Electronic Patient Data Research Articles

Using electronic health records in assessing the prevalence of hypertension in Kyrgyzstan

Abstract Background Non-communicable diseases are an increasing challenge in Kyrgyzstan as well as in other parts of Central Asia. However, there is very little epidemiological data on disease burden in the country. The development of electronic health records in primary health care provides an opportunity to assess the disease incidence and prevalence among patient population. Methods All patients with hypertension (HT) diagnoses (ICD-10 I10-I15) were identified from electronic health records used in primary health care in Kyrgyzstan from years 2021, 2022 and 2023 from all 7 regions and Bishkek and Osh cities (n = 132621-152743). Each year, all patients diagnosed with HT during that year or in previous years were included, while those who had passed away were excluded. The incidence and prevalence of patients was calculated using as denominator both the patients who had visited the services as well as registered patients. Results The prevalence of patients with hypertension calculated using patients with service use as denominator varied by region between 10.2% to 14.0% and was pretty stable during the observed period. The prevalence was the lowest in Chui region and the highest in Bishkek city. When the prevalence was calculated using the registered patient population the prevalence was only between 1.5% to 3.1%. Conclusions The observed prevalences were notably lower compared with data from WHO Steps surveys, which have reported a hypertension prevalence more than 40% among individuals aged 25 to 64. It is known that only approximately 25% of the population in Kyrgyzstan utilize primary healthcare services, with maternity care comprising the largest portion. This factor, coupled with cases that go undiagnosed, could potentially account for the observed low prevalence in electronic patient data. However, the electronic patient records provide useful information on diagnostics and disease burden in the country that can be used in developing the services. Key messages • Electronic health records have developed quickly in Kyrgyzstan and enable extraction of patient data from primary health care. • The low use of primary health care services complicates the use of register data in assessing the disease prevalences.

SATURN: assessing the feasibility of utilising existing registries for real-world evidence data collection to meet patients, regulatory, health technology assessment and payer requirements

BackgroundSATURN (Systematic Accumulation of Treatment practices and Utilisation, Real world evidence, and Natural history data) for the rare condition osteogenesis imperfecta (OI) has the objective to create a common core dataset by utilising existing, well-established data sources to meet the needs of the various stakeholders (physicians, registry/dataset owners, patients and patient associations, OI community leaders, European [EU] policymakers, regulators, health technology assessments [HTA]s, and healthcare systems including payers). This paper describes the steps taken to assess the feasibility of one existing OI registry (i.e., the Registry of OI [ROI]) as a candidate for SATURN. The same methodology will be applied to other existing OI registries in the future and this same concept could be utilised for other rare disease registries.MethodsThe approach to assessing the feasibility of the ROI registry consisted of three steps: (1) an assessment of the registry characteristics using the Registry Evaluation and Quality Standards Tool (REQueST); (2) a gap analysis comparing SATURN required Core Variables to those being captured in the registry’s Case Report Form (CRF); and (3) a compliance check on the data exchange process following the Title 21 of Code of Federal Regulations (CFR) Part 11/EudraLex Annex 11 Compliance Checklist. The first registry that SATURN has assessed is the ROI database at the Istituto Ortopedico Rizzoli (IOR) in Italy.ResultsThe results from the ROI REQueST have demonstrated satisfactory complete responses in terms of methodology, essential standards, interpretability, and interoperability—readiness for data linkage, data sources, and ethics to meet the needs of data customers. However, the ROI data is from a tertiary referral centre which may limit the ability to understand the full patient journey. The gap analysis has revealed that an exact or logical match between SATURN requested variables and the ROI current variables exists for the following items: patient characteristics, treatment of OI (medical and surgical) and treatment of pain (with the exception of frequency of treatment and reasons for discontinuation), fracture history and bone density. However, data on safety was missing. The compliance check has implied that the ROI implemented appropriate controls for the web-based platform (i.e., Genotype–phenotype Data Integration Platform [GeDI]) that is involved in processing the electronic patient data, and GeDI is a validated/compliant application that follows relevant 21 CFR Part 11/EudraLex Annex 11 regulations.ConclusionsThis robust feasibility process highlights potential limitations and opportunities to develop and to refine the collaboration with the ROI as the SATURN programme moves forward. It also ensures that the existing datasets in the rare condition OI are being maximised to respond to the needs of patients, data customers and decision-makers. This feasibility process has allowed SATURN to build a compliant methodology that aligns with the requirements from the European Medicines Agency (EMA) and HTAs. More data variables will continue to be developed and refined along the way with more registries participating in SATURN. As a result, SATURN will become a meaningful and truly collaborative core dataset, which will also contribute to advancing understanding of OI diagnosis, treatment, and care.

