Emergency Department Record Research Articles

Harnessing Moderate-Sized Language Models for Reliable Patient Data De-identification in Emergency Department Records: An Evaluation of Strategies and Performance (Preprint)

Harnessing Moderate-Sized Language Models for Reliable Patient Data De-identification in Emergency Department Records: An Evaluation of Strategies and Performance (Preprint)

Air pollution related adverse respiratory health outcomes in California’s San Joaquin Valley: evidence from 2016 linked emergency department and hospital records

The San Joaquin Valley (SJV) of California has been consistently identified as having one of the highest levels of air pollution in the US. Despite federal and state standards, the SJV has been in non-attainment status for daily PM2.5 concentrations, extreme non-attainment for 8 hr O3 levels, and attainment for NO2. An epidemiological time-stratified case-crossover design was used to estimate the relationship between exposure to NO2, O3, PM2.5 and adverse health outcomes in asthma and upper respiratory infections (URIs). This study compared pollutant exposure effects for each case during limited time intervals and adjusted for seasonality. Elevated concentrations of three criteria outdoor air pollutants are associated with increased asthma and URI-related ED visits and hospitalizations in the SJV for all ages. NO2 exposure increased the odds of having an ED visit by 2.4% in lag 1 (95% CI: 1.017, 1.031). Lags 2, 3, 4, 5, 7, 9, and 14 were statistically significant. O3 modestly increased the odds of ED visits by 0.3% (95% CI:1.000, 1.006) after immediate exposure in the warm season. In the cold season, PM2.5 estimates were significant for all lags except for lags 4 and 12. The two-week lag increased the odds by 28% (95% CI:1.218, 1.345) for ED visits, and 16.5% (95% CI:1.009, 1.345) increased the odds of being hospitalized after cumulative exposure to PM2.5. Findings suggest that SJV residents experience adverse health effects due to elevated exposure to NO2 despite attainment of federal and state pollutant standards. This study provides new evidence about the effects of three criteria air pollutants and adverse health outcomes in the SJV region. The air quality regulatory and public health governing bodies should consider revisions to regional pollutant thresholds and local public health strategies to prevent adverse health outcomes during short and prolonged periods of air pollution exposure.

Using machine learning to mine mental health diagnostic groups from emergency department presentations before and during the COVID-19 pandemic

PurposeThe COVID-19 pandemic had a profound negative effect on mental health worldwide. The hospital emergency department plays a pivotal role in responding to mental health crises. Understanding data trends relating to hospital emergency department usage is beneficial for service planning, particularly around preparing for future pandemics. Machine learning has been used to mine large volumes of unstructured data to extract meaningful data in relation to mental health presentations. This study aims to analyse trends in five mental health-related presentations to an emergency department before and during, the COVID-19 pandemic.MethodsData from 690,514 presentations to two Australian, public hospital emergency departments between April 2019 to February 2022 were assessed. A machine learning-based framework, Mining Emergency Department Records, Evolutionary Algorithm Data Search (MEDREADS), was used to identify suicidality, psychosis, mania, eating disorder, and substance use.ResultsWhile the mental health-related presentations to the emergency department increased during the COVID-19 pandemic compared to pre-pandemic levels, the proportion of mental health presentations relative to the total emergency department presentations decreased. Several troughs in presentation frequency were identified across the pandemic period, which occurred consistently during the public health lockdown and restriction periods.ConclusionThis study implemented novel machine learning techniques to analyse mental health presentations to an emergency department during the COVID-19 pandemic. Results inform understanding of the use of emergency mental health services during the pandemic, and highlight opportunities to further investigate patterns in presentation.

Assessing the Correct Documentation of Time and Physician Information on Medical Records in the Emergency Department of Queen's Hospital: An Audit and Re-audit.

