Knowledge about maternal history is critical for guiding certain aspects of newborn clinical care as well as for research on neonatal issues. However, often the only maternal history available in the newborn record is in the clinical notes. We are using data from the MIMIC-II database for a clinical study on newborns admitted to the intensive care unit. Important maternal data were only available in the newborn notes, so we developed a simple algorithm to extract those data. We manually derived patterns for maternal age, gravida/para status, and laboratory results by reviewing a small set of notes. Using regular expressions and specific filters for notes and results, we extracted maternal data with recall of 0.91–0.99 and precision of 0.95–1.0 for the 289 infants in our study. Our methods could be used with other research datasets and with clinical documentation systems to extract maternal data into a more useful, structured format.
While health IT is thought to be critical to the success of new models of care delivery, we know little about the extent to which those pursuing these models are relying on HIT. We studied a large patient-centered medical home (PCMH) demonstration project, a new model of care delivery that has received substantial policy attention, in order to assess which types of HIT were most widely used, and how adoption rates changed over time as PCMH practices matured. We found that clinically-focused HIT tools were both widely adopted, and increasingly adopted, in PCMH practices compared to non-PCMH practices. In contrast, HIT that supports patient-engagement, patient portals and personal health records, was neither in widespread use nor more likely to be adopted over time by PCMH practices compared to other practices. This suggests that these tools may not yet support the types of patient engagement and interactions that PCMH practices seek.
The National Library of Medicine has published the CORE and the VA/KP problem lists to facilitate the usage of SNOMED CT for encoding diagnoses and clinical data of patients in electronic health records. Therefore, it is essential for the content of the problem lists to be as accurate and consistent as possible. This study assesses the effectiveness of using a concept’s word length and number of parents, two structural indicators for measuring concept complexity, to identify inconsistencies with high probability. The method is able to isolate concepts with over 40% expected of being erroneous. A structural indicator for concepts which is able to identify 52% of the examined concepts as having errors in synonyms is also presented. The results demonstrate that the concepts in problem lists are not free of inconsistencies and further quality assurance is needed to improve the quality of these concepts.
Patient access to electronic health records (EHR) is expected to have a variety of benefits, including enhanced patient involvement in care and access to health information, yet little is known about potential demand. We used the 2007 Health Information and National Trends Survey, a national probability-based survey, to determine which health care users with Internet access are likely to report that electronic access to their health records is important for themselves and their providers. Respondents who represent populations that generally experience health and healthcare disparities (Blacks, Latina/os, and patients with psychological distress) were among the most likely to report that the EHR was very important for them, even after controlling for respondents’ socio-economic status, health status, health care context, and disposition toward health information. Health policies and the designs of EHRs should consider these patterns, which may help address health and health care disparities.
Many Americans are challenged by the tasks of understanding and acting upon their own health data. Low levels of health literacy contribute to poor comprehension and undermine the confidence necessary for health self-management. Visualizations are useful for minimizing comprehension gaps when communicating complex quantitative information. The process of developing visualizations that accommodate the needs of individuals with varying levels of health literacy remains undefined. In this paper we provide detailed descriptions of a) an iterative methodological approach to the development of visualizations, b) the resulting types of visualizations and examples thereof, and c) the types of data the visualizations will be used to convey. We briefly describe subsequent phases in which the visualizations will be tested and refined. Web deployment of the final visualizations will support the ethical obligation to return the data to the research participants and community that contributed it.
This paper reports on a data collection study in a clinical environment to evaluate a new non-invasive monitoring system for people with advanced Multiple Sclerosis (MS) who use powered wheelchairs. The proposed system can acquire respiration and heart activity from ballistocardiogram (BCG) signals, seat and back pressure changes, wheelchair tilt angle, ambient temperature and relative humidity. The data was collected at The Boston Home (TBH), a specialized care residence for adults with advanced MS. The collected data will be used to design algorithms to generate alarms and recommendations for residents and caregivers. These alarms and recommendations will be related to vital signs, low mobility problems and heat exposure. We present different cases where it is possible to illustrate the type of information acquired by our system and the possible alarms we will generate.
