Natural Language Processing Software Research Articles

Understanding treatment preferences and cognitive outcomes in patients with gliomas.

Understanding how glioma patients value cognitive outcomes is essential to personalizing their treatment plans. The purpose of this study was to identify the modifiable cognitive functions most affected by treatment and most important to patient quality of life. Patients with gliomas were prospectively enrolled in focus groups and individual interviews using a standardized guide focusing on cognitive functions until saturation was achieved. Patient values and treatment preferences were elicited and compared to the frequency of reported deficits. NVivo natural language processing software was used to perform thematic qualitative analyses. Quantitative analysis with Fischer's exact test was used for each cognitive function to assess for an association between experiencing a deficit and rating that function as important to quality of life. Twenty participants participated, of whom 60% were female. Racial identification consisted of 75% White, 15% Black/African American, and 10% OtherRacial Identification. The cognitive functions most essential to the quality of life in this cohort were sense of self (80% of participants), memory (70% of participants), and communication (25% of participants). The functions that experienced the most deficits because of treatment were memory (65% of participants), concentration (65% of participants), and special senses (40% of participants). "Dealbreakers" to treatment were complete loss of independence, sense of self, and/orthe ability to interact with loved ones. Fischer's exact test showed no associations between experiencing a cognitive function deficit and rating that function as important to quality of life. Glioma patients in this study prioritized cognitive functions according to memory, personal identity, and their ability to communicate with loved ones independently of experiencing deficits in these functions. Further study should comparepatient prioritization and decision-making between surgically curable and noncurable grade gliomas as well as investigate the quality of life benefits of incorporating the connectomics of highly valued cognitive functions in surgical planning.

Disparities in Documentation: Evidence of Race-Based Biases in the Electronic Medical Record.

Personal implicit biases may contribute to inequitable health outcomes, but the mechanisms of these effects are unclear at a system level. This study aimed to determine whether stigmatizing subjective terms in electronic medical records (EMR) reflect larger societal racial biases. A cross-sectional study was conducted using natural language processing software of all documentation where one or more predefined stigmatizing words were used between January 1, 2019 and June 30, 2021. EMR from emergency care and inpatient encounters in a metropolitan healthcare system were analyzed, focused on the presence or absence of race-based differences in word usage, either by specific terms or by groupings of negative or positive terms based on the common perceptions of the words. The persistence ("stickiness") of negative and/or positive characterizations in subsequent encounters for an individual was also evaluated. Final analyses included 12,238 encounters for 9135 patients, ranging from newborn to 104years old. White (68%) vs Black/African American (17%) were the analyzed groups. Several negative terms (e.g., noncompliant, disrespectful, and curse words) were significantly more frequent in encounters with Black/African American patients. In contrast, positive terms (e.g., compliant, polite) were statistically more likely to be in White patients' documentation. Independent of race, negative characterizations were twice as likely to persist compared with positive ones in subsequent encounters. The use of stigmatizing language in documentation mirrors the same race-based inequities seen in medical outcomes and larger sociodemographic trends. This may contribute to observed healthcare outcome differences by disseminating one's implicit biases to unknown future healthcare providers.

Artificial Intelligence-Based Co-Facilitator (AICF) for Detecting and Monitoring Group Cohesion Outcomes in Web-Based Cancer Support Groups: Single-Arm Trial Study.

Commonly offered as supportive care, therapist-led online support groups (OSGs) are a cost-effective way to provide support to individuals affected by cancer. One important indicator of a successful OSG session is group cohesion; however, monitoring group cohesion can be challenging due to the lack of nonverbal cues and in-person interactions in text-based OSGs. The Artificial Intelligence-based Co-Facilitator (AICF) was designed to contextually identify therapeutic outcomes from conversations and produce real-time analytics. The aim of this study was to develop a method to train and evaluate AICF's capacity to monitor group cohesion. AICF used a text classification approach to extract the mentions of group cohesion within conversations. A sample of data was annotated by human scorers, which was used as the training data to build the classification model. The annotations were further supported by finding contextually similar group cohesion expressions using word embedding models as well. AICF performance was also compared against the natural language processing software Linguistic Inquiry Word Count (LIWC). AICF was trained on 80,000 messages obtained from Cancer Chat Canada. We tested AICF on 34,048 messages. Human experts scored 6797 (20%) of the messages to evaluate the ability of AICF to classify group cohesion. Results showed that machine learning algorithms combined with human input could detect group cohesion, a clinically meaningful indicator of effective OSGs. After retraining with human input, AICF reached an F1-score of 0.82. AICF performed slightly better at identifying group cohesion compared to LIWC. AICF has the potential to assist therapists by detecting discord in the group amenable to real-time intervention. Overall, AICF presents a unique opportunity to strengthen patient-centered care in web-based settings by attending to individual needs. RR2-10.2196/21453.

