BACKGROUND: Heart failure (HF) with preserved or mildly reduced ejection fraction includes a heterogenous group of patients. Reclassification into distinct phenogroups to enable targeted interventions is a priority. This study aimed to identify distinct phenogroups, and compare phenogroup characteristics and outcomes, from electronic health record data.
METHODS: 2,187 patients admitted to five UK hospitals with a diagnosis of HF and a left ventricular ejection fraction ≥ 40% were identified from the NIHR Health Informatics Collaborative database. Partition-based, model-based, and density-based machine learning clustering techniques were applied. Cox Proportional Hazards and Fine-Gray competing risks models were used to compare outcomes (all-cause mortality and hospitalisation for HF) across phenogroups.
RESULTS: Three phenogroups were identified: (1) Younger, predominantly female patients with high prevalence of cardiometabolic and coronary disease; (2) More frail patients, with higher rates of lung disease and atrial fibrillation; (3) Patients characterised by systemic inflammation and high rates of diabetes and renal dysfunction. Survival profiles were distinct, with an increasing risk of all-cause mortality from phenogroups 1 to 3 (p < 0.001). Phenogroup membership significantly improved survival prediction compared to conventional factors. Phenogroups were not predictive of hospitalisation for HF.
CONCLUSIONS: Applying unsupervised machine learning to routinely collected electronic health record data identified phenogroups with distinct clinical characteristics and unique survival profiles.

OBJECTIVE: To describe our methods to compare patient-reported symptoms of acute myeloid leukemia and the corresponding documentation by healthcare providers in the electronic health record.
BACKGROUND: Patients with acute myeloid leukemia experience many distressing symptoms, particularly related to chemotherapy. The timely recognition and provision of evidence-based interventions to manage these symptoms can improve outcomes. However, lack of standardized formatting for symptom documentation within electronic health records leads to challenges for clinicians when accessing and comprehending patients\' symptom information, as it primarily exists in narrative forms in various parts of the electronic health record. This variability raises concerns about over- or under-reporting of symptoms. Consistency between patient-reported symptoms and clinician\'s symptom documentation is important for patient-centered symptom management, but little is known about the degree of agreement between patient reports and their documentation. This is a detailed description of the study\'s methodology, procedures and design to determine how patient-reported symptoms are similar or different from symptoms documented in electronic health records by clinicians.
METHODS: Exploratory, descriptive study.
METHODS: Forty symptoms will be assessed as patient-reported outcomes using the modified version of the Memorial Symptom Assessment Scale. The research team will annotate symptoms from the electronic health record (clinical notes and flowsheets) corresponding to the 40 symptoms. The degree of agreement between patient reports and electronic health record documentation will be analyzed using positive and negative agreement, kappa statistics and McNemar\'s test.
CONCLUSIONS: We present innovative methods to comprehensively compare the symptoms reported by acute myeloid leukemia patients with all available electronic health record documentation, including clinical notes and flowsheets, providing insights into symptom reporting in clinical practice.
CONCLUSIONS: Findings from this study will provide foundational understanding and compelling evidence, suggesting the need for more thorough efforts to assess patients\' symptoms. Methods presented in this paper are applicable to other symptom-intensive diseases.

BACKGROUND: The medical knowledge graph provides explainable decision support, helping clinicians with prompt diagnosis and treatment suggestions. However, in real-world clinical practice, patients visit different hospitals seeking various medical services, resulting in fragmented patient data across hospitals. With data security issues, data fragmentation limits the application of knowledge graphs because single-hospital data cannot provide complete evidence for generating precise decision support and comprehensive explanations. It is important to study new methods for knowledge graph systems to integrate into multicenter, information-sensitive medical environments, using fragmented patient records for decision support while maintaining data privacy and security.
OBJECTIVE: This study aims to propose an electronic health record (EHR)-oriented knowledge graph system for collaborative reasoning with multicenter fragmented patient medical data, all the while preserving data privacy.
METHODS: The study introduced an EHR knowledge graph framework and a novel collaborative reasoning process for utilizing multicenter fragmented information. The system was deployed in each hospital and used a unified semantic structure and Observational Medical Outcomes Partnership (OMOP) vocabulary to standardize the local EHR data set. The system transforms local EHR data into semantic formats and performs semantic reasoning to generate intermediate reasoning findings. The generated intermediate findings used hypernym concepts to isolate original medical data. The intermediate findings and hash-encrypted patient identities were synchronized through a blockchain network. The multicenter intermediate findings were collaborated for final reasoning and clinical decision support without gathering original EHR data.
RESULTS: The system underwent evaluation through an application study involving the utilization of multicenter fragmented EHR data to alert non-nephrology clinicians about overlooked patients with chronic kidney disease (CKD). The study covered 1185 patients in nonnephrology departments from 3 hospitals. The patients visited at least two of the hospitals. Of these, 124 patients were identified as meeting CKD diagnosis criteria through collaborative reasoning using multicenter EHR data, whereas the data from individual hospitals alone could not facilitate the identification of CKD in these patients. The assessment by clinicians indicated that 78/91 (86%) patients were CKD positive.
CONCLUSIONS: The proposed system was able to effectively utilize multicenter fragmented EHR data for clinical application. The application study showed the clinical benefits of the system with prompt and comprehensive decision support.