Medical records are confidential medical and legal documents describing a patient's contact with a healthcare facility. Thequality of documentation has been found to be lower in settings of high patient volume and complex cases, such as the emergency department (ED). The variety and number of healthcare professionals involved in the care of the patient also negatively affect the quality of documentation. The aim of this paper is to present the results of an audit and re-audit conducted in the ED of Queen's Hospital, Romford, to assess ED record documentation against General Medical Council (GMC) and Royal College of Physicians (RCP) standards. For the audit, all records of patients who were discharged from the ED of Queen's Hospital in one day were reviewed and evaluated on whether they have a date, time, the full name of the physician, their GMC number, and signature documented, as per GMC and RCP official guidelines. No medical information or patient data were recorded. After the implementation of the change aiming to raise awareness of ED staff, a new sample was collected two months later, and the same parameters were assessed against the set standards. Results of the audit showed a low percentage of documentation of all parameters, especially of GMC number and signature. After the presentation of the results and implementation of change, the results of the re-audit demonstrated significant raise in all percentages, with a relative improvement of 40% regarding the recording of GMC number and 65% regarding signature. However, the documentation of these two parameters remained low and below acceptable levels. The re-audit results underline that the low compliance was significantly improved after the implementation of measures aiming to increase correct documentation awareness among ED staff. However, to maintain and even raise the level of current practice, additionalsystematic measures need to be put into action.

147 When, What and How Frequently Do Patients Electronically Access Their Emergency Department Records After Discharge?

With the wide adoption of electronic health records (EHR), the 21st century CURES Act aimed to improve patient access and to their personal health information. This “open notes” initiative includes patient access to physician notes and other aspects of the medical record, including those from emergency department (ED) visits. We sought to assess how frequently patients access their ED record, what parts of the record are viewed, and the timing of access following ED discharge.

Age at first alcohol-related hospital separation or emergency department presentation and rate of re-admission: A retrospective data linkage cohort of young Australians.

Alcohol is a leading risk factor for death and disease in young people. We compare age-specific characteristics of young people who experience their first ('index') alcohol-related hospitalisation or emergency department (ED) presentation, and whether age at index predicts 12-month rates of readmission. We used a retrospective linked-data cohort of 10,300 people aged 12-20 years with an index alcohol-related hospital and/or ED record in New South Wales, Australia from 2005 to 2013. Age group (early adolescent [12-14 years], late adolescent [15-17 years], young adult [18-20 years]) and diagnosis fields were used in logistic regression analyses and to calculate incidence rates with adjustment for year of index event, sex, socioeconomic disadvantage and residence remoteness. People who experienced their index event in early adolescence (adjusted relative risk ratio [ARRR] 0.45 [95% confidence interval 0.39, 0.52]) or late adolescence (ARRR 0.82 [0.74, 0.90]) were less likely to be male compared to young adults. Early adolescents (ARRR 0.60 [0.51, 0.70]) and late adolescents (ARRR 0.84 [0.76, 0.93]) were less likely to have a hospitalisation index event. Early adolescents (adjusted incidence rate ratio 1.40 [1.15, 1.71]) and late adolescents (adjusted incidence rate ratio 1.16 [1.01, 1.34]) were more likely than young adults to have a subsequent 12-month non-poisoning injury ED presentation. We identified preventable hospital events in young people who have previously experienced an alcohol-related ED presentation or hospitalisation, with age-specific characteristics and outcomes that can be used to inform future health policy and service planning.

Mastering linked datasets: The future of emergency health care research

Objectives The aim of this work is to describe routine integration of prehospital emergency health records into a health master linkage file, delivering ongoing access to integrated patient treatment and outcome information for ambulance-attended patients in Queensland. Methods The Queensland Ambulance Service (QAS) data are integrated monthly into the Queensland Health master linkage file (MLF) using a linkage algorithm that relies on probabilistic matches in combination with deterministic rules based on patient demographic details, date, time, and facility identifiers. Each ambulance record is assigned an enduring linkage key (unique patient identifier) and further processing determines whether each record matches with a corresponding hospital emergency department, admission, or death registry record. In this study, all QAS electronic ambulance report form (eARF) records from October 2016 to December 2018 where at least one key linkage variable was present (n = 1,771,734) were integrated into the MLF. Results The majority of records (n = 1,456,502; 82.2%) were for transported patients, and 90.1% (n = 1,312,176) of these transports were to public hospital facilities. Of these transport records, 93.9% (n = 1,231,951) matched to ED records and 59.3% (n = 864,394) also linked to admitted patient records. Of ambulance non-transport records integrated into the MLF, 23.6% (n = 74,311) matched with ED records. Conclusion This study demonstrates robust linkage methods, quality assurance processes, and high linkage rates of data across the continuum of care (prehospital/emergency department/admitted patient/death) in Queensland. The resulting infrastructure provides a high-quality linked dataset that facilitates complex research and analysis to inform critical functions such as quality improvement, system evaluation, and design.

