BMJ Health & Care Informatics最新文献

Background: Despite the increasing availability of electronic healthcare record (EHR) data and wide availability of plug-and-play machine learning (ML) Application Programming Interfaces, the adoption of data-driven decision-making within routine hospital workflows thus far, has remained limited. Through the lens of deriving clusters of diagnoses by age, this study investigated the type of ML analysis that can be performed using EHR data and how results could be communicated to lay stakeholders.

Methods: Observational EHR data from a tertiary paediatric hospital, containing 61 522 unique patients and 3315 unique ICD-10 diagnosis codes was used, after preprocessing. K-means clustering was applied to identify age distributions of patient diagnoses. The final model was selected using quantitative metrics and expert assessment of the clinical validity of the clusters. Additionally, uncertainty over preprocessing decisions was analysed.

Findings: Four age clusters of diseases were identified, broadly aligning to ages between: 0 and 1; 1 and 5; 5 and 13; 13 and 18. Diagnoses, within the clusters, aligned to existing knowledge regarding the propensity of presentation at different ages, and sequential clusters presented known disease progressions. The results validated similar methodologies within the literature. The impact of uncertainty induced by preprocessing decisions was large at the individual diagnoses but not at a population level. Strategies for mitigating, or communicating, this uncertainty were successfully demonstrated.

Conclusion: Unsupervised ML applied to EHR data identifies clinically relevant age distributions of diagnoses which can augment existing decision making. However, biases within healthcare datasets dramatically impact results if not appropriately mitigated or communicated.

In the following narrative review, we discuss the potential role of large language models (LLMs) in medical device innovation, specifically examples using generative pretrained transformer-4. Throughout the biodesign process, LLMs can offer prompt-driven insights, aiding problem identification, knowledge assimilation and decision-making. Intellectual property analysis, regulatory assessment and market analysis emerge as key LLM applications. Through case examples, we underscore LLMs' transformative ability to democratise information access and expertise, facilitating inclusive innovation in medical devices as well as its effectiveness with providing real-time, individualised feedback for innovators of all experience levels. By mitigating entry barriers, LLMs accelerate transformative advancements, fostering collaboration among established and emerging stakeholders.

Background and objectives: Patient-clinician communication and shared decision-making face challenges in the perioperative period. Chatbots have emerged as valuable support tools in perioperative care. A simultaneous and complete comparison of overall benefits and harm of chatbot application is conducted.

Materials: MEDLINE, EMBASE and the Cochrane Library were systematically searched for studies published before May 2023 on the benefits and harm of chatbots used in the perioperative period. The major outcomes assessed were patient satisfaction and knowledge acquisition. Untransformed proportion (PR) with a 95% CI was used for the analysis of continuous data. Risk of bias was assessed using the Cochrane Risk of Bias assessment tool version 2 and the Methodological Index for Non-Randomised Studies.

Results: Eight trials comprising 1073 adults from four countries were included. Most interventions (n = 5, 62.5%) targeted perioperative care in orthopaedics. Most interventions use rule-based chatbots (n = 7, 87.5%). This meta-analysis found that the majority of the participants were satisfied with the use of chatbots (mean proportion=0.73; 95% CI: 0.62 to 0.85), and agreed that they gained knowledge in their perioperative period (mean proportion=0.80; 95% CI: 0.74 to 0.87).

Conclusion: This review demonstrates that perioperative chatbots are well received by the majority of patients with no reports of harm to-date. Chatbots may be considered as an aid in perioperative communication between patients and clinicians and shared decision-making. These findings may be used to guide the healthcare providers, policymakers and researchers for enhancing perioperative care.

Objectives: We assessed the feasibility of ChatGPT for patients with type 2 diabetes seeking information about exercise.

Methods: In this pilot study, two physicians with expertise in diabetes care and rehabilitative treatment in Republic of Korea discussed and determined the 14 most asked questions on exercise for managing type 2 diabetes by patients in clinical practice. Each question was inputted into ChatGPT (V.4.0), and the answers from ChatGPT were assessed. The Likert scale was calculated for each category of validity (1-4), safety (1-4) and utility (1-4) based on position statements of the American Diabetes Association and American College of Sports Medicine.