P-685 An artificial intelligence based medication regimen recommendation model for optimizing the treatment outcomes of controlled ovarian stimulation

Abstract Study question Can a machine learning (ML) algorithm suggest an optimal controlled ovarian stimulation (COS) protocol for maximizing the number of retrieved and mature (MII) oocytes? Summary answer Based on historical electronic medical record patient data, using ML algorithm based medication regimen recommendation model may increase the number of retrieved oocytes. What is known already The application of artificial intelligence in the field of reproductive medicine is mainly focused on the image analysis of gamete and embryo, and prediction of treatment outcomes. The research on the optimization of COS process is limited. Previous studies mainly focused on one or several types of COS protocols such as GnRH-a protocol and GnRH-A protocol. And most of them aimed to assist single-node decision-making during COS process such as optimal gonadotropins starting dose and trigger day. Study design, size, duration The data used for developing this model consists of 38,355 COS cycles from women who underwent their first IVF/ICSI treatment in a large fertility center between January 2008 to December 2022, of which 33,853 were used as the model training set and 4,502 were used as the test set. This dataset contains a subset of 8,382 cycles with information about MII oocytes. Participants/materials, setting, methods The goal of model was to recommend the optimal COS protocol based on optimizing COS outcomes. The similarity of patient’s characteristic was used as model input. XGBoost algorithm was constructed to calculate the similarity of COS outcomes. The accuracy was compared with the patient similarity calculation method of KNN+Euclidean distance and KNN+cosine distance using mean absolute error (MAE). The performance of different models in improving the average number of retrieved and MII oocytes was compared. Main results and the role of chance The patient similarity calculation method based on XGBoost had lowest MAE when predicting the number of retrieved oocytes (XGBoost vs KNN+Euclidean distance+k=500 vs KNN+Euclidean distance+k=1000 vs KNN+cosine distance+k=500 vs KNN+cosine distance+k=1000: 3.729 vs 3.810 vs 3.815 vs 3.785 vs 3.788). Including AMH could improve the accuracy of all recommendation models in predicting protocols and the number of retrieved oocytes. When the optimal number of MII oocytes is used as the protocol recommended target, the patient similarity calculation method based on XGBoost had lowest MAE (XGBoost vs KNN+Euclidean distance+k=500 vs KNN+Euclidean distance+k=1000 vs KNN+cosine distance+k=500 vs KNN+cosine distance+k=1000: 2.868 vs 3.397 vs 3.452 vs 3.356 vs 3.395). The COS medication regimen recommended by the patient similarity calculation method based on COS treatment outcomes has resulted in higher numbers of retrieved oocytes in patients compared to those who did not adopt recommendation (P < 0.01). The medication regimen recommended model based on KNN+Euclidean distance algorithm (k = 500) has the best performance in improving the average number of retrieved oocytes（adopt recommendation vs not adopt recommendation: 13.01±6.61 vs 10.09±6.67, P < 0.01）There was no statistical difference in the number of MII oocytes between patients receiving the recommendation and those not adopt recommendation. Limitations, reasons for caution The medication regimen recommendation model is based solely on the predicted number of retrieved and MII oocytes, not including adverse treatment outcomes such as ovarian hyperstimulation syndrome. Moreover, the database used to train and validate the model is from a single fertility center. Wider implications of the findings The recommendation model is helpful to improve the number of retrieved oocytes. Based on the recommended COS protocol, further mining the medication content, medication time distribution and medication sequence transformation from the similar patient set can assist clinicians to develop individualized COS medication regimen for infertile patients. Trial registration number not applicable

How do we leverage data through replication and warehousing to meet blood collection and transfusion service needs.

With the widespread adoption of Blood Establishment Computer Systems and other Blood Collection and Transfusion Service (BCTS) clinical information systems (CIS), electronic blood donor, product, and patient data are now routinely required for clinical, regulatory, operational, and quality needs. That data are often not readily accessible for such secondary use within CIS databases, particularly for applications with significant data availability requirements such as machine learning and artificial intelligence. Data replication provides one avenue by which CIS data can be made more readily available. Members of the AABB's Information Systems Committee along with institutional information technology colleagues provided a multi-institutional viewpoint on data replication through the lens of BCTS specific use cases. Case studies of informatics offerings leveraging such technologies were also elicited. Six distinct use cases describe the potential role of data replication including the creation of data warehouses for frontline laboratory staff. Specific BCTS examples for each use case are presented to highlight the value of data replication, including visualization of critical inventory (O red blood cells, HLA-compatible platelets) and utilization analytics for patient blood management. Two case studies describe the approach to implement such technologies to (1) optimize staffing via laboratory workload reporting and (2) improve access to blood via antigen-negative blood product location services. Data replication and warehousing can empower BCTS analytic offerings not otherwise natively available through one's CIS to improve patient care and laboratory operations.

The effect of vasopressor agents on pressure injury development in intensive care patients