We evaluated the performance of LOINC® and RadLex standard terminologies for covering CT test names from three sites in a health information exchange (HIE) with the eventual goal of building an HIE-based clinical decision support system to alert providers of prior duplicate CTs. Given the goal, the most important parameter to assess was coverage for high frequency exams that were most likely to be repeated. We showed that both LOINC® and RadLex provided sufficient coverage for our use case through calculations of (a) high coverage of 90% and 94%, respectively for the subset of CTs accounting for 99% of exams performed and (b) high concept token coverage (total percentage of exams performed that map to terminologies) of 92% and 95%, respectively. With trends toward greater interoperability, this work may provide a framework for those wishing to map radiology site codes to a standard nomenclature for purposes of tracking resource utilization.
Automated Word Sense Disambiguation in clinical documents is a prerequisite to accurate extraction of medical information. Emerging methods utilizing hyperdimensional computing present new approaches to this problem. In this paper, we evaluate one such approach, the Binary Spatter Code Word Sense Disambiguation algorithm, on 50 ambiguous abbreviation sets derived from clinical notes. This algorithm uses reversible vector transformations to encode ambiguous terms and their context-specific senses into vectors representing surrounding terms. The sense for a new context is then inferred from vectors representing the terms it contains. One-to-one BSC-WSD achieves average accuracy of 94.55% when considering the orientation and distance of neighboring terms relative to the target abbreviation, outperforming Support Vector Machine and Naïve Bayes classifiers. Furthermore, it is practical to deal with all 50 abbreviations in an identical manner using a single one-to-many BSC-WSD model with average accuracy of 93.91%, which is not possible with common machine learning algorithms.
Biomedical ontologies are often large and complex, making ontology development and maintenance a challenge. To address this challenge, scientists use automated techniques to alleviate the difficulty of ontology development. However, for many ontology-engineering tasks, human judgment is still necessary. Microtask crowdsourcing, wherein human workers receive remuneration to complete simple, short tasks, is one method to obtain contributions by humans at a large scale. Previously, we developed and refined an effective method to verify ontology hierarchy using microtask crowdsourcing. In this work, we report on applying this method to find errors in the SNOMED CT CORE subset. By using crowdsourcing via Amazon Mechanical Turk with a Bayesian inference model, we correctly verified 86% of the relations from the CORE subset of SNOMED CT in which Rector and colleagues previously identified errors via manual inspection. Our results demonstrate that an ontology developer could deploy this method in order to audit large-scale ontologies quickly and relatively cheaply.
In this paper we describe a natural language processing system which is able to predict whether or not a patient exhibits a specific phenotype using the information extracted from the narrative reports associated with the patient. Furthermore, the phenotypic annotations from our report dataset were performed at the report level which allows us to perform the prediction of the clinical phenotype at any point in time during the patient hospitalization period. Our experiments indicate that an important factor in achieving better results for this problem is to determine how much information to extract from the patient reports in the time interval between the patient admission time and the current prediction time.
We present a pilot study of an annotation schema representing problems and their attributes, along with their relationship to temporal modifiers. We evaluated the ability for humans to annotate clinical reports using the schema and assessed the contribution of semantic annotations in determining the status of a problem mention as active, inactive, proposed, resolved, negated, or other. Our hypothesis is that the schema captures semantic information useful for generating an accurate problem list. Clinical named entities such as reference events, time points, time durations, aspectual phase, ordering words and their relationships including modifications and ordering relations can be annotated by humans with low to moderate recall. Once identified, most attributes can be annotated with low to moderate agreement. Some attributes – Experiencer, Existence, and Certainty - are more informative than other attributes – Intermittency and Generalized/Conditional - for predicting a problem mention’s status. Support vector machine outperformed Naïve Bayes and Decision Tree for predicting a problem’s status.
Abstraction networks are compact summarizations of terminologies used to support orientation and terminology quality assurance (TQA). Area taxonomies and partial-area taxonomies are abstraction networks that have been successfully employed in support of TQA of small SNOMED CT hierarchies. However, nearly half of SNOMED CT’s concepts are in the large Procedure and Clinical Finding hierarchies. Abstraction network derivation methodologies applied to those hierarchies resulted in taxonomies that were too large to effectively support TQA. A methodology for deriving sub-taxonomies from large taxonomies is presented, and the resultant smaller abstraction networks are shown to facilitate TQA, allowing for the scaling of our taxonomy-based TQA regimen to large hierarchies. Specifically, sub-taxonomies are derived for the Procedure hierarchy and a review for errors and inconsistencies is performed. Concepts are divided into groups within the sub-taxonomy framework, and it is shown that small groups are statistically more likely to harbor erroneous and inconsistent concepts than large groups.