Results: Regarding validity, 4 of 14 ChatGPT (28.6%) responses were scored as 3, indicating accurate but incomplete information. The other 10 responses (71.4%) were scored as 4, indicating complete accuracy with complete information. Safety and utility scored 4 (no danger and completely useful) for all 14 ChatGPT responses.

Conclusion: ChatGPT can be used as supplementary educational material for diabetic exercise. However, users should be aware that ChatGPT may provide incomplete answers to some questions on exercise for type 2 diabetes.

Background: The detrimental repercussions of the COVID-19 pandemic on the quality of care and clinical outcomes for patients with acute coronary syndrome (ACS) necessitate a rigorous re-evaluation of prognostic prediction models in the context of the pandemic environment. This study aimed to elucidate the adaptability of prediction models for 30-day mortality in patients with ACS during the pandemic periods.

Methods: A total of 2041 consecutive patients with ACS were included from 32 institutions between December 2020 and April 2023. The dataset comprised patients who were admitted for ACS and underwent coronary angiography for the diagnosis during hospitalisation. The prediction accuracy of the Global Registry of Acute Coronary Events (GRACE) and a machine learning model, KOTOMI, was evaluated for 30-day mortality in patients with ST-elevation acute myocardial infarction (STEMI) and non-ST-elevation acute coronary syndrome (NSTE-ACS).

Results: The area under the receiver operating characteristics curve (AUROC) was 0.85 (95% CI 0.81 to 0.89) in the GRACE and 0.87 (95% CI 0.82 to 0.91) in the KOTOMI for STEMI. The difference of 0.020 (95% CI -0.098-0.13) was not significant. For NSTE-ACS, the respective AUROCs were 0.82 (95% CI 0.73 to 0.91) in the GRACE and 0.83 (95% CI 0.74 to 0.91) in the KOTOMI, also demonstrating insignificant difference of 0.010 (95% CI -0.023 to 0.25). The prediction accuracy of both models had consistency in patients with STEMI and insignificant variation in patients with NSTE-ACS between the pandemic periods.

Conclusions: The prediction models maintained high accuracy for 30-day mortality of patients with ACS even in the pandemic periods, despite marginal variation observed.

Objective: The study aimed to develop natural language processing (NLP) algorithms to automate extracting patient-centred breast cancer treatment outcomes from clinical notes in electronic health records (EHRs), particularly for women from under-represented populations.

Methods: The study used clinical notes from 2010 to 2021 from a tertiary hospital in the USA. The notes were processed through various NLP techniques, including vectorisation methods (term frequency-inverse document frequency (TF-IDF), Word2Vec, Doc2Vec) and classification models (support vector classification, K-nearest neighbours (KNN), random forest (RF)). Feature selection and optimisation through random search and fivefold cross-validation were also conducted.

Results: The study annotated 100 out of 1000 clinical notes, using 970 notes to build the text corpus. TF-IDF and Doc2Vec combined with RF showed the highest performance, while Word2Vec was less effective. RF classifier demonstrated the best performance, although with lower recall rates, suggesting more false negatives. KNN showed lower recall due to its sensitivity to data noise.

Discussion: The study highlights the significance of using NLP in analysing clinical notes to understand breast cancer treatment outcomes in under-represented populations. The TF-IDF and Doc2Vec models were more effective in capturing relevant information than Word2Vec. The study observed lower recall rates in RF models, attributed to the dataset's imbalanced nature and the complexity of clinical notes.

Conclusion: The study developed high-performing NLP pipeline to capture treatment outcomes for breast cancer in under-represented populations, demonstrating the importance of document-level vectorisation and ensemble methods in clinical notes analysis. The findings provide insights for more equitable healthcare strategies and show the potential for broader NLP applications in clinical settings.

