DNA-T: Deformable Neighborhood Attention Transformer for Irregular Medical Time Series.

The real-world Electronic Health Records (EHRs) present irregularities due to changes in the patient's health status, resulting in various time intervals between observations and different physiological variables examined at each observation point. There have been recent applications of Transformer-based models in the field of irregular time series. However, the full attention mechanism in Transformer overly focuses on distant information, ignoring the short-term correlations of the condition. Thereby, the model is not able to capture localized changes or short-term fluctuations in patients' conditions. Therefore, we propose a novel end-to-end Deformable Neighborhood Attention Transformer (DNA-T) for irregular medical time series. The DNA-T captures local features by dynamically adjusting the receptive field of attention and aggregating relevant deformable neighborhoods in irregular time series. Specifically, we design a Deformable Neighborhood Attention (DNA) module that enables the network to attend to relevant neighborhoods by drifting the receiving field of neighborhood attention. The DNA enhances the model's sensitivity to local information and representation of local features, thereby capturing the correlation of localized changes in patients' conditions. We conduct extensive experiments to validate the effectiveness of DNA-T, outperforming existing state-of-the-art methods in predicting the mortality risk of patients. Moreover, we visualize an example to validate the effectiveness of the proposed DNA.

The use of natural language processing for the identification of ageing syndromes including sarcopenia, frailty and falls in electronic healthcare records: a systematic review.

BACKGROUND: Recording and coding of ageing syndromes in hospital records is known to be suboptimal. Natural Language Processing algorithms may be useful to identify diagnoses in electronic healthcare records to improve the recording and coding of these ageing syndromes, but the feasibility and diagnostic accuracy of such algorithms are unclear. METHODS: We conducted a systematic review according to a predefined protocol and in line with Preferred Reporting Items for Systematic reviews and Meta-Analyses (PRISMA) guidelines. Searches were run from the inception of each database to the end of September 2023 in PubMed, Medline, Embase, CINAHL, ACM digital library, IEEE Xplore and Scopus. Eligible studies were identified via independent review of search results by two coauthors and data extracted from each study to identify the computational method, source of text, testing strategy and performance metrics. Data were synthesised narratively by ageing syndrome and computational method in line with the Studies Without Meta-analysis guidelines. RESULTS: From 1030 titles screened, 22 studies were eligible for inclusion. One study focussed on identifying sarcopenia, one frailty, twelve falls, five delirium, five dementia and four incontinence. Sensitivity (57.1%-100%) of algorithms compared with a reference standard was reported in 20 studies, and specificity (84.0%-100%) was reported in only 12 studies. Study design quality was variable with results relevant to diagnostic accuracy not always reported, and few studies undertaking external validation of algorithms. CONCLUSIONS: Current evidence suggests that Natural Language Processing algorithms can identify ageing syndromes in electronic health records. However, algorithms require testing in rigorously designed diagnostic accuracy studies with appropriate metrics reported.

Dynamic risk prediction of survival in liver cirrhosis: A comparison of landmarking approaches.

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.

Advancing equity in breast cancer care: natural language processing for analysing treatment outcomes in under-represented populations.

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.

Hybrid electronic record: An error reduction strategy for diverse medical prescription formats.

BACKGROUND:  This project is part of a broader effort to develop a new electronic registry for ophthalmology in the KwaZulu-Natal (KZN) province in South Africa. The registry should include a clinical decision support system that reduces the potential for human error and should be applicable for our diversity of hospitals, whether electronic health record (EHR) or paper-based. METHODS:  Post-operative prescriptions of consecutive cataract surgery discharges were included for 2019 and 2020. Comparisons were facilitated by the four chosen state hospitals in KZN each having a different system for prescribing medications: Electronic, tick sheet, ink stamp and handwritten health records. Error types were compared to hospital systems to identify easily-correctable errors. Potential error remedies were sought by a four-step process. RESULTS:  There were 1307 individual errors in 1661 prescriptions, categorised into 20 error types. Increasing levels of technology did not decrease error rates but did decrease the variety of error types. High technology scripts had the most errors but when easily correctable errors were removed, EHRs had the lowest error rates and handwritten the highest. CONCLUSION:  Increasing technology, by itself, does not seem to reduce prescription error. Technology does, however, seem to decrease the variability of potential error types, which make many of the errors simpler to correct.Contribution: Regular audits are an effective tool to greatly reduce prescription errors, and the higher the technology level, the more effective these audit interventions become. This advantage can be transferred to paper-based notes by utilising a hybrid electronic registry to print the formal medical record.

