Manual Review Research Articles

A Framework for Developing and Assessing Custom Case Definitions: A Demonstration Applied to Opioid Overdose in Maryland.

Public health epidemiologists monitor data sources for disease outbreaks and other events of public health concern, but manual review of records to identify cases of interest is slow and labor-intensive and may not reflect evolving data practices. To automatically identify cases from electronic data sources, epidemiologists must use "case definitions" or formal logic that captures the criteria used to identify a record as a case of interest. To establish a methodology for development and evaluation of case definitions. A logical evaluation framework to approach case definitions will allow jurisdictions the flexibility to implement a case definition tailored to their goals and available data. Case definition development is explained as a process with multiple logical components combining free-text and categorical data fields. The process is illustrated with the development of a case definition to identify emergency medical services (EMS) call records related to opioid overdoses in Maryland. The Maryland Department of Health (MDH) installation of the Electronic Surveillance System for Early Notification of Community-Based Epidemics (ESSENCE), which began capturing EMS call records in ESSENCE in 2019 to improve statewide coverage of all-hazards health issues. We describe a case definition evaluation framework and demonstrate its application through development of an opioid overdose case definition to be used in MDH ESSENCE. We show the iterative process of development, from defining how a case can be identified conceptually to examining each component of the conceptual definition and then exploring how to capture that component using available data. We present a framework for developing and qualitatively assessing case definitions and demonstrate an application of the framework to identifying opioid overdose incidents from MDH EMS data. We discuss guidelines to support jurisdictions in applying this framework to their own data and public health challenges to improve local surveillance capability.

A Comparison of Automatically Extracted Quantitative EEG Features for Seizure Risk Stratification in Neonatal Encephalopathy.

Seizures occur in up to 40% of neonates with neonatal encephalopathy. Earlier identification of seizures leads to more successful seizure treatment, but is often delayed because of limited availability of continuous EEG monitoring. Clinical variables poorly stratify seizure risk, and EEG use to stratify seizure risk has previously been limited by need for manual review and artifact exclusion. The goal of this study is to compare the utility of automatically extracted quantitative EEG (qEEG) features for seizure risk stratification. We conducted a retrospective analysis of neonates with moderate-to-severe neonatal encephalopathy who underwent therapeutic hypothermia at a single center. The first 24 hours of EEG underwent automated artifact removal and qEEG analysis, comparing qEEG features for seizure risk stratification. The study included 150 neonates and compared the 36 (23%) with seizures with those without. Absolute spectral power best stratified seizure risk with area under the curve ranging from 63% to 71%, followed by range EEG lower and upper margin, median and SD of the range EEG lower margin. No features were significantly more predictive in the hour before seizure onset. Clinical examination was not associated with seizure risk. Automatically extracted qEEG features were more predictive than clinical examination in stratifying neonatal seizure risk during therapeutic hypothermia. qEEG represents a potential practical bedside tool to individualize intensity and duration of EEG monitoring and decrease time to seizure recognition. Future work is needed to refine and combine qEEG features to improve risk stratification.

Implementation of an Automated Code Checking Algorithm Based on Site Analysis

To date, BIM has been primarily utilized in cost and schedule management, an interference check between architectural and structural models and systems based on geometric data in the process of the construction life cycle. However, there is a lack of research that utilizes the information contained in the BIM model to review whether the proposed architectural model is appropriately designed in accordance with each country’s building regulations or building codes or that proposes a model optimized for laws and standards. ‘Building code checking’ is the step of reviewing whether a building designed based on the building codes is suitable for being constructed as a building. However, this process consumes significant time and money and leads to human errors due to the manual review process. This study included implementation of an algorithm based on the Korean building code. In this study, there was the process of selection of codes when architects interpret building codes in common and implementation based on the codes selected. Next, modeling based on DXF files from NGII (National Geographic Information Institute) was applied to the algorithm developed in this study. Last, it includes case studies that compare the outputs of the algorithm with the real buildings, which had received real code checking, to make sure the algorithm in this paper is working properly. The implementation of such an automated system has the potential to significantly improve the efficiency and effectiveness of the building design and construction process. It can help architects to quickly and accurately identify potential legal issues and provide alternative solutions that meet regulatory requirements. This, in turn, can lead to reduced project costs, improved quality of designs, and faster project delivery times.

Multicenter appraisal of comorbid TANGO2 deficiency disorder in patients with 22q11.2 deletion syndrome.