Assessment of Patient-Reported Naloxone Acquisition and Carrying With an Automated Text Messaging System After Emergency Department Discharge in Philadelphia

A central tenet of harm reduction and prevention of opioid overdose deaths is the distribution and use of naloxone. Patient-centered methods that investigate naloxone acquisition and carrying can guide opioid overdose education and naloxone distribution efforts. To assess patients' self-reported naloxone acquisition and carrying after an emergency department (ED) encounter using automated text messaging. This cohort study investigated self-reported patient behaviors involving naloxone after ED discharge in a large, urban academic health system in Philadelphia, Pennsylvania. Adult patients who were prescribed or dispensed naloxone and who had a mobile phone number listed in the electronic health record provided informed consent after ED discharge, and data were collected prospectively using text messaging from October 10, 2020, to March 19, 2021. Patients who did not respond to the survey or who opted out were excluded. Automated text message-based survey after ED discharge for patients who were prescribed or dispensed naloxone. The primary outcome was patient-reported naloxone acquisition, carrying, and use. Descriptive statistics were used to summarize patient demographic characteristics. Of 205 eligible patients, 41 (20.0%) completed the survey; of those patients, the mean (SD) age was 39.5 (13.7) years, and 21 (51.2%) were women. Fifteen (36.6%) had a personal history of being given naloxone after an overdose. As indicated by the ED record, 27 participants (65.9%) had naloxone dispensed in the ED, and 36 (87.8%) self-reported acquiring naloxone during or after their ED visit. Twenty-four participants (58.5%) were not carrying naloxone in the week before their ED visit. Twenty participants (48.8%) were carrying naloxone after the ED visit, and 27 (65.9%) reported planning to continue carrying naloxone in the future. Of the 24 individuals (58.5%) not carrying naloxone before their ED encounter, 13 (54.2%) reported planning to continue carrying naloxone in the future. In this cohort study of adult patients dispensed or prescribed naloxone from the ED, most reported acquiring naloxone on or after discharge. The ED remains a key point of access to naloxone for individuals at high risk of opioid use and overdose, and text messaging could be a method to engage and motivate patient-reported behaviors in enhancing naloxone acquisition and carrying.

Linking Ambulance Records with Hospital and Death Index Data to Evaluate Patient Outcomes.

ObjectiveLinkage of electronic administrative datasets is becoming increasingly common, offering a powerful resource for research and analysis. However, routine linkage of prehospital data with emergency department (ED) presentation and hospital admission datasets is rare. We describe a methodology used to link ambulance data with hospital ED presentations, admissions, and death records, and examine potential biases between matched and unmatched patients.MethodsIterative deterministic linkage methodologies were employed to link clinical, operational, and secondary triage ambulance data to ED presentations, hospital admissions, and death records in Victoria, Australia. Descriptive analyses and standardised differences were used to examine potential biases between matched and unmatched patients.ResultsA total of 2,813,913 ambulance records were available for linkage. Of the patients that were transported to a public ED (n=1,753,268), 83.3% matched with an ED record. Only small differences were observed between matched and unmatched patients for sex, year, time of day and attending crew type. The data elements with the largest standardised differences were patient age (0.25) and paramedic diagnosis (0.25). Matched patients were older (mean ± standard deviation: 55.6±25.7 vs 49.0±26.0 years) and more likely to have a paramedic-suspected cardiac, respiratory, neurological, or gastrointestinal/genitourinary condition, suspected infection/sepsis, or pain.ConclusionThis linked dataset will facilitate a large body of research into prehospital care and patient outcomes. Although future analysis of matched patients should acknowledge the linkage error rate, our findings suggest that results are likely to be generalisable to the broader ambulance population.

