Free-text Portion Research Articles

The impact of an aging simulation program for dental students in two community dental clinics.

Clinicians who have not experienced the difficulties that come with aging or disability may be unable to relate to the limitations and experiences of afflicted patients, which is necessary to improve patient-provider connection and treatment outcomes. The purpose of this study was to test the effectiveness of an aging-simulation experience on improving dental students' awareness of aging-related limitations, examine the students' perceptions, and assess planned patient-care modifications based on the aging-simulation experience. A total of 78 dental students who rotated through two extramural clinic sites from August 2021 through October 2022 completed pre-simulation surveys, donned aging simulators in the dental operatory, initiated pre-defined tasks and their corresponding modifications, completed a post-simulation survey, and completed a reflection questionnaire containing structured and open-ended questions. Quantitative responses were analyzed using descriptive frequencies and paired sample t-tests, whereas thematic analyses were used to interpret free-text portion of the reflection questionnaire. Student awareness of the impact of the four aging-related disabilities improved after the experience. Visual impairment was identified as the most difficult symptom to experience during simulation. Students reported increased feelings of empathy and acknowledged the effectiveness of clinical practice modifications to accommodate elderly patients with limitations. Students also expressed intentions to make similar modifications in their future clinical practice and the need for exposure to longer periods of simulated experiences to further clinical practice modifications for elderly patients. The aging-simulation experience is an effective tool for raising dental students' awareness of aging-related difficulties, the need for clinical practice modifications, and increasing empathy.

Foresight—a generative pretrained transformer for modelling of patient timelines using electronic health records: a retrospective modelling study

An electronic health record (EHR) holds detailed longitudinal information about a patient's health status and general clinical history, a large portion of which is stored as unstructured, free text. Existing approaches to model a patient's trajectory focus mostly on structured data and a subset of single-domain outcomes. This study aims to evaluate the effectiveness of Foresight, a generative transformer in temporal modelling of patient data, integrating both free text and structured formats, to predict a diverse array of future medical outcomes, such as disorders, substances (eg, to do with medicines, allergies, or poisonings), procedures, and findings (eg, relating to observations, judgements, or assessments). Foresight is a novel transformer-based pipeline that uses named entity recognition and linking tools to convert EHR document text into structured, coded concepts, followed by providing probabilistic forecasts for future medical events, such as disorders, substances, procedures, and findings. The Foresight pipeline has four main components: (1) CogStack (data retrieval and preprocessing); (2) the Medical Concept Annotation Toolkit (structuring of the free-text information from EHRs); (3) Foresight Core (deep-learning model for biomedical concept modelling); and (4) the Foresight web application. We processed the entire free-text portion from three different hospital datasets (King's College Hospital [KCH], South London and Maudsley [SLaM], and the US Medical Information Mart for Intensive Care III [MIMIC-III]), resulting in information from 811 336 patients and covering both physical and mental health institutions. We measured the performance of models using custom metrics derived from precision and recall. Foresight achieved a precision@10 (ie, of 10 forecasted candidates, at least one is correct) of 0·68 (SD 0·0027) for the KCH dataset, 0·76 (0·0032) for the SLaM dataset, and 0·88 (0·0018) for the MIMIC-III dataset, for forecasting the next new disorder in a patient timeline. Foresight also achieved a precision@10 value of 0·80 (0·0013) for the KCH dataset, 0·81 (0·0026) for the SLaM dataset, and 0·91 (0·0011) for the MIMIC-III dataset, for forecasting the next new biomedical concept. In addition, Foresight was validated on 34 synthetic patient timelines by five clinicians and achieved a relevancy of 33 (97% [95% CI 91-100]) of 34 for the top forecasted candidate disorder. As a generative model, Foresight can forecast follow-on biomedical concepts for as many steps as required. Foresight is a general-purpose model for biomedical concept modelling that can be used for real-world risk forecasting, virtual trials, and clinical research to study the progression of disorders, to simulate interventions and counterfactuals, and for educational purposes. National Health Service Artificial Intelligence Laboratory, National Institute for Health and Care Research Biomedical Research Centre, and Health Data Research UK.

