Data Warehouse Research Articles

Leveraging ML for Anomaly Detection in Healthcare Data Warehouses

The rapid emergence of digitalisation leads to unprecedented growth in the generation of the healthcare sector-particularly EHRs and medical equipment data. This extended the way for challenges for integrity in managing data and anomaly detection, including fraudulent transactions, medication errors, and many more system failures. Modern healthcare data poses a challenge to traditional methods of anomaly detection due to high and complex dimensionality. Machine learning provides a strong solution, using algorithms such as Gaussian Mixture Models, One-Class SVM and deep learning algorithms such as Autoencoders, and Recurrent Neural Networks in the detection of anomalies in healthcare data warehouse settings [1]. This study reports how ML can help advance care for patients, enable the validity of the data and reduce costs through real-time monitoring, fraud detection, and early detection of diseases. Applying anomaly detection through ML would most likely bring better operational performance, patient safety, and decision-making in health care for organizations as issues of poor data quality, lack of interpretability of models, and real-time detection would be addressed [2].

Automated MRI Quality Assessment of Brain T1-weighted MRI in Clinical Data Warehouses: A Transfer Learning Approach Relying on Artefact Simulation

The emergence of clinical data warehouses (CDWs), which contain the medical data of millions of patients, has paved the way for vast data sharing for research. The quality of MRIs gathered in CDWs differs greatly from what is observed in research settings and reflects a certain clinical reality. Consequently, a significant proportion of these images turns out to be unusable due to their poor quality. Given the massive volume of MRIs contained in CDWs, the manual rating of image quality is impossible. Thus, it is necessary to develop an automated solution capable of effectively identifying corrupted images in CDWs. This study presents an innovative transfer learning method for automated quality con- trol of 3D gradient echo T1-weighted brain MRIs within a CDW, leveraging artefact sim- ulation. We first intentionally corrupt images from research datasets by inducing poorer contrast, adding noise and introducing motion artefacts. Subsequently, three artefact- specific models are pre-trained using these corrupted images to detect distinct types of artefacts. Finally, the models are generalised to routine clinical data through a transfer learning technique, utilising 3660 manually annotated images. The overall image quality is inferred from the results of the three models, each designed to detect a specific type of artefact. Our method was validated on an independent test set of 385 3D gradient echo T1-weighted MRIs. Our proposed approach achieved excellent results for the detection of bad quality MRIs, with a balanced accuracy of over 87%, surpassing our previous approach by 3.5 percent points. Additionally, we achieved a satisfactory balanced accuracy of 79% for the detection of moderate quality MRIs, outperforming our previous performance by 5 percent points. Our framework provides a valuable tool for exploiting the potential of MRIs in CDWs.

Impact of Institutional Practices and Surgical Complexity on Sarcoma Surgery Costs: Driving Efficiency in Value-Based Healthcare.

Sarcomas present a unique challenge within healthcare systems due to their rarity and complex treatment requirements. This study explores the economic impact of sarcoma surgeries across three Swiss tertiary healthcare institutions, utilizing a consistent surgical approach by a single surgeon to eliminate variability in surgical expertise as a confounding factor. By analyzing data from 356 surgeries recorded in a real-world-time data warehouse, this study assesses surgical and hospital costs relative to institutional characteristics and surgical complexity. Our findings reveal significant cost variations driven more by institutional resource management and pricing strategies than by surgical techniques. Surgical and total hospitalization costs were analyzed in relation to tumor dignity and complexity scores, showing that higher complexity and malignancy significantly increase costs. Interestingly, it was found that surgical costs accounted for only one-third of the total hospitalization costs, highlighting the substantial impact of non-surgical factors on the overall cost of care. The study underscores the need for standardized cost assessment practices and highlights the potential of predictive models in enhancing resource allocation and surgical planning. By advocating for value-based healthcare models and standardized treatment guidelines, this research contributes to more equitable and sustainable healthcare delivery for sarcoma patients. These insights affirm the necessity of including a full spectrum of care costs in value-based models to truly optimize healthcare delivery. These insights prompt a reevaluation of current policies and encourage further research across diverse geographical settings to refine cost management strategies in sarcoma treatment.

Is the Risk of Alzheimer's Disease and Related Dementias Among U.S. Veterans Influenced by the Intersectionality of Housing Status, HIV/AIDS, Hepatitis C, and Psychiatric Disorders?

