Semantic Clinical Process Management Massimo Ruffolo1,2, Marco Manna3, Vittoria Cozza2 ,Raffaello Ursino4 ICAR-CNR 1 Exeura s.r.l. 2 Department of Mathematics3 Orangee s.r.l. 4 c/o University of Calabria 87036, Rende (CS), Italy
[email protected],
[email protected],
[email protected],
[email protected]
Abstract This work describes a clinical process management system aimed to support a processcentred vision of health care practices. The system is founded on knowledge representation and semantic information extraction approaches allowing medical knowledge modelling and acquisition. At the heart of the system there are formalisms and languages well suited for representing, clinical processes as workflows, medical and domain knowledge as ontologies and rules enabling the recognition of semantic patterns representing ontology concepts. The system acquires and stores clinical process instances into a medical knowledge base which parameters are obtained exploiting a semantic information extraction approach enabling automatic medical knowledge acquisition from unstructured clinical documents. The main goal of the system is to assists health care professional in executing and monitoring clinical processes by providing functionalities for automatic knowledge acquisition. Acquired information can be analyzed for identifying main causes of medical errors, high costs and, potentially, to suggest clinical processes restructuring or improvement able to enhance cost control and patient safety.
1. Introduction Nowadays health care costs and risks management is a high priority theme for health care professionals and providers. Across the world the whole issue of patient safety, medical errors prevention and adverse events reporting is a very challenging and widely studied research and development topic that stimulates a growing interest in the computer science researchers community. Health care practices are characterized by complex clinical processes in which high risk activities take place. A clinical process can be seen as a particular workflow where medical (e.g. treatments, drugs administration, guidelines execution, medical examinations, risk evaluation, etc.) and non-medical (e.g. patient enrolment, medical record instantiation, etc.) activities and events occur. A promising approach for reducing cost and risk and enhancing patient safety is a semantic process-oriented vision of health care services and practices. Health care organizations, frequently, store relevant information about clinical processes in electronic medical records having heterogeneous format. Semantic information extraction allows the acquisition of information about the clinical processes from unstructured sources and their storage into structured machine-readable form. This make available large amount of data on which data mining techniques, can be performed to discover patterns related to adverse events, errors and cost dynamics, hidden in the structure of clinical processes, that are cause of risks and of poor performances.
Twentieth IEEE International Symposium on Computer-Based Medical Systems (CBMS'07) 0-7695-2905-4/07 $20.00 © 2007
In the recent past a strong research effort has been taken to provide uniform representations of declarative and procedural clinical knowledge that can be used to create semanticallyaware health care information systems. Interesting results have been obtained in the field of medical ontologies and guidelines. Currently are available ontologies such as UMLS (Unified Medical Language System) [1] and the cancer ontology NCI-EVS [2] fulfilling medical topics. UMLS ontology integrates information coming from heterogeneous sources in a unified medical terminology. Most important common sources for UMLS are HL7, ICD9-CM, MeSH [3,4,5]. In the clinical process field the Evidence Based Medicine movement has stimulated the definition of many guideline representation formalisms such as GLIF and PROForma based on different paradigms [6,7]; e.g. PROforma is a process description language grounded in a logical model whereas GLIF is a specification consisting of an object-oriented model. In this paper is presented a semantic Clinical Process Management System (CPMS) providing functionalities for medical knowledge modelling and automatic information acquisition able to support clinical processes design, acquisition and management. Basically the system allows the definition of Medical Knowledge Models (MKM) and related Medical Knowledge Bases (MKB). The MKM contains schemas of clinical processes represented as workflows, ontologies schemas and concept descriptors. The MKB contains process and ontology instances, the former obtained by means of manual creation the latter obtained by automatic extraction from unstructured sources. In the paper is also presented an application aimed to support the automatic construction of the patient history as a process in which all the occurred events are represented and the related data semi-automatically acquired and stored into an ontology. Acquired process instances can be analyzed to identify main causes of risks, to control costs and, potentially, to suggest clinical processes restructuring or improvement. Thanks to the system health care professionals have knowledge management and decision support functionalities able to enhance cost control and patient safety, reducing risks due to medical errors and adverse events. The paper is organized as follows: section 2 contains a general description of the clinical process management system, section 3 describes an application of the system to the acquisition of information about clinical risks occurring in lung cancer cures.