The Safety INdEx of Prehospital On Scene Triage (SINEPOST) study: the development and validation of a risk prediction model to support ambulance clinical transport decisions on-scene\u2014a protocol

BackgroundDemand for both the ambulance service and the emergency department (ED) is rising every year and when this demand is excessive in both systems, ambulance crews queue at the ED waiting to hand patients over. Some transported ambulance patients are ‘low-acuity’ and do not require the treatment of the ED. However, paramedics can find it challenging to identify these patients accurately. Decision support tools have been developed using expert opinion to help identify these low acuity patients but have failed to show a benefit beyond regular decision-making. Predictive algorithms may be able to build accurate models, which can be used in the field to support the decision not to take a low-acuity patient to an ED.Methods and analysisAll patients in Yorkshire who were transported to the ED by ambulance between July 2019 and February 2020 will be included. Ambulance electronic patient care record (ePCR) clinical data will be used as candidate predictors for the model. These will then be linked to the corresponding ED record, which holds the outcome of a ‘non-urgent attendance’. The estimated sample size is 52,958, with 4767 events and an EPP of 7.48. An XGBoost algorithm will be used for model development. Initially, a model will be derived using all the data and the apparent performance will be assessed. Then internal-external validation will use non-random nested cross-validation (CV) with test sets held out for each ED (spatial validation). After all models are created, a random-effects meta-analysis will be undertaken. This will pool performance measures such as goodness of fit, discrimination and calibration. It will also generate a prediction interval and measure heterogeneity between clusters. The performance of the full model will be updated with the pooled results.DiscussionCreating a risk prediction model in this area will lead to further development of a clinical decision support tool that ensures every ambulance patient can get to the right place of care, first time. If this study is successful, it could help paramedics evaluate the benefit of transporting a patient to the ED before they leave the scene. It could also reduce congestion in the urgent and emergency care system.Trial RegistrationThis study was retrospectively registered with the ISRCTN: 12121281

Low back pain presentations to New South Wales emergency departments: Trends over time and geographical variation.

To determine whether rates of ED presentations because of low back pain (LBP) have increased from 2016 to 2019 in New South Wales and map the geographical distribution of ED presentations because of LBP across New South Wales. We sourced data from the New South Wales Emergency Department Records for Epidemiology. We included all ED presentations aged 15 years and older with a diagnosis of LBP to the 178 public EDs across New South Wales from 1 January 2016 to 31 December 2019. We calculated the ratio (95% confidence interval) between the 2016 and 2019 age-standardised rates to determine whether an increase in the rate of ED LBP presentations has occurred. To assess geographical variation, we aggregated presentations by their home postcode. We calculated age-standardised rates per 100 000 person year for each of those areas using data from 2016 to 2019. We included 188 275 LBP presentations for patients aged 15 years or older. Their mean (standard deviation) age was 51.3 (20.0) years. From 2016 to 2019, we observed a 5.3% increase in the age-adjusted LBP ED presentation rates (age-standardised ratio 1.05, 95% confidence interval 1.04-1.06). We found a 20-fold variation in LBP ED age-standardised presentation rates across the different local government areas of New South Wales. Higher rates were mostly observed in rural and regional areas. The demand for ED services because of LBP has increased in New South Wales over time, and we observed a 20-fold variation in presentation rates across different regions.