A mixed methods evaluation of patient perspectives on the implementation of an electronic health record-integrated patient-reported symptom and needs monitoring program in cancer care.

BACKGROUND: As cancer centers have increased focus on patient-centered, evidenced-based care, implementing efficient programs that facilitate effective patient-clinician communication remains critical. We implemented an electronic health record-integrated patient-reported symptom and needs monitoring program ('cPRO' for cancer patient-reported outcomes). To aid evaluation of cPRO implementation, we asked patients receiving care in one of three geographical regions of an academic healthcare system about their experiences. METHODS: Using a sequential mixed-methods approach, we collected feedback in two waves. Wave 1 included virtual focus groups and interviews with patients who had completed cPRO. In Wave 2, we administered a structured survey to systematically examine Wave 1 themes. All participants had a diagnosed malignancy and received at least 2 invitations to complete cPRO. We used rapid and traditional qualitative methods to analyze Wave 1 data and focused on identifying facilitators and barriers to cPRO implementation. Wave 2 data were analyzed descriptively. RESULTS: Participants (n = 180) were on average 62.9 years old; were majority female, White, non-Hispanic, and married; and represented various cancer types and phases of treatment. Wave 1 participants (n = 37) identified facilitators, including cPRO's perceived value and favorable usability, and barriers, including confusion about cPRO's purpose and various considerations for responding. High levels of clinician engagement with, and patient education on, cPRO were described as facilitators while low levels were described as barriers. Wave 2 (n = 143) data demonstrated high endorsement rates of cPRO's usability on domains such as navigability (91.6%), comprehensibility (98.7%), and relevance (82.4%). Wave 2 data also indicated low rates of understanding cPRO's purpose (56.7%), education from care teams about cPRO (22.5%), and discussing results of cPRO with care teams (16.3%). CONCLUSIONS: While patients reported high value and ease of use when completing cPRO, they also reported areas of confusion, emphasizing the importance of patient education on the purpose and use of cPRO and clinician engagement to sustain participation. These results guided successful implementation changes and will inform future improvements.

Deep Learning-Based Dynamic Risk Prediction of Venous Thromboembolism for Patients With Ovarian Cancer in Real-World Settings From Electronic Health Records.

PURPOSE: Patients with epithelial ovarian cancer (EOC) have an elevated risk for venous thromboembolism (VTE). To assess the risk of VTE, models were developed by statistical or machine learning algorithms. However, few models have accommodated deep learning (DL) algorithms in realistic clinical settings. We aimed to develop a predictive DL model, exploiting rich information from electronic health records (EHRs), including dynamic clinical features and the presence of competing risks. METHODS: We extracted EHRs of 1,268 patients diagnosed with EOC from January 2007 through December 2017 at the National Cancer Center, Korea. DL survival networks using fully connected layers, temporal attention, and recurrent neural networks were adopted and compared with multi-perceptron-based classification models. Prediction accuracy was independently validated in the data set of 423 patients newly diagnosed with EOC from January 2018 to December 2019. Personalized risk plots displaying the individual interval risk were developed. RESULTS: DL-based survival networks achieved a superior area under the receiver operating characteristic curve (AUROC) between 0.95 and 0.98 while the AUROC of classification models was between 0.85 and 0.90. As clinical information benefits the prediction accuracy, the proposed dynamic survival network outperformed other survival networks for the test and validation data set with the highest time-dependent concordance index (0.974, 0.975) and lowest Brier score (0.051, 0.049) at 6 months after a cancer diagnosis. Our visualization showed that the interval risk fluctuating along with the changes in longitudinal clinical features. CONCLUSION: Adaption of dynamic patient clinical features and accounting for competing risks from EHRs into the DL algorithms demonstrated VTE risk prediction with high accuracy. Our results show that this novel dynamic survival network can provide personalized risk prediction with the potential to assist risk-based clinical intervention to prevent VTE among patients with EOC.