2. System Description The system allows to define a Medical Knowledge Model (MKM) by means of the Medical Knowledge Designer (MKD) module and a Medical Knowledge Base (MKB) by means of the Medical Knowledge Acquisition and Storing (MKAS) module. The MKM is composed by medical ontologies, clinical processes represented as workflows and semantic extraction rules enabling the recognition of information into unstructured source like EMR in flat text format. The MKM is defined using functionalities provided by the MKD module made of software modules providing functionalities for the representation of clinical processes, medical guidelines and related ontologies. The MKD module is composed of three sub modules: the OntoDLV [8,9] Ontology Designer (OOD), the Clinical Process Composer (CPC) and the Hilex Semantic Rules Designer (HSRD). The knowledge design can be obtained either by means of direct (on-screen) drawing and specification or importing existing medical ontologies and guidelines. The OOD module allows to define kind of ontologies: the medical ontology and the local ontology. Both the ontologies are written using the OntoDLP language [10]. The medical ontology concern a wide variety of concepts
Twentieth IEEE International Symposium on Computer-Based Medical Systems (CBMS'07) 0-7695-2905-4/07 $20.00 © 2007
related to the medical domain and to specific clinical knowledge areas (e.g. diseases, drugs, medical examinations, medical procedures, laboratory terms, anatomy, etc.). Medical ontologies are obtained importing existing medical ontologies and thesaurus like MeSH and ICD9-CM. The local ontology, obtained by means of direct definition using the MKD module, defines the semantic of administrative data regarding patients, wards, hospital and the structure of medical data regarding diseases, medical practices, treatments, drugs administration, medical examinations, etc. Medical and domain ontologies are used for the semantic description of process activities and data and for the definition of semantic Electronic Medical Records schemas (Meta EMR). The CPC module allows the representation of clinical processes using an ad-hoc workflow representation formalism also based on OntoDLP. This formalism permits the definition of clinical processes schemas through the representation of workflow elements (i.e. subprocesses, activities, events, conditional splits, conditional joins, process participants, triggers, etc.) and the annotation of the workflow elements and their parameters (data) w.r.t. medical and domain ontologies. The module HSRD enables the representation of semantic rules, expressed using the HiLEx language [11], representing everyone a possible way (pattern) of writing a concept represented into the ontology into an unstructured source. The module MKAS yields functionalities for the assisted execution of clinical processes and the acquisition of process instances and their data. The execution, performed through a graphical user interface constituted of web-based forms filled by doctors and nurses, is assisted in two different ways. The clinical process enactment in which designed workflow schema is followed exactly producing clinical process instances coming from the same schema. The dynamic workflow composition in which each activity or sub-process instance is acquired selecting the most appropriate one to execute in a given moment. In this case a specific workflow schema and the related instance are created at each execution. In both cases, for each clinical process instance data related to a given cared patient, are provided by the user or automatically extracted from unstructured clinical document. The extraction is performed exploiting the semantic information extraction approach described in [12]. Thanks to the system a huge amount of clinical process instances can be acquired and stored into the MKB. Acquired data can be analyzed by means of data and process mining techniques to discover pattern enabling adverse events and errors prevention, risks and costs reduction and patients safety enhancement.
3. An e-Health Application for Risk Management The semantic CPMS described in the previous paragraph has been applied to obtain a real world application in the field of health care risk management. The scope of the application is to support hospital to monitor errors and risk causes. The application consists in the representation of a MKM containing: a set of clinical processes related to the lung cancer care, each process includes a set of activities aimed at the acquisition of information related to possible risk causes; an ontology representing concepts related to lung cancer and risk causes; a set of rules enabling the automatic acquisition of information from clinical electronic documents in natural language. The system takes as input documents in unstructured format coming from different hospital wards. In particular, documents consist in clinical Electronic Medical Records (EMR) and risk reports. EMRs used in the experimentation refers to 25 patients with
Twentieth IEEE International Symposium on Computer-Based Medical Systems (CBMS'07) 0-7695-2905-4/07 $20.00 © 2007
lung cancer. An EMR is generally a flat text document (having usually 3 pages) written in Italian language. EMRs are weakly structured, for example, the personal data of the patient are in the top of the document, clinical events (e.g medical exams, surgical operations, diagnosis, etc.) are introduced by a date. In the experiment data to extract consist in personal information, address information, diagnosis data (diagnosis time, kind of tumor, body part affected by the cancer, cancer progression level), care and therapies information. Risk reports, filled at the end of each clinical process, are provided by wards to acquire errors with or without serious outcomes, adverse events, near misses. The 50 report kept under consideration in the experiments refers to a semistructured html template. It contains a key section with information identifying the patient and the ward where lookout event happened. In the experimentation the issue has been to extract information about errors related with oncology therapies and to discriminate if such errors happen in phases of drugs prescription, preparation or administration. The system for each cared patient create an instance of lung cancer clinical process and stores it into the MKB. Data regarding process activities, related to medical examinations, drug administrations and so on, are acquired by means of visual interfaces, whereas information concerning risk management are automatically acquired from unstructured documents. In particular, all relevant extracted process instances information, coming from unstructured EMR and semi-structured risk reports, and semantic concepts they belong to are stored inside an XML file by mean of the MKAS module. Each XML output document, has been defined as shown in the following DTD.