A Usability Survey of a Quality Improvement Data Visualization Tool among Medical Intensive Care Unit Nurses

Abstract Background Cognitive overload is prevalent among intensive care unit (ICU) clinicians. Data visualization may decrease cognitive load by assisting with data interpretation and task prioritization. We developed the Bundle Board to display real-time data from the electronic medical record (EMR), highlighting opportunities for action in standardized ICU patient care. This study evaluates the practical usability of this data visualization tool among nurses in the ICU. Methods The tool is offered as an application separate from the EMR and was available in the medical ICU for 8 months before we surveyed unit nursing staff. To evaluate usability of the tool, we adapted the Health-Information Technology Usability Scale and included an option to provide open-ended feedback. Survey data were analyzed using quantitative and qualitative methods. Results ICU nurses were invited to participate through email and verbal announcements. Of the potential participants, 38% (N = 47) responded. The survey demonstrated that the tool was perceived as usable. For each subscale, mean scores were as follows: Perceived Ease of Use 4.40, Impact 4.14, User Control 4.07, and Perceived Usefulness 3.61. There were no significant differences between core and contracted nurses or after stratifying by duration of Bundle Board use. Fifteen respondents completed the optional free-text portion of the survey. Qualitative analysis revealed six subthemes focusing on perceived impacts on quality and safety, cognitive burden and workload, and emotional impact of the Bundle Board. Conclusion The Bundle Board demonstrated good usability among ICU nurses, who provided substantive feedback for its improvement. These observations may be generalizable to other comparable interventions. Iterative feedback from end users is vital to developing and implementing a digital health intervention. Our study provides a framework for performing a usability analysis within a specific clinician population and environment.

Improving lung cancer screening in a community healthcare system by electronic health record (EHR) optimization.

437 Background: In March 2021, the United States Preventative Services Taskforce updated its lung cancer screening guidelines to improve overall survival rates, decrease disparities and provide a net cost savings to the healthcare system. With many patients having advanced lung cancer at the time of their diagnosis, it is imperative that screening is performed as soon as applicable. The objective of this study was to improve documentation in EPIC to help prompt low-dose CT (LDCT) scan reminders for providers on the EHR. Methods: This was a prospective study that utilized patient data from EPIC from the Baylor Scott & White Healthcare primary care clinic in Round Rock, Texas. Our version of EPIC used the guidelines prior to the 2021 change. To prompt a reminder on EPIC, there needs to be sufficient numerical data in the smoking history entry form to calculate the pack-years for a patient. EPIC will then automatically generate a notification for providers to order screening LDCT scans for patients that fulfill the screening criteria. We utilized “SlicerDicer” on EPIC to create a custom report of patients that qualified for the study in terms of the inclusion and exclusion criteria. From October 2021 to January 2022, we collected a pre-intervention representative sample to assess how well smoking had been documented during this period. We manually reviewed each medical record number (MRN) listed in the report to verify if proper documentation was used. Our intervention was to contact the clinic manager multiple times with specific instructions on proper smoking history charting. This was then distributed to all ancillary clinic staff via team meetings and flyers. We planned an intervention period from February 2022 to May 2022 that involved data totaling 1178 patients with similar abstraction methods as the pre-intervention data set. Results: We were able to increase lung cancer screening reminders on EPIC for this higher risk patient population. Specifically, 37% of patients pre-intervention had complete data compared to 49% of patients post-intervention. Therefore, an absolute difference of 12% (31.80% increase) of proper documentation was achieved to increase LDCT reminders. The most common mistake by staff was documenting smoking history into the free-text portion of the entry-data forms, which EPIC is not able to calculate pack-years from. Conclusions: Following the intervention at the outpatient clinic, we were able to boost LDCT reminders via an already established process. Increased reminders will eventually lead to more screening orders placed, which we plan to assess for in the next phase of our study. This study can be easily replicated at other centers by instilling similar clinical staff information sessions about documentation. Proper documentation by all healthcare providers is a shared responsibility that has both immediate and longer-lasting impacts on an individual and population level.

Linking Free Text Documentation of Functioning and Disability to the ICF With Natural Language Processing.