Homelessness and housing instability disproportionately affect U.S. veterans with psychiatric disorders, human immunodeficiency virus/acquired immunodeficiency syndrome (HIV/AIDS), hepatitis C, and Alzheimer's disease and related disorders (ADRD). We examined housing status and/or HIV/AIDS in relation to ADRD risk and evaluated hepatitis C, substance use, and mental health disorders as mediators and/or moderators of hypothesized relationships, among U.S. veterans ≥50 years of age seeking Department of Veterans Affairs (VA) healthcare services. A retrospective cohort study was conducted using linked VA Homeless Operations Management and Evaluation System and Corporate Data Warehouse databases (2017-2023) on 3 275 098 eligible veterans yielding 133 388 ADRD cases over 5 years of follow-up. Multivariable regression and causal mediation analyses were performed, controlling for demographic and clinical characteristics. Taking stably housed veterans without HIV/AIDS as referent, ADRD risk was higher among veterans with homelessness/housing instability alone (adjusted hazard ratio [aHR] = 1.67, 95% confidence interval [CI]: 1.63,1.72), lower among veterans with HIV/AIDS alone (aHR = 0.65, 95% CI: 0.58,0.73), but similar to veterans with homelessness/housing instability and HIV/AIDS (aHR = 1.01, 95% CI: 0.79,1.29). In adjusted models, hepatitis C and psychiatric disorders were positively related to homelessness/housing instability and ADRD risk, but negatively related to HIV/AIDS. Statistically significant mediation and/or moderation of hepatitis C and psychiatric disorders were observed, although <10% of total effects were explained by these characteristics, controlling for confounders. Among older veterans, ADRD diagnoses over 5 years were less among those with HIV/AIDS, but more among those with homelessness/housing instability, and these relationships were partly explained by hepatitis C and psychiatric disorders.

Time to diagnosis among young patients with cancer.

Sociodemographic and clinical factors associated with diagnostic delays in pediatric, adolescent, and young adult cancers are poorly understood. Using the Optum Labs Data Warehouse's de-identified claims data for commercial health plan enrollees, we identified children (0-14years) and adolescents/young adults (AYAs) (15-39years) diagnosed with one of 10 common cancers from 2001 to 2017, who were continuously enrolled for 6months preceding diagnosis. Time to diagnosis was calculated as days between first medical encounter with possible cancer symptoms and cancer diagnosis date. Median times from first symptom to diagnosis were compared using Wilcoxon rank sum test. Multivariable unconditional logistic regression identified sociodemographic factors associated with longer time (>3months) to cancer diagnosis (from symptom onset). Of 47,296 patients, 87% presented prior to diagnosis with symptoms. Patients with central nervous system (CNS) tumors were most likely to present with symptoms (93%), whereas patients with cervical cancer were least likely (70%). Symptoms varied by malignancy. Of patients with symptoms, thyroid (105days [range: 50-154]) and cervical (104days [range: 41-151]) cancer had the longest median time to diagnosis. Females and patients at either end of the age spectrum were more likely to experience diagnosis delays of more than 3months. In a commercially insured population, time to diagnosis varies by cancer type, age, and sex. Further work is needed to understand the patient, provider, and health system-level factors contributing to time from symptom onset to diagnosis, specifically in the very young children and the young adult patient population going forward.

Racial and Ethnic and Rural Variations in Access to Primary Care for Veterans Following the MISSION Act

The 2018 Veterans Affairs Maintaining Internal Systems and Strengthening Integrated Outside Networks (VA MISSION) Act was implemented to increase timely access to care by expanding veterans' opportunities to receive Veterans Affairs (VA)-purchased care in the community (community care [CC]). Because health equity is a major VA priority, it is important to know whether Black and Hispanic veterans compared with White veterans experienced equitable access to primary care within the VA MISSION Act. To examine whether utilization of and wait times for primary care differed between Black and Hispanic veterans compared with White veterans in rural and urban areas after the implementation of the VA MISSION Act. This cross-sectional study used VA and CC outpatient and consult data from the VA's Corporate Data Warehouse for fiscal years 2021 to 2022 (October 1, 2020, to September 30, 2022). Separate fixed-effects multivariable models were run to predict CC utilization and wait times. Each model was run twice, once comparing Black and White veterans and then comparing Hispanic and White veterans. Adjusted risk ratios (ARRs) were calculated for Black and Hispanic veterans compared with White veterans within rurality status for both outcomes. VA and CC primary care utilization as measured by primary care visits (utilization cohort); VA and CC primary care access as measured by mean wait times (access cohort). A total of 5 046 087 veterans (994 517 [19.7%] Black, 390 870 [7.7%] Hispanic, and 3 660 700 [72.6%] White individuals) used primary care from fiscal years 2021 to 2022. Utilization increased for all 3 racial and ethnicity groups, more so in CC than VA primary care. ARRs were significantly less than 1 regardless of rurality status, indicating Black and Hispanic veterans compared with White veterans were less likely to utilize CC for primary care. There were 468 246 primary care consultations during the study period. The overall mean (SD) wait time was 33.3 (32.4) days. Despite decreases in wait times over time, primary care wait times remained longer in CC than in VA. Black veterans compared with White veterans had significantly longer wait times in CC (ARRs >1) but significantly shorter wait times in VA (ARRS <1) regardless of rurality status in VA and CC. CC wait times for Hispanic veterans compared with White veterans were longer in rural areas only and in VA rural and urban areas (ARRs >1). The results of this cross-sectional study suggest that additional research should explore the determinants and implications of utilization differences among Black and Hispanic veterans compared with White veterans. Efforts to promote equitable primary care access for all veterans are needed so that policy changes can be more effective in ensuring timely access to care for all veterans.

