Physician Review Research Articles

Analysis of gender bias in Yelp reviews for gynecologic oncology practices and physicians across the United States

Analysis of gender bias in Yelp reviews for gynecologic oncology practices and physicians across the United States

37 Adverse events associated with paediatric emergency medicine virtual clinic visits

Abstract Background Emergency medical care inherently carries a potential for adverse events. With the increase in virtual medical care due to its ease of access and ability to limit the spread of infection, there is a need to evaluate the safety of paediatric emergency medicine virtual clinics. Objectives To elucidate the frequency of adverse events occurring within 7 days of accessing medical care provided by a Paediatric Emergency Medicine Virtual Clinic. We then seek to characterize the severity, nature, and preventability of the adverse events and to provide a knowledge base for future work related to patient safety in the virtual emergency care domain. Design/Methods This study was a prospective cohort study where the primary outcome was the proportion of patients who experienced an adverse event within 7 days of accessing care through a virtual emergency clinic. Data was collected through telephone interviews and chart analysis with the use of a validated adverse events trigger tool to determine which charts required independent physician review. The final data was categorized using descriptive statistics, and logistic regression analysis was conducted to examine potential predictors of adverse events. Results Of 593 eligible families, 232 (39%) consented to participate and completed the interview process. Of these families, 89 were flagged for further review, and subsequent analysis concluded that 5 (2.2%) had experienced an adverse event. Of the adverse events, 4 (80%) were considered preventable and the most common type of adverse event was sub-optimal follow up (60%). The median patient age was 3.0 years (IQR 1.0-5.0). There were no clinically significant differences between the patients who experienced an adverse event and those that did not across the parameters studied (age, sex, gender, medical complexity, length of symptoms, language of fluency, and self-identification of visible minority status). Conclusion 2.2% of patients, approximately 1 in 45, experienced adverse events related to the care they received at the virtual clinic. The majority of these adverse events were preventable. That said, the rates of events were within the published norm, suggesting the virtual clinic provides a viable option in paediatric emergency medicine.

How the Menstrual Cycle Can Be Utilized During Sports Training, Performance, and Recovery through Wearable Technology: A Narrative Review for Researchers, Physicians, Coaches, and Athletes.

The menstrual cycle (MC) serves as a vital indicator of overall health and metabolic function, regulated by the hypothalamic-pituitary axis and involving a complex interplay of hormones. Understanding these hormonal dynamics is crucial for deciphering an individual's physiological status and performance potential, particularly in athletes. Studies regarding the MC's impact on athletic performance and training often lack inclusivity, standardized methodologies, and inconsistent biological definitions, hindering comprehensive conclusions. Moreover, societal inequalities contribute to the underrepresentation of female athletes in research, exacerbating the lack of understanding regarding female physiology in sports medicine. Leveraging wearable technology presents a promising avenue for both tracking MCs and optimizing athletic training/recovery. Wearables offer real-time monitoring of biometrics that often correlate with hormonal fluctuations, and lifestyle trends (diet, sleep, stress) aiding in personalized training schedules and performance optimization. Integrating data collected by MC dynamics and wearable technology into athletic training has the potential to decrease the generally perceived negative impacts MC has on athletic performance. Addressing gaps in research methodologies and promoting awareness among athletes, coaches, and healthcare professionals are essential steps toward maximizing the potential of MC-informed training strategies.

Defining, measuring and managing healthcare quality using unstructured physician review comments

PurposePhysician review websites (PRW) and Medicare requirements are pressing administrators to measure, monitor and improve healthcare service delivery. Healthcare service attributes linked to patient satisfaction have received increased attention. Text analysis provides an alternative methodology to capture contemporaneous data on service delivery attributes. A Kano analysis based on these service attributes can help administrators prioritize service delivery and ultimately improve patient satisfaction.Design/methodology/approachHealthcare service attributes were defined from 4,000+ comments on a PRW using latent content text analysis. The resulting 15 attributes were validated by medical professionals using a q-sort methodology and analyzed using a Kano methodology.FindingsThe 15 attributes cover three domains of healthcare service – clinic operations, competency and care. The Kano analysis yields a hierarchy, or pyramid, of healthcare service attributes: (1) must-be’s: establish service operational capabilities and benchmark peer performance; (2) satisfiers: establish and increase trust through: (a) clinical competence, (b) careful management of young patients and (c) delivery of appropriate care and treatment (3) delighters: use service-dominant logic to provide patient-centered care.Originality/valueThis research bridges the gap between the “what” and “how” that is frequently missing in text analysis of online reviews. We provide a methodology coupled with a Kano analysis, a widely used quality improvement tool, which results in a hierarchy of service attributes that can guide administrators and researchers.

