Electronic Healthcare Record Research Articles

Generating Complex Explanations for Artificial Intelligence Models: An Application to Clinical Data on Severe Mental Illness.

We present an explainable artificial intelligence methodology for predicting mortality in patients. We combine clinical data from an electronic patient healthcare record system with factors relevant for severe mental illness and then apply machine learning. The machine learning model is used to predict mortality in patients with severe mental illness. Our methodology uses class-contrastive reasoning. We show how machine learning scientists can use class-contrastive reasoning to generate complex explanations that explain machine model predictions and data. An example of a complex class-contrastive explanation is the following: "The patient is predicted to have a low probability of death because the patient has self-harmed before, and was at some point on medications such as first-generation and second-generation antipsychotics. There are 11 other patients with these characteristics. If the patient did not have these characteristics, the prediction would be different". This can be used to generate new hypotheses, which can be tested in follow-up studies. Diuretics seemed to be associated with a lower probability of mortality (as predicted by the machine learning model) in a group of patients with cardiovascular disease. The combination of delirium and dementia in Alzheimer's disease may also predispose some patients towards a higher probability of predicted mortality. Our technique can be employed to create intricate explanations from healthcare data and possibly other areas where explainability is important. We hope this will be a step towards explainable AI in personalized medicine.

Diagnostics Based Patient Classification for Clinical Decision Support Systems

The widespread adoption of Electronic Healthcare Records has resulted in an abundance of healthcare data. This data holds significant potential for improving healthcare services by providing valuable clinical insights and enhancing clinical decision-making. This paper presents a patient classification methodology that utilizes a multiclass and multilabel diagnostic approach to predict the patient's clinical class. The proposed model effectively handles comorbidities while maintaining a high level of accuracy. The implementation leverages the MIMIC III database as a data source to create a phenotyping dataset and train the models. Various machine learning models are employed in this study. Notably, the natural language processing-based One-Vs-Rest classifier achieves the best classification results, maintaining accuracy and F1 scores even with a large number of classes. The patient diagnostic class prediction model, based on the International Classification of Diseases 9, showcased in this paper, has broad applications in diagnostic support, treatment prediction, clinical assistance, recommender systems, clinical decision support systems, and clinical knowledge discovery engines.

A nurse driven care management program to engage older diabetes patients in personalized goal setting and disease management.

Multiple diabetes care guidelines have called for the personalization of risk factor goals, medication management, and self-care plans among older patients. Study of the implementation of these recommendations is needed. This study aimed to test whether a patient survey embedded in the Electronic Healthcare Record (EHR), coupled with telephonic nurse care management, could engage patients in personalized goal setting and chronic disease management. We conducted a single-center equal-randomization delayed comparator trial at the primary care clinics of the University of Chicago Medicine from 2018.6 to 2019.12. Patients over the age of 65 years with type 2 diabetes with an active patient portal account were recruited and randomized to receive an EHR embedded goal setting and preference survey immediately in the intervention arm or after 6 months in the delayed intervention control arm. In the intervention arm, nurses reviewed American Diabetes Association recommendations for A1C goals based on health status class, established personalized goals, and provided monthly telephonic care management phone calls for a maximum of 6 months. Our primary outcome was the documentation of a personalized A1C goal in the EHR. A total of 100 patients completed the trial (mean age, 72.51 [SD, 5.22] years; mean baseline A1C, 7.14% [SD, 1.06%]; 68% women). The majority were in the Healthy (59%) followed by Complex (30%) and Very Complex (11%) health status classes. Documentation of an A1C goal in the EHR increased from 42% to 90% (p < 0.001) at 6 months in the intervention group and from 54% to 56% in the control group. Across health status classes, patients set similar A1C goals. Older patients can be engaged in personalized goal setting and disease management through an embedded EHR intervention. The clinical impact of the intervention may differ if deployed among older patients with more complex health needs and higher glucose levels. ClinicalTrials.gov Identifier: NCT03692208.

Using Routinely Collected Electronic Healthcare Record Data to Investigate Fibrotic Multimorbidity in England.