The effects of politeness in shaping discourse in online debates

Computer-supported collaborative argumentation is an online activity that can engage students in deep discussion and analysis of complex problems. Given the potentially confrontational nature of argumentation, using polite language becomes a strategic approach to prevent breakdowns in group communication and nurture productive dialogues. This study aims to understand how politeness and argumentation moves influence subsequent conversation dynamics in online debates. Student postings in threaded discussions (from 20 online debates containing 2,008 messages posted by students across five semesters of a graduate-level course on distance education) were coded and scored on politeness and impoliteness using natural language processing software. The scored postings were examined to determine how impoliteness and politeness impact students’ proclivity to engage in and produce more sustained argumentative exchanges to evaluate presented claims thoroughly. The findings show that impoliteness steadily increased by up to 84% as contentious discussion threads reached 7 levels deep, and impoliteness was significantly higher in frequency in postings presenting challenges. Correspondingly, the use of politeness increased steadily by up to 139% as discussion threads reached 6 levels deep, with the frequency of politeness nearly equal across postings presenting challenges, explanations, and evidence.

ChatGPT vs. neurologists: a cross-sectional study investigating preference, satisfaction ratings and perceived empathy in responses among people living with multiple sclerosis.

ChatGPT is an open-source natural language processing software that replies to users' queries. We conducted a cross-sectional study to assess people living with Multiple Sclerosis' (PwMS) preferences, satisfaction, and empathy toward two alternate responses to four frequently-asked questions, one authored by a group of neurologists, the other by ChatGPT. An online form was sent through digital communication platforms. PwMS were blind to the author of each response and were asked to express their preference for each alternate response to the four questions. The overall satisfaction was assessed using a Likert scale (1-5); the Consultation and Relational Empathy scale was employed to assess perceived empathy. We included 1133 PwMS (age, 45.26 ± 11.50years; females, 68.49%). ChatGPT's responses showed significantly higher empathy scores (Coeff = 1.38; 95% CI = 0.65, 2.11; p > z < 0.01), when compared with neurologists' responses. No association was found between ChatGPT' responses and mean satisfaction (Coeff = 0.03; 95% CI = -0.01, 0.07; p = 0.157). College graduate, when compared with high school education responder, had significantly lower likelihood to prefer ChatGPT response (IRR = 0.87; 95% CI = 0.79, 0.95; p < 0.01). ChatGPT-authored responses provided higher empathy than neurologists. Although AI holds potential, physicians should prepare to interact with increasingly digitized patients and guide them on responsible AI use. Future development should consider tailoring AIs' responses to individual characteristics. Within the progressive digitalization of the population, ChatGPT could emerge as a helpful support in healthcare management rather than an alternative.

Psychiatry Pitstop: Enhancing Communication Skills of Medical Students in Mental Health Settings

