Automated detection of altered mental status in emergency department clinical notes: a deep learning approach.

BACKGROUND: Machine learning has been used extensively in clinical text classification tasks. Deep learning approaches using word embeddings have been recently gaining momentum in biomedical applications. In an effort to automate the identification of altered mental status (AMS) in emergency department provider notes for the purpose of decision support, we compare the performance of classic bag-of-words-based machine learning classifiers and novel deep learning approaches. METHODS: We used a case-control study design to extract an adequate number of clinical notes with AMS and non-AMS based on ICD codes. The notes were parsed to extract the history of present illness, which was used as the clinical text for the classifiers. The notes were manually labeled by clinicians. As a baseline for comparison, we tested several traditional bag-of-words based classifiers. We then tested several deep learning models using a convolutional neural network architecture with three different types of word embeddings, a pre-trained word2vec model and two models without pre-training but with different word embedding dimensions. RESULTS: We evaluated the models on 1130 labeled notes from the emergency department. The deep learning models had the best overall performance with an area under the ROC curve of 98.5% and an accuracy of 94.5%. Pre-training word embeddings on the unlabeled corpus reduced training iterations and had performance that was statistically no different than the other deep learning models. CONCLUSION: This supervised deep learning approach performs exceedingly well for the detection of AMS symptoms in clinical text in our environment. Further work is needed for the generalizability of these findings, including evaluation of these models in other types of clinical notes and other environments. The results seem promising for the ultimate use of these types of classifiers in combination with other information derived from the electronic health records as input for clinical decision support.

Health Information Exchange in the ED: What Do ED Clinicians Think?

OBJECTIVES: Our regional health information exchange (HIE), known as Carolina eHealth Alliance (CeHA)-HIE, serves all major hospital systems in our region and is accessible to emergency department (ED) clinicians in those systems. We wanted to understand reasons for low CeHA-HIE utilization and explore options for improving it. METHODS: We implemented a 24-item user survey among ED clinician users of CeHA-HIE to investigate their perceptions of system usability and functionality, the quality of the information available through CeHA-HIE, the value of clinician time spent using CeHA-HIE, the ease of use of CeHA-HIE, and approaches for improving CeHA-HIE. RESULTS: Of the 231 ED clinicians surveyed, 51 responded, and among those, 48 reported having used CeHA-HIE and completed the survey. CONCLUSIONS: Results show most ED clinicians believed that CeHA-HIE was easy to use and added value to their work, but they also desired better integration of information available from CeHA-HIE into their system's electronic medical record.

Primer in health information exchange for the emergency physician: benefits and barriers.

For various reasons, patients seek care at different hospitals within a region, resulting in fragmented medical records at the point of care. In the emergency department, this is a particularly important issue because the emergency department provides open access to all patients and requires rapid high-stakes decision making to function well. To address these issues and as a result of federal initiatives, health information exchanges (HIEs) have been designed and implemented in various regions throughout the United States to promote health information sharing. The use of HIEs has been demonstrated to lower costs and avoid duplicative testing and treatment; however, obstacles such as physician usage characteristics and institutional concerns regarding information sharing exist and must be addressed before full implementation and adoption of HIEs among institutions take place. Further research is needed to describe the benefits of HIEs and how they can affect these barriers.

In-Hospital Cost Comparison of Triple-Rule-Out Computed Tomography Angiography Versus Standard of Care in Patients With Acute Chest Pain.