TANGO2 deficiency disorder (TDD) is a rare, autosomal recessive condition caused by pathogenic variants in TANGO2, a gene residing within the region commonly deleted in 22q11.2 deletion syndrome (22q11.2DS). Although patients with 22q11.2DS are at substantially higher risk for comorbid TDD, it remains underdiagnosed within 22q11.2DS, likely due to overlapping symptomatology and a lack of knowledge about TDD. Initiation of B-vitamin supplementation may provide therapeutic benefit in TDD, highlighting the need for effective screening methods to improve diagnosis rates in this at-risk group. In this retrospective, multicenter study, we evaluated two cohorts of patients with 22q11.2DS (total N = 435) for possible comorbid TDD using two different symptom-based screening methods (free text-mining and manual chart review versus manual chart review alone). The methodology of the cohort 1 screening method successfully identified a known 22q11.2DS patient with TDD. Combined, these two cohorts identified 21 living patients meeting the consensus recommendation for TANGO2 testing for suspected comorbid TDD. Of the nine patients undergoing TANGO2 sequencing with del/dup analysis, none were ultimately diagnosed with TDD. Of the 12 deaths in the suspected comorbid TDD cohort, some of these patients exhibited symptoms (rhabdomyolysis, cardiac arrhythmia, or metabolic crisis) suspicious of comorbid TDD contributing to their death. Collectively, these findings highlight the need for robust prospective screening tools for diagnosing comorbid TDD in patients with 22q11.2DS.

Outcomes of a tele-intensive care unit pharmacist on the management of diabetic ketoacidosis.

In an effort to expedite the publication of articles, AJHP is posting manuscripts online as soon as possible after acceptance. Accepted manuscripts have been peer-reviewed and copyedited, but are posted online before technical formatting and author proofing. These manuscripts are not the final version of record and will be replaced with the final article (formatted per AJHP style and proofed by the authors) at a later time. The tele-intensive care unit (tele-ICU) pharmacist facilitates patient-specific diabetic ketoacidosis (DKA) treatment utilizing guideline-directed therapy. This study was designed to determine how patient-specific interventions by a tele-ICU pharmacist affected patients with DKA compared to the standard of care. This retrospective cohort study utilized custom reports and manual chart review to evaluate the electronic health records of patients 18 years or older who received continuous intravenous insulin and were admitted for DKA between January 2019 and December 2020. The primary endpoint was time to DKA resolution, defined by the patient meeting at least 2 of the following criteria: a serum bicarbonate concentration of at least 18 mEq/L, an arterial pH of greater than 7.3, and closure of the anion gap (less than or equal to 12 mEq/L). Patients treated with tele-ICU pharmacist patient-specific interventions reached DKA resolution 7.32 hours earlier than patients treated with the standard of care (22.16 vs 29.48 hours; P = 0.0019). There was no statistically significant difference between the groups for ICU length of stay, time until subcutaneous insulin administration, incidence of hypoglycemia, incidence of severe hypoglycemia, and sodium bicarbonate use. In patients who received a tele-ICU pharmacist intervention, there was a statistically significant increase in the volume for fluid resuscitation and the amount of total continuous insulin infused after ICU admission and a statistically significant reduction in the time between laboratory draws. Treatment of patients with tele-ICU pharmacist patient-specific interventions was associated with faster DKA resolution, more frequent laboratory monitoring, and a higher volume of insulin and fluids infused than in patients treated with protocol-driven therapy.

Detection of immune-related adverse events among hospitalized patients using large language models.

2638 Background: Immune checkpoint inhibitor (ICI)-induced colitis, hepatitis, and pneumonitis are common immune-related adverse events (irAEs); however, the true incidence for these irAEs remains incompletely understood. Chart review is the gold standard for their detection but is time-consuming and cannot be implemented in large cohorts. The use of ICD codes is limited in sensitivity and specificity. Large language models (LLMs) are a scalable method of answering queries from human-generated text, though there is no data on the use of LLM for the identification of irAEs. Therefore, we investigated the application of a LLM to identify ICI-colitis, hepatitis, and pneumonitis among hospitalized patients, comparing its performance to manual chart review and ICD codes. Methods: Hospital admissions of patients on ICI therapy from February 5th, 2011, to November 3rd, 2021, were manually reviewed by a multidisciplinary immunotoxicity team using established published definitions for the presence of ICI colitis, hepatitis, and pneumonitis. Standard ICD codes and a LLM pipeline with retrieval-augmented generation (RAG) were used to detect irAEs. Performance was measured via sensitivity, specificity, and model runtime. The LLM was validated with a second dataset of inpatients with ICI colitis, hepatitis, and pneumonitis admitted from November 4th, 2021, to September 5th, 2023. Results: Among 5,677 hospitalized patients on ICI therapy in the initial cohort, there were 132 cases adjudicated with ICI colitis, 57 with ICI hepatitis, and 47 with ICI pneumonitis. The LLM was more sensitive in detecting all three irAEs compared to ICD codes (94.2% vs. 71.8%), achieving significance for ICI hepatitis (p<0.001) and pneumonitis (p=0.006), while having similar specificities (92.5% vs 91.1%, Table 1). The LLM approach was also efficient, spending an average of 9.42s per chart, compared to an estimated 15 minutes per chart for individual chart review. The mean sensitivity and specificity of the LLM on the validation dataset for adjudicated ICI colitis (n=20), hepatitis (n=24), and pneumonitis (n=6) were 96.9% and 93.2%, respectively. Conclusions: LLMs serve as a useful tool for the detection of ICI colitis, hepatitis, and pneumonitis, significantly outperforming ICD-codes in accuracy and manual chart review in efficiency.[Table: see text]

Performance evaluation of ChatGPT in detecting diagnostic errors and their contributing factors: an analysis of 545 case reports of diagnostic errors

BackgroundManual chart review using validated assessment tools is a standardised methodology for detecting diagnostic errors. However, this requires considerable human resources and time. ChatGPT, a recently developed artificial intelligence chatbot...