Are all cancer survivors included when using electronically administered patient reported outcomes in primary healthcare cancer rehabilitation? A cross-sectional study.

BACKGROUND: Patient reported outcomes (PROs) are being used frequently in clinical practice. PROs often serve several purposes, such as increasing patient involvement, assessing health status, and monitoring and improving the quality-of-care at an aggregated level. However, the lack of representative PRO-data may have implications for all these purposes. This study aims to assess the association of non-administration of (not sending an electronic invite to PRO) and non-response to (not responding to PRO) electronically administered PROs with social inequality in a primary healthcare cancer rehabilitation setting. Furthermore, it examines whether the workflows surrounding PRO have an impact on non-administration and non-response. METHODS: This is a cross sectional study using routinely collected data from electronic health records and registers including cancer survivors (CSs) over 18 years booked for an initial consultation in a primary healthcare cancer rehabilitation setting using PROs for systematic health status assessment. During the study period two different PRO platforms were used, each associated with different workflows. Non-administration and non-response rates were calculated for sociodemographic characteristics for each PRO platform. Crude and adjusted odds ratios were calculated using univariate and multivariate logistic regression. RESULTS: In total, 1868 (platform 1) and 1446 (platform 2) CSCSs were booked for an initial consultation. Of these, 233 (12.5%) (platform 1) and 283 (19.6%) (platform 2) were not sent a PRO (non-administration). Among those who received a PRO, 157 (9.6%) on platform 1 and 140 (12.0%) on platform 2 did not respond (non-response). Non-administration of and non-response to PROs were significantly associated with lower socioeconomic status. Moreover, the workflows surrounding PROs seem to have an impact on non-inclusion in and non-response to PROs. CONCLUSIONS: Non-administration of and non-response to PROs in clinical practice is associated with determinants of social inequality. Clinical workflows and the PRO platforms used may potentially worsen this inequality. It is important to consider these implications when using PROs at both the individual and aggregated levels. A key aspect of implementing PROs in clinical practice is the ongoing focus on representativeness, including a focus on monitoring PRO administration and response.

Blockchain technology and raft consensus for secure physician prescriptions and improved diagnoses in electronic healthcare systems.

With electronic healthcare systems undergoing rapid change, optimizing the crucial process of recording physician prescriptions is a task with major implications for patient care. The power of blockchain technology and the precision of the Raft consensus algorithm are combined in this article to create a revolutionary solution for this problem. In addition to addressing these issues, the proposed framework, by focusing on the challenges associated with physician prescriptions, is a breakthrough in a new era of security and dependability for the healthcare sector. The Raft algorithm is a cornerstone that improves the diagnostic decision-making process, increases confidence in patients, and sets a new standard for robust healthcare systems. In the proposed consensus algorithm, a weighted sum of two influencing factors including the physician acceptability and inter-physicians' reliability is used for selecting the participating physicians. An investigation is conducted to see how well the Raft algorithm performs in overcoming prescription-related roadblocks that support a compelling argument for improved patient care. Apart from its technological benefits, the proposed approach seeks to revolutionize the healthcare system by fostering trust between patients and providers. Raft's ability to communicate presents the proposed solution as an effective way to deal with healthcare issues and ensure security.