BackgroundVasopressors are life-saving agents that increase mean arterial pressure. The pharmacodynamic features of these agents and previous studies suggest that vasopressors may be an essential risk factor in developing pressure injuries. ObjectiveThis study aimed to examine the effect of vasopressors in medical-surgical intensive care patients on pressure injury development. Design and settingsThis retrospective and correlational study was conducted between March 2021- May 2022. The electronic patient data were obtained from 148 surgical and medical patients exposed to vasopressor agents in the intensive care unit. Data on patients' demographic and clinical characteristics were evaluated using descriptive statistical methods (number, percentage, mean, standard deviation). Logistic regression modelling was used to assess independent relationships with pressure injury risk; results are reported as odds ratios (OR) and 95% confidence intervals (CI). ResultsAll patients were given norepinephrine agents, and dopamine infusion secondary to norepinephrine was found in only 28.3 % of patients (n = 42). Pressure injury incidence was 43.2 % (n = 64). Duration of norepinephrine infusion was recognized as an independent risk factor for ICU-acquired pressure injury development (OR 1.22, 95 % CI 1.11–1.35), while a medical admission diagnosis (instead of surgical) was protective against pressure injury risk (OR 0.24, 95 % CI 0.10–0.59). ConclusionThis study indicated that duration of norepinephrine infusion is a significant risk factor for pressure injury development. Implications for clinical practiceAlthough norepinephrine use cannot be avoided entirely, its administration may be an early warning for nurses to increase pressure injury prevention strategies. Skin evaluation should be performed more frequently, and preventive strategies should be implemented meticulously.This study was registered on clincialtrials.gov (Identifier: NCT06163352)

A systematic evaluation of the performance of GPT‐4 and PaLM2 to diagnose comorbidities in MIMIC‐IV patients

AbstractBackgroundGiven the strikingly high diagnostic error rate in hospitals, and the recent development of Large Language Models (LLMs), we set out to measure the diagnostic sensitivity of two popular LLMs: GPT‐4 and PaLM2. Small‐scale studies to evaluate the diagnostic ability of LLMs have shown promising results, with GPT‐4 demonstrating high accuracy in diagnosing test cases. However, larger evaluations on real electronic patient data are needed to provide more reliable estimates.MethodsTo fill this gap in the literature, we used a deidentified Electronic Health Record (EHR) data set of about 300,000 patients admitted to the Beth Israel Deaconess Medical Center in Boston. This data set contained blood, imaging, microbiology and vital sign information as well as the patients' medical diagnostic codes. Based on the available EHR data, doctors curated a set of diagnoses for each patient, which we will refer to as ground truth diagnoses. We then designed carefully‐written prompts to get patient diagnostic predictions from the LLMs and compared this to the ground truth diagnoses in a random sample of 1000 patients.ResultsBased on the proportion of correctly predicted ground truth diagnoses, we estimated the diagnostic hit rate of GPT‐4 to be 93.9%. PaLM2 achieved 84.7% on the same data set. On these 1000 randomly selected EHRs, GPT‐4 correctly identified 1116 unique diagnoses.ConclusionThe results suggest that artificial intelligence (AI) has the potential when working alongside clinicians to reduce cognitive errors which lead to hundreds of thousands of misdiagnoses every year. However, human oversight of AI remains essential: LLMs cannot replace clinicians, especially when it comes to human understanding and empathy. Furthermore, a significant number of challenges in incorporating AI into health care exist, including ethical, liability and regulatory barriers.

Acceptance of Digital Health Technologies in Palliative Care Patients.

Digital health technologies have potential to transform palliative care (PC) services. The global aging population poses unique challenges for PC, which digital health technologies may help overcome. Evaluation of attitudes and perceptions combined with quantification of prior use habits favor an understanding of psychological barriers to PC patient acceptance of digital health technologies including artificial intelligence (AI). We aimed to evaluate the attitudes and perceptions of PC patients regarding a broad range of digital health technologies used in their routine monitoring and treatment and identify barriers to use. We used a 39-item questionnaire to evaluate acceptance and use of smartphone-based electronic patient report outcome measures, wearables, AI, data privacy, and virtual reality (VR) in 29 female and male PC inpatients. A majority of patients indicated an interest in (69.0%) and positive attitude toward (75.9%) digital health technologies. Nearly all (93.1%) patients believe that digital health technologies will become more important in medicine in the future. Most patients would consider using their smartphone (79.3%) or wearable (69.0%) more often for their health. The most feasible technologies were smartphones, wearables, and VR. Barriers to acceptance included unfamiliarity, data security, errors in data interpretation, and loss of personal interaction through AI. In this patient survey, acceptance of new technologies in a PC patient population was high, encouraging its use also at the end-of-life.

The Iowa Health Data Resource (IHDR): an innovative framework for transforming the clinical health data ecosystem.

This manuscript will be of interest to most Clinical and Translational Science Awards (CTSA) as they retool for the increasing emphasis on translational science from translational research. This effort is an extension of the EDW4R work that most CTSAs have done to deploy infrastructure and tools for researchers to access clinical data. The Iowa Health Data Resource (IHDR) is a strategic investment made by the University of Iowa to improve access to real-world health data. The goals of IHDR are to improve the speed of translational health research, to boost interdisciplinary collaboration, and to improve literacy about health data. The first objective toward this larger goal was to address gaps in data access, data literacy, lack of computational environments for processing Personal Health Information (PHI) and the lack of processes and expertise for creating transformative datasets. A three-pronged approach was taken to address the objective. The approach involves integration of an intercollegiate team of non-informatics faculty and staff, a data enclave for secure patient data analyses, and novel comprehensive datasets. To date, all five of the health science colleges (dentistry, medicine, nursing, pharmacy, and public health) have had at least one staff and one faculty member complete the two-month experiential learning curriculum. Over the first two years of this project, nine cohorts totaling 36 data liaisons have been trained, including 18 faculty and 18 staff. IHDR data enclave eliminated the need to duplicate computational infrastructure inside the hospital firewall which reduced infrastructure, hardware and human resource costs while leveraging the existing expertise embedded in the university research computing team. The creation of a process to develop and implement transformative datasets has resulted in the creation of seven domain specific datasets to date. The combination of people, process, and technology facilitates collaboration and interdisciplinary research in a secure environment using curated data sets. While other organizations have implemented individual components to address EDW4R operational demands, the IHDR combines multiple resources into a novel, comprehensive ecosystem IHDR enables scientists to use analysis tools with electronic patient data to accelerate time to science.