Disrupting how we recruit to cancer clinical trials.

11072 Background: Traditional methods to identify eligible patients for cancer clinical trials rely heavily on manual review, a time-consuming bottle neck limiting access to novel treatments, particularly for underserved populations. This study presents an automated informatics platform designed to simplify and streamline patient identification and trial matching. Methods: Our platform leverages three tools: 1. Parser:Extracts and categorizes trial data based on cancer type, stage, and receptor status from the National Cancer Institute's ClinicalTrials.gov database, storing it in a structured JSON format; 2. Screener: Utilizes Epic Clarity electronic health record data to identify upcoming patients approaching treatment decision points, times when clinical trials are more relevant. Patients are classified as "newly diagnosed" or "potentially progressed" based on progress notes and imaging reports. Eligibility criteria, including stage and receptor status, are assessed using regular expressions; 3. Matcher: Pairs eligible patients with relevant trials based on compatible cancer stage and receptor status. One Cancer Center (CC) is prospectively using the platform to identify patients. The 2nd CC is refining the code using retrospective data. Accuracy was measured comparing automated to manual review by experienced trained abstractors. Results: Implementation across two cancer centers yielded: 1. Reduced manual review burden: Between June 2023 – Jan 2024, of 674 breast cancer patients seen at CC #1, 371 were new patients (N=371) and thus, at a treatment decision point; 3 did not undergo imaging. Of the 300 established patients undergoing imaging studies, the algorithm identified 50/300 (17%) scans as “positive,” and 61 (20%) with mixed disease progression/ regression, requiring manual review and trial matching. 189 scans were read as “negative”: 96% True Negatives; 4% were False Negatives. Automated screening reduced manual review from 300 to 111 patients, significantly alleviating workload on healthcare personnel. 2. Variable staging accuracy: While receptor status classification proved effective, stage accuracy varied between centers, highlighting a need for further refinement in this area. 3. Accurate receptor status classification: The system achieved high accuracy in demonstrating reliable assessment of this vital factor. Conclusions: This automated informatics platform demonstrates significant potential to streamline case identification for cancer clinical trials and thus, foster broader access to novel therapies. Its adaptability makes it suitable for implementation across diverse healthcare settings, empowering wider participation in advancements in cancer care. Ongoing efforts to refine the staging module are underway. Clinical trial information: NCT05146297 . [Table: see text]

Improving quality of oncology (onc) documentation and enhancing structured data collection using a standardized onc note template.

11128 Background: To improve documentation quality in Medical Onc outpatient clinics, a standardized template was created. In this template we embedded: 1. An Epic SmartForm to collect structured data (SD) about patient, disease, and response status at each encounter. Minimal Common Oncology Data Elements (mCODE) compatible SD elements were used to enhance interoperability. 2. Several clinically impactful quality metrics (QM) selected based on the Quality Onc Practice Initiative (QOPI) guidelines. In this analysis, we aim to assess how template use impacted documentation quality. Methods: 113,376 outpatient encounters occurred between 01/2018 – 12/2022 (41 providers: 35 physicians, 6 APPs). The template go live was 3/19/2019. 2,520 randomly selected encounters were manually reviewed. Distinct documentation of 8 items served as QMs (Table). Of note, using the template was voluntary and data entry in the embedded form is not mandated. Documentation of each QM was compared between notes authored with/without template. For each note, a completeness score was calculated based on total number of QMs present/note. Categorical comparisons were made using chi-squared or Fisher’s exact tests. Numerical/continuous values were summarized using means and standard deviations. A template “user” was defined as a provider using the template in >10% of encounters after go live. This QI project was exempted by IRB. Results: 154/2520 encounters were excluded after manual review (visits not for solid tumor malignancy). The template was used in 38.7% (917/2366) of the reviewed notes. Cancer diagnosis was documented in >99.8% of notes regardless of template use. Template use was associated with increased documentation of 7 QMs (Table). Documentation overall completeness improved from 62% to 90% with a very large effect size (Table). Template use positively correlated with female gender and fewer years since last training (p<0.001). In the subgroup of notes written by “users” (21 physicians, 4 APPs), individual QM and overall completeness scores improved with template use (p<0.001). Conclusions: Template use profoundly improved QM documentation (statistically significant and clinically meaningful) by Onc clinic providers. This strategy enhances mCODE compatible structured data collection and semantic interoperability between systems, strengthening the available real-world data (RWD) in EHR. [Table: see text]

Quality of telehealth-provided cancer care in VA.