Background:Invaluable information on patient functioning and the complex interactions that define it is recorded in free text portions of the Electronic Health Record (EHR). Leveraging this information to improve clinical decision-making and conduct research requires natural language processing (NLP) technologies to identify and organize the information recorded in clinical documentation.Methods:We used natural language processing methods to analyze information about patient functioning recorded in two collections of clinical documents pertaining to claims for federal disability benefits from the U.S. Social Security Administration (SSA). We grounded our analysis in the International Classification of Functioning, Disability, and Health (ICF), and used the Activities and Participation domain of the ICF to classify information about functioning in three key areas: mobility, self-care, and domestic life. After annotating functional status information in our datasets through expert clinical review, we trained machine learning-based NLP models to automatically assign ICF categories to mentions of functional activity.Results:We found that rich and diverse information on patient functioning was documented in the free text records. Annotation of 289 documents for Mobility information yielded 2,455 mentions of Mobility activities and 3,176 specific actions corresponding to 13 ICF-based categories. Annotation of 329 documents for Self-Care and Domestic Life information yielded 3,990 activity mentions and 4,665 specific actions corresponding to 16 ICF-based categories. NLP systems for automated ICF coding achieved over 80% macro-averaged F-measure on both datasets, indicating strong performance across all ICF categories used.Conclusions:Natural language processing can help to navigate the tradeoff between flexible and expressive clinical documentation of functioning and standardizable data for comparability and learning. The ICF has practical limitations for classifying functional status information in clinical documentation but presents a valuable framework for organizing the information recorded in health records about patient functioning. This study advances the development of robust, ICF-based NLP technologies to analyze information on patient functioning and has significant implications for NLP-powered analysis of functional status information in disability benefits management, clinical care, and research.

Strategies of Advanced Practice Providers to Reduce Stress at Work.

Background: High levels of stress and burnout, documented among healthcare professionals, result in high levels of job turnover. However, little is known about personal strategies employed by advanced practice providers (APPs) to mitigate stress. Methods: 3,939 APPs were invited to complete an online, anonymous, cross-sectional survey to examine work stress and burnout among APPs using quantitative and qualitative analyses. The Maslach Burnout Inventory (MBI) served as the measure for burnout. Work-stress reduction strategies were identified by an open-ended question. Findings: 854 APPs (70% of 1,218 respondents) (nurse practitioners, physician assistants, nurse midwives, anesthetists, and clinical nurse specialists) from four different United States health systems completed the free text portion of the survey. Qualitative analyses revealed that almost all of the APPs (94%) reported at least one stress reduction strategy. Four main themes were derived from the free-text responses: "Self-focused" (67.5%), "relational-focused"(16.1%), "job-focused" (11.5%) and "nothing" (4.9%). Quantitative results from the MBI revealed high levels of burnout among APPs with insignificant differences between those who provided a response to the qualitative question and those who did not: emotional exhaustion (p = .188); depersonalization (p = .265); personal accomplishment (p = .213). Conclusion/Application to Practice: Qualitative results highlighted strategies, many evidence-based, that APPs use to mitigate stress. Further research is needed to determine the frequency and consistency of APPs' enactment of personal strategies. These initial findings provide insights for occupational health practitioners and researchers planning primary prevention and secondary interventions for improving workplace health, enhancing personal wellness, and reducing job-related stressors.

Assessing Provider-Generated Free-Text Quality in EHR-Integrated Handoff Notes.

Handoff notes are increasingly integrated within electronic health record (EHR) systems and often contain data automatically generated from the EHR and free-text narratives. We examined the quality of data entered by providers in the free-text portion of our institutional EHR handoff tool. Overall, 65% of handoff notes contained at least one error (average 1.7 errors per note). Most errors were omissions in information around patient plan/management or assessment/diagnosis rather than entry of false data. Factors associated with increased error rate were increasing hospital day number; weekend note; medical (vs. surgical) service team; and authorship by a medical student, first or fourth year resident physician, or attending physician. Our findings suggest that errors are common in handoff notes, and while these errors are not completely false data, they may provide individuals caring for patients an inaccurate understanding of patient status.

Variability in Threshold for Medication Error Reporting Between Physicians, Nurses, Pharmacists, and Families.