PURPOSE: The purpose of this study was to evaluate the utilization of invasive and noninvasive tests and compare cost in patients presenting with chest pain to the emergency department (ED) who underwent either triple-rule-out computed tomography angiography (TRO-CTA) or standard of care. MATERIALS AND METHODS: We performed a retrospective single-center analysis of 2156 ED patients who presented with acute chest pain with a negative initial troponin and electrocardiogram for myocardial injury. Patient cohorts matched by patient characteristics who had undergone TRO-CTA as a primary imaging test (n=1139) or standard of care without initial CTA imaging (n=1017) were included in the study. ED visits, utilization of tests, and costs during the initial episode of hospital care were compared. RESULTS: No significant differences in the diagnosis of coronary artery disease, pulmonary embolism, or aortic dissection were observed. Median ED waiting time (4.5 vs. 7.0 h, P<0.001), median total length of hospital stay (5.0 vs. 32.0 h, P<0.001), hospital admission rate (12.6% vs. 54.2%, P<0.001), and ED return rate to our hospital within 30 days (3.5% vs. 14.6%, P<0.001) were significantly lower in the TRO-CTA group. Moreover, reduced rates of additional testing and invasive coronary angiography (4.9% vs. 22.7%, P<0.001), and ultimately lower total cost per patient (11,783$ vs. 19,073$, P<0.001) were observed in the TRO-CTA group. CONCLUSIONS: TRO-CTA as an initial imaging test in ED patients presenting with acute chest pain was associated with shorter ED and hospital length of stay, fewer return visits within 30 days, and ultimately lower ED and hospitalization costs.

Impact of De-Identification on Clinical Text Classification Using Traditional and Deep Learning Classifiers.

Clinical text de-identification enables collaborative research while protecting patient privacy and confidentiality; however, concerns persist about the reduction in the utility of the de-identified text for information extraction and machine learning tasks. In the context of a deep learning experiment to detect altered mental status in emergency department provider notes, we tested several classifiers on clinical notes in their original form and on their automatically de-identified counterpart. We tested both traditional bag-of-words based machine learning models as well as word-embedding based deep learning models. We evaluated the models on 1,113 history of present illness notes. A total of 1,795 protected health information tokens were replaced in the de-identification process across all notes. The deep learning models had the best performance with accuracies of 95% on both original and de-identified notes. However, there was no significant difference in the performance of any of the models on the original vs. the de-identified notes.

Implementation of the HEART Pathway: Using the Consolidated Framework for Implementation Research.

OBJECTIVE: The HEART Pathway is an evidence-based decision tool for identifying emergency department (ED) patients with acute chest pain who are candidates for early discharge, to reduce unhelpful and potentially harmful hospitalizations. Guided by the Consolidated Framework for Implementation Research, we sought to identify important barriers and facilitators to implementation of the HEART Pathway. STUDY SETTING: Data were collected at 4 academic medical centers. STUDY DESIGN: We conducted semi-structured interviews with 25 key stakeholders (e.g., health system leaders, ED physicians). We conducted interviews before implementation of the HEART Pathway tool to identify potential barriers and facilitators to successful adoption at other regional academic medical centers. We also conducted postimplementation interviews at 1 medical center, to understand factors that contributed to successful adoption. DATA COLLECTION: Interviews were recorded and transcribed verbatim. We used a Consolidated Framework for Implementation Research framework-driven deductive approach for coding and analysis. PRINCIPAL FINDINGS: Potential barriers to implementation include time and resource burden, challenges specific to the electronic health record, sustained communication with and engagement of stakeholders, and patient concerns. Facilitators to implementation include strength of evidence for reduced length of stay and unnecessary testing and iatrogenic complications, ease of use, and supportive provider climate for evidence-based decision tools. CONCLUSIONS: Successful dissemination of the HEART Pathway will require addressing institution-specific barriers, which includes engaging clinical and financial stakeholders. New SMART-FHIR technologies, compatible with many electronic health record systems, can overcome barriers to health systems with limited information technology resources.

Prognostic value of CT-derived left atrial and left ventricular measures in patients with acute chest pain.