Automated, Real-Time Integration of Biometric Data From Wearable Devices With Electronic Medical Records: A Feasibility Study.

A major barrier to the incorporation of biometric data into clinical practice is the lack of device integration with electronic medical records (EMRs). We developed infrastructure to transmit biometric data from an Apple Watch into the EMR for physician review. The study objective was to test feasibility of using this infrastructure for patients undergoing radiotherapy. The study included patients with breast or prostate cancer receiving ≥3 weeks of radiotherapy who reported owning an Apple Watch. Daily resting heart rate (HR), HR variability, step count, and exercise minutes were automatically transferred to our EMR using a custom app installed on each patient's iPhone. Biometric data were presented to the treating radiation oncologist for review on a weekly basis during creation of the on-treatment note. Feasibility was defined a priori as physician review of biometric data for at least 90% of patients. Time trends in biometric data were tested using the Jonckheere-Terpstra test. Patient satisfaction was assessed using the System Usability Scale (SUS), with scores above 80 considered above-average user experience. Of the 20 patients enrolled, biometric data were successfully transmitted to the EMR and reviewed by the radiation oncologist for 95% (n = 19) of patients, thus meeting the a priori feasibility threshold. For patients with radiation courses ≥4 weeks, exercise minutes decreased over time (P = .01) and daily mean HR variability increased over time (P = .02). The median SUS was 82.5 (IQR, 70-87.5). Our study demonstrates the feasibility of real-time integration of biometric data collected from an Apple Watch into the EMR with subsequent physician review. The high rates of physician review and patient satisfaction provide support for further development of large-scale collection of wearable device data.

Minimal variation between surgeons in online patient-reported ratings: A meaningless metric? – A retrospective analysis of physician reviews

Minimal variation between surgeons in online patient-reported ratings: A meaningless metric? – A retrospective analysis of physician reviews

Venoarterial Extracorporeal Membrane Oxygenation Therapy in Patients with Sickle Cell Disease: Case Series and Review for Intensive Care Physicians.

Sickle cell disease (SCD) is associated with substantial morbidity and early mortality in afflicted adults. Cardiopulmonary complications that occur at increased frequency in SCD such as pulmonary embolism, pulmonary arterial hypertension, and acute chest syndrome can acutely worsen right ventricular function and lead to cardiogenic shock. Mechanical circulatory support including venoarterial extracorporeal membrane oxygenation (VA ECMO) is being increasingly utilized to treat hemodynamic collapse in various patient populations. However, a paucity of literature exists to guide the use of mechanical circulatory support in adults with SCD where disease-related sequela and unique hematologic aspects of this disorder may complicate extracorporeal therapy and must be understood. Here, we review the literature and describe three cases of adult patients with SCD who developed cardiogenic shock from acute decompensated right heart failure and were treated clinically with VA ECMO. Using an in vitro ECMO system, we investigate a potential increased risk of systemic fat emboli in patients with SCD who may be experiencing vaso-occlusive events with bone marrow involvement given the high-volume shunting of blood from venous to arterial systems with VA ECMO. The purpose of this study is to describe available extracorporeal life support experiences, review potential complications, and discuss the special considerations needed to further our understanding of the utility of VA ECMO in those with SCD.

Examining the Role of Physician Characteristics in Web-Based Verified Primary Care Physician Reviews: Observational Study.