1633 Background: As the largest integrated health care system in the United States, the Veterans Health Administration (VA) has extensive experience in utilizing telehealth-(TH) to deliver care to the 10 million enrolled Veterans, including the~50,000 patients newly diagnosed with cancer annually and ~500,000 cancer survivors. The VA National TeleOncology (NTO) service was established in 2020 to provide specialized treatment regardless of geographical location. We sought to compare quality in TH-delivered care compared to traditional (TR) in-person VA care. Methods: Using the VA EHR, we identified patients who had an ICD-10 diagnostic code for an incident malignancy from Dec 2016 to March 2021 at three medical centers providing both TR and TH care. We used EHR review to classify patients as TH users, if they received TH services at least once for their cancer care, or TR care if no telehealth was used. Using both structured EHR data and manual review by two trained abstractors, we gathered demographic, clinical, and treatment characteristics to calculate 25 Quality Oncology Practice Initiative (QOPI) measures in the domains of symptoms and toxicity management (2), end of life and palliative care (10), and core measures (13). We report QOPI measures descriptively; and used chi-squared tests with alpha = 0.05 to compare TH and TR. Results: We identified 972 patients with lymphoma, prostate, lung or colorectal cancer. 427 (44%) were TH users and 545 (56%) were TR, and they were predominately white (n=819, 84.3%) men (n=930, 95.7%). TH users had better (n=13), worse (n=9), same (n=2) and unevaluable (n=1) descriptive performance on measures. For example, TH and TR were similar in administration of appropriate antiemetic therapy for high and moderate-risk antineoplastic agents (44.8% vs 42.4%). TH performed better than TR for chemotherapy summary being completed within three months of chemotherapy ending (83.4% vs 75.6%) and for emotional well-being being assessed by the second office visit (92.3% vs 89.9%). Appropriate tobacco use cessation administration within the previous year was statistically higher in TH (85.3% vs 76.2% p=0.0021). No differences were statistically significant for any other QOPI measures. Conclusions: VA is a leader in TH cancer care because of both its volume and quality. VA-provided TH cancer care quality is similar to or better than that of TR in-person care. NTO specifically, and VA teleoncology broadly, provides another option to Veterans for cancer care.

EHR documentation of cost discussions and financial interventions for a cohort of patients with advanced cancer: Findings from a retrospective chart review-based study.

e23150 Background: Financial Toxicity (FT), the adverse impact of high medical costs on quality of life, notably affects many oncology patients. In a prior initiative to improve quality of life for advanced cancer patients, a palliative care-trained nurse practitioner (PCNP) was embedded in an academic oncology practice. Patients completed the Edmonton Symptom Assessment System (ESAS) and Canadian Problem Checklist (CPC) to gauge physical symptoms, psychosocial issues, and financial concerns. However, the incidence of self-reported financial issues and electronic health record (EHR) documentation of cost discussions and interventions remain unclear for this patient group. Methods: The study included adult oncology patients with advanced cancer seen by the PCNP from January 1, 2020, to January 1, 2022. We assessed incidence of self-reported financial concerns via CPC checklists in this cohort. Ten charts underwent detailed manual review for evidence of cost discussions or referral to finance-related interventions for up to 1 year from the initial consult with the PCNP or until death. During the review, we identified keywords in documentation of cost discussions and finance-related interventions to create a keyword library. We iteratively compared the library to our reference standard manual chart review to ensure 100% sensitivity. We then employed this keyword library to evaluate the remaining charts. Descriptive statistics were employed to assess correlations between self-reported financial concerns, chart documentation of cost discussions or financial interventions, and patient demographics. Results: Out of 109 eligible patients, 71% (77) completed pre-visit questionnaires, with 30% (23) reporting financial issues. Of this group, 43% (10) died during the study period. Average follow-up was 5.0 months among the decedents. Among those reporting financial concerns, 65% (15) had documented cost discussions or financial assistance referrals in the EHR, involving physicians, social workers, and other clinic staff. Most interventions addressed medication costs and referrals to financial assistance as indicated. Patient demographics (age, sex, insurance type) showed no apparent association with chart documentation of cost discussions or financial assistance. Conclusions: We used retrospective chart review to evaluate the extent of documentation of cost discussions and finance-related intervention in a cohort of patients with advanced cancer. In our patient cohort, 30% self-reported financial concerns, a proportion consistent with previously reported incidence of FT among oncology patients. Notably, 35% of the patients with known financial concerns had no other relevant documentation in the EHR. More work is needed to establish best practices for systematic screening for FT and interventions to support patients.