Electronic healthcare records (EHRs) are used to document diagnoses, symptoms, tests, and prescriptions. Though not primarily collected for research purposes, owing to the size of the data as well as the depth of information collected, they have been used extensively to conduct epidemiological research. The Clinical Practice Research Datalink (CPRD) is an EHR database containing representative data of the UK population with regard to age, sex, race, and social deprivation measures. Fibrotic conditions are characterised by excessive scarring, contributing towards organ dysfunction and eventual organ failure. Fibrosis is associated with ageing as well as many other factors, it is hypothesised that fibrotic conditions are caused by the same underlying pathological mechanism. We calculated the prevalence of fibrotic conditions (as defined in a previous Delphi survey of clinicians) as well as the prevalence of fibrotic multimorbidity (the proportion of people with multiple fibrotic conditions). We included a random sample of 993,370 UK adults, alive, and enrolled at a UK general practice, providing data to the CPRD Aurum database as of 1st of January 2015. Individuals had to be eligible for linkage to hospital episode statistics (HES) and ONS death registration. We calculated the point prevalence of fibrotic conditions and multi-morbid fibrosis on the 1st of January 2015. Using death records of those who died in 2015, we investigated the prevalence of fibrosis associated death. We explored the most commonly co-occurring fibrotic conditions and determined the settings in which diagnoses were commonly made (primary care, secondary care or after death). The point prevalence of any fibrotic condition was 21.46%. In total, 6.00% of people had fibrotic multimorbidity. Of the people who died in 2015, 34.82% had a recording of a fibrotic condition listed on their death certificate. The key finding was that fibrotic multimorbidity affects approximately 1 in 16 people.

High-fidelity synthetic patient data applications and privacy considerations

This paper explores the potential applications of high-fidelity synthetic patient data in the context of healthcare research, including challenges and benefits. The paper starts by defining synthetic data, types of synthetic data and approaches to generating synthetic data. It then discusses the potential applications of synthetic data in addition to as a privacy enhancing technology and current debates around whether synthetic data should be considered personal data and,therefore, should be subjected to privacy controls to minimise reidentification risks. This will be followed by a discussion of privacy preservation approaches and privacy metrics that can be applied in the context of synthetic data. The paper includes a case study based on synthetic electronic healthcare record data from the Clinical Practice Research Datalink on how privacy concerns due to reidentification have been addressed in order to make this data available for research purposes. The authors conclude that synthetic data, particularly high-fidelity synthetic patient data, has the potential to add value over and above real data for public health and that it is possible to address privacy concerns to make synthetic data available via a combination of privacy measures applied during the synthetic data generation process and post-generation reidentification risk assessments as part of data protection impact assessments.

Prevalence of Mild Cognitive Impairment and Alzheimer's Disease Identified in Veterans in the United States.

Diagnostic codes can be instrumental for case identification in Alzheimer's disease (AD) research; however, this method has known limitations and cannot distinguish between disease stages. Clinical notes may offer more detailed information including AD severity and can complement diagnostic codes for case identification. To estimate prevalence of mild cognitive impairment (MCI) and AD using diagnostics codes and clinical notes available in the electronic healthcare record (EHR). This was a retrospective study in the Veterans Affairs Healthcare System (VAHS). Health records from Veterans aged 65 years or older were reviewed during Fiscal Years (FY) 2010-2019. Overall, 274,736 and 469,569 Veterans were identified based on a rule-based algorithm as having at least one clinical note for MCI and AD, respectively; 201,211 and 149,779 Veterans had a diagnostic code for MCI and AD, respectively. During FY 2011-2018, likely MCI or AD diagnosis was defined by≥2 qualifiers (i.e., notes and/or codes)≥30 days apart. Veterans with only 1 qualifier were considered as suspected MCI/AD. Over the 8-year study, 147,106 and 207,225 Veterans had likely MCI and AD, respectively. From 2011 to 2018, yearly MCI prevalence increased from 0.9% to 2.2%; yearly AD prevalence slightly decreased from 2.4% to 2.1%; mild AD changed from 22.9% to 26.8%, moderate AD changed from 26.5% to 29.1%, and severe AD changed from 24.6% to 30.7. The relative distribution of AD severities was stable over time. Accurate prevalence estimation is critical for healthcare resource allocation and facilitating patients receiving innovative medicines.

Mapping patient education encounters in elective surgery: a cohort study and cross-sectional survey

ObjectiveDevelop a process map of when patients learn about their proposed surgery and what resources patients use to educate themselves.DesignA mixed methods design, combining semistructured stakeholder interviews, quantitative validation using...