IntroductionPsychiatry Pitstop is a role-play-based program for medical students aimed to improve communication skills in the framework of mental health. The workshop involved amateur actors who simulated different clinical scenarios and psychiatry residents, who facilitated the sessions and provided constructive feedback following the Pendleton method. Psychiatry Pitstop was originally developed in the United Kingdom and it was expanded to Lisbon, Portugal, in 2019. The authors adapted the course to the Portuguese context, adjusting the number of sessions and altering the scenarios to match common clinical situations faced by junior doctors in Portugal. By now, we conducted four courses.ObjectivesOur study aims to describe the Portuguese adaptation of the program and to learn insights from the students feedback.MethodsThe course was assessed using satisfaction questionnaires, completed by the students after each session. These included a Likert scale ranging from 1 to 5, with items pertaining to Future Importance, Overall Quality, Theoretical Quality, and Practical Quality. Quantitative data was analyzed using Excel and standard descriptive statistics to summarize the results. The open questions invited students to articulate the main positive aspects, suggestions for improvement and future topics. A Natural Language Processing (NLP) software was used to evaluate open-ended responses and extract the main concepts.ResultsWe obtained a total of 39 single-answers from 4 different courses. Evaluation results yielded a mean score of 4.7 for Future Importance, 4.9 for Overall Quality, 4.3 for Theoretical Teaching, and 4.9 for Practical Teaching. Notable positive aspects included students’ appreciation of the immersive interview environment, the dedication exhibited by actors and doctors, well-prepared case scenarios, and engaging interactions with participants. Suggestions for improvement encompass enhanced theoretical introductions, comprehensive topic coverage, universal participation in simulations, and expanded workshop days. Future prospects for the program include practicing interviews with other psychiatric diagnosis, addressing difficult patients, delivering bad news and covering topics related to sexuality, grief and moral dilemmas.ConclusionsOur study shows that Psychiatry Pitstop adaptation to the Portuguese context was successful. Overall, the feedback from medical students has been consistently positive. Subsequent editions will draw upon the findings of this study to enhance overall program quality.Disclosure of InterestNone Declared

Information heterogeneity between progress notes by physicians and nurses for inpatients with digestive system diseases

This study focused on the heterogeneity in progress notes written by physicians or nurses. A total of 806 days of progress notes written by physicians or nurses from 83 randomly selected patients hospitalized in the Gastroenterology Department at Kagawa University Hospital from January to December 2021 were analyzed. We extracted symptoms as the International Classification of Diseases (ICD) Chapter 18 (R00–R99, hereinafter R codes) from each progress note using MedNER-J natural language processing software and counted the days one or more symptoms were extracted to calculate the extraction rate. The R-code extraction rate was significantly higher from progress notes by nurses than by physicians (physicians 68.5% vs. nurses 75.2%; p = 0.00112), regardless of specialty. By contrast, the R-code subcategory R10–R19 for digestive system symptoms (44.2 vs. 37.5%, respectively; p = 0.00299) and many chapters of ICD codes for disease names, as represented by Chapter 11 K00–K93 (68.4 vs. 30.9%, respectively; p < 0.001), were frequently extracted from the progress notes by physicians, reflecting their specialty. We believe that understanding the information heterogeneity of medical documents, which can be the basis of medical artificial intelligence, is crucial, and this study is a pioneering step in that direction.

Development and Practical Applications of Computational Intelligence Technology

Computational intelligence (CI) uses applied computational methods for problem-solving inspired by the behavior of humans and animals. Biological systems are used to construct software to solve complex problems, and one type of such system is an artificial immune system (AIS), which imitates the immune system of a living body. AISs have been used to solve problems that require identification and learning, such as computer virus identification and removal, image identification, and function optimization problems. In the body’s immune system, a wide variety of cells work together to distinguish between the self and non-self and to eliminate the non-self. AISs enable learning and discrimination by imitating part or all of the mechanisms of a living body’s immune system. Certainly, some deep neural networks have exceptional performance that far surpasses that of humans in certain tasks, but to build such a network, a huge amount of data is first required. These networks are used in a wide range of applications, such as extracting knowledge from a large amount of data, learning from past actions, and creating the optimal solution (the optimization problem). A new technique for pre-training natural language processing (NLP) software ver.9.1by using transformers called Bidirectional Encoder Representations (BERT) builds on recent research in pre-training contextual representations, including Semi-Supervised Sequence Learning, Generative Pre-Training, ELMo (Embeddings from Language Models), which is a method for obtaining distributed representations that consider context, and ULMFit (Universal Language Model Fine-Tuning). BERT is a method that can address the issue of the need for large amounts of data, which is inherent in large-scale models, by using pre-learning with unlabeled data. An optimization problem involves “finding a solution that maximizes or minimizes an objective function under given constraints”. In recent years, machine learning approaches that consider pattern recognition as an optimization problem have become popular. This pattern recognition is an operation that associates patterns observed as spatial and temporal changes in signals with classes to which they belong. It involves identifying and retrieving predetermined features and rules from data; however, the features and rules here are not logical information, but are found in images, sounds, etc. Therefore, pattern recognition is generally conducted by supervised learning. Based on a new theory that deals with the process by which the immune system learns from past infection experiences, the clonal selection of immune cells can be viewed as a learning rule of reinforcement learning.