Would Shared Health Visitor and Emergency Department Records Improve Recognition of Child Maltreatment within the Emergency Department? A Prospective Multicentre Study

Burns are common causes of paediatric emergency care attendance; approximately ten per cent result from maltreatment. Following emergency department (ED) attendance with a burn by 232 under five‐year‐olds, 11 risk factors for maltreatment were collected via health visitor (HV) telephone surveys. Three of these risk factors (domestic violence, social care involvement and developmental impairment) were also available in ED records, and information collected was compared between the two. Non‐parametric Fisher's exact tests were applied. Fifty‐nine per cent of children lived in families with risk factors for maltreatment. Prominent risk factors known by HVs included: prior injuries (n = 55, 23.7%), carer/parent mental health problems (n = 48, 20.7%), domestic violence (n = 47, 20.3%) and social care involvement (n = 45, 19.4%). A total of 158 cases had complete data for all 11 risk factors: 49 (31.0%) lived in households with one factor, 22 (13.9%) with two factors and 27 (17.1%) with three or more risk factors. In cases where HVs recorded the following risk factors as present, EDs recorded five of 47 (10.6%) for domestic violence, ten of 45 (22.2%) with social care involvement and four of 23 (17.4%) with developmental impairment. Many risk factors that were known to HVs were not identified by EDs staff despite being part of a standardised proforma. Maltreatment risk assessment could be improved if EDs staff had access to HV information.Key Practitioner Messages Fifty‐nine per cent of pre‐school children who attended an ED with a burn live in a family with one or more maltreatment risk factors, as identified by HV records. ED staff should be able to access a child's HV record electronically at the time of presentation, to identify known maltreatment risk factors. The lack of integration of community and acute setting health records is a barrier to comprehensive assessment and treatment decisions for children in EDs, especially in relation to safeguarding risks.

Serious infections in patients with myasthenia gravis: population-based cohort study.

To characterize the frequency and risk of serious infections in patients with myasthenia gravis (MG) relative to age/sex/area-matched comparators. This was a population-based cohort study in Ontario, Canada of patients with newly-diagnosed MG and 1:4 age/sex/area-matched general population comparators accrued from 1 April 2002 to 31 December 2015. The main outcome was a serious infection, defined by a primary diagnosis code on a hospitalization or emergency department record. We computed crude overall and sex-specific rates of infection among patients with MG and comparators, and the frequency of specific types of infection. Adjusted hazard ratios and 95% confidence intervals were estimated using Cox regression. Among 3823 patients with MG, 1275 (33.4%) experienced a serious infection compared with 2973/15292 (19.4%) of comparators over a mean follow-up of over 5years. Crude infection rates among patients with MG were twice those in comparators (72.5 vs. 35.0 per 1000 person-years, respectively). The most common infection types were respiratory infections, particularly bacterial pneumonia. After adjustment for potential confounders, MG was associated with a 39% increased infection risk (adjusted hazard ratio, 1.39; 95% confidence intervals, 1.28-1.51). Patients with MG are at a significantly higher absolute and relative risk of serious infections compared with age/sex/area-matched comparators. This needs to be considered when selecting MG treatments and when planning vaccination/prophylaxis. Determining whether this risk is due to the use of immunosuppressive medications (vs. MG itself) is an important area for future research.

Patient and health-care factors associated with potentially missed acute deterioration in primary care: a retrospective observational study of linked primary and secondary care data

BackgroundIn the UK, the majority of patient contact with health services occurs in primary care. Most of these contacts are uncomplicated; however, patient safety incidents (eg, failure to recognise patient deterioration) can occur. We aimed to explore patient and health-care factors associated with a self-referred admission, in patients with deteriorating health who consulted a general practitioner (GP). MethodsIn this observational study, we identified patients who had consulted a GP in the 3 days before an unplanned admission (indication of deterioration) between April 1, 2014, and Dec 31, 2017, in England, using the Clinical Practice Research Datalink with linkage to inpatient hospital admissions and emergency department data. We applied a multivariable, multilevel logistic regression model (generalised estimating equations) to investigate factors associated with self-referral (ie, patient age and existing health conditions, GP consultation, deteriorating health condition, and previous health service use) compared with other-referred unplanned admissions (eg, GP-referrals). Self-referred admission, as a composite measure, was defined as an unplanned admission via the emergency department (inpatient data) recorded as a self-referral in the corresponding emergency department record. We investigated all diagnoses and a subset of commonly reported missed conditions: sepsis, pulmonary embolism, urinary tract infections, and ectopic pregnancies in women. FindingsOf 405 878 unplanned admissions, 116 094 (28%) patients had contact with a GP 3 days before admission. The proportion of self-referred admissions varied by region (4189 [31%] of 13 639 inpatient admissions in London vs 1721 [12%] of 14 641 inpatient admissions in south west England), age, deteriorating health, and existing health conditions. Patients with sepsis or a urinary tract infection were more likely to self-refer than patients with other conditions (adjusted odds ratio [OR] 1·10, 95% CI 1·02–1·19 for sepsis; 1·09, 1·04–1·14, for urinary tract infection). GP appointment length was associated with a self-referred admission: a 5 min increase in consultation duration decreased the risk of self-referral by 6% (OR 0·94, 0·91–0·97). Telephone consultations, comorbidity, and previous health service use were also associated with self-referred admission. InterpretationDifferentiating deterioration from self-limiting conditions is difficult for GPs, particularly in patients with sepsis, urinary tract infections, or long-term conditions. The negative association between GP consultation duration and self-referral supports demand for longer GP consultations. However, more research is needed to investigate the underlying mechanism between GP consultation time and referral. FundingNational Institutes for Health Research.