Traceable Research Data Sharing in a German Medical Data Integration Center With FAIR (Findability, Accessibility, Interoperability, and Reusability)-Geared Provenance Implementation: Proof-of-Concept Study.

Secondary investigations into digital health records, including electronic patient data from German medical data integration centers (DICs), pave the way for enhanced future patient care. However, only limited information is captured regarding the integrity, traceability, and quality of the (sensitive) data elements. This lack of detail diminishes trust in the validity of the collected data. From a technical standpoint, adhering to the widely accepted FAIR (Findability, Accessibility, Interoperability, and Reusability) principles for data stewardship necessitates enriching data with provenance-related metadata. Provenance offers insights into the readiness for the reuse of a data element and serves as a supplier of data governance. The primary goal of this study is to augment the reusability of clinical routine data within a medical DIC for secondary utilization in clinical research. Our aim is to establish provenance traces that underpin the status of data integrity, reliability, and consequently, trust in electronic health records, thereby enhancing the accountability of the medical DIC. We present the implementation of a proof-of-concept provenance library integrating international standards as an initial step. We adhered to a customized road map for a provenance framework, and examined the data integration steps across the ETL (extract, transform, and load) phases. Following a maturity model, we derived requirements for a provenance library. Using this research approach, we formulated a provenance model with associated metadata and implemented a proof-of-concept provenance class. Furthermore, we seamlessly incorporated the internationally recognized Word Wide Web Consortium (W3C) provenance standard, aligned the resultant provenance records with the interoperable health care standard Fast Healthcare Interoperability Resources, and presented them in various representation formats. Ultimately, we conducted a thorough assessment of provenance trace measurements. This study marks the inaugural implementation of integrated provenance traces at the data element level within a German medical DIC. We devised and executed a practical method that synergizes the robustness of quality- and health standard-guided (meta)data management practices. Our measurements indicate commendable pipeline execution times, attaining notable levels of accuracy and reliability in processing clinical routine data, thereby ensuring accountability in the medical DIC. These findings should inspire the development of additional tools aimed at providing evidence-based and reliable electronic health record services for secondary use. The research method outlined for the proof-of-concept provenance class has been crafted to promote effective and reliable core data management practices. It aims to enhance biomedical data by imbuing it with meaningful provenance, thereby bolstering the benefits for both research and society. Additionally, it facilitates the streamlined reuse of biomedical data. As a result, the system mitigates risks, as data analysis without knowledge of the origin and quality of all data elements is rendered futile. While the approach was initially developed for the medical DIC use case, these principles can be universally applied throughout the scientific domain.

Sociodemographic Inequities in Telemedicine Use Among US Patients Initiating Treatment in Community Cancer Centers During the Ongoing COVID-19 Pandemic, 2020-2022.

Although telemedicine was seen as a way to improve cancer care during the coronavirus disease (COVID-19) pandemic, there is limited information regarding inequities in its uptake. This study assessed sociodemographic factors associated with telemedicine use among patients initiating treatment for 20 common cancers. This retrospective cohort study used deidentified electronic health record-derived patient data from a nationwide network of community cancer practices, linked to area-level Census information. We included adults (age 18 years and older) who initiated first-line systemic cancer treatment between March 2020 and December 2022 (follow-up through March 2023). Exposures include race/ethnicity, insurance status, and area-level social determinants of health (eg, block group socioeconomic status [SES]). The outcome was telemedicine use within 90 days after treatment initiation. Associations were examined using logistic regression models adjusted for age, sex, performance status, stage, and cancer type. This study included 36,993 patients (48.6% women; median age, 69 years), of whom 15.1% used telemedicine services. Black (12.2%; odds ratio [OR], 0.78 [95% CI, 0.70 to 0.88]) and uninsured (9.2%; OR, 0.59 [95% CI, 0.48 to 0.73]) patients were less likely to use telemedicine services than their White and well-insured counterparts (14.5% and 15.0%, respectively). Patients in rural (9.7%; OR, 0.54 [95% CI, 0.46 to 0.57]), suburban (11.8%; OR, 0.67 [95% CI, 0.61 to 0.74]), and low SES areas (9.9%; OR, 0.39 [95% CI, 0.35 to 0.43]) were less also likely to use telemedicine than their counterparts in urban (16.6%) or high SES (21.6%) areas. Nearly one sixth of patients initiating cancer treatment during the pandemic used telemedicine, but there were substantial inequities. The proliferation of telemedicine may perpetuate cancer care inequities if marginalized populations do not have equitable access.