Improve the efficiency and accuracy of ophthalmologists' clinical decision-making based on AI technology.

BACKGROUND: As global aging intensifies, the prevalence of ocular fundus diseases continues to rise. In China, the tense doctor-patient ratio poses numerous challenges for the early diagnosis and treatment of ocular fundus diseases. To reduce the high risk of missed or misdiagnosed cases, avoid irreversible visual impairment for patients, and ensure good visual prognosis for patients with ocular fundus diseases, it is particularly important to enhance the growth and diagnostic capabilities of junior doctors. This study aims to leverage the value of electronic medical record data to developing a diagnostic intelligent decision support platform. This platform aims to assist junior doctors in diagnosing ocular fundus diseases quickly and accurately, expedite their professional growth, and prevent delays in patient treatment. An empirical evaluation will assess the platform's effectiveness in enhancing doctors' diagnostic efficiency and accuracy. METHODS: In this study, eight Chinese Named Entity Recognition (NER) models were compared, and the SoftLexicon-Glove-Word2vec model, achieving a high F1 score of 93.02%, was selected as the optimal recognition tool. This model was then used to extract key information from electronic medical records (EMRs) and generate feature variables based on diagnostic rule templates. Subsequently, an XGBoost algorithm was employed to construct an intelligent decision support platform for diagnosing ocular fundus diseases. The effectiveness of the platform in improving diagnostic efficiency and accuracy was evaluated through a controlled experiment comparing experienced and junior doctors. RESULTS: The use of the diagnostic intelligent decision support platform resulted in significant improvements in both diagnostic efficiency and accuracy for both experienced and junior doctors (P < 0.05). Notably, the gap in diagnostic speed and precision between junior doctors and experienced doctors narrowed considerably when the platform was used. Although the platform also provided some benefits to experienced doctors, the improvement was less pronounced compared to junior doctors. CONCLUSION: The diagnostic intelligent decision support platform established in this study, based on the XGBoost algorithm and NER, effectively enhances the diagnostic efficiency and accuracy of junior doctors in ocular fundus diseases. This has significant implications for optimizing clinical diagnosis and treatment.

Adaptation of risk prediction equations for cardiovascular outcomes among patients with type 2 diabetes in real-world settings: a cross-institutional study using common data model approach.

OBJECTIVE: To adapt risk prediction equations for myocardial infarction (MI), stroke, and heart failure (HF) among patients with type 2 diabetes in real-world settings using cross-institutional electronic health records (EHRs) in Taiwan. METHODS: The EHRs from two medical centers, National Cheng Kung University Hospital (NCKUH; 11,740 patients) and National Taiwan University Hospital (NTUH; 20,313 patients), were analyzed using the common data model approach. Risk equations for MI, stroke, and HF from UKPDS-OM2, RECODe, and CHIME models were adapted for external validation and recalibration. External validation was assessed by (1) discrimination, evaluated by the area under the receiver operating characteristic curve (AUROC) and (2) calibration, evaluated by calibration slopes and intercepts and the Greenwood-Nam-D'Agostino (GND) test. Recalibration was conducted for unsatisfactory calibration (p-value of GND test < 0.05) by adjusting the baseline hazards of original equations to address variations in patients' cardiovascular risks across institutions. RESULTS: The CHIME risk equations had acceptable discrimination (AUROC: 0.71-0.79) and better calibration than that for UKPDS-OM2 and RECODe, although the calibration remained unsatisfactory. After recalibration, the calibration slopes/intercepts of the CHIME-MI, CHIME-stroke, and CHIME-HF risk equations were 0.9848/- 0.0008, 1.1003/- 0.0046, and 0.9436/0.0063 in the NCKUH population and 1.1060/- 0.0011, 0.8714/0.0030, and 1.0476/- 0.0016 in the NTUH population, respectively. All the recalibrated risk equations showed satisfactory calibration (p-values of GND tests ≥ 0.05). CONCLUSIONS: We provide valid risk prediction equations for MI, stroke, and HF outcomes in Taiwanese type 2 diabetes populations. A framework for adapting risk equations across institutions is also proposed.