Doctor review websites have become increasingly popular as a source of information for patients looking to select a primary care provider. Zocdoc is one such platform that allows patients to not only rate and review their experiences with doctors but also directly schedule appointments. This study examines how several physician characteristics including gender, age, race, languages spoken in a physician's office, education, and facial attractiveness impact the average numerical rating of primary care doctors on Zocdoc. The aim of this study was to investigate the association between physician characteristics and patient satisfaction ratings on Zocdoc. A data set of 1455 primary care doctor profiles across 30 cities was scraped from Zocdoc. The profiles contained information on the physician's gender, education, and languages spoken in their office. Age, facial attractiveness, and race were imputed from profile pictures using commercial facial analysis software. Each doctor profile listed an average overall satisfaction rating, bedside manner rating, and wait time rating from verified patients. Descriptive statistics, the Wilcoxon rank sum test, and multivariate logistic regression were used to analyze the data. The average overall rating on Zocdoc was highly positive, with older age, lower facial attractiveness, foreign degrees, allopathic degrees, and speaking more languages negatively associated with the average rating. However, the effect sizes of these factors were relatively small. For example, graduates of Latin American medical schools had a mean overall rating of 4.63 compared to a 4.77 rating for US graduates (P<.001), a difference roughly equivalent to a 2.8% decrease in appointments. On multivariate analysis, being Asian and having a doctor of osteopathic medicine degree were positively associated with higher overall ratings, while attending a South Asian medical school and speaking more European and Middle Eastern languages in the office were negatively associated with higher overall ratings. Overall, the findings suggest that age, facial attractiveness, education, and multilingualism do have some impact on web-based doctor reviews, but the numerical effect is small. Notably, bias may play out in many forms. For example, a physician's appearance or accent may impact a patient's trust, confidence, or satisfaction with their physician, which could in turn influence their take-up of preventative services and lead to either better or worse health outcomes. The study highlights the need for further research in how physician characteristics influence patient ratings of care.

Should I See You Again Soon? Multispecialty Assessment of Impact and Burden of Preoperative History and Physical Update Visits.

Federal regulations require a history and physical (H&P) update performed 30 days or less before a planned procedure. We evaluated the use and burdens of H&P update visits by determining impact on operative management, suitability for telehealth, and visit time and travel burden. We identified H&P update visits performed in our health system during 2019 for 8 surgical specialties. As available, up to 50 visits per specialty were randomly selected. Primary outcomes were interval changes in history, examination, or operative plan between the initial and updated H&P notes, and visit suitability for telehealth, as determined by 2 independent physician reviewers. Clinic time was captured, and round-trip driving time and distance between patients' home and clinic ZIP codes were estimated. We identified 8,683 visits and 362 were randomly selected for review. Documented changes were most commonly identified in histories (60.8%), but rarely in physical examinations (11.9%) and operative plans (11.6%). Of 362 visits, 359 (99.2%) visits were considered suitable for telehealth. Median clinic time was 52 minutes (interquartile range 33.8 to 78), driving time was 55.6 minutes (interquartile range 35.5 to 85.5), and driving distance was 20.2 miles (interquartile range 8.5 to 38.4). At the health system level, patients spent an estimated aggregate 7,000 hours (including 4,046 hours of waiting room and travel time) and drove 142,273 miles to attend in-person H&P update visits in 2019. Given their minimal impact on operative management, regulatory requirements for in-person H&P updates should be reconsidered. Flexibility in update timing and modality might help defray the substantial burdens these visits impose on patients.

Recovery of arterial blood pressure after chest compression pauses in patients with out-of-hospital cardiac arrest