Femoral anteversion linked to the inability to squat: Analysis of CT images in the patient and control groups.

Many patients who cannot squat well in a neutral toe position can only squat in an excessively out-toeing position. This excessive out-toeing squat is thought to be caused by rotational problems of the lower extremities. In this study, we aimed to identify the cause for the inability to squat by measuring and comparing femoral and tibial torsion between an excessive out-toeing squat patient group and a control group representing the general population. Between 2008 and 2022, a patient group comprising 50 lower extremities with excessive out-toeing squats was established. A control group representing the general population was selected from patients aged 0 to 29 years, who underwent lower-extremity CT angiography between 2012 and 2022, using the Clinical Data Warehouse with exclusion criteria applied. A total of 94 lower extremities were included in the control group. The femoral torsional angle (FTA) and tibial torsional angle (TTA) of both groups were measured and compared using Student t test. Additionally, 30 each of those with the highest and lowest 30 FTA values were selected from the patient and control groups, and the TTA was compared between the high- and low-FTA groups using Student t test. The mean FTA was 0.34° (SD, 11.11°) in the patient group and 10.14° (SD, 11.85°) in the control group, with a mean difference of 9.8° and P < .001. The mean TTA was 27.95° (SD, 7.82°) in the patient group and 32.67 ° (SD, 7.58°) in the control group, with a mean difference of 4.72° (P = .001). The mean TTA was 34.3° (SD, 7.72°) in the high-FTA group and 28.17° (SD, 8.35°) in the low-FTA group, with a mean difference of 6.13° (P = .005). Patients with excessive out-toeing squat showed lower FTA and TTA values than the general population. Furthermore, although a correlation between FTA and TTA was not established through Pearson correlation analysis, a tendency was observed where a decrease in FTA was associated with a decrease in TTA. Based on these results, decreased FTA was demonstrated to be one of the major causes of excessive out-toeing squats.

Spinal infection: Assessing comorbidities and costs to inform patient management and resource use strategies

BackgroundSpinal Infection (SI) is associated with various comorbidities. The interaction of these comorbidities and their impact on costs and complexity of care has not been fully assessed. MethodsThis is a retrospective cohort study of SI patients in an urban hospital system to characterize comorbidities and outcomes in adult patients with SI. Adult patients in our hospital system who were hospitalized with an initial diagnosis of SI between July 1, 2017 and June 30, 2019 were included. Outcomes measures included length of stay (LOS) of the index hospitalization for SI, charges and payments for the index hospitalization, and hospital readmissions within one year after discharge from the index hospitalization Data was obtained by querying our Electronic Data Warehouse (EDW) using ICD-10-CM and CPT procedure codes. Spearman's correlation was used to summarize the relationships between LOS, charges, and payments. Multivariable linear regression were used to evaluate associations of demographics, comorbidities, and other factors with LOS. Multivariable Cox regression were used to evaluate associations of demographics, comorbidities, and other factors with hospital readmissions. Results403 patients with a first diagnosis of SI were identified. The average number of comorbidities per patient was 1.3, 294 (73%) had at least one medical comorbidities, and 54 (13%) had three or more comorbidities. The most common medical comorbidities were diabetes mellitus (26%), intravenous drug use (IVDU, 26%), and malnutrition (20%). Twenty-eight percentage had a surgical site infection (SSI). DM (p < 0.001) and SSI (p = 0.016) were more common among older patients while IVDU was more common among younger patients (p < 0.001). Median LOS was 12 days. A larger number of medical comorbidities was associated with a longer LOS (p < 0.001) while the presence of a SSI infection was associated with a shorter LOS (p = 0.007) after multivariable adjustment. LOS was positive correlated with both charges (r = 0.83) and payments (r = 0.61). Among 389 patients discharged alive after the index hospitalization, 36% had a readmission within 1 year. The rate of readmission was twice as high for patients with three or more comorbidities than patients with zero comorbidities (hazard ratio: 1.95, p = 0.017). ConclusionsPatients with SI often have multiple comorbidities, and the specific type of comorbidity is associated with the patient's age. The presence of multiple comorbidities correlates with initial LOS, cost of care, and readmission rate. Readmission in the first year post-discharge is high.