Hereditary Hemorrhagic Telangiectasia May be the Most Clinically Significant and Morbid Inherited Bleeding Disorder of Women

Background Hereditary hemorrhagic telangiectasia (HHT) is the second most common inherited bleeding disorder (1 in 5000 people) behind von Willebrand disease (VWD) and has no FDA- or EMA-approved therapies. HHT causes serious bleeding resulting in severe iron deficiency anemia, major psychosocial complications, and visceral arteriovenous malformations in brain, lung, and liver that can cause life-threatening hemorrhage, stroke, and other morbid complications. While HHT prevalence is equal in men and women, manifestations may be worse in women (as evidenced by a female predominance in HHT clinical trials). Due to the focus on hemophilia (an X-linked disease nearly exclusive to men) in the hemostasis research community, no study has examined the relative morbidity of the most common inherited bleeding disorders in women: HHT and VWD. Methods We performed an observational cohort study of women with HHT or VWD cared for at our institutional Comprehensive Hemophilia and Von Willebrand Disease Treatment Center or HHT Center of Excellence. Using an electronic patient data registry, a representative sample of 100 randomly selected women with HHT were age-matched to 100 randomly selected women with VWD for analysis. All data was extracted and confirmed using manual chart review; we quantified and compared bleeding and healthcare utilization outcomes. Results Patient Population. The mean (range) age in both groups was 49 (14-87) years. Mean BMI was 27 in both groups, and racial/ethnic breakdown was similar in both groups. Mean follow-up durations were 9.07 years (HHT) and 14.83 years (VWD) per patient. The VWD group included 86 patients with type 1 VWD and 14 with type 2 VWD. Bleeding Outcomes. Bleeding outcomes are detailed in TABLE 1. In women with HHT compared with VWD, recurrent epistaxis and GI bleeding were more likely (odds ratio [OR] [95% CI] = 32.73 [13.81-71.80], P<0.0001 and 5.69 [2.59-12.89], P<0.0001, respectively), and heavy menstrual bleeding was less likely (OR 0.32 [0.18-0.57], P<0.0001). Iron deficiency anemia was significantly more likely, and lowest hemoglobin significantly lower, in HHT versus VWD. Requirements for IV iron and RBC transfusion were significantly more likely in HHT versus VWD. Women with HHT had 17-fold higher odds of iron infusion dependence and 8-fold higher odds of requiring hemostatic surgical procedures than women with VWD. Healthcare Utilization for HHT or VWD Management. Healthcare utilization is detailed in TABLE 2. In women with HHT compared with VWD, emergency department (ED) visits and hospital admissions specifically to manage disease complications were significantly more likely (OR [95% CI] = 3.83 [1.67-8.48], P=0.001 and 7.33 [3.57-14.73], P<0.0001, respectively). Rates of ED visitation, admission, and outpatient encounters specifically for HHT or VWD management were significantly higher in women with HHT vs. women with VWD: ED visits, 10.3 per 100 patient-years (pt-yrs) vs. 0.74 per 100 pt-yrs, P=0.001; admissions, 14.4 per 100 pt-yrs vs. 1.0 per 100 pt-yrs, P<0.0001; outpatient encounters, 386.5 per 100 pt-yrs vs. 43.7 per 100 pt-yrs, P<0.0001. Adjusting these rates for the prevalence of HHT (1 in 5000) and VWD (1 in 1000) at the population level, all rates were still higher in the HHT group. Infusion visits, radiology encounters, and nearly all subspecialty provider visit types were much higher in women with HHT ( TABLE 2). Conclusions Analyzing matched populations of women at a health system that serves as a tertiary referral center for both diseases, HHT, a historically neglected bleeding disorder, was responsible for much higher morbidity and healthcare utilization in women than VWD, certainly at the individual patient level and possibly at the population level (prevalence-adjusted). Overall incidence of disease complications in each group was consistent with the published literature for both HHT and VWD, supporting the generalizability of our findings. These findings are critically important in achieving gender equity in the hemostasis field, given that both government and industry funding for VWD research is many times greater than for HHT research and such funding for hemophilia research, a disease of men, is at least 10-fold greater than for HHT research. Continued neglect of funding for clinical care and research of HHT within the hemostasis academic community and industry will only sustain and exacerbate gender inequities in hematology.

Excellent agreement of Norwegian trauma registry data compared to corresponding data in electronic patient records