Background and aimsChest compressions generating good perfusion during cardiopulmonary resuscitation (CPR) in cardiac arrest patients are critical for positive patient outcomes. Conventional wisdom advises minimizing compression pauses because several compressions are required to recover arterial blood pressure (ABP) back to pre-pause values. Our study examines how compression pauses influence ABP recovery post-pause in out-of-hospital cardiac arrest. MethodsWe analyzed data from a subset of a prospective, randomized LUCAS 2 Active Decompression trial. Patients were treated by an anesthesiologist-staffed rapid response car program in Oslo, Norway (2015–2017) with mechanical chest compressions using the LUCAS device at 102 compressions/min. Patients with an ABP signal during CPR and at least one compression pause >2 sec were included. Arterial cannulation, compression pauses, and ECG during the pause were verified by physician review of patient records and physiological signals. Pauses were excluded if return of spontaneous circulation occurred during the pause (pressure pulses associated with ECG complexes). Compression, mean, and decompression ABP for 10 compressions before/after each pause and the mean ABP during the pause were measured with custom MATLAB code. The relationship between pause duration and ABP recovery was investigated using linear regression. ResultsWe included 56 patients with a total of 271 pauses (pause duration: median = 11 sec, Q1 = 7 sec, Q3 = 18 sec). Mean ABP dropped from 53 ± 10 mmHg for the last pre-pause compression to 33 ± 7 mmHg during the pause. Compression and mean ABP recovered to >90% of pre-pause pressure within 2 compressions, or 1.7 sec. Pause duration did not affect the recovery of ABP post-pause (R2: 0.05, 0.03, 0.01 for compression, mean, and decompression ABP, respectively). ConclusionsABP generated by mechanical CPR recovered quickly after pauses. Recovery of ABP after a pause was independent of pause duration.

Validation and Use of a Measure in Critical Access Hospitals to Reduce Treatment of Asymptomatic Bacteriuria

Background: Inappropriate diagnosis and treatment of urinary tract infections (UTI) contributes to antibiotic overuse. The Inappropriate Diagnosis of UTI (ID-UTI) measure—which quantifies the percentage of treated bacteriuria that is asymptomatic (ASB) using a standard definition of UTI—has been validated in acute care hospitals, but not in critical access hospitals (CAHs) where resources differ. Methods: From October 2022-July 2023, ten CAHs participated in a program to reduce ASB treatment, including education, coaching, and hospital feedback using the ID-UTI measure. Our primary aim was to assess performance characteristics of the ID-UTI standard definition and measure in CAHs (Figure 1). Non-physician abstractors from each CAH submitted clinical data (e.g., signs/symptoms) via REDCap for consecutive adult patients who were admitted or discharged from the emergency department and received antibiotics for bacteriuria. The case abstraction goal for each CAH was 5-6 cases/month. To assess feasibility, we evaluated the ability of each CAH to achieve goal case abstractions. To assess validity and reliability of the ID-UTI standard definition, two physicians reviewed 10% of submitted cases using deidentified patient notes and assessed agreement with the standard ID-UTI definition and consensus clinical opinion. Based on submitted data, we provided bi-monthly feedback reports to CAHs including the hospital-level ID-UTI measure to benchmark progress and for peer comparison. We measured monthly change in the ID-UTI measure using a mixed-effects logistic regression model (Figure 1, Figure 2). Results: Among 10 CAHs, 4 (40%) submitted &gt;59 cases over 10 months (goal) while 3 (30%) submitted &gt;35 cases (secondary goal). Physician reviewers assessed 9.5% (58/608) of cases. Utilizing the ID-UTI standard definition, there was high agreement (93%) in ASB vs UTI designation between each physician reviewer and the CAH’s REDCap assessment (Figure 3). Compared to clinical opinion, the ID-UTI standard definition identified 48% (16/33) of ASB cases and 100% (25/25) of UTI cases (Figure 4). Over the program, the percentage of cases treated for UTI that were ASB decreased from 28.4% (range 0-63%) to 18.6% (range, 0-33%; p=0.055) (Figure 5). Conclusions: Case abstraction with use of the ID-UTI measure was feasible and reliable to implement with modifications for CAHs. Data collection by untrained staff was as reliable as physician adjudication. Though the ID-UTI standard definition undercounted ASB cases (low sensitivity), cases reported as ASB were always ASB when adjudicated (high specificity). The program, including performance feedback using the ID-UTI measure, was associated with a trend toward lower treatment of ASB.