Promoting equity in clinical research: The role of social determinants of health

ObjectiveThis study aims to investigate the association between social determinants of health (SDoH) and clinical research recruitment outcomes and recommends evidence-based strategies to enhance equity. Materials and MethodsData were collected from the internal clinical study manager database, clinical data warehouse, and clinical research registry. Study characteristics (e.g., study phase) and sociodemographic information were extracted. Median neighborhood income, distance from the study location, and Area Deprivation Index (ADI) were calculated. Mixed effect generalized regression was used for clustering effects and false discovery rate adjustment for multiple testing. A stratified analysis was performed to examine the impact in distinct medical departments. ResultsThe study sample consisted of 3,962 individuals, with a mean age of 61.5 years, 53.6 % male, 54.2 % White, and 49.1 % non-Hispanic or Latino. Study characteristics revealed a variety of protocols across different departments, with cardiology having the highest percentage of participants (46.4 %). Industry funding was the most common (74.5 %), and digital advertising and personal outreach were the main recruitment methods (58.9 % and 90.8 %). DiscussionThe analysis demonstrated significant associations between participant characteristics and research participation, including biological sex, age, ethnicity, and language. The stratified analysis revealed other significant associations for recruitment strategies. SDoH is crucial to clinical research recruitment, and this study presents evidence-based solutions for equity and inclusivity. Researchers can tailor recruitment strategies to overcome barriers and increase participant diversity by identifying participant characteristics and research involvement status. ConclusionThe findings highlight the relevance of clinical research inequities and equitable representation of historically underrepresented populations. We need to improve recruitment strategies to promote diversity and inclusivity in research.

Comorbidities in Hispanic/Latino Veterans with epilepsy

BackgroundHispanic/Latino people with epilepsy may be at a differential risk of medical and psychiatric comorbidities given genetic, environmental, sociocultural, and quality of care factors. In people with epilepsy, comorbidities are especially crucial to investigate given the well-known impact on quality of life and risk of adverse outcomes. Yet, Hispanic/Latino Veterans with Epilepsy (HL-VWE) remain an understudied population. The present nationwide population study sought to investigate medical and psychiatric comorbidities in this group. MethodsData from the Veterans Health Administration (VHA) Corporate Data Warehouse administrative data were used to identify 56,556 VWE (5.7 % HL-VWE) using a one-year cross-sectional analysis of ICD codes. Elixhauser Comorbidity Index scores and psychiatric diagnoses were calculated based on ICD-9/ICD-10-CM diagnoses using a lookback period. Comparisons were made between HL-VWE and non-HL-VWE using chi-squared and student t-tests. Regression analyses were then performed to examine group differences while accounting for age. ResultsHL-VWE had higher probability of being diagnosed with several psychiatric conditions when accounting for age, including depression (OR 1.21, 95 % CI 1.13–1.31) and schizophrenia (OR 1.56, 95 % CI 1.31–1.84). There were no significant differences in medical comorbidities between the HL-VWE and non-HL-VWE groups. ConclusionsWe present results from the largest known study of HL people with epilepsy examining their psychiatric and medical comorbidities and one of the first to specifically study HL-VWE. Compared to non-HL-VWE, the Hispanic/Latino group had comparable medical comorbidity, but higher rates of multiple psychiatric conditions. Results indicate a need for increased screening and interventions in this population to reduce psychiatric disease burden.

Trends in Veteran hospitalizations and associated readmissions and emergency department visits during the MISSION Act era.