Novel approach to implementing natural language processing for clinical staging of non-small-cell lung cancer.

e13624 Background: Cancer staging is instrumental in driving clinical management and trial enrollment, but staging data are generally unreliable and unstructured in the electronic health record (EHR). Advances in natural language processing (NLP) may facilitate clinical staging and documentation [1], but challenges to real-world implementation include (1) automatically identifying appropriate patients and reports from the EHR and (2) developing an unbiased dataset for training and validation [2]. We describe our institution’s novel approach to overcome these barriers while building an in-house NLP pipeline for clinical tumor staging of non-small-cell lung cancer (NSCLC). Methods: We identified patients by searching our EHR (Epic) for a molecular analysis test ordered specifically for pathological diagnoses of NSCLC at our institution. We used the test order date as the diagnosis proxy date (DPD). For each patient, we extracted imaging reports up to 16 weeks before and 6 weeks after the DPD. To derive primary tumor size, we analyzed the CT Chest or PET/CT report closest to the DPD using an oncology-trained NLP text extraction and labeling tool (John Snow Labs). We cleaned all extracted tumor size entities and identified the largest measurement linked to the lungs. We compared primary tumor measurements from the NLP pipeline to those in a preexisting, manually compiled cancer registry (CNEXT). We manually analyzed discrepancies through chart review. Results: 542 patients with a DPD between 11/2016 - 9/2023 were processed through the NLP pipeline. Of 443 patients with valid values in both the pipeline and CNEXT, 53% (234) were exact matches, and 20% (90) had a close match (within 0-5mm), yielding a 73% accuracy rate for values within 5mm. When mismatched values were manually reviewed, several cases in CNEXT were found to have a DPD differing by more than 3 months and tumor sizes derived from external reports. When these cases were excluded, 320 of the remaining 349 patients had valid values in both the pipeline and the updated manual review. In this refined population, 66% (213) were exact matches, and 15% (48) had a close match, yielding an 82% accuracy rate for values within 5mm. Conclusions: To our knowledge, this is the first report of a pathology-based method to automatically and reliably identify patients with NSCLC and their relevant imaging reports directly from the EHR. We used a prebuilt NLP tool to derive primary tumor sizes with relatively high accuracy and found that adding flags for timeline discrepancies and external reports can further improve validity. As we near completion of analogous pipelines for node and metastasis staging, we will develop methodology to identify subgroups of patients that can be clinically staged with near-perfect accuracy, ultimately aiming to substantially limit manual staging of uncomplicated cases. 1. Puts 2023. 2. Wang 2022.

Preparedness training programs improved the emergency response and crucial outcomes among nursing students: a scoping review

Abstract Objective The study aims to review and discuss the outcomes of emergency preparedness training programs among nursing students. Methods A scoping review was conducted based on the guidelines of Arskey and O’Malley. Databases through PubMed, Science Direct, and Google Scholar were used to obtain the relevant articles within the scoping review. Results The researchers found 1175 articles. After manual review, 28 articles were identified and assessed. The review reported that the emergency preparedness training program significantly affects preparedness, knowledge, performance, self-efficacy, learning motivation, self-confidence, satisfaction, critical decision-making, and problem-solving. Conclusions Emergency training effectively increases knowledge, skills, confidence, satisfaction, and team performance.

Effect of tumor burden on survival in patients with stage IV EGFR-mutant NSCLC receiving osimertinib monotherapy.

8611 Background: Osimertinib (osi) has traditionally been given as monotherapy for patients (pts) with stage IV EGFR-mutant (EGFRm) non-small cell lung cancer (NSCLC), although acquired resistance remains a challenge. Recent data from the FLAURA2 study demonstrated a progression free survival (PFS) benefit of approximately 9 months (mos) when osi is combined with platinum-pemetrexed chemotherapy. Despite this improvement, combination therapy is associated with higher rates of adverse events, making it challenging to decide which pts should receive TKI + chemo vs TKI monotherapy. Tumor burden (TB) in pts with stage IV EGFRm NSCLC, which may be associated with risk of developing resistant clones on TKI monotherapy, and its association with clinical outcomes has not been examined. Methods: Retrospective, single institution manual case review of pts who received osi monotherapy. Primary objectives were PFS (i.e., oligoprogression requiring local therapy, disease progression (PD) with osi discontinuation, or death) and overall survival (OS) related to extent of TB. Those who did not receive osi monotherapy for stage IV EGFRm NSCLC or discontinued osi for reasons other than PD were excluded. Demographics, comorbidities, performance status, PD-L1 expression, histology, EGFRm subtype, prior TKI use, total number of visceral tumors (TTn), and total number/sites of metastatic disease (MDn) collected for analysis. Extent of TB categorized by TTn (1-4, 5-9, ≥10) and MDn (1-2, 3-4, ≥5). PFS and OS estimated by Kaplan-Meier method and compared using log-rank test. Results: 325 pts eligible for analysis. Median follow-up time was 22.7 mos. Median PFS (mPFS) for the overall cohort was 16.3 mos (95% CI; 13.7 - 18.9) and median OS (mOS) was 31.1 mos (95% CI; 27.4 - 39.3). Pts with a TTn of 1-4 had significantly better mPFS compared to those with a TTn of 5-9 and ≥10 (mPFS: 21.5 vs 16.3 vs 13.4 mos, respectively; P = 0.001). Pts with a TTn ≥10 had significantly worse mOS compared to those with a TTn of 5-9 and 1-4 (23.7 vs 35.1 vs 57.7 mos, respectively; P < 0.0001). Similarly, pts with a MDn of 1-2 had significantly better mPFS and mOS compared to those with a MDn of 3-4 and ≥5 (mPFS: 21.6 vs 11.4 vs 9.0 mos, respectively [P < 0.0001]; mOS: 44.3 vs 20.8 vs 22.4, respectively [P < 0.0001]). Pts with liver and bone mets had significantly worse mPFS and mOS, whereas those with lung and brain mets had worse mOS without affecting PFS. Conclusions: The extent of baseline TB, categorized by TTn and MDn, impacts both PFS and OS in pts with stage IV EGFRm NSCLC treated with osi monotherapy. Pts with “low-TB” (i.e., TTn 1-4, MDn 1-2) appear to have a survival advantage compared to those with “intermediate- or high-TB” (i.e., TTn 5-9 and ≥10, MDn 3-4 and ≥5) and may derive less benefit from the addition of chemotherapy to osi. Prospective, matched real-world treatment cohorts are warranted to optimize treatment strategies for this patient population.