Aguhyper: a hyperledger-based electronic health record management framework

The increasing importance of healthcare records, particularly given the emergence of new diseases, emphasizes the need for secure electronic storage and dissemination. With these records dispersed across diverse healthcare entities, their physical maintenance proves to be excessively time-consuming. The prevalent management of electronic healthcare records (EHRs) presents inherent security vulnerabilities, including susceptibility to attacks and potential breaches orchestrated by malicious actors. To tackle these challenges, this article introduces AguHyper, a secure storage and sharing solution for EHRs built on a permissioned blockchain framework. AguHyper utilizes Hyperledger Fabric and the InterPlanetary Distributed File System (IPFS). Hyperledger Fabric establishes the blockchain network, while IPFS manages the off-chain storage of encrypted data, with hash values securely stored within the blockchain. Focusing on security, privacy, scalability, and data integrity, AguHyper’s decentralized architecture eliminates single points of failure and ensures transparency for all network participants. The study develops a prototype to address gaps identified in prior research, providing insights into blockchain technology applications in healthcare. Detailed analyses of system architecture, AguHyper’s implementation configurations, and performance assessments with diverse datasets are provided. The experimental setup incorporates CouchDB and the Raft consensus mechanism, enabling a thorough comparison of system performance against existing studies in terms of throughput and latency. This contributes significantly to a comprehensive evaluation of the proposed solution and offers a unique perspective on existing literature in the field.

Quality of heart failure registration in primary care: observations from 1 million electronic health records in the Amsterdam Metropolitan Area.

Proper coding of heart failure (HF) in electronic health records (EHRs) is an important prerequisite for adequate care and research towards this vulnerable patient population. We set out to evaluate the accuracy of registration of HF diagnoses in primary care EHRs. In a routine primary care database covering the Amsterdam Metropolitan Area, we identified all episodes of care with International Classification of Primary Care (ICPC) codes K77 (decompensatio cordis) or K84.03 (cardiomyopathy) up to 31/12/2021. We also performed two text-based searches toidentify HF episodes without an appropriate ICPC-code. Anexpert panel evaluated all ICPC and text matches for congruence between the assigned codes and notes. From a database of 968,433 records we identified 19,106 patients (2.0 %) with a total of 24,011 ICPC-coded HF episodes. Removal of 1,324 episodes found to concern other or uncertain diagnoses and inclusion of 4,582 validated HF episodes identified through text search led to exclusion of 909 (overregistration: 4.8 %) and inclusion of 2,266 additional patients (underregistration: 11.1 %). The inclusion of miscoded HF episodes advanced the first known date of HFdiagnosis in 3.9 % of records, with a median shift of 3.45 years. Episode-level underregistration decreased significantly over time, from 23.8 % in 2006 to 10.0 % in 2021. While there is improvement over time, there are still substantial levels of over- and underregistration ofHF, emphasizing the need for cautious interpretation ofICPC-coded data. The findings contribute to the understanding of HF registration issues in primary care and provide insights for improving registration practices.

Concept Recognition and Characterization of Patients Undergoing Resection of Vestibular Schwannoma Using Natural Language Processing

Abstract Background Natural language processing (NLP), a subset of artificial intelligence (AI), aims to decipher unstructured human language. This study showcases NLP's application in surgical health care, focusing on vestibular schwannoma (VS). By employing an NLP platform, we identify prevalent text concepts in VS patients' electronic health care records (EHRs), creating concept panels covering symptomatology, comorbidities, and management. Through a case study, we illustrate NLP's potential in predicting postoperative cerebrospinal fluid (CSF) leaks. Methods An NLP model analyzed EHRs of surgically managed VS patients from 2008 to 2018 in a single center. The model underwent unsupervised (trained on one million documents from EHR) and supervised (300 documents annotated in duplicate) learning phases, extracting text concepts and generating concept panels related to symptoms, comorbidities, and management. Statistical analysis correlated concept occurrences with postoperative complications, notably CSF leaks. Results Analysis included 292 patients' records, yielding 6,901 unique concepts and 360,929 occurrences. Concept panels highlighted key associations with postoperative CSF leaks, including “antibiotics,” “sepsis,” and “intensive care unit admission.” The NLP model demonstrated high accuracy (precision 0.92, recall 0.96, macro F1 0.93). Conclusion Our NLP model effectively extracted concepts from VS patients' EHRs, facilitating personalized concept panels with diverse applications. NLP shows promise in surgical settings, aiding in early diagnosis, complication prediction, and patient care. Further validation of NLP's predictive capabilities is warranted.