To examine changes in hospitalization trends and healthcare utilization among Veterans following Maintaining Internal Systems and Strengthening Integrated Outside Networks (MISSION) Act implementation. VA Corporate Data Warehouse and Centers for Medicare and Medicaid Services datasets. Retrospective cohort study to compare 7- and 30-day rates for unplanned readmission and emergency department visits following index hospital stays based on payor type (VHA facility stay, VA-funded stay in community facility [CC], or Medicare-funded community stay [CMS]). Segmented regression models were used to compare payors and estimate changes in outcome levels and slopes following MISSION Act implementation. Veterans with active VA primary care utilization and ≥1 acute hospitalization between January 1, 2016 and December 31, 2021. Monthly index stays increased for all payors until MISSION Act implementation, when VHA and CMS admissions declined while CC admissions accelerated and overtook VHA admissions. In December 2021, CC admissions accounted for 54% of index admissions, up from 25% in January 2016. From adjusted models, just prior to implementation (May 2019), Veterans with CC admissions had 47% greater risk of 7-day readmission (risk ratio [RR]: 1.47, 95% confidence interval [CI]: 1.43, 1.51) and 20% greater risk of 30-day readmission (RR: 1.20, 95% CI: 1.19, 1.22) compared with those with VHA admissions; both effects persisted post-implementation. Pre-implementation CC admissions were also associated with higher 7- and 30-day ED visits, but both risks were substantially lower by study termination (RR: 0.90, 95% CI: 0.88, 0.91) and (RR: 0.89, 95% CI: 0.87, 0.90), respectively. MISSION Act implementation was associated with substantial shifts in treatment site and federal payor for Veteran hospitalizations. Post-implementation readmission risk was estimated to be higher for those with CC and CMS index admissions, while post-implementation risk of ED utilization following CC admissions was estimated to be lower compared with VHA index admissions. Reasons for this divergence require further investigation.

IMPROVING THE PROCESSES OF ENSURING THE AIRWORTHINESS OF AIRCRAFT BY CREATING A HIGH-SPEED HETEROGENEOUS INFORMATION SYSTEM

The problem of forming the appearance of a high-speed heterogeneous information system for monitoring airworthiness during the operation of aviation equipment is considered. A mixed-integer model has been formed for forming the optimal data transmission path for verifying the authenticity of airway components, where integer variables determine the set of external systems from where and where data is transmitted, a set of data warehouses and connections between them, and continuous variables – characteristics of data transfer rates by web services and bandwidth of data transmission channels. The criterion for choosing the optimal solution is proposed – the time of data transfer between external data source systems and external data consumer systems, taking into account data warehouses. An algorithm for optimizing the structure and parameters of a heterogeneous information system has been developed, which ensures the organization of information interaction between heterogeneous data sources/consumers in a minimum time. It is shown that the problem under consideration relates to difficult combinatorial optimization problems, that is, there is no polynomial algorithm for solving it in a reasonable time. However, taking into account the specifics of the task, in some cases, allows you to build calculation schemes that provide sufficient accuracy of the solution and an acceptable time. After conducting a brief analysis of possible solutions, including Dijkstra, Johnson, Floyd – Warshall, the authors implemented computational procedures based on the Bellman – Ford method, the main advantages of which are the possibility of obtaining an accurate solution, no restrictions when working with negative edges of the network graph, low value of the algorithm execution time. The results of calculations are presented and the possibility of reducing the data transfer time by 20,2 % is shown by choosing the optimal connections between data sources and consumers, taking into account the data formats of the sources used and the compatibility of web services providing their processing.

Differences in baseline characteristics and healthcare costs among insured patients with advanced cancer with vs without clinical trial participation.

11154 Background: Despite concerted efforts to improve clinical trial participation (CTP), enrollment remains low. Perceived additional costs to the patient or health plan may be barriers to CTP; low biomarker testing rates may be another. This study assessed differences in characteristics of patients with CTP versus patients without CTP in an insured population with advanced cancer. Methods: A retrospective analysis was conducted using de-identified administrative claims data from commercially insured and Medicare Advantage (MA) enrollees in the Optum Labs Data Warehouse. Patients ≥18y with claims evidence of advanced cancer and systemic therapy between 01/01/2018 and 02/28/2022 were stratified into 2 groups: 1) with CTP: ≥1 claim with a CTP diagnosis (ICD10 Z00.6) and ≥1 claim with a Q modifier on or within 90 days after index date; and 2) without CTP: no claims with a CTP diagnosis or Q modifier. Patients with &gt;1 primary cancer, T-cell therapy, or &lt;360 days baseline or follow-up (unless death) enrollment were excluded. Biomarker testing, targeted therapy use, and healthcare costs per patient per month (PPPM) were assessed in the baseline period. Results: Of 61,490 patients identified, 1753 (3%) had CTP and 59,737 (97%) did not have CTP; 4% of commercial and 3% of MA patients had CTP. Most common tumor types were breast (27%; 2% had CTP), lung (16%; 3% had CTP), digestive tract (15%; 2% had CTP), and prostate (15%; 3% had CTP). Patients with CTP were younger and more frequently had documented biomarker test use in the baseline period than patients without CTP (Table). In unadjusted analyses, the CTP group had higher healthcare costs in the baseline period prior to CTP regardless of insurance type, driven by 51% higher systemic cancer therapy costs and 35% higher ambulatory visit costs. An ongoing propensity score matched analysis will evaluate the impact of CTP on cost of care in the follow-up period. Conclusions: Consistent with prior studies, the overall CTP rate among patients with advanced cancers was low (3%). Baseline costs (prior to CTP) were higher for patients that later enrolled in clinical trials than for patients without CTP. Further research is ongoing to assess differences in follow-up healthcare costs for patients with vs without CTP adjusting for baseline characteristics. [Table: see text]

Lung cancer screening among women Veterans within the Veterans’ Health Administration.