Leveraging artificial intelligence in neurosurgery—unveiling ChatGPT for neurosurgical discharge summaries and operative reports

PurposeChat generative pre-trained transformer (GPT) is a novel large pre-trained natural language processing software that can enable scientific writing amongst a litany of other features. Given this, there is a growing interest in exploring the use of ChatGPT models as a modality to facilitate/assist in the provision of clinical care.MethodsWe investigated the time taken for the composition of neurosurgical discharge summaries and operative reports at a major University hospital. In so doing, we compared currently employed speech recognition software (i.e., SpeaKING) vs novel ChatGPT for three distinct neurosurgical diseases: chronic subdural hematoma, spinal decompression, and craniotomy. Furthermore, factual correctness was analyzed for the abovementioned diseases.ResultsThe composition of neurosurgical discharge summaries and operative reports with the assistance of ChatGPT leads to a statistically significant time reduction across all three diseases/report types: p < 0.001 for chronic subdural hematoma, p < 0.001 for decompression of spinal stenosis, and p < 0.001 for craniotomy and tumor resection. However, despite a high degree of factual correctness, the preparation of a surgical report for craniotomy proved to be significantly lower (p = 0.002).ConclusionChatGPT assisted in the writing of discharge summaries and operative reports as evidenced by an impressive reduction in time spent as compared to standard speech recognition software. While promising, the optimal use cases and ethics of AI-generated medical writing remain to be fully elucidated and must be further explored in future studies.

Teen-adult interactions during the co-design of data literacy activities for the public library: insights from a natural language processing analysis of linguistic patterns

PurposeThe purpose of this study is to explore teen-adult dialogic interactions during the co-design of data literacy activities in order to determine the nature of teen thinking, their emotions, level of engagement, and the power of relationships between teens and adults in the context of data literacy. This study conceives of co-design as a learning space for data literacy. It investigates the teen–adult dialogic interactions and what these interactions say about the nature of teen thinking, their emotions, level of engagement and the power relationships between teens and adults.Design/methodology/approachThe study conceives of co-design as a learning space for teens. Linguistic Inquiry and Word Count (LIWC-22), a natural language processing (NLP) software tool, was used to examine the linguistic measures of Analytic Thinking, Clout, Authenticity, and Emotional Tone using transcriptions of recorded Data Labs with teens and adults. Linguistic Inquiry and Word Count (LIWC-22), a natural language processing (NLP) software tool, was used to examine the linguistic measures of Analytic Thinking, Clout, Authenticity and Emotional Tone using transcriptions of recorded Data Labs with teens and adults.FindingsLIWC-22 scores on the linguistic measures Analytic Thinking, Clout, Authenticity and Emotional Tone indicate that teens had a high level of friendly engagement, a relatively low sense of power compared with the adult co-designers, medium levels of spontaneity and honesty and the prevalence of positive emotions during the co-design sessions.Practical implicationsThis study provides a concrete example of how to apply NLP in the context of data literacy in the public library, mapping the LIWC-22 findings to STEM-focused informal learning. It adds to the understanding of assessment/measurement tools and methods for designing data literacy education, stimulating further research and discussion on the ways to empower youth to engage more actively in informal learning about data.Originality/valueThis study applies a novel approach for exploring teen engagement within a co-design project tasked with the creation of youth-oriented data literacy activities.