Ethical Review of Physicians' Self Referral

Ethical Review of Physicians' Self Referral

Peripheral artery disease, chronic kidney disease, and recurrent admissions for acute decompensated heart failure: The ARIC study

Background and aimsPeripheral artery disease (PAD) has not only been associated with recurrent hospitalization for acute decompensated heart failure (ADHF) but is also associated with chronic kidney disease (CKD), a known risk factor for worse heart failure outcomes. The interaction of CKD with PAD in post-discharge ADHF outcomes is not well known. MethodsSince 2005, hospitalizations for ADHF were sampled from 4 US regions by the Atherosclerosis Risk in Communities (ARIC) study and classified by physician review. We examined the adjusted association of PAD with 1-year ADHF readmissions, in patients with and without CKD (defined by glomerular filtration rate [GFR] ≤60 mL/min/1.73 m2 [stage 3a or worse]). ResultsFrom 2005 to 2018, there were 1049 index hospitalizations for patients with ADHF (mean age 77 years, 66 % white) with creatinine data, who were discharged alive. Of these, 155 (15 %) had PAD and 66 % had CKD. In comparison to those without PAD, patients with PAD had more comorbid conditions and higher 1-year ADHF readmission rates, irrespective of CKD status. After adjustment, PAD was associated with a greater risk of 1-year ADHF readmissions, both for patients with concomitant CKD (HR, 1.70; 95 % CI: 1.29–2.24) and those without CKD (HR, 1.97; 95 % CI: 1.14–3.40); p-interaction = 0.8. ConclusionAmong patients hospitalized with ADHF, those with concurrent PAD have more prevalent cardiovascular comorbidities and higher likelihood of 1-year ADHF readmission, irrespective of CKD status. Integrating a more holistic approach in management of patients with concomitant heart failure, PAD and CKD may be an important strategy to improve the prognosis in this vulnerable population.

Preoperative survey to evaluate the patients' allergy list and its relevance to perioperative care.

Perioperative hypersensitivity and allergic reactions can result in significant morbidity and mortality. For routine anesthetic care, allergies are determined from a review of the electronic medical record supplemented by a detailed patient history. Although the electronic medical record is generally assumed to be accurate, it may be that allergies are erroneously listed or not based on sound medical practice. The purpose of the current study is to evaluate allergies listed in the electronic medical record of children presenting for surgery and determine their origin, authenticity, and impact on perioperative care. Eligible patients included those presenting for a surgical procedure in the main operating room, who were ≤ 21 years of age, with a drug allergy listed on the EMR. Prior to intraoperative care, an electronic survey questionnaire containing questions related to medication allergies was provided to a guardian or parent. Two anesthesiology physicians reviewed the survey responses to determine the validity of any reported allergies. A second electronic survey was given postoperatively to the attending anesthesiologist to determine whether the documented allergy impacted anesthetic care. The study cohort included 250 patients, ranging in age from 5 to 14 years (median age 9 years). All of the patients had at least one allergy listed on the electronic medical record. Seventy of the 250 patients (28%) had more than one drug allergy listed for a total of 351 medication allergies. The majority of the listed allergies were related to antibiotics including 155 (44%) from the penicillin family, 26 (7%) cephalosporins, 16 (5%) sulfonamides, and 36 (10%) other antimicrobial agents. Other commonly listed allergies were 27 (8%) nonsteroidal anti-inflammatory agents and 15 (4%) opioids. The remaining 76 (22%) included a miscellaneous list of other medications. On further review of the allergies, the survey was completed for 301 medications. After physician review, 135 of 301 (45%) responses were considered consistent with IgE reactions "true allergy," 73 (24%) were deemed less relevant to IgE reactions "unlikely true allergy," and 93 (31%) were not related to IgE reactions "not an allergy." Care alterations during surgery were uncommon regardless of whether the issue was assessed as a true allergy (11%), unlikely to be a true allergy (3%), or not a true allergy (13%). A significant portion of the documented allergies in children are not true allergies, but rather recognized adverse effects (apnea from an opioid, renal failure from an NSAIDs) or other nonallergic concerns (gastrointestinal upset such as nausea). Erroneously listed allergies may lead to unnecessary alterations in patient care during perioperative care. A careful analysis of the allergy list on the EMR should be supplemented by a thorough patient history with specific questions related to the drug allergy. Once this is accomplished, the allergy listed should be updated to avoid its erroneous impact on perioperative care.