PURPOSE: To determine which left atrial (LA) and left ventricular (LV) parameters are associated with future major adverse cardiac event (MACE) and whether these measurements have independent prognostic value beyond risk factors and computed tomography (CT)-derived coronary artery disease measures. MATERIALS AND METHODS: This retrospective analysis was performed under an IRB waiver and in HIPAA compliance. Subjects underwent coronary CT angiography (CCTA) using a dual-source CT system for acute chest pain evaluation. LV mass, LV ejection fraction (EF), LV end-systolic volume (ESV) and LV end-diastolic volume (EDV), LA ESV and LA diameter, septal wall thickness and cardiac chamber diameters were measured. MACE was defined as cardiac death, non-fatal myocardial infarction, unstable angina, or late revascularization. The association between cardiac CT measures and the occurrence of MACE was quantified using Cox proportional hazard analysis. RESULTS: 225 subjects (age, 56.2±11.2; 140 males) were analyzed, of whom 42 (18.7%) experienced a MACE during a median follow-up of 13 months. LA diameter (HR:1.07, 95%CI:1.01-1.13permm) and LV mass (HR:1.05, 95%CI:1.00-1.10perg) remained significant prognostic factor of MACE after controlling for Framingham risk score. LA diameter and LV mass were also found to have prognostic value independent of each other. The other morphologic and functional cardiac measures were no significant prognostic factors for MACE. CONCLUSION: CT-derived LA diameter and LV mass are associated with future MACE in patients undergoing evaluation for chest pain, and portend independent prognostic value beyond traditional risk factors, coronary calcium score, and obstructive coronary artery disease.

Computed Tomography-Derived Parameters of Myocardial Morphology and Function in Black and White Patients With Acute Chest Pain.

Blacks have higher mortality and hospitalization rates because of congestive heart failure compared with white counterparts. Differences in cardiac structure and function may contribute to the racial disparity in cardiovascular outcomes. Our aim was to compare computed tomography (CT)-derived cardiac measurements between black patients with acute chest pain and age- and gender-matched white patients. We performed a retrospective analysis under an institutional review board waiver and in Health Insurance Portability and Accountability Act compliance. We investigated patients who underwent cardiac dual-source CT for acute chest pain. Myocardial mass, left ventricular (LV) ejection fraction, LV end-systolic volume, and LV end-diastolic volume were quantified using an automated analysis algorithm. Septal wall thickness and cardiac chamber diameters were manually measured. Measurements were compared by independent t test and linear regression. The study population consisted of 300 patients (150 black-mean age 54 ± 12 years; 46% men; 150 white-mean age 55 ± 11 years; 46% men). Myocardial mass was larger for blacks compared with white (176.1 ± 58.4 vs 155.9 ± 51.7 g, p = 0.002), which remained significant after adjusting for age, gender, body mass index, and hypertension. Septal wall thickness was slightly greater (11.9 ± 2.7 vs 11.2 ± 3.1 mm, p = 0.036). The LV inner diameter was moderately larger in black patients in systole (32.3 ± 9.0 vs 30.1 ± 5.4 ml, p = 0.010) and in diastole (50.1 ± 7.8 vs 48.9 ± 5.2 ml, p = 0.137), as well as LV end-diastolic volume (134.5 ± 42.7 vs 128.2 ± 30.6 ml, p = 0.143). Ejection fraction was nonsignificantly lower in blacks (67.1 ± 13.5% vs 69.0 ± 9.6%, p = 0.169). In conclusion, CT-derived myocardial mass was larger in blacks compared with whites, whereas LV functional parameters were generally not statistically different, suggesting that LV mass might be a possible contributing factor to the higher rate of cardiac events in blacks.

Impact of a health information exchange on resource use and Medicare-allowable reimbursements at 11 emergency departments in a midsized city.