Further Validation of a Self‐administered Smartphone‐based Verbal Learning and Memory Assessment

AbstractBackgroundThe International Shopping List test (ISLT) is a verbal word list learning test set within a realistic shopping list context, supporting ease of cultural adaptation, and addressing content validity and patient relevance. Improvements to and the wide adoption of natural language processing (NLP) software has enabled the possibility for self‐administered tests of verbal learning. LilaTM a self‐administered, smartphone‐application (app) was developed, using the virtual assistant and NLP. We have previously reported acceptability and usability data from qualitative analysis of concept elicitation interviews. We now seek to provide preliminary validation data around its use.MethodA convenience sample of healthy young adults (n = 40), M(SD)age = 35 (16), were invited to install and complete one assessment of the LilaTM app. A usability survey was also administered to explore participants’ experiences. The sample was predominantly female (70%) and of Asian ethnicity (62.5%). Data were collected in Australia, using an English word pool developed for the USA.ResultAcceptability was generally high with N = 38 completing immediate recall rounds and N = 32 completing the delayed recall round. Eight‐two percent found it extremely or somewhat easy to hear and understand what items were on the shopping list and 95% indicated that it was extremely or somewhat easy to understand what they had to do to complete the assessment. Scoring accuracy was 88% at the word item level, with reprocessing to address accent, context setting, and other NLP approaches to improve accuracy underway e.g., homophones biased to the grocery item.ConclusionWhilst acceptability was high, the importance of localization of test stimuli was confirmed as well as the need for additional data processing options to account for accent. Continued optimization of data processing to score participant responses and word item pools to reduce ambiguity is expected to further improve accuracy e.g., accounting for regional accents and consideration of homophones and compound words. However, specific characteristics of speech including in those for whom English may be a second/foreign language will be an area of focus for future development.

Further Validation of a Self‐administered Smartphone‐based Verbal Learning and Memory Assessment

AbstractBackgroundThe International Shopping List test (ISLT) is a verbal word list learning test set within a realistic shopping list context, supporting ease of cultural adaptation, and addressing content validity and patient relevance. Improvements to and the wide adoption of natural language processing (NLP) software has enabled the possibility for self‐administered tests of verbal learning. LilaTM a self‐administered, smartphone‐application (app) was developed, using the virtual assistant and NLP. We have previously reported acceptability and usability data from qualitative analysis of concept elicitation interviews. We now seek to provide preliminary validation data around its use.MethodA convenience sample of healthy young adults (n = 40), M(SD)age = 35 (16), were invited to install and complete one assessment of the LilaTM app. A usability survey was also administered to explore participants’ experiences. The sample was predominantly female (70%) and of Asian ethnicity (62.5%). Data were collected in Australia, using an English word pool developed for the USA.ResultAcceptability was generally high with N = 38 completing immediate recall rounds and N = 32 completing the delayed recall round. Eight‐two percent found it extremely or somewhat easy to hear and understand what items were on the shopping list and 95% indicated that it was extremely or somewhat easy to understand what they had to do to complete the assessment. Scoring accuracy was 88% at the word item level, with reprocessing to address accent, context setting, and other NLP approaches to improve accuracy underway e.g., homophones biased to the grocery item.ConclusionWhilst acceptability was high, the importance of localization of test stimuli was confirmed as well as the need for additional data processing options to account for accent. Continued optimization of data processing to score participant responses and word item pools to reduce ambiguity is expected to further improve accuracy e.g., accounting for regional accents and consideration of homophones and compound words. However, specific characteristics of speech including in those for whom English may be a second/foreign language will be an area of focus for future development.

Racial Discrepancies in the Use of Stigmatizing Language Is Evident in Hematology/Oncology Electronic Medical Records