Reliability of YouTube content targeting Arabic-speaking patients with breast cancer regarding post-mastectomy reconstruction.

e13626 Background: The dynamic and easily accessible nature of social media platforms, particularly YouTube (YT), has revolutionized the way breast cancer (BC) patients seek and process information. This study aims to evaluate the quality of YT videos addressing post-mastectomy reconstruction (PMR) that target Arabic-speaking women (ASW), a heavily digitally connected segment of society that may experience challenges in accessing mainstream medical information in Arabic Language (AL). Methods: YT was searched incognito, using the terms: "post mastectomy reconstruction" in (AL) in July 2023. Upload date, video length, number of views, poster identity and other relevant information were compiled. Video understandability and actionability were assessed by physician-reviewers (PR) with the Patient Education Materials Assessment Tool for audiovisual materials (PEMAT A/V), while the DISCERN tool appraised quality. Sponsorship presence, patient vs health care professional narratives, and intended audience were documented. Thematic analysis of most discussed topics was performed. Additionally, cluster analysis was employed to categorize the videos based on their quality. Results: Of 109 videos (mean length: 7 mins.), 67% had sponsorship (of which 45% corporate); 92% were information-based (vs experience-based); 82% targeted general population over BC patients; 93.6% were in AL, and 5.5% had translation. Mean PEMAT scores were 61.3% (understandability) and 20% (actionability). Mean DISCERN score was 2.6/5. Only 21% of videos featured women. Predominant themes were awareness (82.6%), appearance, and body image (68.8%). Less prevalent: sexuality and fertility (4.6%). The likelihood that the PR would recommend the video to a patient was low at 43.1%. Cluster analysis showed that high-quality videos had a higher understandability (74%), actionability (40.5%) and greater likelihood of being recommended by a PR (90.5%). Conclusions: YT content on PMR for ASW is easily understandable but lacks actionability and has moderate overall quality. Videos are highly sponsored and questionably representative of patients’ and women’s perspective. Therefore, Improving YT content, especially in AL, is crucial for supporting BC patients exploring PMR. The identified themes and insights into gender representation provide valuable guidance for content creators aiming to improve digital resources for this digitally connected demographic.

AI-Assisted Sentiment Analysis of ACFAS Fellowship-Trained Foot and Ankle Surgeon Online Reviews

Online physician ratings and reviews prove useful among patients when selecting a provider. Analyzing such reviews across medical and surgical specialties to determine their emotional tone through sentiment analysis yielded varying levels of positivity, negativity, and neutrality. To provide insight into what patients are saying, this study similarly analyzes the sentiment of physician ratings and reviews among foot and ankle surgeons. Healthgrades ratings and reviews, entered as of February 2024, were collected among the American College of Foot and Ankle Surgeons (ACFAS) fellows along with each surgeon's demographic information. ChatGPT was used to perform a sentiment analysis to describe the positivity, negativity, and neutrality of online physician reviews. Ratings and review sentiment were described among the sample and between sexes. Among 268 fellows, men received higher average rating scores than women (p = 0.02), From the 2,339 reviews, women received a greater proportion of negative reviews compared to men (p<0.001). The overall sentiment scores among men were higher than women (p < 0.001). There existed a very weak inverse relationship between ratings and years in practice (R = -0.16; p = 0.01). Fellowship-trained foot and ankle surgeons received predominantly positive reviews. When comparing sexes, males received higher ratings with higher sentiment scores. As patients place high credence in online reviews when selecting their provider, surgeons should remain mindful of and monitor or manage their online reputation. Level of EvidenceIV

Interpretable artificial intelligence-based analysis for morphologic classification of neuroblastic tumors.