In a previous paper, we presented initial findings from a study on the feasibility and challenges of implementing teleretinal screening for diabetic retinopathy in an urban safety net setting facing eyecare specialist shortages. This paper presents some final results from that study, which involved six South Los Angeles safety net clinics. A total of 2,732 unique patients were screened for diabetic retinopathy by three ophthalmologist readers, with 1035 receiving a recommendation for referral to specialty care. Referrals included 48 for proliferative diabetic retinopathy, 115 for severe non-proliferative diabetic retinopathy (NPDR), 247 for moderate NPDR, 246 for mild NPDR, 97 for clinically significant macular edema, and 282 for a non-diabetic condition, such as glaucoma. Image quality was also assessed, with ophthalmologist readers grading 4% to 13% of retinal images taken at the different clinics as being inadequate for any diagnostic interpretation.
While some published research indicates a fairly high frequency of Intravenous (IV) medication errors associated with the use of smart infusion pumps, the generalizability of these results are uncertain. Additionally, the lack of a standardized methodology for measuring these errors is an issue. In this study we iteratively developed a web-based data collection tool to capture IV medication errors using a participatory design approach with interdisciplinary experts. Using the developed tool, a prevalence study was then conducted in an academic medical center. The results showed that the tool was easy to use and effectively captured all IV medication errors. Through the prevalence study, violation errors of hospital policy were found that could potentially place patients at risk, but no critical errors known to contribute to patient harm were noted.
While Electronic Medical Records (EMR) contain detailed records of the patient-clinician encounter — vital signs, laboratory tests, symptoms, caregivers’ notes, interventions prescribed and outcomes — developing predictive models from this data is not straightforward. These data contain systematic biases that violate assumptions made by off-the-shelf machine learning algorithms, commonly used in the literature to train predictive models. In this paper, we discuss key issues and subtle pitfalls specific to building predictive models from EMR. We highlight the importance of carefully considering both the special characteristics of EMR as well as the intended clinical use of the predictive model and show that failure to do so could lead to developing models that are less useful in practice. Finally, we describe approaches for training and evaluating models on EMR using early prediction of septic shock as our example application.
Three-dimensional models are being extensively used in the medical area in order to improve clinical medical examinations and diagnosis. The Cardiology field handles with several types of image slices to compose the diagnosis. MRI (Magnetic Resonance Imaging) is a non-invasive technique to detect anomalies from internal images of the human body that generates hundreds of images, which takes long for the specialist to analyze frame by frame and the diagnosis precision can be affected. Many cardiac diseases could be identified through shape deformation, but systems aimed to aid diagnosis usually identify shapes in two-dimensional (2D) images. Our aim is to apply a shape descriptor to retrieve three-dimensional cardiac models, obtained from a set of 2D slices, which were segmented and reconstructed from MRI images using their geometry information. Preliminary results show that the shape deformation in 3D models can be a good indicator to detect Congestive Heart Failure, a very common heart disease.
The biomedical literature presents a uniquely challenging text mining problem. Sentences are long and complex, the subject matter is highly specialized with a distinct vocabulary, and producing annotated training data for this domain is time consuming and expensive. In this environment, unsupervised text mining methods that do not rely on annotated training data are valuable. Here we investigate the use of random indexing, an automated method for producing vector-space semantic representations of words from large, unlabeled corpora, to address the problem of term normalization in sentences describing drugs and genes. We show that random indexing produces similarity scores that capture some of the structure of PHARE, a manually curated ontology of pharmacogenomics concepts. We further show that random indexing can be used to identify likely word candidates for inclusion in the ontology, and can help localize these new labels among classes and roles within the ontology.