Healthcare inequities are often attributed to large-scale structural barriers for access to medical care or patient-medical staff interactions. However, the role of the electronic medical record (EMR) in perpetuating hidden biases has been poorly characterized. This passive form of sharing implicit biases with unknown readers may have profound ripple effects asynchronous in place and time from the note writer. This study aims to illustrate racial imbalances in ad hominem patient characterizations as documented by medical staff, regardless of role, in a large metro healthcare system. The use of certain potentially stigmatizing language in emergency department (ED) and inpatient notes for hematology/oncology patients was investigated. Retrospective chart review of four hospitals within a university network in a major metro area between 1/1/2019 - 6/30/2021 was undertaken. Patients ≥22 years old were included with two primary hypotheses in mind: 1) Negative terms are used more frequently for underrepresented minorities, and 2) words with shock value (i.e. curse words) are used more often than more general descriptors of these terms (e.g. “cursing” or “profanity”). Nine positive and twelve negative terms were evaluated, based on the authors' observations of their use in the EMR in past encounters (Table 1). Some were personal descriptors in which the ED or inpatient note writer would be less likely than outpatient providers to have a full sense of outpatient care or social determinants of health (e.g. compliant). Others were behavioral descriptions (e.g. hostile) or words with minimal medical relevance (e.g. curse words). If multiple notes in an encounter included the same word(s), the encounter was only analyzed once for each word to avoid word count inflation, though re-occurrences for the same patient in subsequent encounters were counted. Authors' manual data review using natural language processing software confirmed appropriate use of terms. Generalized estimating equations were used for race-based comparisons to account for same-patient clustered encounters. Odds ratios (OR) and 95% confidence intervals (CI) were calculated for self-reported demographic comparisons with White as the reference category. P≤0.05 was considered significant. A total of 1,059 encounters for 889 hematology/oncology patients were analyzed. Mean age was 61 years (standard deviation 16.7) and 49.4% were female. White race was reported for 80.2% of patients (72% system-wide prevalence during same timeframe) and Black for 11.4% (15.2% system-wide); all others were ≤4% so only White/Black comparisons were included. Positive terms were more prevalent than negative ones (n=535 vs 459). However, negative terms were found more often for Black individuals (OR 2.24, CI 1.41 - 3.56, P&amp;lt;0.001), while the positive terms were less frequently documented in Black patients relative to White patients (OR 0.51, CI 0.33 - 0.77, P=0.001), and neither category was concordant with the overall patient racial demographics. Other significant racial differences existed for the terms “Argumentative”, “Disrespectful”, “Noncompliant”, and “Polite”, all discrepancies which favored White patients (Table 2). Specific curse words were documented nearly as often as “cursing”, “swearing”, or “profanity”, but no racial differences were observed. Some words in the initial search (“Brave”, “Courageous”, “Empathetic”, and “Ambitious”) were not found in hematology/oncology notes in this timeframe. EMR documentation in hospital-based hematology/oncology care demonstrated evidence of racial discrepancies in language that may carry stigma, positively or negatively. Importantly, many of these terms have questionable, if any, medical relevance. The use of stigmatizing terminology in the EMR may reflect implicit biases that extend past the direct patient encounter, passively and subconsciously prejudicing future care from other staff members. Time-saving heuristics and documentation shortcuts like copy-forwarding may further ossify healthcare inequities. This phenomenon is undoubtedly not limited to this hospital system or the hematology/oncology field, but quantitative real-world evidence has been scant. Improved awareness of stigmatizing language in documentation may be one step towards mitigating this contributor to unequal care for hematology/oncology providers and their patients.

Abstract 17253: Concepts Associated With 30-day Rehospitalization From Unstructured Data Among Patients With Heart Failure