10023 Background: Rapid and accurate identification of morphologic features of neuroblastic tumors (NTs) is critical for risk stratification and therapeutic decision making. The prognostic value of features like neuroblast differentiation, mitosis-karyorrhexis index (MKI), and Schwannian stromal presence is well established. Deep learning permits objective histopathological analysis, streamlining workflows for pathologists, notably in rare cancers. In rare cancers, our method minimizes bias and optimizes limited data using transfer and self-supervised learning (SSL) for feature extraction, with improved explainability. Here, we used an artificial intelligence-based model to morphologically classify NT tumors and MYCN-amplification. Methods: Annotated H&amp;E-stained slides of diagnostic NT tumor biopsies from the University of Chicago and the Children’s Oncology Group were digitalized. Pathologists defined three binarized measures including diagnostic category (ganglioneuroblastoma/neuroblastoma), grade (differentiating/poorly differentiating), and MKI (low and intermediate/high). MYCN status was abstracted from patient records (amplified/non-amplified). Using Slideflow, our open-source pipeline, we developed an attention-based multiple instance learning model with features extracted by CTransPath, a SSL model pretrained on pan-cancer images from The Cancer Genome Atlas. For each measure, model performance was evaluated using 5-fold cross validation by aggregating k-fold model predictions across multiple metrics. Patients were excluded from a model if the measure of interest was unknown. Feature significance was assessed visually using Class Activation Mapping (Grad-CAM). Results: The mean age of the study cohort (n = 172) was 3.66 years. Of patients with clinical information, 84 of 138 (60.2%) had metastatic disease and 94 of 133 (70.7%) were high-risk. Of the 148 tumors with a diagnostic category of neuroblastoma, 93.2% were poorly differentiated and 25% had high MKI. Of the 135 tumors with known MYCN status, 40 were amplified (29.6%). The final models excelled across all outcomes, performing best for diagnostic category, grade, and MYCN status (Table 1). Physician review of the attention-based heatmaps for all measures highlighted biologically relevant regions such as neuropil. Conclusions: We created a deep learning pipeline for auto-characterization of digitized H&amp;E-stained NT pathology slides. Our approach may also aid in identifying molecular features including MYCN-amplification. Review of heatmaps showed pertinent biological tissue, boosting model reliability.[Table: see text]

A web-based, LLM-powered AI symptom summarization tool (ASST) for monitoring of breast cancer treatment toxicity.

e13622 Background: Traditional methods for capturing patient-reported outcomes (PROs) are often time-consuming and may lead to underreporting of symptoms by physicians. With advancements in artificial intelligence (AI), particularly large language models (LLMs) like GPT-4, there is potential to revolutionize this aspect of patient care. Methods: This study developed and tested a GPT-4-based web application for monitoring cancer treatment toxicity in breast cancer patients. The application utilized 35 items from the PRO-CTCAE scale to create an interactive form for patients to report treatment-related symptoms. Upon form completion, a natural language summary of the patient’s responses was generated using the GPT-4 API for physician review. Nine radiation oncologists (5 attendings, 4 residents) evaluated the summaries of virtual patients using the adapted Physician Documentation Quality Index (PDQI-9). PDQI-9 consists of 9 items scored using a 5-point Likert scale (1 -not at all- to 5. -extremely-). IRB approval was not required because this study did not use real patient data, in accordance with the 2018 Revised Common Rule Requirements of the NIH. All data used were researcher-generated and non-identifiable. Results: The mean time for textual summary generation was 7 (5.7-9.2) seconds. The AI-Symptom Summarization Tool (ASST) mean scores were 4.25 for accuracy and 4 for thoroughness. Usefulness, organization, comprehensibility, concision, synthesis quality and consistency were all rated between 3 and 4. Overall score was 42/50. No hallucinations (false information made up by the LLM) were found. Conclusions: The GPT-4-based application for cancer treatment toxicity monitoring demonstrates significant promise in enhancing the quality and efficiency of patient care. By automating the documentation of patient-reported symptoms, this tool could allow physicians to focus more time on patient interaction and individualized care. As AI technologies continue to evolve, their integration into clinical practice must be approached with caution, emphasizing augmentation over replacement and the importance of verification to maintain trust in these new tools.