During a running of the simulation many document can be concurrently processed, thanks to parallel MKAS module architecture, and all the final extraction results will be stored in once XML document_list. For each document this DTD allows to store the document_type, the input document doc_path and especially the ontology_path. This last attribute gives the namespace with respect to the extraction process has been performed. Anytime an user specify an extraction query, a tag table with the query name is generated in each document, any instance of this query generate an XML row with inside required fields. Each field has three attributes: fieldname is the name the final user chooses to nominate the field, concept is the ontology name for the concept that allowed to identify the field, concept_instance corresponds to tokens extracted by the original documents exactly as they appeared in it. The default value for a concept is the HiLEx super-class element. The default value for concept_instance is obviously an empty string. The MKM for this application is founded on a domain ontology describing patient information and a medical ontology describing the medical risk. Both the ontologies are written in OntoDLP language by using the visual editor OntoDLV. To each concept described inside the ontology corresponds semantic extraction patterns written in the HiLEx language. A portion of the domain ontology describing patient name, surname, age and gender is represented in the following: class patient_element isa {domain_element}.
Twentieth IEEE International Symposium on Computer-Based Medical Systems (CBMS'07) 0-7695-2905-4/07 $20.00 © 2007
class name isa {patient_element}. class family_name isa {name}. class given_name isa {name}. class gender_code {patient_element}. class age {patient_element}.
The medical ontology is inherent lung cancer and represents concepts related to this tumor, its diagnosis, cares in term of surgical operations and chemotherapies with the associated side effects. All the concepts related to the cancer come from the ICD9-CM diseases classification system. The chemotherapy drugs taxonomy, inspired at the Anatomic Therapeutic Chemical (ATC) classification system [14]. High level concepts about chemotherapy drugs contained into patients EMRs has been defined by the following ontology classes: class anatomic_main_group isa {body_organ_group}. class therapeutic_main_group isa {body_organ_group}. class chemical_therapeutic_group isa {therapeutic_main_group}. class chemical_group isa {therapeutic_main_group}.
A section of the medical ontology represents concepts related to medical errors declared by means of patients risk reports. An example of different kinds of errors is shown in the following taxonomy. class medical_error isa {domain_element}. class error_in_verbal_communication isa {medical_error}. class error_in_written_communication isa {medical_error}. class error_in_drug_somministration isa {medical_error}.
Beside the domain and medical ontologies the MKM contains the extraction rules representing all the way in which represented concepts can be expressed (patterns) into unstructured document in natural language. Extraction rules are constituted by regular expressions enabling the recognition of simple words or complex structures describing semantically concepts which meaning depends from the context. An example of rule aimed at the recognition of a simple word (the name of a drug) is shown in the following: daunorobicin_chemical_group_drug: chemical_group ( type: regexp_type, expression: [Dd]aunorobicin| DAUNOROBICIN").
An example of rule enabling the recognition of a complex pattern expressing complex concept such us cancer typology is the following: patient_cancer_typology: patient_cancer_typology_in_diagnosis( type:hilex_type, expression:"heldIn_generalizationOf( arg:@cancer_typology, caseArg:diagnosis_paragraph, condition: notCoincident)").
The above rule enables to identify a diagnosis_paragraph and to extract the cancer_typology from it. The use of the @ operator means that all the instances of the cancer_typology taxonomy are considered. The MKB is created storing XML documents according to the DTD previously described. Documents are obtained exploiting the semantic information extraction approach cited above. The extraction mechanism, allowed by HiLEx rules, can be considered in a WOXM fashion: Write Once eXtract Many. In the simple case of personal patient information, the extraction process generates the following portion of output document:
