Automatic generation of natural language nursing shift summaries in neonatal intensive care: BT-Nurse

A rapid review on current and potential uses of large language models in nursing

BACKGROUND

The application of large language models across commercial and consumer contexts has grown exponentially in recent years. However, a gap exists in the literature on how large language models can support nursing practice, education, and research. This study aimed to synthesize the existing literature on current and potential uses of large language models across the nursing profession.

METHODS

A rapid review of the literature, guided by Cochrane rapid review methodology and PRISMA reporting standards, was conducted. An expert health librarian assisted in developing broad inclusion criteria to account for the emerging nature of literature related to large language models. Three electronic databases (i.e., PubMed, CINAHL, and Embase) were searched to identify relevant literature in August 2023. Articles that discussed the development, use, and application of large language models within nursing were included for analysis.

RESULTS

The literature search identified a total of 2028 articles that met the inclusion criteria. After systematically reviewing abstracts, titles, and full texts, 30 articles were included in the final analysis. Nearly all (93 %; n = 28) of the included articles used ChatGPT as an example, and subsequently discussed the use and value of large language models in nursing education (47 %; n = 14), clinical practice (40 %; n = 12), and research (10 %; n = 3). While the most common assessment of large language models was conducted by human evaluation (26.7 %; n = 8), this analysis also identified common limitations of large language models in nursing, including lack of systematic evaluation, as well as other ethical and legal considerations.

DISCUSSION

This is the first review to summarize contemporary literature on current and potential uses of large language models in nursing practice, education, and research. Although there are significant opportunities to apply large language models, the use and adoption of these models within nursing have elicited a series of challenges, such as ethical issues related to bias, misuse, and plagiarism.

CONCLUSION

Given the relative novelty of large language models, ongoing efforts to develop and implement meaningful assessments, evaluations, standards, and guidelines for applying large language models in nursing are recommended to ensure appropriate, accurate, and safe use. Future research along with clinical and educational partnerships is needed to enhance understanding and application of large language models in nursing and healthcare.

Patient Information Summarization in Clinical Settings: Scoping Review

BACKGROUND

Information overflow, a common problem in the present clinical environment, can be mitigated by summarizing clinical data. Although there are several solutions for clinical summarization, there is a lack of a complete overview of the research relevant to this field.

OBJECTIVE

This study aims to identify state-of-the-art solutions for clinical summarization, to analyze their capabilities, and to identify their properties.

METHODS

A scoping review of articles published between 2005 and 2022 was conducted. With a clinical focus, PubMed and Web of Science were queried to find an initial set of reports, later extended by articles found through a chain of citations. The included reports were analyzed to answer the questions of where, what, and how medical information is summarized; whether summarization conserves temporality, uncertainty, and medical pertinence; and how the propositions are evaluated and deployed. To answer how information is summarized, methods were compared through a new framework "collect-synthesize-communicate" referring to information gathering from data, its synthesis, and communication to the end user.

RESULTS

Overall, 128 articles were included, representing various medical fields. Exclusively structured data were used as input in 46.1% (59/128) of papers, text in 41.4% (53/128) of articles, and both in 10.2% (13/128) of papers. Using the proposed framework, 42.2% (54/128) of the records contributed to information collection, 27.3% (35/128) contributed to information synthesis, and 46.1% (59/128) presented solutions for summary communication. Numerous summarization approaches have been presented, including extractive (n=13) and abstractive summarization (n=19); topic modeling (n=5); summary specification (n=11); concept and relation extraction (n=30); visual design considerations (n=59); and complete pipelines (n=7) using information extraction, synthesis, and communication. Graphical displays (n=53), short texts (n=41), static reports (n=7), and problem-oriented views (n=7) were the most common types in terms of summary communication. Although temporality and uncertainty information were usually not conserved in most studies (74/128, 57.8% and 113/128, 88.3%, respectively), some studies presented solutions to treat this information. Overall, 115 (89.8%) articles showed results of an evaluation, and methods included evaluations with human participants (median 15, IQR 24 participants): measurements in experiments with human participants (n=31), real situations (n=8), and usability studies (n=28). Methods without human involvement included intrinsic evaluation (n=24), performance on a proxy (n=10), or domain-specific tasks (n=11). Overall, 11 (8.6%) reports described a system deployed in clinical settings.

CONCLUSIONS

The scientific literature contains many propositions for summarizing patient information but reports very few comparisons of these proposals. This work proposes to compare these algorithms through how they conserve essential aspects of clinical information and through the "collect-synthesize-communicate" framework. We found that current propositions usually address these 3 steps only partially. Moreover, they conserve and use temporality, uncertainty, and pertinent medical aspects to varying extents, and solutions are often preliminary.

Research Trends in Artificial Intelligence-Associated Nursing Activities Based on a Review of Academic Studies Published From 2001 to 2020

The present study referred to the technology-based learning model to conduct a systematic review of the dimensions of nursing activities, research samples, research methods, roles of artificial intelligence, applied artificial intelligence algorithms, evaluation measure of algorithms, and research foci. Based on the Preferred Reporting Items for Systematic Reviews and Meta-Analyses procedure, this study obtained and analyzed a total of 102 high-quality artificial intelligence-associated nursing activities studies published from 2001 to 2020 in the Web of Science database. The results showed: (1) In terms of nursing activities, nursing management was explored the most, followed by nursing assessment; (2) quantitative methods were most frequently adopted in artificial intelligence-associated nursing activities studies to investigate issues related to patients, followed by nursing staff; (3) the most adopted roles of artificial intelligence in artificial intelligence-associated nursing activities studies were profiling and prediction, followed by assessment and evaluation; (4) artificial intelligence-associated nursing activities studies frequently mixed applied artificial intelligence algorithms and evaluation measure of algorithms; (5) in the dimension of research foci, most studies mainly paid attention to the design or evaluation of the artificial intelligence systems/instruments, followed by investigating the correlation and affect issues. Based on the findings, several recommendations are raised as a reference for future researchers, educators, and policy makers.

Deep Multi-Objective Learning from Low-Dose CT for Automatic Lung-RADS Report Generation

Radiology report generation through chest radiography interpretation is a time-consuming task that involves the interpretation of images by expert radiologists. It is common for fatigue-induced diagnostic error to occur, and especially difficult in areas of the world where radiologists are not available or lack diagnostic expertise. In this research, we proposed a multi-objective deep learning model called CT2Rep (Computed Tomography to Report) for generating lung radiology reports by extracting semantic features from lung CT scans. A total of 458 CT scans were used in this research, from which 107 radiomics features and 6 slices of segmentation related nodule features were extracted for the input of our model. The CT2Rep can simultaneously predict position, margin, and texture, which are three important indicators of lung cancer, and achieves remarkable performance with an F1-score of 87.29%. We conducted a satisfaction survey for estimating the practicality of CT2Rep, and the results show that 95% of the reports received satisfactory ratings. The results demonstrate the great potential in this model for the production of robust and reliable quantitative lung diagnosis reports. Medical personnel can obtain important indicators simply by providing the lung CT scan to the system, which can bring about the widespread application of the proposed framework.

Building a Persuasive Virtual Dietitian

This paper describes the Multimedia Application for Diet Management (MADiMan), a system that supports users in managing their diets while admitting diet transgressions. MADiMan consists of a numerical reasoner that takes into account users’ dietary constraints and automatically adapts the users’ diet, and of a natural language generation (NLG) system that automatically creates textual messages for explaining the results provided by the reasoner with the aim of persuading users to stick to a healthy diet. In the first part of the paper, we introduce the MADiMan system and, in particular, the basic mechanisms related to reasoning, data interpretation and content selection for a numeric data-to-text NLG system. We also discuss a number of factors influencing the design of the textual messages produced. In particular, we describe in detail the design of the sentence-aggregation procedure, which determines the compactness of the final message by applying two aggregation strategies. In the second part of the paper, we present the app that we developed, CheckYourMeal!, and the results of two human-based quantitative evaluations of the NLG module conducted using CheckYourMeal! in a simulation. The first evaluation, conducted with twenty users, ascertained both the perceived usefulness of graphics/text and the appeal, easiness and persuasiveness of the textual messages. The second evaluation, conducted with thirty-nine users, ascertained their persuasive power. The evaluations were based on the analysis of questionnaires and of logged data of users’ behaviour. Both evaluations showed significant results.

Personality-dependent content selection in natural language generation systems

This paper focuses on the computer side of human-computer interaction through natural language, which is the domain of natural language generation (NLG) studies. From a given (usually non-linguistic) input, NLG systems will in principle generate the same fixed text as an output and in order to attain more natural or human-like interaction will often resort to a wide range of strategies for stylistic variation. Among these, the use of computational models of human personality has emerged as a popular alternative in the field and will be the focus of the present work as well. More specifically, the present study describes two machine learning experiments to establish possible relations between personality and content selection (as opposed to the more well-documented relation between personality and surface realisation), and it is, to the best of our knowledge, the first of its kind to address this issue at both macro and micro planning levels, which may arguably pave the way for the future development of more robust personality-dependent systems of this kind.

Natural language generation for electronic health records

One broad goal of biomedical informatics is to generate fully-synthetic, faithfully representative electronic health records (EHRs) to facilitate data sharing between healthcare providers and researchers and promote methodological research. A variety of methods existing for generating synthetic EHRs, but they are not capable of generating unstructured text, like emergency department (ED) chief complaints, history of present illness, or progress notes. Here, we use the encoder–decoder model, a deep learning algorithm that features in many contemporary machine translation systems, to generate synthetic chief complaints from discrete variables in EHRs, like age group, gender, and discharge diagnosis. After being trained end-to-end on authentic records, the model can generate realistic chief complaint text that appears to preserve the epidemiological information encoded in the original record-sentence pairs. As a side effect of the model’s optimization goal, these synthetic chief complaints are also free of relatively uncommon abbreviation and misspellings, and they include none of the personally identifiable information (PII) that was in the training data, suggesting that this model may be used to support the de-identification of text in EHRs. When combined with algorithms like generative adversarial networks (GANs), our model could be used to generate fully-synthetic EHRs, allowing healthcare providers to share faithful representations of multimodal medical data without compromising patient privacy. This is an important advance that we hope will facilitate the development of machine-learning methods for clinical decision support, disease surveillance, and other data-hungry applications in biomedical informatics.

Natural language generation for electronic health records

One broad goal of biomedical informatics is to generate fully-synthetic, faithfully representative electronic health records (EHRs) to facilitate data sharing between healthcare providers and researchers and promote methodological research. A variety of methods existing for generating synthetic EHRs, but they are not capable of generating unstructured text, like emergency department (ED) chief complaints, history of present illness, or progress notes. Here, we use the encoder-decoder model, a deep learning algorithm that features in many contemporary machine translation systems, to generate synthetic chief complaints from discrete variables in EHRs, like age group, gender, and discharge diagnosis. After being trained end-to-end on authentic records, the model can generate realistic chief complaint text that appears to preserve the epidemiological information encoded in the original record-sentence pairs. As a side effect of the model's optimization goal, these synthetic chief complaints are also free of relatively uncommon abbreviation and misspellings, and they include none of the personally identifiable information (PII) that was in the training data, suggesting that this model may be used to support the de-identification of text in EHRs. When combined with algorithms like generative adversarial networks (GANs), our model could be used to generate fully-synthetic EHRs, allowing healthcare providers to share faithful representations of multimodal medical data without compromising patient privacy. This is an important advance that we hope will facilitate the development of machine-learning methods for clinical decision support, disease surveillance, and other data-hungry applications in biomedical informatics.

Evaluation of an automated knowledge-based textual summarization system for longitudinal clinical data, in the intensive care domain

OBJECTIVES

To examine the feasibility of the automated creation of meaningful free-text summaries of longitudinal clinical records, using a new general methodology that we had recently developed; and to assess the potential benefits to the clinical decision-making process of using such a method to generate draft letters that can be further manually enhanced by clinicians.

METHODS

We had previously developed a system, CliniText (CTXT), for automated summarization in free text of longitudinal medical records, using a clinical knowledge base. In the current study, we created an Intensive Care Unit (ICU) clinical knowledge base, assisted by two ICU clinical experts in an academic tertiary hospital. The CTXT system generated free-text summary letters from the data of 31 different patients, which were compared to the respective original physician-composed discharge letters. The main evaluation measures were (1) relative completeness, quantifying the data items missed by one of the letters but included by the other, and their importance; (2) quality parameters, such as readability; (3) functional performance, assessed by the time needed, by three clinicians reading each of the summaries, to answer five key questions, based on the discharge letter (e.g., "What are the patient's current respiratory requirements?"), and by the correctness of the clinicians' answers.

RESULTS

Completeness: In 13/31 (42%) of the letters the number of important items missed in the CTXT-generated letter was actually less than or equal to the number of important items missed by the MD-composed letter. In each of the MD-composed letters, at least two important items that were mentioned by the CTXT system were missed (a mean of 7.2±5.74). In addition, the standard deviation in the number of missed items in the MD letters (STD=15.4) was much higher than the standard deviation in the CTXT-generated letters (STD=5.3). Quality: The MD-composed letters obtained a significantly better grade in three out of four measured parameters. However, the standard variation in the quality of the MD-composed letters was much greater than the standard variation in the quality of the CTXT-generated letters (STD=6.25 vs. STD=2.57, respectively). Functional evaluation: The clinicians answered the five questions on average 40% faster (p<0.001) when using the CTXT-generated letters than when using the MD-composed letters. In four out of the five questions the clinicians' correctness was equal to or significantly better (p<0.005) when using the CTXT-generated letters than when using the MD-composed letters.

CONCLUSIONS

An automatic knowledge-based summarization system, such as the CTXT system, has the capability to model complex clinical domains, such as the ICU, and to support interpretation and summarization tasks such as the creation of a discharge summary letter. Based on the results, we suggest that the use of such systems could potentially enhance the standardization of the letters, significantly increase their completeness, and reduce the time to write the discharge summary. The results also suggest that using the resultant structured letters might reduce the decision time, and enhance the decision quality, of decisions made by other clinicians.

Effects of Individual Differences in Working Memory on Plan Presentational Choices

This paper addresses research questions that are central to the area of visualization interfaces for decision support: (RQ1) whether individual user differences in working memory should be considered when choosing how to present visualizations; (RQ2) how to present the visualization to support effective decision making and processing; and (RQ3) how to evaluate the effectiveness of presentational choices. These questions are addressed in the context of presenting plans, or sequences of actions, to users. The experiments are conducted in several domains, and the findings are relevant to applications such as semi-autonomous systems in logistics. That is, scenarios that require the attention of humans who are likely to be interrupted, and require good performance but are not time critical. Following a literature review of different types of individual differences in users that have been found to affect the effectiveness of presentational choices, we consider specifically the influence of individuals' working memory (RQ1). The review also considers metrics used to evaluate presentational choices, and types of presentational choices considered. As for presentational choices (RQ2), we consider a number of variants including interactivity, aggregation, layout, and emphasis. Finally, to evaluate the effectiveness of plan presentational choices (RQ3) we adopt a layered-evaluation approach and measure performance in a dual task paradigm, involving both task interleaving and evaluation of situational awareness. This novel methodology for evaluating visualizations is employed in a series of experiments investigating presentational choices for a plan. A key finding is that emphasizing steps (by highlighting borders) can improve effectiveness on a primary task, but only when controlling for individual variation in working memory.

Textual summarization of events leading to health alerts

Extracting information from the sensors installed in the homes of elderly pose a unique set of challenges. Add to it the short amount of time the clinicians and nurses have to analyze this data, and the problem becomes more complicated. A system already in place at an "Aging in Place" facility monitors the activities of residents through multiple non-intrusive sensors and sends alerts on detecting an unusual event. We present an approach to generate textual summaries of events leading to the alerts. We analyze our system using four case studies and also list the comments provided by collaborators in healthcare domain. The system was then iterated to take some of those suggestions into account to give a glimpse of what an ideal system should look like.

Applying artificial intelligence technology to support decision-making in nursing: A case study in Taiwan

This study applied artificial intelligence to help nurses address problems and receive instructions through information technology. Nurses make diagnoses according to professional knowledge, clinical experience, and even instinct. Without comprehensive knowledge and thinking, diagnostic accuracy can be compromised and decisions may be delayed. We used a back-propagation neural network and other tools for data mining and statistical analysis. We further compared the prediction accuracy of the previous methods with an adaptive-network-based fuzzy inference system and the back-propagation neural network, identifying differences in the questions and in nurse satisfaction levels before and after using the nursing information system. This study investigated the use of artificial intelligence to generate nursing diagnoses. The percentage of agreement between diagnoses suggested by the information system and those made by nurses was as much as 87 percent. When patients are hospitalized, we can calculate the probability of various nursing diagnoses based on certain characteristics.

Electronic health records and transgender patients--practical recommendations for the collection of gender identity data

Transgender (Trans, Trans*) persons may have a gender identity and a preferred name that differ from those assigned at birth, and/or those listed on their current legal identification (Gender ID, Birth-assigned Sex, Legal Sex). Transgender people who are referred to in a clinical setting using the wrong pronoun or name may suffer distress, ridicule or even assault by others in the waiting area, and may not return for further care. Furthermore, failure to accurately document (and therefore count) transgender identities has negative implications on quality improvement and research efforts, funding priorities and policy activities. The recent announcement that gender identity data may be included in Meaningful Use Stage 3 has accelerated the need for guidance for both vendors and local implementation teams on how to best record and store these data. A recent study demonstrated wide variation in current practices. This manuscript provides a description of identifiers associated with gender identity, and makes practical and evidence based recommendations for implementation and front-end functionality.

Nursing clinical handover in neonatal care

BACKGROUND

Clinical handover ensures continuity of care, providing the opportunity to transfer responsibility and accountability for the care of a patient from nurse to nurse.

AIM

The aim of this study was to examine afternoon nursing clinical handover from the perspective of nurses, exploring the quality of information, the interactions and support, the efficiency and the involvement of parents in a private neonatal unit.

METHODS/DESIGN

An exploratory, descriptive, prospective quantitative survey with qualitative elements was undertaken using The Handover Evaluation Scale (O'Connell, MacDonald, & Kelly, 2008). All nurses working in the Neonatal unit who attend afternoon handover, were invited to participate in the study (N = 22), with N = 16 responses received.

RESULTS/FINDINGS

The quantitative and qualitative results indicate that the quality of the information handed over in neonatal care units can be maintained despite intrinsic limitations. Additionally, high levels of support and interaction between nursing staff in this stressful practice environment occur during the handover period.

CONCLUSION

Given the vulnerability of neonates it is important that accurate information is efficiently handed over. In order to do this distractions should be minimised. Finally, parental involvement in handover should be actively encouraged whenever feasible.

Data mining for wearable sensors in health monitoring systems: a review of recent trends and challenges

The past few years have witnessed an increase in the development of wearable sensors for health monitoring systems. This increase has been due to several factors such as development in sensor technology as well as directed efforts on political and stakeholder levels to promote projects which address the need for providing new methods for care given increasing challenges with an aging population. An important aspect of study in such system is how the data is treated and processed. This paper provides a recent review of the latest methods and algorithms used to analyze data from wearable sensors used for physiological monitoring of vital signs in healthcare services. In particular, the paper outlines the more common data mining tasks that have been applied such as anomaly detection, prediction and decision making when considering in particular continuous time series measurements. Moreover, the paper further details the suitability of particular data mining and machine learning methods used to process the physiological data and provides an overview of the properties of the data sets used in experimental validation. Finally, based on this literature review, a number of key challenges have been outlined for data mining methods in health monitoring systems.

Data-to-text summarisation of patient records: using computer-generated summaries to access patient histories

OBJECTIVE

We assess the efficacy and utility of automatically generated textual summaries of patients' medical histories at the point of care.

METHOD

Twenty-one clinicians were presented with information about two cancer patients and asked to answer key questions. For each clinician, the information on one of the patients comprised their official hospital records, and for the other patient it comprised summaries that were computer-generated by a natural language generation system from data extracted from the official records. We measured the accuracy of the clinicians' responses to the questions, the time they took to complete them, and recorded their attitude to the computer-generated summaries.

RESULTS

Results showed no significant difference in the accuracy of responses to the computer-generated records over the official records, but a significant difference in the time taken to assess the patients' condition from the computer-generated records. Clinicians expressed a positive attitude towards the computer-generated records.

CONCLUSION

AI-based computer-generated textual summaries of patient histories can be as accurate as, and more efficient than, human-produced patient records for clinicians seeking to accurately identify key information about a patients overall history.

PRACTICE IMPLICATIONS

Computer-generated textual summaries of patient histories can contribute to the management of patients at the point-of-care.

Implementation of artifact detection in critical care: a methodological review

Artifact detection (AD) techniques minimize the impact of artifacts on physiologic data acquired in critical care units (CCU) by assessing quality of data prior to clinical event detection (CED) and parameter derivation (PD). This methodological review introduces unique taxonomies to synthesize over 80 AD algorithms based on these six themes: 1) CCU; 2) physiologic data source; 3) harvested data; 4) data analysis; 5) clinical evaluation; and 6) clinical implementation. Review results show that most published algorithms: a) are designed for one specific type of CCU; b) are validated on data harvested only from one OEM monitor; c) generate signal quality indicators (SQI) that are not yet formalized for useful integration in clinical workflows; d) operate either in standalone mode or coupled with CED or PD applications; e) are rarely evaluated in real-time; and f) are not implemented in clinical practice. In conclusion, it is recommended that AD algorithms conform to generic input and output interfaces with commonly defined data: 1) type; 2) frequency; 3) length; and 4) SQIs. This shall promote: a) reusability of algorithms across different CCU domains; b) evaluation on different OEM monitor data; c) fair comparison through formalized SQIs; d) meaningful integration with other AD, CED and PD algorithms; and e) real-time implementation in clinical workflows.