Introduction: Heart failure (HF) is a diagnosis for which the Centers for Medicare &amp; Medicaid Services mandated public reporting on 30-day risk-standardized rehospitalization rates. Examining risk factors predicting rehospitalization has been widely studied, and most of studies utilized structured data from electronic health records (EHRs) in HF. However, those studies did not use patient-level information that could be captured from unstructured data. Hypothesis: Unified Medical Language System (UMLS) concepts that may be used as factors that could be associated with the likelihood of 30-day rehospitalization among patients with HF by using natural language processing technique. Methods: We obtained notes from patients diagnosed with HF from EHRs from a tertiary hospital between June 1, 2015 and December 31, 2019. We analyzed 14 note types including physician notes, nursing notes and discharge summaries. We used the publicly available natural language processing software MetaMap to extract UMLS concepts that may be used as factors associated with the liekhood of 30-day rehospitalization and the Pandas package in Python for processing and analysis. Results: We identified 350,272 notes among 4,773 patients discharged with HF. The median age of the cohort was 76 years, 54.3% were male, and 89% white, and the 30 day-rehospitalization rate was 12.8%.. A total of 350,272 notes were for 4,164 patients who were not rehospitalized, and 57,212 notes for 609 patients who were rehospitalized. The UMLS concepts identified at a higher rate in notes of rehospitalized patients compared to non-rehospitalized patients included “personal appearance” (7.0% vs 5.5%), “power (psychology)” (3.5% vs 2.5%), and “intrinsic drive” (2.1% vs 1.1%). The UMLS Concepts that were mentioned less in rehospitalized patient notes include “family” (26.3% vs 30.8%), “home environment” (59.9% vs 64%), and “independence” (20.6% vs 24.6%). Conclusions: We identified six UMLS concepts associated with 30-day rehospitalization in patients with HF. Further study is warranted to determine whether incorporation of concepts from unstructured data could improve prediction of 30-day rehospitalization and provide opportunities for patient-level intervention.

An Insight into Patients' Perspectives of Ulcerative Colitis Flares via Analysis of Online Public Forum Posts.

The knowledge of patients' perceptions of factors contributing to ulcerative colitis (UC) flares is limited; however, online patient communications could offer insight. This analysis aimed to identify the most frequent patient-reported triggers and symptoms of UC flares, which could highlight potential interventions for outcome improvement. Online posts written pre- and postflare by patients with UC on 8 public forums in 6 countries between January 1, 2019, and February 14, 2021, were identified using flare-related keywords. Flare-related posts were captured and Netbase Quid™ artificial intelligence text analytics and natural language processing software were used to semantically map and identify commonly discussed themes and topics (subsets of themes). Of >27 000 patient posts, 12 900 were identified as flare related. The most frequent themes were treatment experiences and side effects (28.5% of posts), followed by flare symptoms (22.9% of posts). The most frequent topic was emotional/peer support (9.4% of posts), followed by experiences with mesalamine (and other oral/rectal formulations; 8.0% of posts), and dietary recommendations (6.0% of posts). Stress and anxiety were the most frequently reported flare triggers (37.9% of posts), followed by diet (28.4% of posts). Stress and anxiety were frequently identified as both triggers for, and general symptoms of, flare. Blood in the stool was the most discussed flare indicator (57.8% of posts). Frequently discussed patient-perceived triggers of UC flares included diet, stress, and anxiety. These results suggest that physicians could incorporate a broader and more holistic approach to UC monitoring and management than is currently practiced.

Forms or Free-Text? Measuring Advance Care Planning Activity Using Electronic Health Records.

Advance care planning (ACP) discussions seek to guide future serious illness care. These discussions may be recorded in the electronic health record by documentation in clinical notes, structured forms and directives, and physician orders. Yet, most studies of ACP prevalence have only examined structured electronic health record elements and ignored data existing in notes. We sought to investigate the relative comprehensiveness and accuracy of ACP documentation from structured and unstructured electronic health record data sources. We evaluated structured and unstructured ACP documentation present in the electronic health records of 435 patients with cancer drawn from three separate healthcare systems. We extracted structured ACP documentation by manually annotating written documents and forms scanned into the electronic health record. We coded unstructured ACP documentation using a rule-based natural language processing software that identified ACP keywords within clinical notes and was subsequently reviewed for accuracy. The unstructured approach identified more instances of ACP documentation (238, 54.7% of patients) than the structured ACP approach (187, 42.9% of patients). Additionally, 16.6% of all patients with structured ACP documentation only had documents that were judged as misclassified, incomplete, blank, unavailable, or a duplicate of a previously entered erroneous document. ACP documents scanned into electronic health records represent a limited view of ACP activity. Research and measures of clinical practice with ACP should incorporate information from unstructured data.