Comparing natural language processing representations of coded disease sequences for prediction in electronic health records.

Natural language processing (NLP) algorithms are increasingly being applied to obtain unsupervised representations of electronic health record (EHR) data, but their comparative performance at predicting clinical endpoints remains unclear. Our objective was to compare the performance of unsupervised representations of sequences of disease codes generated by bag-of-words versus sequence-based NLP algorithms at predicting clinically relevant outcomes. This cohort study used primary care EHRs from 6286233 people with Multiple Long-Term Conditions in England. For each patient, an unsupervised vector representation of their time-ordered sequences of diseases was generated using 2 input strategies (212 disease categories versus 9462 diagnostic codes) and different NLP algorithms (Latent Dirichlet Allocation, doc2vec, and 2 transformer models designed for EHRs). We also developed a transformer architecture, named EHR-BERT, incorporating sociodemographic information. We compared the performance of each of these representations (without fine-tuning) as inputs into a logistic classifier to predict 1-year mortality, healthcare use, and new disease diagnosis. Patient representations generated by sequence-based algorithms performed consistently better than bag-of-words methods in predicting clinical endpoints, with the highest performance for EHR-BERT across all tasks, although the absolute improvement was small. Representations generated using disease categories perform similarly to those using diagnostic codes as inputs, suggesting models can equally manage smaller or larger vocabularies for prediction of these outcomes. Patient representations produced by sequence-based NLP algorithms from sequences of disease codes demonstrate improved predictive content for patient outcomes compared with representations generated by co-occurrence-based algorithms. This suggests transformer models may be useful for generating multi-purpose representations, even without fine-tuning.

Accuracy of the recording of pneumonia events in English electronic healthcare record data in patients with chronic obstructive pulmonary disease

BackgroundIn primary care, identifying pneumonia events in people with chronic obstructive pulmonary disease (COPD) may be challenging due to similarities in symptoms with COPD exacerbations and lack of diagnostic testing. This study explored the accuracy of pneumonia diagnosis coded in primary care by comparing diagnosis in primary care with diagnosis in hospital.MethodsA study population of people with COPD in England was created using the Clinical Practice Research Datalink Aurum database linked with Hospital Episode Statistics inpatient data. Pneumonia codes only, and pneumonia code with associated clinical and/or treatment codes (chest x-ray, symptoms, antibiotics, sputum and blood culture) were used to determine pneumonia events in primary care. Events that were followed by hospitalisation within 7 days were used to estimate the positive predictive value (PPV) of pneumonia coding in primary care, using primary diagnosis of pneumonia in secondary care as the gold standard. The PPV of primary care recording of hospitalised pneumonia was also calculated.ResultsTwo hundred seventy-four thousand one hundred fifty-six COPD patients were eligible for inclusion, of whom 7,560 had an eligible pneumonia event in primary care diagnosed between 2015–2019 which was not ‘hospital-acquired’ and was diagnosed and entered on the same day. Of the 2,094 events which were followed by hospitalisation within 7 days, 1,208 had a primary diagnosis of pneumonia in hospital, representing a PPV of pneumonia coding in primary care of 57.7% (95% CI 55.6%-59.8%). Another 284 (13.6%) were diagnosed as a COPD exacerbation and 114 (5.4%) were diagnosed as another respiratory disease. Use of additional pneumonia clinical and treatment codes had a modest effect on the PPV but substantially lowered the number of events. Of the 33,603 eligible pneumonia events identified in secondary care, only 11,445 were recorded in primary care within 42 days, representing a sensitivity of 34.1% (95% CI 33.6%-34.6%).ConclusionsUse of primary care pneumonia codes and associated clinical and treatment codes to determine pneumonia is not recommended due to significant levels of misdiagnosis and many hospitalised events failing to be recorded in primary care.