10537 Background: Despite advancements in survival rates, lung cancer remains the leading cause of cancer-related deaths among women in the United States (US). In response to this challenge, the United States Preventive Services Task Force (USPSTF) issued updated guidelines in 2021 to broaden the pool of eligible candidates for lung cancer screening (LCS). Pioneering initiatives, such as the Lung Cancer Screening Demonstration Project (LCSDP) launched by the Veterans Affairs (VA) in 2012, have played a pivotal role in evaluating the feasibility of implementing Low-Dose Computed Tomography (LDCT) at eight Veterans Affairs Medical Centers (VAMC). Subsequently, the program has undergone substantial expansion. Projections within the veteran community indicate a notable increase in the proportion of women veterans, from 11% in 2023 to an anticipated 18% in 2048. However, scant information is currently available regarding the lung cancer screening patterns within this expanding demographic. Methods: Women veterans ages 55 -80, current or former smokers from 2015- 2021 were identified through the VA corporate data warehouse (CDW). ICD 9/10 codes were used to identify women veterans with a lung cancer diagnosis. Diagnoses were verified via chart review. Data were analyzed using R Studio software. Results: Of the women veterans identified (n=44,342), 10,623 received LCS (24%) and 942 (2%) had a primary diagnosis of lung cancer. In the cohort of 942 women diagnosed, 286 (30%) had received lung cancer. Women veterans diagnosed with lung cancer tended to be aged &gt;65 (55%), white (72%), single (72%), unemployed (71%), and urban dwelling (70%) within the southeastern US (35%). Compared to women veterans who did not receive LCS, those who did had higher rates of localized disease (47% vs 34%) and lower rates of metastatic disease (12% vs 19%). Factors associated with higher rates of LCS included mammography (60% vs 49%), a family history of lung cancer (4% vs 2%) and living in the southeastern US (46% vs 35%). It is to be noted that 23% of the women veterans who were screened lived in rural areas and 18.53% identified as black. 98.9% of the women had reported current or former tobacco use but the data on pack year history was missing in 87.6% of charts. Conclusions: While further progress can be made, it is notable that 24% of women veterans received LCS, surpassing national averages. This discrepancy underscores the potential to advance efforts in establishing a comprehensive and structured lung cancer screening initiative, thereby facilitating the equitable inclusion of historically marginalized groups, particularly minorities and rural populations. The findings emphasize the imperative to overcome existing impediments to nationwide enrollment and retention in LCS. Lastly, the study underscores the necessity for increased resource allocation towards tobacco cessation initiatives as an integral component of the broader expansion strategy.

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]

A Machine Learning Algorithm Avoids Unnecessary Paracentesis for Exclusion of SBP in Cirrhosis in Resource-limited Settings

Despite the poor prognosis associated with missed or delayed spontaneous bacterial peritonitis (SBP) diagnosis, <15% get timely paracentesis, which persists despite guidelines/education in the United States. Measures to exclude SBP non-invasively where timely paracentesis cannot be performed could streamline this burden. Using Veterans Health Administration Corporate Data Warehouse (VHA-CDW) we included patients with cirrhosis between 2009 and 2019 who underwent timely paracentesis and collected relevant clinical information (demographics, cirrhosis severity, medications, vitals, and comorbidities). XGBoost-models were trained on 75% of the primary cohort, with 25% reserved for testing. The final model was further validated in 2 cohorts: Validation cohort #1: In VHA-CDW, those without prior SBP who received 2nd early paracentesis, and Validation cohort #2: Prospective data from 276 non-electively admitted University hospital patients. Negative predictive values (NPVs) at 5%,10%, and 15% probability cutoffs were examined. Primary cohort: n= 9643 (mean age, 63.1 ± 8.7 years; 97.2% men; SBP, 15.0%) received first early paracentesis. Testing-set NPVs for SBP were 96.5%, 93.0%, and 91.6% at the 5%, 10%, and 15% probability thresholds, respectively. In Validation cohort #1: n= 2844 (mean age, 63.14 ± 8.37 years; 97.1% male; SBP, 9.7%) with NPVs were 98.8%, 95.3%, and 94.5%. In Validation cohort #2: n= 276 (mean age, 56.08 ± 9.09; 59.6% male; SBP, 7.6%) with NPVs were 100%, 98.9%, and 98.0% The final machine learning model showed the greatest net benefit on decision-curve analyses. A machine learning model generated using routinely collected variables excluded SBP with high NPV. Applying this model could ease the need to provide paracentesis in resource-limited settings by excluding those unlikely to have SBP.