Automating incidence and prevalence analysis in open cohorts.

MOTIVATION: 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. IMPLEMENTATION: 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. GENERAL FEATURES: 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. AVAILABILITY: The command line interface is freely available from https://github.com/THINKINGGroup/analogy_publication .

Preventive Service Usage and New Chronic Disease Diagnoses: Using PCORnet Data to Identify Emerging Trends, United States, 2018-2022.

Background: Data modernization efforts to strengthen surveillance capacity could help assess trends in use of preventive services and diagnoses of new chronic disease during the COVID-19 pandemic, which broadly disrupted health care access. Methods: This cross-sectional study examined electronic health record data from US adults aged 21 to 79 years in a large national research network (PCORnet), to describe use of 8 preventive health services (N = 30,783,825 patients) and new diagnoses of 9 chronic diseases (N = 31,588,222 patients) during 2018 through 2022. Joinpoint regression assessed significant trends, and health debt was calculated comparing 2020 through 2022 volume to prepandemic (2018 and 2019) levels. Results: From 2018 to 2022, use of some preventive services increased (hemoglobin A1c and lung computed tomography, both P < .05), others remained consistent (lipid testing, wellness visits, mammograms, Papanicolaou tests or human papillomavirus tests, stool-based screening), and colonoscopies or sigmoidoscopies declined (P < .01). Annual new chronic disease diagnoses were mostly stable (6% hypertension; 4% to 5% cholesterol; 4% diabetes; 1% colonic adenoma; 0.1% colorectal cancer; among women, 0.5% breast cancer), although some declined (lung cancer, cervical intraepithelial neoplasia or carcinoma in situ, cervical cancer, all P < .05). The pandemic resulted in health debt, because use of most preventive services and new diagnoses of chronic disease were less than expected during 2020; these partially rebounded in subsequent years. Colorectal screening and colonic adenoma detection by age group aligned with screening recommendation age changes during this period. Conclusion: Among over 30 million patients receiving care during 2018 through 2022, use of preventive services and new diagnoses of chronic disease declined in 2020 and then rebounded, with some remaining health debt. These data highlight opportunities to augment traditional surveillance with EHR-based data.

Phenogrouping heart failure with preserved or mildly reduced ejection fraction using electronic health record data.

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.

Citizen Views on an Opt-Out Approach to National Electronic Health Records in Germany: A Small-Scale Qualitative Study.

Objectives: Electronic health records (German: elektronische Patientenakte - ePA) are an important healthcare tool. However, in Germany, current participation remains low for their national ePA. To rectify this, the German government recently adopted an opt-out approach to their national ePA system. The objective of this study is to investigate and provide a brief overview of German public attitudes towards this approach to inform policymakers with evidence-based insights. Methods: Four public focus groups were conducted with 12 German citizens to discuss their opinions on the German governments new opt-out approach to the ePA. Results: Three major thematic categories were identified (Contributors to Opt-Out Implementation, Barriers to Opt-Out Implementation, and Contingent Factors) to describe citizen views on the opt-out approach for the ePA. Conclusion: The public is generally supportive of an opt-out approach to ePAs in Germany, as they see the benefits ePAs can provide to German society; but they are skeptical on how successful this approach might be due to extant issues that policymakers must be aware of in order to successfully implement an opt-out approach for Germany's national ePA system.

An Approach to Potentially Increasing Adoption of an Artificial Intelligence-Enabled Electronic Medical Record Encounter in Canadian Primary Care: Protocol for a User-Centered Design.