Bt-GAN: Generating Fair Synthetic Healthdata via Bias-transforming Generative Adversarial Networks

Synthetic data generation offers a promising solution to enhance the usefulness of Electronic Healthcare Records (EHR) by generating realistic de-identified data. However, the existing literature primarily focuses on the quality of synthetic health data, neglecting the crucial aspect of fairness in downstream predictions. Consequently, models trained on synthetic EHR have faced criticism for producing biased outcomes in target tasks. These biases can arise from either spurious correlations between features or the failure of models to accurately represent sub-groups. To address these concerns, we present Bias-transforming Generative Adversarial Networks (Bt-GAN), a GAN-based synthetic data generator specifically designed for the healthcare domain. In order to tackle spurious correlations (i), we propose an information-constrained Data Generation Process (DGP) that enables the generator to learn a fair deterministic transformation based on a well-defined notion of algorithmic fairness. To overcome the challenge of capturing exact sub-group representations (ii), we incentivize the generator to preserve sub-group densities through score-based weighted sampling. This approach compels the generator to learn from underrepresented regions of the data manifold. To evaluate the effectiveness of our proposed method, we conduct extensive experiments using the Medical Information Mart for Intensive Care (MIMIC-III) database. Our results demonstrate that Bt-GAN achieves state-of-the-art accuracy while significantly improving fairness and minimizing bias amplification. Furthermore, we perform an in-depth explainability analysis to provide additional evidence supporting the validity of our study. In conclusion, our research introduces a novel and professional approach to addressing the limitations of synthetic data generation in the healthcare domain. By incorporating fairness considerations and leveraging advanced techniques such as GANs, we pave the way for more reliable and unbiased predictions in healthcare applications.

Associations between long-term exposure to air pollution and kidney function utilizing electronic healthcare records: a cross-sectional study

BackgroundChronic kidney disease (CKD) affects more than 38 million people in the United States, predominantly those over 65 years of age. While CKD etiology is complex, recent research suggests associations with environmental exposures.MethodsOur primary objective is to examine creatinine-based estimated glomerular filtration rate (eGFRcr) and diagnosis of CKD and potential associations with fine particulate matter (PM2.5), ozone (O3), and nitrogen dioxide (NO2) using a random sample of North Carolina electronic healthcare records (EHRs) from 2004 to 2016. We estimated eGFRcr using the serum creatinine-based 2021 CKD-EPI equation. PM2.5 and NO2 data come from a hybrid model using 1 km2 grids and O3 data from 12 km2 CMAQ grids. Exposure concentrations were 1-year averages. We used linear mixed models to estimate eGFRcr per IQR increase of pollutants. We used multiple logistic regression to estimate associations between pollutants and first appearance of CKD. We adjusted for patient sex, race, age, comorbidities, temporality, and 2010 census block group variables.ResultsWe found 44,872 serum creatinine measurements among 7,722 patients. An IQR increase in PM2.5 was associated with a 1.63 mL/min/1.73m2 (95% CI: -1.96, -1.31) reduction in eGFRcr, with O3 and NO2 showing positive associations. There were 1,015 patients identified with CKD through e-phenotyping and ICD codes. None of the environmental exposures were positively associated with a first-time measure of eGFRcr < 60 mL/min/1.73m2. NO2 was inversely associated with a first-time diagnosis of CKD with aOR of 0.77 (95% CI: 0.66, 0.90).ConclusionsOne-year average PM2.5 was associated with reduced eGFRcr, while O3 and NO2 were inversely associated. Neither PM2.5 or O3 were associated with a first-time identification of CKD, NO2 was inversely associated. We recommend future research examining the relationship between air pollution and impaired renal function.

Non-imaging Medical Data Synthesis for Trustworthy AI: A Comprehensive Survey

Data quality is a key factor in the development of trustworthy AI in healthcare. A large volume of curated datasets with controlled confounding factors can improve the accuracy, robustness, and privacy of downstream AI algorithms. However, access to high-quality datasets is limited by the technical difficulties of data acquisition, and large-scale sharing of healthcare data is hindered by strict ethical restrictions. Data synthesis algorithms, which generate data with distributions similar to real clinical data, can serve as a potential solution to address the scarcity of good quality data during the development of trustworthy AI. However, state-of-the-art data synthesis algorithms, especially deep learning algorithms, focus more on imaging data while neglecting the synthesis of non-imaging healthcare data, including clinical measurements, medical signals and waveforms, and electronic healthcare records (EHRs). Therefore, in this article, we will review synthesis algorithms, particularly for non-imaging medical data, with the aim of providing trustworthy AI in this domain. This tutorial-style review article will provide comprehensive descriptions of non-imaging medical data synthesis, covering aspects such as algorithms, evaluations, limitations, and future research directions.