BackgroundThe Norwegian Trauma Registry (NTR) is designed to monitor and improve the quality and outcome of trauma care delivered by Norwegian trauma hospitals. Patient care is evaluated through specific quality indicators, which are constructed of variables reported to the registry by certified registrars. Having high-quality data recorded in the registry is essential for the validity, trust and use of data. This study aims to perform a data quality check of a subset of core data elements in the registry by assessing agreement between data in the NTR and corresponding data in electronic patient records (EPRs).MethodsWe validated 49 of the 118 variables registered in the NTR by comparing those with the corresponding ones in electronic patient records for 180 patients with a trauma diagnosis admitted in 2019 at eight public hospitals. Agreement was quantified by calculating observed agreement, Cohen’s Kappa and Gwet’s first agreement coefficient (AC1) with 95% confidence intervals (CIs) for 27 nominal variables, quadratic weighted Cohen’s Kappa and Gwet’s second agreement coefficient (AC2) for five ordinal variables. For nine continuous, one date and seven time variables, we calculated intraclass correlation coefficient (ICC).ResultsAlmost perfect agreement (AC1 /AC2/ ICC > 0.80) was observed for all examined variables. Nominal and ordinal variables showed Gwet’s agreement coefficients ranging from 0.85 (95% CI: 0.79–0.91) to 1.00 (95% CI: 1.00–1.00). For continuous and time variables there were detected high values of intraclass correlation coefficients (ICC) between 0.88 (95% CI: 0.83–0.91) and 1.00 (CI 95%: 1.00–1.00). While missing values in both the NTR and EPRs were in general negligeable, we found a substantial amount of missing registrations for a continuous “Base excess” in the NTR. For some of the time variables missing values both in the NTR and EPRs were high.ConclusionAll tested variables in the Norwegian Trauma Registry displayed excellent agreement with the corresponding variables in electronic patient records. Variables in the registry that showed missing data need further examination.

O026 The surgical datathon - a high-yield learning model for students: a mixed methods analysis

Abstract Introduction Datathons are high-yield research events where cross-disciplinary researchers collaborate to investigate pertinent healthcare problems by analysing multimodal electronic patient data. This paper aims to demonstrate the value of a surgical datathon as a learning model in students using both objective and qualitative assessments. Methods Fifteen individuals were recruited for the 24-hour datathon with representation from medicine, psychology, and clinical neurosciences including two postgraduate neurosurgery faculty. All reported interest in learning more academic skills. The aim was to solve a pertinent neurosurgical challenge, namely the prediction of intracranial pressure based on brain MR-imaging, with the goal of completing a full manuscript within 24 hours. Participants were trained and split into smaller open working groups, supervised allowing real-time specialist feedback. Groups intermittently merged to cross-collaborate. Surveys with mixed-methods analysis explored participant opinions and the value of developing research skills at multiple time points. Results Participants’ knowledge were tested showing an increased mean test score post-datathon (69.92%) compared to pre-datathon scores (57.31%). Wilcoxon signed-rank test results demonstrated a significant change for these results (p=0.04, 95% CI -27.21 - -0.04). Likert-scales showed that majority of participants strongly agreed that the datathon accelerated their productivity (58.3%). Data collection was rated particularly useful (100%) despite majority being less than confident in analysing neuro-radiological images (84.6%) pre-datathon. Time pressures were the largest barrier participants faced (47.1%). Conclusion A full manuscript was produced and the datathon goal was met with conference acceptance. Surgical datathons represent a pragmatic learning model for students of all backgrounds.

Resuscitation room management of patients with non-traumatic critical illness in the emergency department (OBSERvE-DUS-study)

BackgroundFew studies address the care of critically ill non-traumatic patients in the emergency department (ED). The aim of this study was to assess the epidemiology, management, and outcome of these patients.MethodsIn this retrospective study, we identified and analyzed data from all consecutive adult critically ill non-traumatic ED patients treated from March 2018 to February 2019. Patient characteristics, major complaint leading to admission, out-of-hospital, and in-hospital interventions and 30-day mortality were extracted from medical records of the electronic patient data management system.ResultsDuring the study period, we analyzed 40,764 patients admitted to the ED. Of these, 621 (1.5%) critically ill non-traumatic patients were admitted for life-threatening emergencies to the resuscitation room (age: 70 ± 16 years, 52% male). Leading problem on admission was disability/unconsciousness (D), shock (C), respiratory failure (B), airway obstruction (A), and environment problems (E) in 41%, 31%, 25%, 2%, and 1%, respectively. Out-of-hospital and in-hospital measures included: intravenous access (61% vs. 99%), 12-lead ECG (55% vs. 87%), invasive airway management (21% vs. 34%) invasive ventilation (21% vs. 34%), catecholamines (9% vs. 30%), arterial access (0% vs. 52%), and cardiopulmonary resuscitation (11% vs. 6%). The underlying diagnoses were mainly neurological (29%), followed by cardiological (28%), and pulmonological (20%) emergencies. The mean length of stay (LOS) in the resuscitation room and ED was 123 ± 122 and 415 ± 479 min, respectively. The 30-day mortality was 18.5%.ConclusionThe data describe the care of critically ill non-traumatic patients in the resuscitation room. Based on these data, algorithms for the structured care of critically ill non-traumatic patients need to be developed.

Type 1 diabetes management: Room for improvement.