Electronic health records (EHR) data provides the researcher and physician with the opportunity to improve risk prediction by employing newer, more sophisticated modeling techniques. Rather than treating the impact of predictor variables on health trajectories as static, we explore the use of time-dependent variables in dynamically modeling time-to-event data through the use of landmarking (LM) data sets. We compare several different dynamic models presented in the literature that utilize LM data sets as the basis of their approach. These techniques include using pseudo-means, pseudo-survival probabilities, and the traditional Cox model. The models are primarily compared with their static counterparts using appropriate measures of model discrimination and calibration based on what summary measure is employed for the response variable.

Motivated by improving the prediction of the human immunodeficiency virus (HIV) suppression status using electronic health records (EHR) data, we propose a functional multivariable logistic regression model, which accounts for the longitudinal binary process and continuous process simultaneously. Specifically, the longitudinal measurements for either binary or continuous variables are modeled by functional principal components analysis, and their corresponding functional principal component scores are used to build a logistic regression model for prediction. The longitudinal binary data are linked to underlying Gaussian processes. The estimation is done using penalized spline for the longitudinal continuous and binary data. Group-lasso is used to select longitudinal processes, and the multivariate functional principal components analysis is proposed to revise functional principal component scores with the correlation. The method is evaluated via comprehensive simulation studies and then applied to predict viral suppression using EHR data for people living with HIV in South Carolina.

BACKGROUND: Data is increasingly used for improvement and research in public health, especially administrative data such as that collected in electronic health records. Patients enter and exit these typically open-cohort datasets non-uniformly; this can render simple questions about incidence and prevalence time-consuming and with unnecessary variation between analyses. We therefore developed methods to automate analysis of incidence and prevalence in open cohort datasets, to improve transparency, productivity and reproducibility of analyses.
METHODS: We provide both a code-free set of rules for incidence and prevalence that can be applied to any open cohort, and a python Command Line Interface implementation of these rules requiring python 3.9 or later.
UNASSIGNED: The Command Line Interface is used to calculate incidence and point prevalence time series from open cohort data. The ruleset can be used in developing other implementations or can be rearranged to form other analytical questions such as period prevalence.
BACKGROUND: The command line interface is freely available from https://github.com/THINKINGGroup/analogy_publication .

UNASSIGNED: Accurately identifying clinical phenotypes from Electronic Health Records (EHRs) provides additional insights into patients\' health, especially when such information is unavailable in structured data. This study evaluates the application of OpenAI\'s Generative Pre-trained Transformer (GPT)-4 model to identify clinical phenotypes from EHR text in non-small cell lung cancer (NSCLC) patients. The goal was to identify disease stages, treatments and progression utilizing GPT-4, and compare its performance against GPT-3.5-turbo, Flan-T5-xl, Flan-T5-xxl, Llama-3-8B, and 2 rule-based and machine learning-based methods, namely, scispaCy and medspaCy.
UNASSIGNED: Phenotypes such as initial cancer stage, initial treatment, evidence of cancer recurrence, and affected organs during recurrence were identified from 13 646 clinical notes for 63 NSCLC patients from Washington University in St. Louis, Missouri. The performance of the GPT-4 model is evaluated against GPT-3.5-turbo, Flan-T5-xxl, Flan-T5-xl, Llama-3-8B, medspaCy, and scispaCy by comparing precision, recall, and micro-F1 scores.
UNASSIGNED: GPT-4 achieved higher F1 score, precision, and recall compared to Flan-T5-xl, Flan-T5-xxl, Llama-3-8B, medspaCy, and scispaCy\'s models. GPT-3.5-turbo performed similarly to that of GPT-4. GPT, Flan-T5, and Llama models were not constrained by explicit rule requirements for contextual pattern recognition. spaCy models relied on predefined patterns, leading to their suboptimal performance.
UNASSIGNED: GPT-4 improves clinical phenotype identification due to its robust pre-training and remarkable pattern recognition capability on the embedded tokens. It demonstrates data-driven effectiveness even with limited context in the input. While rule-based models remain useful for some tasks, GPT models offer improved contextual understanding of the text, and robust clinical phenotype extraction.