Determining households from patient addresses and unique property reference numbers in general practitioner electronic health records

IntroductionHouseholds are increasingly studied in population health research as an important context for understanding health and social behaviours and outcomes. Identifying household units of analysis in routinely collected data rather than traditional surveys requires innovative and standardised tools, which do not currently exist. ObjectivesTo design a utility that identifies households at a point in time from pseudonymised Unique Property Reference Numbers (UPRNs) known as Residential Anonymised Linkage Fields (RALFs) assigned to general practitioner (GP) patient addresses in electronic health records (EHRs) in north east London (NEL). MethodsRule-based logic was developed to identify households based on GP registration, address date, and RALF validity. The logic was tested on a use case on the household clustering of childhood weight status, and bias in success of identifying households was examined in the use case cohort and in a full population cohort. Results92.1% of the use case cohort was assigned a household. The most frequent dominant reason (55.3%) for a household not assigned was that a person had no valid household RALFs available across their patient registration address records. Other reasons are having none or multiple valid household RALFs, or not being alive at the event date. In the use case, children not assigned to a household were more likely to attend schools in City &amp; Hackney and living in the third most deprived quintile of lower super output areas. 88.9% of the population cohort was assigned a household. Patients not assigned to a household were more likely to be aged 18 to 45 years, living in City &amp; Hackney, and living in the second quintile of most deprived lower super output areas. ConclusionsWe have developed a method for deriving households from primary care EHRs that can be implemented quickly and in real-time, providing timely data to support population health research on households.

A Harmonised Approach to Curating Research-Ready Datasets for Asthma, Chronic Obstructive Pulmonary Disease (COPD) and Interstitial Lung Disease (ILD) in England, Wales and Scotland Using Clinical Practice Research Datalink (CPRD), Secure Anonymised Information Linkage (SAIL) Databank and DataLoch.

Electronic healthcare records (EHRs) are an important resource for health research that can be used to improve patient outcomes in chronic respiratory diseases. However, consistent approaches in the analysis of these datasets are needed for coherent messaging, and when undertaking comparative studies across different populations. We developed a harmonised curation approach to generate comparable patient cohorts for asthma, chronic obstructive pulmonary disease (COPD) and interstitial lung disease (ILD) using datasets from within Clinical Practice Research Datalink (CPRD; for England), Secure Anonymised Information Linkage (SAIL; for Wales) and DataLoch (for Scotland) by defining commonly derived variables consistently between the datasets. By working in parallel on the curation methodology used for CPRD, SAIL and DataLoch for asthma, COPD and ILD, we were able to highlight key differences in coding and recording between the databases and identify solutions to enable valid comparisons. Codelists and metadata generated have been made available to help re-create the asthma, COPD and ILD cohorts in CPRD, SAIL and DataLoch for different time periods, and provide a starting point for the curation of respiratory datasets in other EHR databases, expediting further comparable respiratory research.

An explainable long short-term memory network for surgical site infection identification

BackgroundCurrently, surgical site infection surveillance relies on labor-intensive manual chart review. Recently suggested solutions involve machine learning to identify surgical site infections directly from the medical record. Deep learning is a form of machine learning that has historically performed better than traditional methods while being harder to interpret. We propose a deep learning model, a long short-term memory network, for the identification of surgical site infection from the medical record with an attention layer for explainability. MethodsWe retrieved structured data and clinical notes from the University of Utah Health System’s electronic health care record for operative events randomly selected for manual chart review from January 2016 to June 2021. Surgical site infection occurring within 30 days of surgery was determined according to the National Surgical Quality Improvement Program definition. We trained the long short-term memory model along with traditional machine learning models for comparison. We calculated several performance metrics from a holdout test set and performed additional analyses to understand the performance of the long short-term memory, including an explainability analysis. ResultsSurgical site infection was present in 4.7% of the total 9,185 operative events. The area under the receiver operating characteristic curve and sensitivity of the long short-term memory was higher (area under the receiver operating characteristic curve: 0.954, sensitivity: 0.920) compared to the top traditional model (area under the receiver operating characteristic curve: 0.937, sensitivity: 0.736). The top 5 features of the long short-term memory included 2 procedure codes and 3 laboratory values. ConclusionSurgical site infection surveillance is vital for the reduction of surgical site infection rates. Our explainable long short-term memory achieved a comparable area under the receiver operating characteristic curve and greater sensitivity when compared to traditional machine learning methods. With explainable deep learning, automated surgical site infection surveillance could replace burdensome manual chart review processes.