Derivation and validation of a machine learning record linkage algorithm between emergency medical services and the emergency department.

Linking emergency medical services (EMS) electronic patient care reports (ePCRs) to emergency department (ED) records can provide clinicians access to vital information that can alter management. It can also create rich databases for research and quality improvement. Unfortunately, previous attempts at ePCR and ED record linkage have had limited success. In this study, we use supervised machine learning to derive and validate an automated record linkage algorithm between EMS ePCRs and ED records. All consecutive ePCRs from a single EMS provider between June 2013 and June 2015 were included. A primary reviewer matched ePCRs to a list of ED patients to create a gold standard. Age, gender, last name, first name, social security number, and date of birth were extracted. Data were randomly split into 80% training and 20% test datasets. We derived missing indicators, identical indicators, edit distances, and percent differences. A multivariate logistic regression model was trained using 5-fold cross-validation, using label k-fold, L2 regularization, and class reweighting. A total of 14032 ePCRs were included in the study. Interrater reliability between the primary and secondary reviewer had a kappa of 0.9. The algorithm had a sensitivity of 99.4%, a positive predictive value of 99.9%, and an area under the receiver-operating characteristic curve of 0.99 in both the training and test datasets. Date-of-birth match had the highest odds ratio of 16.9, followed by last name match (10.6). Social security number match had an odds ratio of 3.8. We were able to successfully derive and validate a record linkage algorithm from a single EMS ePCR provider to our hospital EMR.

Finding Chances to Intervene Before the Fatal Overdose: Linking ED and Mortality Data