Distinguishing the Job Market Across Aerospace and Aviation: A Natural Language Processing Approach

This study dives into the intricate landscape of the aerospace and aviation job market. While these two markets are often conflated as being similar, if not the same, we propose that the differences are important to recent graduates of educational institutions and career programs. The research utilized a custom-written Natural Language Processing (NLP) software tool to distinguish the differences in 6,000 job offerings between the two industries with the hope of illuminating nuances to those in positions involved in placing professionals into careers. This research not only reveals the dynamic employment landscape of aerospace and aviation but also highlights the power of NLP in more clearly discerning emerging trends in job data.

Using the Pan American Health Organization Digital Conversational Agent to Educate the Public on Alcohol Use and Health: Preliminary Analysis.

There is widespread misinformation about the effects of alcohol consumption on health, which was amplified during the COVID-19 pandemic through social media and internet channels. Chatbots and conversational agents became an important piece of the World Health Organization (WHO) response during the COVID-19 pandemic to quickly disseminate evidence-based information related to COVID-19 and tobacco to the public. The Pan American Health Organization (PAHO) seized the opportunity to develop a conversational agent to talk about alcohol-related topics and therefore complement traditional forms of health education that have been promoted in the past. This study aimed to develop and deploy a digital conversational agent to interact with an unlimited number of users anonymously, 24 hours a day, about alcohol topics, including ways to reduce risks from drinking, that is accessible in several languages, at no cost, and through various devices. The content development was based on the latest scientific evidence on the impacts of alcohol on health, social norms about drinking, and data from the WHO and PAHO. The agent itself was developed through a nonexclusive license agreement with a private company (Soul Machines) and included Google Digital Flow ES as the natural language processing software and Amazon Web Services for cloud services. Another company was contracted to program all the conversations, following the technical advice of PAHO staff. The conversational agent was named Pahola, and it was deployed on November 19, 2021, through the PAHO website after a launch event with high publicity. No identifiable data were used and all interactions were anonymous, and therefore, this was not considered research with human subjects. Pahola speaks in English, Spanish, and Portuguese and interacts anonymously with a potentially infinite number of users through various digital devices. Users were required to accept the terms and conditions to enable access to their camera and microphone to interact with Pahola. Pahola attracted good attention from the media and reached 1.6 million people, leading to 236,000 clicks on its landing page, mostly through mobile devices. Only 1532 users had a conversation after clicking to talk to Pahola. The average time users spent talking to Pahola was 5 minutes. Major dropouts were observed in different steps of the conversation flow. Some questions asked by users were not anticipated during programming and could not be answered. Our findings showed several limitations to using a conversational agent for alcohol education to the general public. Improvements are needed to expand the content to make it more meaningful and engaging to the public. The potential of chatbots to educate the public on alcohol-related topics seems enormous but requires a long-term investment of resources and research to be useful and reach many more people.

AI and Incidental Findings

AI and Incidental Findings

Implementation of Peer Messengers to Deliver Feedback: An Observational Study to Promote Professionalism in Nursing.

The Co-Worker Observation System (CORS) is a tool and a process to address disrespectful behavior through feedback from trained peer messengers. First used by physicians and advanced practice providers (APPs), CORS has been shown to decrease instances of unprofessional behaviors among physicians and APPs. The research team assessed the feasibility and fidelity of implementing CORS for staff nurses. CORS was implemented at three academic medical centers using a project bundle with 10 essential implementation elements. Reports of unprofessional behavior among staff nurses that were submitted through the institution's electronic reporting system were screened through natural language processing software, coded by trained CORS coders using the Martinez taxonomy, and referred to a trained peer messenger to share the observations with the nurse. A mixed methods, observational design assessed feasibility and fidelity. A total of 590 reports from three sites were identified by the Center for Patient and Professional Advocacy from September 1, 2019, through August 31, 2021. Most reports included more than one problematic behavior, each of which was coded. Of the peer messages, 76.5% were successfully documented using the debriefing survey as complete, 2.2% as awaiting messenger feedback, and 0.2% as awaiting messenger assignments (total of 78.9 % considered delivered). A total of 21.1% were not shared; 4.7% of reports were intentionally not shared because the issue stemmed from a new system or policy implementation (4.0%) or because of known factors affecting the nurse (0.7%). CORS can be implemented with staff nurses efficiently when nursing infrastructure is adequate.