BACKGROUND: Primary care physicians are at the forefront of the clinical process that can lead to diagnosis, referral, and treatment. With electronic medical records (EMRs) being introduced and, over time, gaining acceptance by primary care users, they have now become a standard part of care. EMRs have the potential to be further optimized with the introduction of artificial intelligence (AI). There has yet to be a widespread exploration of the use of AI in primary health care and how clinicians envision AI use to encourage further uptake. OBJECTIVE: The primary objective of this research is to understand if the user-centered design approach, rooted in contextual design, can lead to an increased likelihood of adoption of an AI-enabled encounter module embedded in a primary care EMR. In this study, we use human factor models and the technology acceptance model to understand the results. METHODS: To accomplish this, a partnership has been established with an industry partner, TELUS Health, to use their EMR, the collaborative health record. The overall intention is to understand how to improve the user experience by using user-centered design to inform how AI should be embedded in an EMR encounter. Given this intention, a user-centered approach will be used to accomplish it. The approach of user-centered design requires qualitative interviewing to gain a clear understanding of users' approaches, intentions, and other key insights to inform the design process. A total of 5 phases have been designed for this study. RESULTS: As of March 2024, a total of 14 primary care clinician participants have been recruited and interviewed. First-cycle coding of all qualitative data results is being conducted to inform redesign considerations. CONCLUSIONS: Some limitations need to be acknowledged related to the approach of this study. There is a lack of market maturity of AI-enabled EMR encounters in primary care, requiring research to take place through scenario-based interviews. However, this participant group will still help inform design considerations for this tool. This study is targeted for completion in the late fall of 2024. INTERNATIONAL REGISTERED REPORT IDENTIFIER (IRRID): DERR1-10.2196/54365.

Exploring the uses of digital health in palliative care in Southeast Asia.

BACKGROUND: This integrative review explores the use of digital health technologies in palliative care within Southeast Asia. Despite extensive documentation of digital health in palliative care in Western nations, its application in Southeast Asia remains underdeveloped. METHOD: The review includes a total of four papers meeting the eligibility criteria. FINDINGS: The findings reveal limited studies of digital health adoption in palliative care. Key technologies include mobile health applications, electronic health records and telemedicine platforms. Challenges, such as health inequities, data security and the need for technology validation were identified. The review underscores the necessity for region-specific research to address these challenges and improve the integration of digital health in palliative care. CONCLUSION: This study highlights the potential of digital health to enhance palliative care delivery and patient outcomes in Southeast Asia, advocating for increased adoption and tailored implementation strategies.

Large Language Model-Based Responses to Patients' In-Basket Messages.

Importance: Virtual patient-physician communications have increased since 2020 and negatively impacted primary care physician (PCP) well-being. Generative artificial intelligence (GenAI) drafts of patient messages could potentially reduce health care professional (HCP) workload and improve communication quality, but only if the drafts are considered useful. Objectives: To assess PCPs' perceptions of GenAI drafts and to examine linguistic characteristics associated with equity and perceived empathy. Design, Setting, and Participants: This cross-sectional quality improvement study tested the hypothesis that PCPs' ratings of GenAI drafts (created using the electronic health record [EHR] standard prompts) would be equivalent to HCP-generated responses on 3 dimensions. The study was conducted at NYU Langone Health using private patient-HCP communications at 3 internal medicine practices piloting GenAI. Exposures: Randomly assigned patient messages coupled with either an HCP message or the draft GenAI response. Main Outcomes and Measures: PCPs rated responses' information content quality (eg, relevance), using a Likert scale, communication quality (eg, verbosity), using a Likert scale, and whether they would use the draft or start anew (usable vs unusable). Branching logic further probed for empathy, personalization, and professionalism of responses. Computational linguistics methods assessed content differences in HCP vs GenAI responses, focusing on equity and empathy. Results: A total of 16 PCPs (8 [50.0%] female) reviewed 344 messages (175 GenAI drafted; 169 HCP drafted). Both GenAI and HCP responses were rated favorably. GenAI responses were rated higher for communication style than HCP responses (mean [SD], 3.70 [1.15] vs 3.38 [1.20]; P = .01, U = 12 568.5) but were similar to HCPs on information content (mean [SD], 3.53 [1.26] vs 3.41 [1.27]; P = .37; U = 13 981.0) and usable draft proportion (mean [SD], 0.69 [0.48] vs 0.65 [0.47], P = .49, t = -0.6842). Usable GenAI responses were considered more empathetic than usable HCP responses (32 of 86 [37.2%] vs 13 of 79 [16.5%]; difference, 125.5%), possibly attributable to more subjective (mean [SD], 0.54 [0.16] vs 0.31 [0.23]; P < .001; difference, 74.2%) and positive (mean [SD] polarity, 0.21 [0.14] vs 0.13 [0.25]; P = .02; difference, 61.5%) language; they were also numerically longer (mean [SD] word count, 90.5 [32.0] vs 65.4 [62.6]; difference, 38.4%), but the difference was not statistically significant (P = .07) and more linguistically complex (mean [SD] score, 125.2 [47.8] vs 95.4 [58.8]; P = .002; difference, 31.2%). Conclusions: In this cross-sectional study of PCP perceptions of an EHR-integrated GenAI chatbot, GenAI was found to communicate information better and with more empathy than HCPs, highlighting its potential to enhance patient-HCP communication. However, GenAI drafts were less readable than HCPs', a significant concern for patients with low health or English literacy.