ObjectiveLink emergency department (ED) with death certificate mortality data in order to examine the prior medical history of opioid overdose victims leading up to their death.IntroductionIn 2017, 951 Missouri residents died from an opioid overdose—a record number for the state.1 This continues the trend from 2016, which saw an increase of over 30% in opioid overdose deaths compared to 2015. The Missouri Department of Health and Senior Services (MDHSS) manages several public health surveillance data sources that can be used to inform about the opioid epidemic. Opioid overdose deaths are identified through death certificates which are collected through the vital records system. MDHSS also manages the Patient Abstract System (PAS), which contains ED and inpatient hospitalization data from approximately 132 non-federal Missouri hospitals. PAS contains about 130 variables, which include demographic data, diagnoses codes, procedures codes, and other visit information. Records can have up to 23 diagnosis fields, which are coded using ICD-10-CM (International Classification of Diseases, Clinically Modified). The first diagnosis field is the primary reason for a visit.MethodsLinkage and analysis of the data was performed using SAS Enterprise Guide 6.1. Opioid overdose deaths were identified through ICD-10 analysis looking for drug poisoning underlying cause of death codes and opioid-specific codes found in the multiple cause (contributing cause) of death fields. Table 1, below, summarizes the ICD-10 codes used. Mortality data from the 951 decedents were linked to ED data from 2016 and 2017. Records were linked using multiple passes over the ED records. Records were first linked on social security number. Following this linkage, ED records with no initial match went through a second pass and linked on name and date of birth. Finally, a third pass for records still without a match was conducted using date of birth, census tract, and sex. After these passes, the linkages were reviewed to identify any false positives. The 23 diagnosis fields contained in PAS were analyzed to look for patterns in diagnosis coding. ICD-10-CM codes were too broad so CCS (Clinical Classifications Software) categories were utilized.ResultsIn total, 3,500 ED records were linked to the 951 decedents. After removing false positives, the total number of ED records was 3,357. Approximately 70% (687) of decedents were linked to at least one ED record. One hundred and eighty-eight visits were due to drug overdose (153 opioid overdoses). The most common primary diagnosis CCS categories (category numbers in parentheses) were: substance-related disorders (661), Spondylosis; intervertebral disc disorders; other back problems (205), abdominal pain (251), and other nervous system disorders (95). Collectively, these four categories represented over 20% of all primary diagnoses. Across all 23 diagnosis fields there were similar results. The most common CCS categories were as follows: substance-related disorders (661), other aftercare (257), essential hypertension (98), and mood disorders (657). Pie charts (Fig. 1 and 2) below show proportions of CCS categories across all diagnoses fields and primary diagnosis broken into three major categories: pain/injury, substance abuse/mental health, and other. In order to reduce the impact of CCS categories with small numbers, these graphics represent only CCS categories that made up 1% or more of the total collection of diagnoses codes. Of the 687 decedents that were matched successfully to ED records, 96% had at least one pain/injury or one substance abuse/mental health ICD-CM code in at least one record, and 68% had both.ConclusionsThese findings suggest that many overdose decedents visited the ED in the years prior to death. Many of these visits were not due to an overdose; however, they could be indicative of a problem with opioids (i.e. pain, drug-seeking, substance use-related). ED staff and public health professionals could utilize these opportunities to refer patients to recovery services and recommend they heed caution when using opioids.

Usability Evaluation of Visual Representation Formats for Emergency Department Records.

Background Integration of electronic information is a challenge for multitasking emergency providers, with implications for patient safety. Visual representations can assist sense-making of complex data sets; however, benefit and acceptability in emergency care is unproven. Objectives This article evaluates visually focused alternatives to lists or tabular formats, to better understand possible usability in Emergency Department Information System (EDIS). Methods A counterbalanced, repeated-measures experiment, satisfaction surveys, and narrative content analysis was conducted remotely by Web platform. Participants were 37 American emergency physicians; they completed 16 clinical cases comparing 4 visual designs to the control formats from a commercially available EDIS. They then evaluated two additional chart overview representations without controls. Results Visual designs provided benefit in several areas compared to controls. Task correctness (90% to 76%; p = 0.003) and completion time (median: 49–74 seconds; p < 0.001) were superior for a medication history timeline with class and schedule highlighting. Completion time (median: 45–60 seconds; p = 0.03) was superior for a past medical history design, using pertinent diagnosis codes in highlighting rules. Less mental effort was reported for visual allergy ( p = 0.04), past medical history ( p < 0.001), and medication timeline ( p < 0.001) designs. Most of the participants agreed with statements of likeability, preference, and benefit for visual designs; nonetheless, contrary opinions were seen, and more complex designs were viewed less favorably. Conclusion Physician performance with visual representations of clinical data can in some cases exceed standard formats, even in absence of training. Highlighting of priority clinical categories was rated easier-to-use on average than unhighlighted controls. Perceived complexity of timeline representations can limit desirability for a subset of users, despite potential benefit.

Analysis of Unstructured Text-Based Data Using Machine Learning Techniques: The Case of Pediatric Emergency Department Records in Nicaragua.

