Margarita Sordo

Extracting principal diagnosis, co-morbidity and smoking status for asthma research: evaluation of a natural language processing system

BackgroundThe text descriptions in electronic medical records are a rich source of information. We have developed a Health Information Text Extraction (HITEx) tool and used it to extract key findings for a research study on airways disease.MethodsThe principal diagnosis, co-morbidity and smoking status extracted by HITEx from a set of 150 discharge summaries were compared to an expert-generated gold standard.ResultsThe accuracy of HITEx was 82% for principal diagnosis, 87% for co-morbidity, and 90% for smoking status extraction, when cases labeled Insufficient Data by the gold standard were excluded.ConclusionWe consider the results promising, given the complexity of the discharge summaries and the extraction tasks.

Rapid identification of myocardial infarction risk associated with diabetes medications using electronic medical records.

OBJECTIVE To assess the ability to identify potential association(s) of diabetes medications with myocardial infarction using usual care clinical data obtained from the electronic medical record. RESEARCH DESIGN AND METHODS We defined a retrospective cohort of patients (n = 34,253) treated with a sulfonylurea, metformin, rosiglitazone, or pioglitazone in a single academic health care network. All patients were aged >18 years with at least one prescription for one of the medications between 1 January 2000 and 31 December 2006. The study outcome was acute myocardial infarction requiring hospitalization. We used a cumulative temporal approach to ascertain the calendar date for earliest identifiable risk associated with rosiglitazone compared with that for other therapies. RESULTS Sulfonylurea, metformin, rosiglitazone, or pioglitazone therapy was prescribed for 11,200, 12,490, 1,879, and 806 patients, respectively. A total of 1,343 myocardial infarctions were identified. After adjustment for potential myocardial infarction risk factors, the relative risk for myocardial infarction with rosiglitazone was 1.3 (95% CI 1.1–1.6) compared with sulfonylurea, 2.2 (1.6–3.1) compared with metformin, and 2.2 (1.5–3.4) compared with pioglitazone. Prospective surveillance using these data would have identified increased risk for myocardial infarction with rosiglitazone compared with metformin within 18 months of its introduction with a risk ratio of 2.1 (95% CI 1.2–3.8). CONCLUSIONS Our results are consistent with a relative adverse cardiovascular risk profile for rosiglitazone. Our use of usual care electronic data sources from a large hospital network represents an innovative approach to rapid safety signal detection that may enable more effective postmarketing drug surveillance.

The tell-tale heart: population-based surveillance reveals an association of rofecoxib and celecoxib with myocardial infarction

Background COX-2 selective inhibitors are associated with myocardial infarction (MI). We sought to determine whether population health monitoring would have revealed the effect of COX-2 inhibitors on population-level patterns of MI. Methodology/Principal Findings We conducted a retrospective study of inpatients at two Boston hospitals, from January 1997 to March 2006. There was a population-level rise in the rate of MI that reached 52.0 MI-related hospitalizations per 100,000 (a two standard deviation exceedence) in January of 2000, eight months after the introduction of rofecoxib and one year after celecoxib. The exceedence vanished within one month of the withdrawal of rofecoxib. Trends in inpatient stay due to MI were tightly coupled to the rise and fall of prescriptions of COX-2 inhibitors, with an 18.5% increase in inpatient stays for MI when both rofecoxib and celecoxib were on the market (P<0.001). For every million prescriptions of rofecoxib and celecoxib, there was a 0.5% increase in MI (95%CI 0.1 to 0.9) explaining 50.3% of the deviance in yearly variation of MI-related hospitalizations. There was a negative association between mean age at MI and volume of prescriptions for celecoxib and rofecoxib (Spearman correlation, −0.67, P<0.05). Conclusions/Significance The strong relationship between prescribing and outcome time series supports a population-level impact of COX-2 inhibitors on MI incidence. Further, mean age at MI appears to have been lowered by use of these medications. Use of a population monitoring approach as an adjunct to pharmacovigilence methods might have helped confirm the suspected association, providing earlier support for the market withdrawal of rofecoxib.

Description and status update on GELLO: a proposed standardized object-oriented expression language for clinical decision support.

A major obstacle to sharing computable clinical knowledge is the lack of a common language for specifying expressions and criteria. Such a language could be used to specify decision criteria, formulae, and constraints on data and action. Al-though the Arden Syntax addresses this problem for clinical rules, its generalization to HL7s object-oriented data model is limited. The GELLO Expression language is an object-oriented language used for expressing logical conditions and computations in the GLIF3 (GuideLine Interchange Format, v. 3) guideline modeling language. It has been further developed under the auspices of the HL7 Clinical Decision Support Technical Committee, as a proposed HL7 standard., GELLO is based on the Object Constraint Language (OCL), because it is vendor-independent, object-oriented, and side-effect-free. GELLO expects an object-oriented data model. Although choice of model is arbitrary, standardization is facilitated by ensuring that the data model is compatible with the HL7 Reference Information Model (RIM).

On sample size and classification accuracy: a performance comparison

We investigate the dependency between sample size and classification accuracy of three classification techniques: Naive Bayes, Support Vector Machines and Decision Trees over a set of 8500 text excerpts extracted automatically from narrative reports from the Brigham & Womens Hospital, Boston, USA. Each excerpt refers to the smoking status of a patient as: current, past, never a smoker or, denies smoking. Our empirical results, consistent with [1], confirm that size of the training set and the classification rate are indeed correlated. Even though these algorithms perform reasonably well with small datasets, as the number of cases increases, both SMV and Decision Trees show a substantial improvement in performance, suggesting a more consistent learning process. Unlike the majority of evaluations, ours were carried out specifically in a medical domain where the limited amount of data is a common occurrence [13][14]. This study is part of the I2B2 project, Core 2.