Background: The learning health system (LHS) concept is a potential solution to the challenges currently faced by primary care. There are few descriptions of the barriers and facilitators to achieving an LHS in general practice, and even fewer that are underpinned by implementation science. This study aimed to describe the barriers and facilitators to achieving an LHS in primary care and provide practical recommendations for general practices on their journey towards an LHS.

Methods: This study is a secondary data analysis from a qualitative investigation of an LHS in a university-based general practice in Sydney, Australia. A framework analysis was conducted using transcripts from semistructured interviews with clinic staff. Data were coded according to the theoretical domains framework, and then to an LHS framework.

Results: 91% (n=32) of practice staff were interviewed, comprising general practitioners (n=15), practice nurses (n=3), administrative staff (n=13) and a psychologist. Participants reported that the practice alignment with LHS principles was influenced by many behavioural determinants, some of which were applicable to healthcare in general, for example, some staff lacked knowledge about practice policies and skills in using software. However, many were specific to the general practice environment, for example, the environmental context of general practice meant that administrative staff were an integral part of the LHS, particularly in facilitating partnerships with patients.

Conclusions: The LHS journey in general practice is influenced by several factors. Mapping the LHS domains in relation to the theoretical domains framework can be used to generate a roadmap to hasten the journey towards LHS in primary care settings.

Background: Referring providers are often critiqued for writing poor-quality referrals. This study characterised clinical referral guidelines and forms to understand which data consultant providers require. These data were then used to codesign an evidence-based, high-quality referral form.

Methods: This study used both observational and quality improvement approaches. Canadian referral guidelines were reviewed and summarised. Referral data fields from 150 randomly selected Ontario referral forms were categorised and counted. The referral guideline summary and referral data were then used by referring providers, consultant providers and administrators to codesign a referral form.

Results: Referral guidelines recommended 42 types of referral data be included in referrals. Referral data were categorised as patient demographics, provider demographics, reason for referral, clinical information and administrative information. The percentage of referral guidelines recommending inclusion of each type of referral data varied from 8% to 77%. Ontario referral forms requested 264 different types of referral data. Digital referral forms requested more referral data types than paper-based referral forms (55.0±10.6 vs 30.5±8.1; 95% CI p<0.01). A codesigned referral form was created across two sessions with 29 and 21 participants in each.

Discussion: Referral guidelines lack consistency and specificity, which makes writing high-quality referrals challenging. Digital referral forms tend to request more referral data than paper-based referrals, which creates administrative burdens for referring and consultant providers. We created the first codesigned referral form with referring providers, consultant providers and administrators. We recommend clinical adoption of this form to improve referral quality and minimise administrative burdens.

Objectives: Risk stratification tools that predict healthcare utilisation are extensively integrated into primary care systems worldwide, forming a key component of anticipatory care pathways, where high-risk individuals are targeted by preventative interventions. Existing work broadly focuses on comparing model performance in retrospective cohorts with little attention paid to efficacy in reducing morbidity when deployed in different global contexts. We review the evidence supporting the use of such tools in real-world settings, from retrospective dataset performance to pathway evaluation.

Methods: A systematic search was undertaken to identify studies reporting the development, validation and deployment of models that predict healthcare utilisation in unselected primary care cohorts, comparable to their current real-world application.

Results: Among 3897 articles screened, 51 studies were identified evaluating 28 risk prediction models. Half underwent external validation yet only two were validated internationally. No association between validation context and model discrimination was observed. The majority of real-world evaluation studies reported no change, or indeed significant increases, in healthcare utilisation within targeted groups, with only one-third of reports demonstrating some benefit.

Discussion: While model discrimination appears satisfactorily robust to application context there is little evidence to suggest that accurate identification of high-risk individuals can be reliably translated to improvements in service delivery or morbidity.

Conclusions: The evidence does not support further integration of care pathways with costly population-level interventions based on risk prediction in unselected primary care cohorts. There is an urgent need to independently appraise the safety, efficacy and cost-effectiveness of risk prediction systems that are already widely deployed within primary care.