Comparing the clinical practice and prescribing safety of locum and permanent doctors: observational study of primary care consultations in England

BackgroundTemporary doctors, known as locums, are a key component of the medical workforce in the NHS but evidence on differences in quality and safety between locum and permanent doctors is limited. We aimed to examine differences in the clinical practice, and prescribing safety for locum and permanent doctors working in primary care in England.MethodsWe accessed electronic health care records (EHRs) for 3.5 million patients from the CPRD GOLD database with linkage to Hospital Episode Statistics from 1st April 2010 to 31st March 2022. We used multi-level mixed effects logistic regression to compare consultations with locum and permanent GPs for several patient outcomes including general practice revisits; prescribing of antibiotics; strong opioids; hypnotics; A&E visits; emergency hospital admissions; admissions for ambulatory care sensitive conditions; test ordering; referrals; and prescribing safety indicators while controlling for patient and practice characteristics.ResultsConsultations with locum GPs were 22% more likely to involve a prescription for an antibiotic (OR = 1.22 (1.21 to 1.22)), 8% more likely to involve a prescription for a strong opioid (OR = 1.08 (1.06 to 1.09)), 4% more likely to be followed by an A&E visit on the same day (OR = 1.04 (1.01 to 1.08)) and 5% more likely to be followed by an A&E visit within 1 to 7 days (OR = 1.05 (1.02 to 1.08)). Consultations with a locum were 12% less likely to lead to a practice revisit within 7 days (OR = 0.88 (0.87 to 0.88)), 4% less likely to involve a prescription for a hypnotic (OR = 0.96 (0.94 to 0.98)), 15% less likely to involve a referral (OR = 0.85 (0.84 to 0.86)) and 19% less likely to involve a test (OR = 0.81 (0.80 to 0.82)). We found no evidence that emergency admissions, ACSC admissions and eight out of the eleven prescribing safety indicators were different if patients were seen by a locum or a permanent GP.ConclusionsDespite existing concerns, the clinical practice and performance of locum GPs did not appear to be systematically different from that of permanent GPs. The practice and performance of both locum and permanent GPs is likely shaped by the organisational setting and systems within which they work.

Development and external validation of the eFalls tool: a multivariable prediction model for the risk of ED attendance or hospitalisation with a fall or fracture in older adults.

Falls are common in older adults and can devastate personal independence through injury such as fracture and fear of future falls. Methods to identify people for falls prevention interventions are currently limited, with high risks of bias in published prediction models. We have developed and externally validated the eFalls prediction model using routinely collected primary care electronic health records (EHR) to predict risk of emergency department attendance/hospitalisation with fall or fracture within 1 year. Data comprised two independent, retrospective cohorts of adults aged ≥65 years: the population of Wales, from the Secure Anonymised Information Linkage Databank (model development); the population of Bradford and Airedale, England, from Connected Bradford (external validation). Predictors included electronic frailty index components, supplemented with variables informed by literature reviews and clinical expertise. Fall/fracture risk was modelled using multivariable logistic regression with a Least Absolute Shrinkage and Selection Operator penalty. Predictive performance was assessed through calibration, discrimination and clinical utility. Apparent, internal-external cross-validation and external validation performance were assessed across general practices and in clinically relevant subgroups. The model's discrimination performance (c-statistic) was 0.72 (95% confidence interval, CI: 0.68 to 0.76) on internal-external cross-validation and 0.82 (95% CI: 0.80 to 0.83) on external validation. Calibration was variable across practices, with some over-prediction in the validation population (calibration-in-the-large, -0.87; 95% CI: -0.96 to -0.78). Clinical utility on external validation was improved after recalibration. The eFalls prediction model shows good performance and could support proactive stratification for falls prevention services if appropriately embedded into primary care EHR systems.