Voluntary medication error reporting is an imperfect resource used to improve the quality of medication administration. It requires judgment by front-line staff to determine how to report enough to identify opportunities to improve patients' safety but not jeopardize that safety by creating a culture of "report fatigue." This study aims to provide information on interpretability of medication error and the variability between the subgroups of caregivers in the hospital setting. Survey participants included nursing, physician (trainee and graduated), patient/families, pharmacist across a large academic health system, including an attached free-standing pediatric hospital. Demographics and survey questions were collected and analyzed using Fischer's exact testing with SAS v9.3. Statistically significant variability existed between the four groups for a majority of the questions. This included all cases designated as administration errors and many, but not all, cases of prescribing events. Commentary provided in the free-text portion of the survey was sub-analyzed and found to be associated with medication allergy reporting and lack of education surrounding report characteristics. There is significant variability in the threshold to report specific medication errors in the hospital setting. More work needs to be done to further improve the education surrounding error reporting in hospitals for all noted subgroups.

Detecting Overlapping Outbreaks of Influenza

ntroductionInfluenza is a contagious disease that causes epidemics in manyparts of the world. The World Health Organization estimates thatinfluenza causes three to five million severe illnesses each year and250,000-500,000 deaths [1]. Predicting and characterizing outbreaksof influenza is an important public health problem and significantprogress has been made in predicting single outbreaks. However,multiple temporally overlapping outbreaks are also common.These may be caused by different subtypes or outbreaks in multipledemographic groups. We describe ourMultiple Outbreak DetectionSystem(MODS) and its performance on two actual outbreaks.This work extends previous work by our group [2,3,4] by using model-averaging and a new method to estimate non-influenza influenza-likeillness (NI-ILI). We also apply MODS to a real dataset with a doubleoutbreak.MethodsMODS is part of a framework for disease surveillance developedby our group. In this framework, a natural language processing systemextracts symptoms from emergency department patient-care reports.These features are combined with laboratory results and passed to acase detection system that infers a probability distribution over thediseases each patient may have. These diseases include influenza,NI-ILI, and other (appendicitis, trauma, etc.). This distribution isexpressed in terms of the likelihoods of the patients’ data. These aregiven to MODS which searches a space of multiple outbreak models,computes the likelihood of each model, and calculates the expectednumber of influenza cases day-by-day. This work differs from pastwork in three important ways. First, we address the problem ofdetecting and characterizing multiple, overlapping outbreaks. Second,we do not rely on simple counts, but use likelihoods given evidencein the free-text portion of patient-care reports as well as laboratoryfindings. Third, we explicitly account for non-influenza influenza-like illnesses. This is important because some forms of influenza-likeillness (such as respiratory syncytial virus) are contagious and exhibitoutbreak activity. This research was approved by the University ofPittsburgh and Intermountain Healthcare IRBs.ResultsWe conducted a set of experiments with simulated outbreaks.MODS is able to detect a single outbreak six to eight weeks beforethe peak. It is also able to recognize a second outbreak approximatelyhalfway between peaks for simulated double outbreaks. Weconducted experiments using real outbreaks and compared ourresults to thermometer sales [5]. Using data from Allegheny CountyPennsylvania for the 2009-2010 influenza season, on September1 MODS predicted an outbreak with a peak on October 5. Thethermometer peak was October 21. The figure “Prediction on October1 for Allegheny County” compares MODS’ prediction on October 1to thermometer sales. Using data from Salt Lake City Utah for the2010-2011 influenza season, on November 1 MODS predicted anoutbreak with peak on December 7. The first thermometer peak wasDecember 29. On January 20 MODS predicted a second outbreakwith peak on February 9. The second thermometer peak was March5. The figure “Prediction on January 20 for Salt Lake City” comparesMODS’ prediction on January 20 to thermometer sales.ConclusionsWe have built aMultiple Outbreak Detection Systemthat candetect and characterize overlapping outbreaks of influenza. Althoughthe system currently predicts outbreaks of influenza, it is built on ageneral Bayesian framework that can be extended to other diseases.Future work includes incorporating multiple forms of evidence,modeling other known contagious diseases, and detecting outbreaksof new previously unknown diseases.Prediction on October 1 for Allegheny County 2009-2010Prediction on January 20 for Salt Lake City 2010-2011

JOURNAL CLUB: Structured Feedback From Patients on Actual Radiology Reports: A Novel Approach to Improve Reporting Practices.