Optimal diabetes care and risk factor management are important to delay micro- and macrovascular complications in individuals with type 1 diabetes (T1D). Ongoing improvement of management strategies requires the evaluation of target achievement and identification of risk factors in individuals who do (or do not) achieve these targets. Cross-sectional data were collected from adults with T1D visiting six diabetes centers in the Netherlands in 2018. Targets were defined as glycated hemoglobin (HbA1c) <53 mmol/mol, low-density lipoprotein-cholesterol (LDL-c) <2.6 mmoL/L (no cardiovascular disease [CVD] present) or <1.8 mmoL/L (CVD present), or blood pressure (BP) <140/90 mm Hg. Target achievement was compared for individuals with and without CVD. Data from 1737 individuals were included. Mean HbA1c was 63 mmol/mol (7.9%), LDL-c was 2.67 mmoL/L, and BP 131/76 mm Hg. In individuals with CVD, 24%, 33%, and 46% achieved HbA1c, LDL-c, and BP targets respectively. In individuals without CVD these percentages were 29%, 54%, and 77%, respectively. Individuals with CVD did not have any significant risk factors for HbA1c, LDL-c, and BP target achievement. In comparison, individuals without CVD were more likely to achieve glycemic targets if they were men and insulin pump users. Smoking, microvascular complications, and the prescription of lipid-lowering and antihypertensive medication were negatively associated with glycemic target achievement. No characteristics were associated with LDL-c target achievement. Microvascular complications and antihypertensive medication prescription were negatively associated with BP target attainment. Opportunities for improvement of diabetes management exist for the achievement of glycemic, lipid, and BP targets but may differ between individuals with and without CVD.

Exploratory pharmacovigilance with machine learning in big patient data: A focused scoping review.

Machine learning can operationalize the rich and complex data in electronic patient records for exploratory pharmacovigilance endeavours. The objective of this review is to identify applications of machine learning and big patient data in exploratory pharmacovigilance. We searched PubMed and Embase and included original articles with an exploratory pharmacovigilance purpose, focusing on medicinal interventions and reporting the use of machine learning in electronic patient records with ≥1000 patients collected after market entry. Of 2557 studies screened, seven were included. Those covered six countries and were published between 2015 and 2021. The most prominent machine learning methods were random forests, logistic regressions, and support vector machines. Two studies used artificial neural networks or naive Bayes classifiers. One study used formal concept analysis for association mining, and another used temporal difference learning. Five studies compared several methods against each other. The numbers of patients in most data sets were in the order of thousands; two studies used what can more reasonably be considered big data with >1 000 000 patients records. Despite years of great aspirations for combining machine learning and clinical data for exploratory pharmacovigilance, only few studies still seem to deliver somewhat on these expectations.

Type-2 Neutrosophic Set and Their Applications in Medical Databases Deadlock Resolution

Electronic patient data gives many advantages, but also new difficulties. Deadlocks may delay procedures like acquiring patient information. Distributed deadlock resolution solutions introduce uncertainty due to inaccurate transaction properties. Soft computing-based solutions have been developed to solve this challenge. In a single framework, ambiguous, vague, incomplete, and inconsistent transaction attribute information has received minimal attention. The work presented in this paper employed type-2 neutrosophic logic, an extension of type-1 neutrosophic logic, to handle uncertainty in real-time deadlock-resolving systems. The proposed method is structured to reflect multiple types of knowledge and relations among transactions’ features that include validation factor degree, slackness degree, degree of deadline-missed transaction based on the degree of membership of truthiness, degree of membership of indeterminacy, and degree of membership of falsity. Here, the footprint of uncertainty (FOU) for truth, indeterminacy, and falsity represents the level of uncertainty that exists in the value of a grade of membership. We employed a distributed real-time transaction processing simulator (DRTTPS) to conduct the simulations and conducted experiments using the benchmark Pima Indians diabetes dataset (PIDD). As the results showed, there is an increase in detection rate and a large drop in rollback rate when this new strategy is used. The performance of Type-2 neutrosophic-based resolution is better than the Type-1 neutrosophic-based approach on the execution ratio scale. The improvement rate has reached 10% to 20%, depending on the number of arrived transactions.

619: THE IMPACT OF NEIGHBORHOOD FACTORS ON PATIENT DISEASE SEVERITY ON PRESENTATION TO THE PICU

Introduction: Health disparities yield worse health outcomes for critically ill patients. Neighborhoods capture social and physical attributes experienced by residents. Increasingly, neighborhood-level geographic variations in health disparities have been reported. Individuals are commonly clustered into neighborhoods based on socioeconomic status. The impact of sociodemographic factors on outcomes for PICU patients is understudied. Few studies delve into the specific neighborhood characteristics affecting disease severity on PICU admission. To address this gap, we conducted a retrospective study exploring the association between geography and disease severity on PICU presentation, with sociodemographic neighborhood data and clinical data as additional predictors. We anticipate patients with higher disease severity will be clustered in low opportunity neighborhoods. Methods: Electronic medical record patient data were collected for non-elective admissions to the PICU from Jan 2013 to Dec 2019. The peak Pediatric Logistic Organ Dysfunction-2 (PELOD-2) score for the first 72 hours of admission was calculated for each encounter. Addresses were geocoded into census tracts. The Childhood Opportunity Index (COI), Social Vulnerability Index (SVI), and Neighborhood Disorder Index (NDI) were matched to each census tract. We report descriptive analyses, Spearman correlation coefficients for continuous variables, and Kruskal-Wallis test results for categorical variables. Results: There were 8497 unique patient encounters for 5874 distinct patients. Data were skewed right with a PELOD-2 score mean of 3.2 and median of 3.0. There was no correlation between the SVI (p=0.129), COI (p=0.957) or NDI (p=0.363) and peak PELOD-2 score. There was a correlation between age (rho 0.102, p< 0.001) and number of comorbidities (rho 0.373, p< 0.001) and peak PELOD-2 scores. There was a difference between median PELOD-2 scores for varying races and insurance types (p< 0.001). Conclusions: There was no association between neighborhood resources and peak PELOD-2 scores; however, there was a difference in mean PELOD-2 scores across varying races and insurance types. This indicates a need for more analyses to evaluate the association between neighborhood indices and peak PELOD-2 scores when accounting for other sociodemographic variables.