The Tele Intensive Care Unit (tele-ICU) supports a high volume, high acuity population of patients. There is a high-volume of incoming and outgoing calls, especially during the evening and night hours, through the tele-ICU hubs. The tele-ICU clinicians must be able to communicate effectively to team members in order to support the care of complex and critically ill patients while supporting and maintaining a standard to improve time to intervention. This study describes a software communication tool that will improve the time to intervention, over the paper-driven communication format presently used in the tele-ICU. The software provides a multi-relational database of message instances to mine information for evaluation and quality improvement for all entities that touch the tele-ICU. The software design incorporates years of critical care and software design experience combined with new skills acquired in an applied Health Informatics program. This software tool will function in the tele-ICU environment and perform as a front-end application that gathers, routes, and displays internal communication messages for intervention by priority and provider.
Medication reconciliation is an important and complex task for which careful user interface design has the potential to help reduce errors and improve quality of care. In this paper we focus on the hospital discharge scenario and first describe a novel interface called Twinlist. Twinlist illustrates the novel use of spatial layout combined with multi-step animation, to help medical providers see what is different and what is similar between the lists (e.g., intake list and hospital list), and rapidly choose the drugs they want to include in the reconciled list. We then describe a series of variant designs and discuss their comparative advantages and disadvantages. Finally we report on a pilot study that suggests that animation might help users learn new spatial layouts such as the one used in Twinlist.
Temporal abstraction, a method for specifying and detecting temporal patterns in clinical databases, is very expressive and performs well, but it is difficult for clinical investigators and data analysts to understand. Such patterns are critical in phenotyping patients using their medical records in research and quality improvement. We have previously developed the Analytic Information Warehouse (AIW), which computes such phenotypes using temporal abstraction but requires software engineers to use. We have extended the AIW’s web user interface, Eureka! Clinical Analytics, to support specifying phenotypes using an alternative model that we developed with clinical stakeholders. The software converts phenotypes from this model to that of temporal abstraction prior to data processing. The model can represent all phenotypes in a quality improvement project and a growing set of phenotypes in a multi-site research study. Phenotyping that is accessible to investigators and IT personnel may enable its broader adoption.
This paper describes our approach for fostering and facilitating communication among patients and caregivers in the context of shared decision making, i.e., when decisions must be taken not only on the basis of scientific evidence but also of the patient’s preferences and context. This happens because clinical practice guidelines cannot provide recommendations for every possible situation, and cannot foresee every change in a patient’s context, which might imply the deviation from a previously acknowledged recommendation. Within the EU-funded project MobiGuide (www.mobiguide-project.eu), supporting remote patient management, we propose decision theory as a methodological framework for a tool that, during face to face encounters, is used to tailor pre-defined, generic decision models to the individual patient, by involving the patient himself in the customization of the model parameters. Although this approach is not appropriate for all patients, it leads, in well-chosen cases, to a more informed choice, with potentially better treatment compliance.
Efficiency and quality of documentation are critical in surgical settings because operating rooms are a major source of revenue, and because adverse events may have enormous consequences. Electronic health records (EHRs) have potential to impact surgical volume, quality, and documentation time. Ophthalmology is an ideal domain to examine these issues because procedures are high-throughput and demand efficient documentation. This time-motion study examines nursing documentation during implementation of an EHR operating room management system in an ophthalmology department. Key findings are: (1) EHR nursing documentation time was significantly worse during early implementation, but improved to a level near but slightly worse than paper baseline, (2) Mean documentation time varied significantly among nurses during early implementation, and (3) There was no decrease in operating room turnover time or surgical volume after implementation. These findings have important implications for ambulatory surgery departments planning EHR implementation, and for research in system design.
Graduate training in biomedical informatics (BMI) is evolving rapidly. BMI graduate programs differ in informatics domain, delivery method, degrees granted, as well as breadth and depth of curricular competencies. Using the current American Medical Informatics Association (AMIA) definition of BMI core competencies as a framework, we identified and labeled course offerings within graduate programs. From our qualitative analysis, gaps between defined competencies and curricula emerged. Topics missing from existing graduate curricula include community health, translational and clinical research, knowledge representation, data mining, communication and evidence-based practice.
To determine whether HIT currently supports care transitions we interviewed clinicians from several healthcare settings. We learned about HIT tools to help nurses facilitate transitions, but discovered that there are few tools to promote high quality, safe transitions of care. We also found that HIT is rarely employed for patient-centered care coordination mechanisms. In conclusion, HIT tools are typically used within one healthcare setting to prepare for a transition, rather than across healthcare settings.
electronic health record; meaningful use; care coordination; care transitions