A unique model: Partnership between community oncology and pharma for clinical trial enrichment.

1535 Background: Clinical trials are the foundation of oncologic drug development. Historically, only 23% of oncology trials are conducted in the community despite &gt;70% of patients being treated this setting, leading to a disconnect between the therapy being tested and target population. Also, patients are hesitant to enroll in trials- in 2020, only 49% were willing, down from 85% in 2019. One way to improve this is participation by community oncologists, allowing patients to stay close to home. Targeted therapies are increasingly being approved by the FDA, with 89 small molecule drugs approved by 2020. Given the above landscape, Precision Health Informatics (PHI) collaborated with Mirati Therapeutics to pilot a clinical trial enrichment program focused on increasing enrollment of patients in the community related to efficacy and outcomes of adagrasib (KRAS G12C inhibitor) for NSCLC and CRC. Methods: This pilot was launched in 2021, with a prescreening study (DetECT) to find eligible patients with KRASG12C mutations. The pilot was then expanded across Mirati ‘Krystal’ studies in 2022 and to “1719” studies in 2023. These studies were available to all investigators throughout Texas Oncology (TxO) via STAR model (clinical trial access to all TxO investigators/sites through a regulated but rapid approval process). Eligible patients were identified from PHI’s proprietary molecular data warehouse (that stores molecular results for &gt;400,000 patients treated at TxO). A data analyst identified patients with KRAS G12C mutation from the warehouse, and then cross-referenced patient-centric data in EMR to assess eligibility. The treating physician was then alerted of their patient’s eligibility. Patient demographic information was also collected from the EMR. Results: This collaboration resulted in accrual of 118 patients, engagement of 40 different PIs, across 8 different trials. 6 studies remain open across the practice. 21 patients were enrolled on Krystal-001, 7 on Krystal-007, 10 on Krystal-010 and 56 on DetECT. Enrollment was inclusive of patients from urban areas of DFW as well as rural populations in NE Texas, South and West Texas. In DetECT, TxO provided 84% of the total US enrollment, with a diverse population – 54% white, 23% Hispanic, 12% black, 9% Asian. Texas Oncology is the 2nd highest screening and enrolling site on Krystal studies. Conclusions: This pilot successfully demonstrates that community oncology can lead the way in clinical trial enrollment, even in early phase targeted therapy studies. Community physicians can be engaged so patients can benefit from research participation close to home. This model shows that a diverse population was accrued and may serve as a blueprint to help Pharma satisfy the diversity requirements for clinical trials. PHI, with its molecular database, may partner with Pharma to identify targetable mutations and drive rapid trial enrollment for targeted therapies.

Association of polygenic-based breast cancer risk prediction with patient management.

10527 Background: Guidelines recommend individuals with ≥20% lifetime risk of breast cancer (BC) undergo enhanced management, including annual screening mammography (SM) as early as age 30, annual breast MRI, and genetic counseling (GC). Lifetime BC risk can be estimated using a validated risk model, such as Tyrer-Cuzick (TC). A risk predictor that combines TC with a polygenic risk score (“combined risk score” or CRS) significantly improves risk prediction over TC alone. Little is known about how clinicians manage patients based on CRS. Here, we describe management following receipt of CRS results. Methods: De-identified administrative claims data from the Optum Labs Data Warehouse were linked with de-identified TC and CRS results originally provided to ordering clinicians. Patients were divided into four subgroups based on their lifetime risk predicted by CRS and by TC alone (high risk: ≥20%, average risk: &lt;20%): A (CRS-high/TC-high), B (CRS-high/TC-avg), C (CRS-avg/TC-high), and D (CRS-avg/TC-avg). Patient management claims were compared 360 days before and after the CRS test date and differences evaluated with McNemar’s test. Differences in post-test management were analyzed using multivariable logistic regression with adjustment for clinical factors. Results: Before receiving risk results, 11.9%, 7.7%, 8.8%, and 5.0% of patients in A, B, C, and D received a SM before age 40, respectively; 3.4%, 2.6%, 2.3%, and 0.9% received an MRI, respectively; and 4.3%, 4.7%, 6.0%, and 3.6% received GC, respectively. After receiving risk results, SM in those under age 40 significantly increased 1.6-2.2-fold in A, B, and C; MRI significantly increased 4.7-5.6-fold in A, B, and C; and GC significantly increased 1.9-2.3-fold in A and B. SM, MRI, and GC did not increase in D. Prophylactic mastectomy occurred in only 0.17% of all patients after receiving results. After adjustment for confounders, those in A, B, and C were 3.8-5.2 times more likely to undergo SM; 11.6-23.1 times more likely to undergo MRI; and 2.0-2.9 times more likely to undergo GC, compared to those in Group D (all p&lt;0.001). Conclusions: Patients with a ≥20% lifetime risk for BC were more likely to undergo enhanced management compared to those with a lifetime risk &lt;20%, regardless of whether their risk was based on the CRS or on TC. These results suggest that clinicians recommended management aligned with guidelines for those with ≥20% lifetime risk, even when such risk was predicted by the CRS. [Table: see text]