Remote patient monitoring for early detection of immune checkpoint inhibitor therapy-related pneumonitis in high-risk patients: A pilot-feasibility study.

e13806 Background: Immune checkpoint inhibitor therapy-related (ICI) pneumonitis poses a significant challenge for patients with cancer. It is associated with increased clinical morbidity and mortality, often resulting in prolonged hospital stays and delays and/or discontinuation of cancer-directed treatment. Early detection and prompt treatment of ICI pneumonitis is critical to improve patient outcomes. Therefore, our institution implemented a remote patient monitoring (RPM) program leveraging in-home, electronic health record-integrated technology and virtual centralized nursing to monitor vital signs and patient symptoms in “high-risk” patients treated with immunotherapy. Methods: Patients receiving immunotherapy deemed to be “high-risk” for ICI pneumonitis seen at Mayo Clinic Rochester, Minnesota were offered enrollment in the RPM program beginning in June 2023. High-risk patients included those with pre-existing lung disease (chronic obstructive pulmonary disease, interstitial lung disease); those with a history of resolved ICI pneumonitis being rechallenged with immunotherapy; and those with grade 1 ICI pneumonitis continuing immunotherapy. Eligible patients who agreed to participate in the program were provided a kit consisting of RPM technology and electronic symptom questionnaires. The kit included an armband for continuous vital sign monitoring, a blood pressure monitor, a scale to measure weight, and a pre-connected cellular-enabled tablet for questionnaire answering. RPM data utilized real-time electronic health system integration and predetermined parameters to alert RPM nurses. Pre-specified escalation care pathways were then used to respond and escalate care, as necessary. The primary outcome of this pilot study was to assess the feasibility of implementing an RPM program in this patient population with a secondary outcome to determine the ability to detect ICI pneumonitis before significant clinical symptoms. Results: Sixteen patients were enrolled of whom 5 declined participation (2 due to inconvenience and 3 for unclear reasons) resulting in 11 patients who underwent monitoring. Five patients graduated monitoring without detection of ICI pneumonitis within a 4-month timeframe. Six patients are currently undergoing RPM. 372 alerts occurred of which 8 (2%) were escalated to physician review. These alerts included elevated heart rate in 4, shortness of breath in 3, and elevated blood pressure in 1. One patient was triaged to an emergency department and diagnosed with grade 2 ICI pneumonitis. Conclusions: We show that RPM in patients who are at high-risk for developing ICI pneumonitis can be effectively integrated into a care model and monitor pneumonitis related symptoms. Further studies are needed to prove whether RPM is beneficial in this population including other ICI toxicities.

Evaluation of multiple-vendor AI autocontouring solutions

BackgroundMultiple artificial intelligence (AI)-based autocontouring solutions have become available, each promising high accuracy and time savings compared with manual contouring. Before implementing AI-driven autocontouring into clinical practice, three commercially available CT-based solutions were evaluated.Materials and methodsThe following solutions were evaluated in this work: MIM-ProtégéAI+ (MIM), Radformation-AutoContour (RAD), and Siemens-DirectORGANS (SIE). Sixteen organs were identified that could be contoured by all solutions. For each organ, ten patients that had manually generated contours approved by the treating physician (AP) were identified, totaling forty-seven different patients. CT scans in the supine position were acquired using a Siemens-SOMATOMgo 64-slice helical scanner and used to generate autocontours. Physician scoring of contour accuracy was performed by at least three physicians using a five-point Likert scale. Dice similarity coefficient (DSC), Hausdorff distance (HD) and mean distance to agreement (MDA) were calculated comparing AI contours to “ground truth” AP contours.ResultsThe average physician score ranged from 1.00, indicating that all physicians reviewed the contour as clinically acceptable with no modifications necessary, to 3.70, indicating changes are required and that the time taken to modify the structures would likely take as long or longer than manually generating the contour. When averaged across all sixteen structures, the AP contours had a physician score of 2.02, MIM 2.07, RAD 1.96 and SIE 1.99. DSC ranged from 0.37 to 0.98, with 41/48 (85.4%) contours having an average DSC ≥ 0.7. Average HD ranged from 2.9 to 43.3 mm. Average MDA ranged from 0.6 to 26.1 mm.ConclusionsThe results of our comparison demonstrate that each vendor’s AI contouring solution exhibited capabilities similar to those of manual contouring. There were a small number of cases where unusual anatomy led to poor scores with one or more of the solutions. The consistency and comparable performance of all three vendors’ solutions suggest that radiation oncology centers can confidently choose any of the evaluated solutions based on individual preferences, resource availability, and compatibility with their existing clinical workflows. Although AI-based contouring may result in high-quality contours for the majority of patients, a minority of patients require manual contouring and more in-depth physician review.