BACKGROUND: No routinely recommended cardiovascular disease (CVD) risk prediction equations have adjusted for CVD preventive medications initiated during follow-up (treatment drop-in) in their derivation cohorts. This will lead to underestimation of risk when equations are applied in clinical practice if treatment drop-in is common. We aimed to quantify the treatment drop-in in a large contemporary national cohort to determine whether equations are likely to require adjustment.
METHODS: Eight de-identified individual-level national health administrative datasets in Aotearoa New Zealand were linked to establish a cohort of almost all New Zealanders without CVD and aged 30-74 years in 2006. Individuals dispensing blood-pressure-lowering and/or lipid-lowering medications between 1 July 2006 and 31 December 2006 (baseline dispensing), and in each 6-month period during 12 years\' follow-up to 31 December 2018 (follow-up dispensing), were identified. Person-years of treatment drop-in were determined.
RESULTS: A total of 1 399 348 (80%) out of the 1 746 695 individuals in the cohort were not dispensed CVD medications at baseline. Blood-pressure-lowering and/or lipid-lowering treatment drop-in accounted for 14% of follow-up time in the group untreated at baseline and increased significantly with increasing predicted baseline 5-year CVD risk (12%, 31%, 34% and 37% in <5%, 5-9%, 10-14% and ≥15% risk groups, respectively) and with increasing age (8% in 30-44 year-olds to 30% in 60-74 year-olds).
CONCLUSIONS: CVD preventive treatment drop-in accounted for approximately one-third of follow-up time among participants typically eligible for preventive treatment (≥5% 5-year predicted risk). Equations derived from cohorts with long-term follow-up that do not adjust for treatment drop-in effect will underestimate CVD risk in higher risk individuals and lead to undertreatment. Future CVD risk prediction studies need to address this potential flaw.

BACKGROUND: COVID-19 disrupted consulting behaviour, healthcare delivery and cancer diagnostic services. This study quantifies the cancer incidence coded in UK general practice electronic health records and deviations from historical trends after the March 2020 national lockdown. For comparison, we study the coded incidence of type-2 diabetes mellitus, which is diagnosed almost entirely within primary care.
METHODS: Poisson interrupted time series models investigated the coded incidence of diagnoses in adults aged ≥ 18 years in the Clinical Practice Research Datalink before (01/03/2017-29/02/2020) and after (01/03/2020-28/02/2022) the first lockdown. Datasets were stratified by age, sex, and general practice per 28-day aggregation period. Models captured incidence changes associated with lockdown, both immediately and over time based on historical trends.
RESULTS: We studied 189,457 incident cancer and 191,915 incident diabetes records in 1480 general practices over 52,374,197 person-years at risk. During 01/03/2020-28/02/2022, there were fewer incident records of cancer (n = 22,199, 10.49 %, 10.44-10.53 %) and diabetes (n = 15,709, 7.57 %, 7.53-7.61 %) than expected. Within cancers, impacts ranged from no effect (e.g. unknown primary, pancreas, and ovary), to small effects for lung (n = 773, 3.11 %, 3.09-3.13 % fewer records) and female breast (n = 2686, 6.77 %, 6.73-6.81 %), to the greatest effect for bladder (n = 2874, 31.15 %, 31.00-31.31 %). Diabetes and cancer records recovered maximally to 86 % (95 %CI 80.3-92.7 %) and 74 % (95 %CI 70.3-78.6 %) in July 2021 and May 2021, respectively, of their expected values, declining again until the study end.
CONCLUSIONS: The \"missing\" cancer and diabetes diagnoses in primary care may comprise delayed or missed diagnoses, reduced incidence associated with excess deaths from COVID-19, and potentially increased non-coded recording of diagnoses. Future validation studies must quantify the concordance between primary care and National Cancer Registration Data and Hospital Episode Statistics over the pandemic era.

暂无摘要。