Real world outcomes of patients with small cell lung cancer (SCLC) presenting with brain metastasis at diagnosis in a single institution health system.

e20136 Background: Extensive-stage small cell lung cancer (ES-SCLC) is an aggressive lung cancer subtype with high propensity for brain metastases (mets). Despite this, there is no consensus on the treatment of brain mets especially in the setting of frontline chemoimmunotherapy and advanced radiation techniques like Gamma Knife radiosurgery (GKRS). Methods: In this single-institution retrospective study, we queried the Yale Tumor registry for all patients diagnosed with ES-SCLC in 1/2016-4/2023. Patients with brain mets at the time of presentation were identified and underwent manual chart review for abstraction of clinical information specifically pertaining to brain mets and treatment modalities. Survival outcomes including median progression free survival (mPFS), median central nervous system-progression free survival (mcPFS) and median overall survival (mOS) measured in months were also captured. Results: 423 ES-SCLC patients were identified. Of those, 92 (22%) had brain mets at initial presentation and were reviewed. 44 (48%) of patients with brain mets were treated with frontline chemotherapy (platinum doublet) while 42 (46%) were treated with chemoimmunotherapy (platinum doublet with atezolizumab or durvalumab). Patients treated with upfront chemotherapy had mPFS 6.4 (95% CI -1.2, 14.1), mcPFS 7.7 (95% CI -1.2, 17.1), and mOS 9.3 (95% CI -13.7, 36.4), compared to patients treated with chemoimmunotherapy, who had mPFS 4.9 (p=0.23; 95% CI -3.5, 15.4), mcPFS 6.7 (p=0.99; 95% CI 3.3, 20.8), and mOS 7.7 (p=0.89; 95% CI -11.4, 33.9). 63 (68%) of patients underwent upfront brain radiation. 44 (48%) had whole brain radiation (WBRT) while 19 (30%) underwent GKRS. Of the patients undergoing GKRS, 8 (42%) were symptomatic. 7 patients (37%) presented with 1 brain met, and 12 (63%) patients presented with 2-5 brain mets. For patients treated with GKRS, mPFS 6.8 (95% CI 2.3, 10.4), mcPFS 7.2 (95% CI 2.9, 12.6), and mOS 11.1 (95% CI -13.0, 40.9). For patients undergoing WBRT, mPFS 6.5 (p = 1.0; 95% CI -4.5, 20.2), mcPFS 6.5 (p = 0.96; 95% CI -3.9, 21.6), and mOS 8.6 (p = 0.21; 95% CI -17.7, 39.9). Conclusions: GKRS is a viable option for SCLC patients presenting with brain metastases, especially in those with a low number of mets. Further studies are required to determine the optimal ES-SCLC candidates for upfront GKRS. [Table: see text]

Evaluating the impact of translational research enrollment on the patient care experience of Veterans with prostate cancer who receive care through the Veterans Affairs (VA) healthcare system.