Patient perceptions of radiology reports are largely unknown. The objective of the present study is to describe our experience receiving structured feedback from patients on actual radiology reports as a means of improving reporting practices. Eight reports (two for radiographs, two for ultrasound images, two for CT scans, and two for MR images) were randomly selected from our system for review. For each report, patients were asked to rate their level of comprehension, identify any problems in the report, and, in the free-text portion of the feedback form, indicate any questions about the report that they may have. Potentially confounding factors were also examined. A total of 104 patients (46 men and 58 women) participated in the study (for a total of 832 evaluations). The median score for report comprehension was 2.5 (on a scale of 1-5), with the most common problems affecting comprehension identified as "unclear or technical language" (mentioned in 59.6% of evaluations) and the report being "too long" (mentioned in 10.2% of evaluations). A request for an explanation of the report in lay terms (noted in 20.1% of evaluations) was the most common request mentioned in the free-text portion of the feedback form. An inverse relationship existed between report length and patient comprehension (p < 0.001). Patients who had prior experience with their own radiology reports indicated having greater comprehension than did patients with no prior experience (p = 0.003). No correlation between the educational status and report comprehension of the patients was identified (p = 0.488). Radiology reports are not well understood by patients, who identify technical language and the long length of reports as the most common problems affecting their comprehension. Longer reports tend to be less well understood.

Language, Structure, and Reuse in the Electronic Health Record.

Medical language is at the heart of the electronic health record (EHR), with up to 70 percent of the information in that record being recorded in the natural language, free-text portion. In moving from paper medical records to EHRs, we have opened up opportunities for the reuse of this clinical information through automated search and analysis. Natural language, however, is challenging for computational methods. This paper examines the tension between the nuanced, qualitative nature of medical language and the logical, structured nature of computation as well as the way in which these have interacted with each other through the medium of the EHR. The paper also examines the potential for the computational analysis of natural language to overcome this tension.

Pharmacovigilance using clinical notes.

With increasing adoption of electronic health records (EHRs), there is an opportunity to use the free-text portion of EHRs for pharmacovigilance. We present novel methods that annotate the unstructured clinical notes and transform them into a deidentified patient-feature matrix encoded using medical terminologies. We demonstrate the use of the resulting high-throughput data for detecting drug-adverse event associations and adverse events associated with drug-drug interactions. We show that these methods flag adverse events early (in most cases before an official alert), allow filtering of spurious signals by adjusting for potential confounding, and compile prevalence information. We argue that analyzing large volumes of free-text clinical notes enables drug safety surveillance using a yet untapped data source. Such data mining can be used for hypothesis generation and for rapid analysis of suspected adverse event risk.

What Lies Beneath: The Linguistic Traces of Deception in Online Dating Profiles

This article investigates whether deceptions in online dating profiles correlate with changes in the way daters write about themselves in the free-text portion of the profile, and whether these changes are detectable by both computerized linguistic analyses and human judges. Computerized analyses (Study 1) found that deceptions manifested themselves through linguistic cues pertaining to (a) liars' emotions and cognitions and (b) liars' strategic efforts to manage their self-presentations. Technological affordances (i.e., asynchronicity and editability) affected the production of cognitive cues more than that of emotional cues. Human judges (Study 2) relied on different and nonpredictive linguistic cues to assess daters' trustworthiness. The findings inform theories concerned with deception, media, and self-presentation, and also expound on how writing style influences perceived trustworthiness.

Inverted Index Support for Numeric Search

Today's search engines are increasingly required to broaden their capabilities beyond free-text search. More complex features, such as supporting range constraints over numeric data, are becoming common; structured search over XML data will soon follow. This is particularly true in the enterprise search domain, where engines attempt to integrate data from the web and corporate knowledge portals with data residing in proprietary databases. In this paper we extend previous schemes by which an inverted-index-based search engine can efficiently support queries that contain numeric restrictions in addition to standard, free-text portions. Furthermore, we analyze both the known schemes and our extensions in terms of index-build time, index space, and query processing time. We show how to maximize query processing performance while respecting limits on index size and build time, or conversely, how to minimize index space and build time while maintaining guarantees on runtime performance. Thus, we concisely analyze the trade-off between index size and build time, and runtime performance. Finally, we present experimental results that demonstrate significant performance benefits attained by our method, as compared to alternative approaches.