INTRODUCTION: Use clinician perceptions to estimate the impact of a health information exchange (HIE) on emergency department (ED) care at four major hospital systems (HS) within a region. Use survey data provided by ED clinicians to estimate reduction in Medicare-allowable reimbursements (MARs) resulting from use of an HIE. METHODS: We conducted the study during a one-year period beginning in February 2012. Study sites included eleven EDs operated by four major HS in the region of a mid-sized Southeastern city, including one academic ED, five community hospital EDs, four free-standing EDs and 1 ED/Chest Pain Center (CPC) all of which participated in an HIE. The study design was observational, prospective using a voluntary, anonymous, online survey. Eligible participants included attending emergency physicians, residents, and mid-level providers (PA & NP). Survey items asked clinicians whether information obtained from the HIE changed resource use while caring for patients at the study sites and used branching logic to ascertain specific types of services avoided including laboratory/microbiology, radiology, consultations, and hospital admissions. Additional items asked how use of the HIE affected quality of care and length of stay. The survey was automated using a survey construction tool (REDCap Survey Software © 2010 Vanderbilt University). We calculated avoided MARs by multiplying the numbers and types of services reported to have been avoided. Average cost of an admission from the ED was based on direct cost trends for ED admissions within the region. RESULTS: During the 12-month study period we had 325,740 patient encounters and 7,525 logons to the HIE (utilization rate of 2.3%) by 231 ED clinicians practicing at the study sites. We collected 621 surveys representing 8.25% of logons of which 532 (85.7% of surveys) reported on patients who had information available in the HIE. Within this group the following services and MARs were reported to have been avoided [type of service: number of services; MARs]: Laboratory/Microbiology:187; $2,073, Radiology: 298; $475,840, Consultations: 61; $6,461, Hospital Admissions: 56; $551,282. Grand total of MARs avoided: $1,035,654; average $1,947 per patient who had information available in the HIE (Range: $1,491 - $2,395 between HS). Changes in management other than avoidance of a service were reported by 32.2% of participants. Participants stated that quality of care was improved for 89% of patients with information in the HIE. Eighty-two percent of participants reported that valuable time was saved with a mean time saved of 105 minutes. CONCLUSION: Observational data provided by ED clinicians practicing at eleven EDs in a mid-sized Southeastern city showed an average reduction in MARs of $1,947 per patient who had information available in an HIE. The majority of reduced MARs were due to avoided radiology studies and hospital admissions. Over 80% of participants reported that quality of care was improved and valuable time was saved.

Does targeted education of emergency physicians improve their comfort level in treating psychiatric patients?

INTRODUCTION: We determined if targeted education of emergency physicians (EPs) regarding the treatment of mental illness will improve their comfort level in treating psychiatric patients boarding in the emergency department (ED) awaiting admission. METHODS: We performed a pilot study examining whether an educational intervention would change an EP's comfort level in treating psychiatric boarder patients (PBPs). We identified a set of psychiatric emergencies that typically require admission or treatment beyond the scope of practice of emergency medicine. Diagnoses included major depression, schizophrenia, schizoaffective disorder, bipolar affective disorder, general anxiety disorder, suicidal ideation, and criminal behavior. We designed equivalent surveys to be used before and after an educational intervention. Each survey consisted of 10 scenarios of typical psychiatric patients. EPs were asked to rate their comfort levels in treating the described patients on a visual analogue scale. We calculated summary scores for the non intervention survey group (NINT) and intervention survey group (INT) and compared them using Student's t-test. RESULTS: Seventy-nine percent (33/42) of eligible participants completed the pre-intervention survey (21 attendings, 12 residents) and comprised the NINT group. Fifty-five percent (23/42) completed the post-intervention survey (16 attendings, 7 residents) comprising the INT group. A comparison of summary scores between 'NINT' and 'INT' groups showed a highly significant improvement in comfort levels with treating the patients described in the scenarios (P = 0.003). Improvements were noted on separate analysis for faculty (P = 0.039) and for residents (P = 0.012). Results of a sensitivity analysis excluding one highly significant scenario showed decreased, but still important differences between the NINT and INT groups for all participants and for residents, but not for faculty (all: P = 0.05; faculty: P = 0.25; residents: P = 0.03). CONCLUSION: This pilot study suggests that the comfort level of EPs, when asked to treat PBPs, may be improved with education. We believe our data support further study of this idea and of whether an improved comfort level will translate to a willingness to treat.