e23210 Background: Veterans with cancer represent a unique subset of patients with distinct exposure histories and cancer epidemiology. Translational research studies that incorporate Veteran-derived biospecimens are therefore critical to advance our understanding of Veteran cancer biology and improve treatment options for Veterans with cancer. Biospecimen-based translational studies also offer an opportunity to enhance the Veteran care experience by enabling Veterans to directly contribute to cancer research efforts. While the biologic benefits of biospecimen research have been well established, the benefit of translational research on the Veteran care experience has yet to be defined. In this study, we evaluated: 1. Whether Veteran enrollment in a biospecimen-based prostate cancer research study improved their VA patient care experience and satisfaction and 2. Whether a VA-based data analyst tool could be used to optimize patient enrollment in translational research studies. Methods: Beginning in January, 2022, Veterans with a diagnosis of prostate cancer receiving care at the William S. Middleton VA were enrolled in a translational research study that collected blood and tissue biospecimens for subsequent investigation and analysis. After at least 30 days of enrollment, subjects were sent an anonymous survey inquiring about how participation in the translational research study impacted their VA care experience and healthcare satisfaction. To optimize enrollment in the study, an ICD-9 code search function was employed within the medical record database to readily identify eligible VA clinic patients based on an active prostate cancer diagnosis. Descriptive analyses were performed using Prism statistical software. Results: Nearly all Veterans who responded to the questionnaire agreed that participation in translational research was a meaningful experience that increased their satisfaction with their VA medical care. In addition, a vast majority of Veterans surveyed felt that their research participation could help improve care for other Veterans, and most expressed willingness to participate in future VA research opportunities. In a secondary analysis, our ICD-9 search function accurately identified eligible VA clinic patients with a prostate cancer diagnosis, and implementation of this tool has saved approximately 30 minutes of weekly physician and research coordinator work compared to manual chart review. Conclusions: Veteran enrollment in translational research studies provides a meaningful experience for Veterans that can lead to increased satisfaction with the VA healthcare system. Future VA initiatives to improve the Veteran care experience should focus on expanding translational research opportunities for Veterans as well as addressing potential barriers to enrollment.

Constructing adverse event timelines for patients receiving CAR-T therapy using large language models.

2555 Background: Chimeric antigen receptor T (CAR-T) therapy is associated with a high risk of severe adverse events often only detailed in clinical notes. Monitoring them demands significant time and effort for manual chart review. Recent developments in large language modeling (LLMs) show promise for large-scale information extraction from clinical text. We performed a pilot study to evaluate the capability of the GPT-4 LLM to extract adverse events documented in the progress reports. Methods: We extracted progress notes within 30 days of any CAR-T administration from the UCSF deidentified clinical data warehouse. GPT-4, accessed through a HIPAA compliant Microsoft Azure Studio API, was used to extract CAR-T adverse events resulting in clinical intervention. A random sample of adverse events from 10% of patient notes were evaluated by a clinical reviewer (JG, PharmD). Topic modeling using BERTopic was used to cluster all adverse events to identify trends over time. Results: We identified 4183 clinical notes written within 30 days of CAR-T administration from 253 patients (39.1% women, 60.9% men). Mean age was 60.6 (SD:17.7). Manual validation of clinical notes from 10% of patients with CAR-T therapies (n=25) demonstrated that GPT4 was able to extract CAR-T related adverse events with 64% accuracy. We used BERTopic to cluster all extracted adverse events into 19 topics. Clusters with key terms “hyponatremia, leukocytosis, encephalopathy, toxicities, and neurologic” occurred most often (n=277), and primarily documented 12.9 days after CAR-T administration (Table). Conclusions: Although limited by use of de-identified data and absence of prompt engineering, this pilot supports the further investigation of LLMs for extraction of adverse events from unstructured clinical text. [Table: see text]

Human-AI teams to improve accuracy and timeliness of oncology trial prescreening: Preplanned interim analysis of a randomized trial.

1524 Background: Identifying eligible patients for oncology clinical trials (“prescreening”) relies on manual chart review by clinical research coordinators (CRCs), which is time-consuming and often inaccurate. Consequently, 70% of patients with cancer who meet trial eligibility criteria are not offered participation. Natural language processing (NLP) may improve the accuracy and timeliness of prescreening. Methods: This was a preplanned interim analysis of a paired-design noninferiority trial, powered to assess timeliness. We adapted NLP algorithms to identify 13 common eligibility criteria related to cancer type/stage, prior systemic therapy, actionable biomarkers, and response criteria. NLP systems performed optical character recognition, entity and relationship extraction, and natural language inference through deep learning and symbolic AI techniques. Deidentified unstructured electronic health records (EHRs) from real-world patients with non-small cell lung (NSCLC) or colorectal (CrCa) cancer were presented to CRCs via a secure platform. Two CRCs were randomized 1:1 to view blocks of 20 charts each with (Human+AI) or without (Human-alone) NLP annotations. In pre-trial assessment, the CRCs had 86% inter-rater agreement. The primary outcome was overall chart-level accuracy, defined as percent of the 13 CRC-coded eligibility items matching a gold standard set, determined by 2-3 clinicians blinded to experimental arms. Paired t-tests compared overall accuracy (noninferiority margin 5%) and criteria-specific accuracy. Significant differences were assessed for superiority using two-sided paired t-tests. Mann-Whitney tests compared the secondary outcome of timeliness (time per chart review). Results: Among 74 (40 NSCLC; 34 CrCa) of a planned 400 patients, overall accuracy for Human+AI was noninferior to Human-alone (78.7% vs. 76.7%, mean difference 2.0%, p<0.001 rejecting inferiority); both were greater than AI-alone (63.5%). Superiority for Human+AI was demonstrated for RECIST response, but not for overall nor other criteria-specific accuracy (Table). Median time per review was lower for Human+AI than Human-alone (34.1 vs 43.9 min, adjusted p=0.05). Conclusions: Human+AI teams can improve timeliness of trial prescreening with noninferior accuracy. This platform is being used for eligibility assessment in an ongoing clinical trial. [Table: see text]