Free-text information is still widely used in emergency department (ED) records. Machine learning techniques are useful for analyzing narratives, but they have been used mostly for English-language data sets. Considering such a framework, the performance of an ML classification task of a Spanish-language ED visits database was tested. ED visits collected in the EDs of nine hospitals in Nicaragua were analyzed. Spanish-language, free-text discharge diagnoses were considered in the analysis. Five-hundred random forests were trained on a set of bootstrap samples of the whole data set (1,789 ED visits) to perform the classification task. For each one, after having identified optimal parameter value, the final validated model was trained on the whole bootstrapped data set and tested. The classification accuracies had a median of 0.783 (95% CI [0.779, 0.796]). Machine learning techniques seemed to be a promising opportunity for the exploitation of unstructured information reported in ED records in low- and middle-income Spanish-speaking countries.

The association between insurance status and diagnostic imaging for acute abdominal pain among emergency department patients in the United States, 2005–2014

Introduction and objectives: Acute abdominal pain (AAP) is one of the most common complaints in the emergency department (ED). Rapid diagnosis is essential and is often achieved through imaging. Computed tomography (CT) is widely considered an exemplary test in the diagnosis of AAP in adult patients. As previous studies show disparities in healthcare treatment based on insurance status, our objective was to assess the association between insurance status and frequency of CT ordered for adult patients presenting to the ED with AAP from 2005 to 2014.Methods: This study used the National Hospital and Ambulatory Medical Care Survey: Emergency Department Record (NHAMCS) database, which collects data over a randomly assigned 4 week period in the 50 states and DC, to perform an observational retrospective analysis of patients presenting to the ED with AAP. Patients with Medicaid, Medicare or no insurance were compared to patients with private insurance. The association between insurance status and frequency of CT ordered was measured by obtaining odds ratios along with 95% CIs adjusted for age, gender and race/ethnicity.Results: Individuals receiving Medicaid are 20% less likely to receive CT than those with private insurance (OR 0.8, CI 0.6–0.99, p = .046). Those on Medicare or who are uninsured have no difference in odds of obtaining a CT scan compared to patients with private insurance. Additional findings are that black patients are 42% less likely to receive a CT scan than white patients.Conclusions and implications: Patients on Medicaid are significantly less likely to receive a CT when presenting to the ED with AAP. Differences in diagnostic care may correlate to inferior health outcomes in patients without private insurance.

Abstract TP396: Accuracy of Administrative Hospital, Emergency Department and Death Records

Background: Hospital, emergency presentation and death databases are frequently used to research outcomes after stroke. The reliability of diagnostic coding for cardiovascular disease (CVD) in these administrative data remains uncertain. We aimed to determine the reliability of these data in Australia using an existing clinical trial cohort. Methods: Patients with stroke/TIA who participated in the Shared Team Approach between Nurses and Doctors For Improved Risk factor Management (STAND FIRM) trial (n = 563, recruited from 4 hospitals within Victoria). We used diagnostic ICD-10-AM coded data from hospital, emergency department and death databases within 2 years after stroke/TIA. Medical records for these potential CVD-related events were reviewed by two independent stroke specialists and adjudicated according to strict criteria. We then estimated sensitivity and specificity of using either primary or both primary and secondary diagnoses fields (obtained for all adjudicated records), against the events adjudicated by the specialists (gold standard). False positives were CVD-events defined by ICD-10-AM diagnostic codes that were adjudicated as not being a CVD-event. False negatives were true CVD-events that were misclassified as not being CVD-related when using ICD-10-AM codes. Results: We identified 261 events for medical review. After adjudication, 65 were classified as CVD-events (cases) and 196 were not CVD-events. Using both primary and secondary diagnoses, 55 true positives were correctly identified among the cases (sensitivity = 84.6%) and 129 true negatives among the non-cases (specificity = 65.8%). Using only primary diagnoses, 48 true positives were identified (sensitivity = 73.9%) and 171 true negatives (specificity = 87.2%). Using both primary and secondary diagnoses had an increased sensitivity, but decreased specificity (area under the Receiver Operating Characteristic curve (AUC) = 0.75; 95% CI, 0.70, 0.81) when compared with using only primary diagnoses (AUC = 0.81; 95% CI, 0.75, 0.86; p = 0.028). Conclusion: Both primary and secondary diagnoses should be used to identify true CVD-events and minimise misclassifying these in administrative databases.