Identification and Characterization of a Retrospective Cohort of Secondary Hemophagocytic Lymphohistiocytosis (sHLH) Patients in the US

Introduction: Hemophagocytic lymphohistiocytosis (HLH) is a rare and life-threatening hyperinflammatory syndrome induced by aberrantly activated myeloid and T cells. HLH may be primary (i.e., due to genetic mutations) which typically presents in children, or secondary (i.e., triggered by an infection, malignancy, or rheumatologic condition) which is most common in adults. Without treatment, HLH is universally fatal; even with existing treatment, overall mortality ranges from 50-75%. Most of our understanding of HLH comes from studies in the pediatric population, and limited information is available on the diagnosis and treatment of patients with secondary HLH (sHLH). The aim of this study is to compile a historical cohort of sHLH patients from large US medical systems to better understand and refine the sHLH clinical definition, describe the epidemiology (including incidence, prevalence, and natural history of the disease), and characterize demographic, clinical, treatment characteristics, survival, and healthcare resource utilization (HRU) of the cohort compared with a matched comparison group. Methods: A historical cohort study was designed to identify sHLH patients aged ≥12 years from de-identified Electronic Medical Record (EMR) systems at health care centers in the US. The EMR systems track patients longitudinally over multiple years and are linked at the patient level by a unique identifier that is consistent across service locations and time. Three different methods (Figure 1) and chart reviews were used to enable identification of all possible sHLH patients and to better understand initial recorded diagnosis, differences in treatment and outcomes, and to determine a simpler system for accurate identification of sHLH cases in the healthcare system. As sHLH is a rare disorder that can be complex to diagnose and treat, these algorithms were developed to identify patients that may or may not have been assigned an ICD code for HLH. Approximately 5.4 million unique patient records were reviewed. Comparison of patients without sHLH were also identified from the EMR systems, matched on birth year, sex, and county of residence. Results: Patient electronic data from 2009-2021 were reviewed using all three methods. Of >800,000 unique inpatient records, 903 unique de-duplicated patients were identified for chart review. 57 patients were identified using HLH-related ICD codes (Method 1). An additional 846 patients were identified using HLH-related treatments (dexamethasone and etoposide) and elevated ferritin (≥1,000 to <10,000 ng/mL) in patient charts (Method 2) and/or by elevated ferritin (≥10,000 ng/mL) or sCD25 lab test ordered (with cohorts stratified by sCD25 lab values ≤3,900 U/mL and >3,900 U/mL) (Method 3). Potential HLH patient charts were then adjudicated via clinician review using HLH-2004 criteria and clinical judgment. Conclusion: Since sHLH is a rare, heterogenous, and complex disease to diagnosis and treat, HLH-related ICD codes (Method 1) may not be sufficient to capture the spectrum of sHLH patients. Both HLH-related treatments in charts (Method 2) and elevated ferritin and sCD25 values (Method 3) could prove useful in identifying additional sHLH patients. Moreover, Method 3 could show promise as a simplified strategy for identifying and diagnosing patients with sHLH. This process could inform the clinical definition and true burden of sHLH, particularly those patients who never received an sHLH diagnosis, as well as the clinical, demographic, and treatment characteristics and outcomes of this unique patient population. Figure 1. Flow Diagram of Patient Eligibility and MethodsFigure 1View largeDownload PPTFigure 1View largeDownload PPT Close modal

35. The development, testing, and validation of an automatic LIMS import method for NGS myeloid panel results

For oncology genetic testing, there is a critical need for patient results to be readily available and accessible for the ordering pathologist's review. After an initial genetic testing workup, emergency treatment or follow-up testing may be required, thus accessibility of results is a key feature in providing patient-focused laboratory testing. Ideally, genetic testing results would be incorporated into a LIMS data management system in order to have a permanent record of testing and to allow for the results to be easily accessible, searchable, and transferable by the clinical or pathology team working on behalf of the patient. For maximum impact, Next Generation Sequencing (NGS) reports should be electronically integrated into the patient's electronic health record. NGS reports uploaded as PDF attachments constrain health record searches. Additionally, data in PDF reports often are not able to be transferred to a comprehensive pathology report and may require cumbersome manual entry. Our team has recently completed a project to create an automatic import method for our Myeloid Complete NGS Panel report into our electronic patient data management system. Our data management system is interfaced with our client's LIMS platform enabling the direct and automatic transfer of orders coming in as well as delivery of the resultant NGS report into the patients' health records. We will present the steps taken to create, test, troubleshoot, and validate the automatic upload system. We will also detail critical components of the NGS LIMS report and challenges that were addressed during this project.