Design of a standards-based external rules engine for decision support in a variety of application contexts: report of a feasibility study at Partners HealthCare System.

This project explored functional requirements for an institution-wide method, at Partners HealthCare, for interpreting clinical knowledge for decision support. Such knowledge is currently incorporated in a variety of clinical applications, yet the methods of representation and of execution vary and the ability to author/edit the rules by human experts is limited. We expanded on a 2002 Knowledge Inventory at Partners to evaluate feasibility of designing a single representation approach entailing: (a) exploration of specific needs of different applications, in terms of kinds of response required (synchronous/asynchronous, time criticality, etc.), context (e.g., implied patient, time frame, or episode), and kinds of actions to be triggered; (b) kind of representation of knowledge and feasibility of casting knowledge in the form of if em leader then statements; and (c) data and knowledge resources used (implied data model, and particular knowledge sources and terminology sources). The result of analysis was to design an architecture to accomplish this goal. We also did preliminary analysis of requirements for authoring for such a representation, and for implementation.

A PSO/ACO approach to knowledge discovery in a pharmacovigilance context

We propose and evaluate the use of a Particle Swarm Optimization/Ant Colony Optimization (PSO/ACO) methodology for classification and rule discovery in the context of medication postmarketing surveillance or pharmacovigilance. Our study considers a large data set of diabetic patients on two widely used antidiabetic drugs (rosiglitazone and pioglitazone), and the risk of myocardial infarction as an adverse effect. The goal is to determine the presence of previously undetected causal relationships between therapeutics, patient characteristics, and adverse medication outcomes. Since the proposed approach is able to discover classification rules, the elicited knowledge may suggest new hypotheses regarding associations between risk factors and an adverse event. Our classification results show high accuracy. Furthermore, several medication-related rules were discovered and analyzed. The elicited rules support previous studies from the medical literature. Moreover, one of the studied antidiabetic drugs (rosiglitazone) was found to have a significant higher risk of an adverse event on diabetic, hypertensive patients, as compared to the other drug. This last finding suggests that pioglitazone may have a protective effect against myocardial infarction on diabetic, hypertensive patients.

Grouped Knowledge Elements

Two frequently occurring tasks in the clinical workflow where the health care provider and the computer communicate are during clinical and in computerized-provider order entry (CPOE). These tasks are facilitated by grouped knowledge elements, i.e. structured documentation templates and order sets, respectively. A structured documentation template is an organized collection of data items relevant to a particular clinical context that can be used to collect or present codified information about a patient. An order set, similarly, is an organized collection of actions that can be ordered by a health care provider for the care of a patient in a specific clinical context. These grouped knowledge elements may be considered as vehicles for providing clinical decision support. Those who design the groupings of knowledge elements can use them to drive the behavior of the health care professional user. Additionally, order items and documentation items can be dynamically presented based on the context. By anticipating needs for data entry or access, or for orders, such grouping not only provides CDS but also facilitates workflow, by eliminating extra steps that would otherwise be needed. The specification of an order set’s or a document template’s structure and content is a form of knowledge. Standards are being developed for such specification to encourage the collection of higher quality, more interpretable, more comprehensive data, and to encourage reuse of document specifications, or parts thereof, where appropriate. This chapter reviews those efforts, in terms of their degree of maturity and harmonization, and how they relate to clinical decision support.

Partners healthcare order set schema: An information model for management of clinical content

Developed by the Clinical Knowledge Management and Decision Support Group at Partners HealthCare and the Decision Systems Group at Harvard Medical School, the XML-based Order Set Schema presented in this chapter is the result of a broader enterprise-wide knowledge management effort to enhance quality, safety, and efficiency of provided care at Partners HealthCare while maximizing the use of new clinical information technology. We are in the process of deploying the Order Set Schema at two Partners-based, Harvard-affiliated academic medical centers the Brigham & Women’s Hospital (BWH) and Massachusetts General Hospital (MGH), Boston, MA, so that existing content in the Computerized Physician Order Entry (CPOE) systems at these two institutions can be successfully extracted and mapped into the proposed schema. In this way, “hardwired” knowledge could be mapped into taxonomies of relevant terms, definitions and associations, resulting in formalized conceptual models and ontologies with explicit, consistent, user-meaningful relationships among concepts to support collaboration, and content management that will promote systematic (a) conversion of reference content into a form that approaches specifications for decision support content; (b) development and reuse of clinical content while ensuring consistency in the information; and (c) support an open and distributed review process among leadership, content matter experts, and end-users. Further, incorporating metadata into our unified content strategy will improve workflow by enabling timely review and updating of content, knowledge life-cycle management, and knowledge encoding; reduce costs and; aid authors to identify relevant elements for reuse while reducing redundant and spurious content. Ultimately, we view our knowledge management infrastructure as a key element for knowledge discovery.

Knowledge management infrastructure: Evolution at partners healthcare system

Publisher Summary This chapter represents a snapshot of a multiyear undertaking to develop a knowledge management infrastructure that must, by necessity, serve the needs of a large, extremely heterogeneous application environment with an enormous inventory of homegrown content in production. The infrastructure needed to support knowledge transparency and content governance have been built, and the focus is now on tackling the deeper challenges of knowledge engineering that undermine the ability to expand or change content in production once it is determined that such changes are necessary. To accomplish this, new integration approaches are being pursued among vendor solutions for content management, business rules management, terminology management, and ontology management. This approach is essential to support managing knowledge at the speed of change anticipated, particularly with the advent of molecular medicine now. It is common in health care to focus on the computing power required for analytics or for meeting the event management requirements of patient data transactions. A knowledge management infrastructure can be viewed as a knowledge-event management framework that will support structured knowledge discovery, acquisition, and maintenance for the era of personalized medicine.