneous information should be secured and distributed, and made available to physicians in the right form, at the right time, in order to support decision making: in our view, therefore, a decision support system cannot be conceived as an independent tool, able to substitute the human expert on demand, but should be integrated with the knowledge management (KM) task. From the methodological viewpoint, case based reasoning (CBR) has proved to be a very well suited reasoning paradigm for managing knowledge of the operative type. On the other hand, rule based reasoning (RBR) is historically one of the most successful approaches to deal with formalised knowledge. To take advantage of all the available knowledge types, we propose a multi modal reasoning (MMR) methodology, that integrates CBR and RBR, for supporting context detection, information retrieval and decision support. Our methodology has been successfully tested on an application in the field of diabetic patients management.

Telemedicine in the management of young patients with type 1 diabetes mellitus: a follow-up study.

DCCT (Diabetes Control and Complications Trial) study showed that tight metabolic control of diabetes mellitus can delay the onset and/or reduce the frequency of vascular complications. Telemedicine, i.e. telecommunications and information technologies in health care, is a useful tool to achieve the DCCT goals. Our European Community (EC) sponsored Telematic management of Insulin-Dependent Diabetes Mellitus (T-IDDM) project implements a telemedicine service through on a careful analysis of current medical practice. The system is based on two components: Patient Unit (PU) and Medical Unit (MU) connected by a Telecommunication system (TS). PU allows data collection and transmission from the patient's house to the hospital, assists self-monitoring activity and suggests insulin variations. PU communicates patient's current metabolic state the MU. MU assists the physician in periodic evaluation and suggests the prescriptions to communicate back defining a treatment protocol. TS system is based on telephone lines, relying on the Intranet technology. To test the system functionality and potential impact in type 1 diabetes clinical practice, we enrolled 6 patients (4 males and 2 females), aged 9.9-15.8 yrs, with disease duration 2.1-6.4 yrs, intensively treated. One girl run out after a 1-year follow-up HbA1c levels decreased, but not significantly. Insulin requirement reduced, significantly in 2 patients (p = 0.02 and p = 0.07). A positive correlation was between number of links and protocol changes (p = 0.01), between number of protocols changes and HbA1c decrease (p = 0.02). In pediatric patients periodical visits are necessary, but T-IDDM enables continuity of care improving access and activities. An index is represented by the high number of messages between the 2 Units, seeming weekly exchange.

The GLARE approach to clinical guidelines: main features.

In this paper, we present GLARE, a domain-independent prototypical system for acquiring, representing and executing clinical guidelines. GLARE has been built within a 7-year project with Azienda Ospedaliera San Giovanni Battista in Turin (one of the largest hospitals in Italy) and has been successfully tested on clinical guidelines in different domains, including bladder cancer, reflux esophagitis, and heart failure. GLARE is characterized by the adoption of advanced Artificial Intelligence (AI) techniques, to support medical decision making and to manage temporal knowledge.

A context-adaptable approach to clinical guidelines.

One of the most relevant obstacles to the use and dissemination of clinical guidelines is the gap between the generality of guidelines (as defined, e.g., by physicians' committees) and the peculiarities of the specific context of application. In particular, general guidelines do not take into account the fact that the tools needed for laboratory and instrumental investigations might be unavailable at a given hospital. Moreover, computer-based guideline managers must also be integrated with the Hospital Information System (HIS), and usually different DBMS are adopted by different hospitals. The GLARE (Guideline Acquisition, Representation and Execution) system addresses these issues by providing a facility for automatic resource-based adaptation of guidelines to the specific context of application, and by providing a modular architecture in which only limited and well-localised changes are needed to integrate the system with the HIS at hand.

Improving structural medical process comparison by exploiting domain knowledge and mined information.

OBJECTIVES: Process model comparison and similar process retrieval is a key issue to be addressed in many real-world situations, and a particularly relevant one in medical applications, where similarity quantification can be exploited to accomplish goals such as conformance checking, local process adaptation analysis, and hospital ranking. In this paper, we present a framework that allows the user to: (i) mine the actual process model from a database of process execution traces available at a given hospital; and (ii) compare (mined) process models. The tool is currently being applied in stroke management. METHODS: Our framework relies on process mining to extract process-related information (i.e., process models) from data. As for process comparison, we have modified a state-of-the-art structural similarity metric by exploiting: (i) domain knowledge; (ii) process mining outputs and statistical temporal information. These changes were meant to make the metric more suited to the medical domain. RESULTS: Experimental results showed that our metric outperforms the original one, and generated output closer than that provided by a stroke management expert. In particular, our metric correctly rated 11 out of 15 mined hospital models with respect to a given query. On the other hand, the original metric correctly rated only 7 out of 15 models. The experiments also showed that the framework can support stroke management experts in answering key research questions: in particular, average patient improvement decreased as the distance (according to our metric) from the top level hospital process model increased. CONCLUSIONS: The paper shows that process mining and process comparison, through a similarity metric tailored to medical applications, can be applied successfully to clinical data to gain a better understanding of different medical processes adopted by different hospitals, and of their impact on clinical outcomes. In the future, we plan to make our metric even more general and efficient, by explicitly considering various methodological and technological extensions. We will also test the framework in different domains.