Molecular and clinical features of pancreatic acinar cell carcinoma: A single institutional experience.

e16335 Background: Acinar Cell Carcinoma (ACC) accounts for around 1% of pancreatic cancer. The molecular and clinical features of acinar cell carcinoma are less characterized. We sought to evaluate the clinical and molecular features of patients with ACC at MD Anderson Cancer Center. Methods: Patients with ACC who had germline and/or somatic molecular testing at MD Anderson Cancer Center from 2008 to 2022 were identified retrospectively. After obtaining approval of approval of the institutional review board genomic alterations including germline or somatic BRCA mutations, KRAS mutation, demographic, clinical, and pathological information were extracted from institutional databases using the Palantir Foundry software platform (Syntropy); this included natural language processing with subsequent validation by manual chart review. Overall Survival (OS) was calculated from the date of diagnosis to death or last follow-up. The Kaplan-Meier method was used for survival analysis. Results: 16 patients were identified with molecular testing results available. There were 13 males and 3 females. The median age of diagnosis was 62.5 years (range 49 -72). 14/16 patients (87.5%) had metastatic disease, one (6.25%) patient had borderline resectable disease and one (6.25%) had localized resectable disease at initial diagnosis. Median OS was for 24 months for patients with metastatic disease. There were 2 patients with mixed pathology, 1 with acinar cell carcinoma with neuroendocrine features and the 2nd had mixed acinar cell-ductal carcinoma. 15/16 (93.75%) patients were tested for KRAS and all 15 were KRAS wildtype. Somatic mutations in DNA damage repair genes (BRCA2, PALB2) were present in 3 patients (25%) and 2 patients had germline mutation (BRCA and ATM) (16%) out of 12 patients tested for those genes. 6/16 (37.5%) patients were treated with FOLFIRINOX at first line. 3/6 (50%) of these patients had partial response, 2/6 (33%) had stable disease and only 1/6 (16%) had progression. 5 patients were treated with FOLFOX based regimen and 1/5 (20%) had response, 2/5 (40%) had stable disease and 2/5 (40%) had progression of disease. 5 patients were treated with gemcitabine-based regimen and only 1 had response which received gemcitabine/nab-paclitaxel combined with cisplatin. Conclusions: ACC were KRAS wild type and had higher rates of germline/somatic mutations in DNA damage repair genes. Better treatment outcome of FOLFIRNIOX/ platinum-based regimens was suggested based on this limited patient dataset. Larger studies are needed to further understand the molecular features of ACC and guide the treatment of this rare entity of pancreatic cancer.

Examination of irAE and treatment discontinuation irAE in patients with RCC with T effector phenotype.

4549 Background: The standard of care for renal cell carcinoma (RCC) is physician-choice of dual immunotherapy with ipilimumab/nivolumab (IO/IO) or a combination a VEGF inhibitor with immunotherapy (VEGF/IO). The IMmotion 151 trial identified gene expression signatures that differentiate likelihood of response to IO/IO (cluster 4 and 5 – T effector) versus VEGF/IO (cluster 1 and 2 - Angiogenic). Given that the T effector RNA-seq signature consists of higher expression of genes associated with inflammation, we hypothesized that patients with the T effector phenotype would have higher rates of immune-related adverse events (irAE) on IO-based therapy. Methods: Patients with metastatic RCC treated with systemic IO-based therapy had RNA sequencing completed on the primary tumor or metastatic site. Charts of patients who had a T effector RNA-seq signature were manually curated to identify development of irAE by reading the most recent clinic note and searching ‘steroids’, ‘irAE’, ‘rash’, ‘thyroid’. Hits from searches were investigated by manual chart review. Results: 118 patients underwent RNA sequencing and 105 passed quality control. Nineteen patients (18%) were assigned the T effector phenotype. Of the 17 metastatic patients, twelve of these patients received at least one dose of IO/IO in the first or second line of therapy, five were treated with VEGF/IO Among patients treated with IO/IO, 8 developed any grade irAE (66%; 41% grade 3+), 5 (41%) required steroid treatment, 5 (41%) required hospitalization, and 4 (33%) discontinued treatment due to irAE. Grade 3 toxicities in this group included colitis, adrenal insufficiency, and hypophysitis. Of the 5 patients treated with VEGF/IO, 3 developed any grade irAE (60%; 20% grade 3+), none were treated with steroids, and none were hospitalized. The grade 3 toxicity in this group was adrenal insufficiency. Conclusions: Patients with a T effector RNA-seq signature had higher rates ofirAE than historic controls when treated with IO/IO leading to high rates of steroid use and treatment discontinuation. VEGF/IO-treated patients with the T effector RNA-seq signature had similar rates of irAE to historic controls. The ongoing OPTIC clinical will expand on these results by studying treatment assignment in a prospective manner.