Affordable and real-time antimicrobial resistance prediction from multimodal electronic health records.

The spread of antimicrobial resistance (AMR) leads to challenging complications and losses of human lives plus medical resources, with a high expectancy of deterioration in the future if the problem is not controlled. From a machine learning perspective, data-driven models could aid clinicians and microbiologists by anticipating the resistance beforehand. Our study serves as the first attempt to harness deep learning (DL) techniques and the multimodal data available in electronic health records (EHR) for predicting AMR. In this work, we utilize and preprocess the MIMIC-IV database extensively to produce separate structured input sources for time-invariant and time-series data customized to the AMR task. Then, a multimodality fusion approach merges the two modalities with clinical notes to determine resistance based on an antibiotic or a pathogen. To efficiently predict AMR, our approach builds the foundation for deploying multimodal DL techniques in clinical practice, leveraging the existing patient data.

Application of information from external data to correct for collider bias in a Covid-19 hospitalised cohort.

BACKGROUND: Throughout the Covid-19 pandemic, researchers have made use of electronic health records to research this disease in a rapidly evolving environment of questions and discoveries. These studies are prone to collider bias as they restrict the population of Covid-19 patients to only those with severe disease. Inverse probability weighting is typically used to correct for this bias but requires information from the unrestricted population. Using electronic health records from a South London NHS trust, this work demonstrates a method to correct for collider bias using externally sourced data while examining the relationship between minority ethnicities and poor Covid-19 outcomes. METHODS: The probability of inclusion within the observed hospitalised cohort was modelled based on estimates from published national data. The model described the relationship between patient ethnicity, hospitalisation, and death due to Covid-19 - a relationship suggested to be susceptible to collider bias. The obtained probabilities (as applied to the observed patient cohort) were used as inverse probability weights in survival analysis examining ethnicity (and covariates) as a risk factor for death due to Covid-19. RESULTS: Within the observed cohort, unweighted analysis of survival suggested a reduced risk of death in those of Black ethnicity - differing from the published literature. Applying inverse probability weights to this analysis amended this aberrant result to one more compatible with the literature. This effect was consistent when the analysis was applied to patients within only the first wave of Covid-19 and across two waves of Covid-19 and was robust against adjustments to the modelled relationship between hospitalisation, patient ethnicity, and death due to Covid-19 made as part of a sensitivity analysis. CONCLUSIONS: In conclusion, this analysis demonstrates the feasibility of using external publications to correct for collider bias (or other forms of selection bias) induced by the restriction of a population to a hospitalised cohort using an example from the recent Covid-19 pandemic.