Effect of age on delivering chemotherapy to older adults with advanced pancreatic adenocarcinoma.

e13767 Background: Age is one of the most important risk factors for developing pancreatic adenocarcinoma (PDAC), 65% of the newly diagnosed cancer cases occur in individuals above 65 years of age (yo). The only potential cure for PDAC includes surgical resection, however, the majority of patients (70%) present with locally advanced/unresectable or metastatic PDAC (APDAC). Many oncologists remain hesitant to administer treatment to older adults for fear of perceived increased chemotoxicity risk and the thought of treatment intolerability. Methods: This retrospective cohort study included patients who were diagnosed and treated for APDAC within the Veterans Integrated Services Network (VISN) 8 which includes Veterans Affairs medical facilities in Florida, southern Georgia, Puerto Rico, and the U.S. Virgin Islands. Demographic data such as age, sex, race, ethnicity and treatment data such as chemotherapy and palliative care use were pulled from the VA Cancer Registry, available from Corporate Data Warehouse Data. We conducted a detailed chart review in the VA electronic health record to confirm stage and the eastern cooperative oncology group (ECOG) status at the time of diagnosis. Statistical analyses were conducted using GraphPad Prism 10.1.2 software. Analyses consisted of t-test for normally distributed continuous variables, chi-squared test for categorical variables and the Mann Whitney U test for the ECOG score of both groups. Results: Ninety-nine patients with APDAC above the of age 65 were identified from the period of June 1, 2017 till July 31, 2021. 46 received chemotherapy (C group) and 53 patients did not receive chemotherapy (NC group), characteristics (Table). The mean age for those in the C group were 72 years old (yo) vs 76 yo from the NC group with a P value of 0.0017 which was statistically significant. Looking at the age subgroups, the majority of the patients that received C, were of the 65-74 age subgroup (71%). The median ECOG for the NC group was 2 and that for C was 1 group. 39% of the C group received palliative care as opposed to 32% of the NC group with a P value of 0.532. Conclusions: Despite an acceptable ECOG range between 1 and 2, Age played an important factor in determining who received chemotherapy in APDAC. Studies other than ECOG are needed to help oncologists determine the fitness of APDAC patients and minimize hesitancy to deliver treatment. The involvement of palliative care in patients with APDAC was suboptimal and a push for more involvement of palliative care is needed. [Table: see text]

Longitudinal Trajectories of Antiretroviral Treatment Adherence and Associations With Durable Viral Suppression Among Adolescents Living With HIV in South Africa.

Compared with other age groups, adolescents living with HIV (ALHIV) are estimated to have lower levels of adherence to antiretroviral treatment. Despite this, we lack evidence on adolescents' adherence patterns over time to inform the customization of intervention strategies. Eastern Cape province, South Africa. We analyzed data from a cohort of ALHIV (N = 1046, aged 10-19 years at baseline) recruited from 53 public health facilities. The cohort comprised 3 waves of data collected between 2014 and 2018 and routine viral load data from the National Institute for Communicable Disease data warehouse (2014-2019). Durable viral suppression was defined as having suppressed viral load (<1000 copies/mL) at ≥2 consecutive study waves. Group-based multitrajectory model was used to identify adherence trajectories using 5 indicators of self-reported adherence. Logistic regression modeling evaluated the associations between adherence trajectories and durable viral suppression. Overall, 933 ALHIV (89.2%) completed all 3 study waves (55.1% female, mean age: 13.6 years at baseline). Four adherence trajectories were identified, namely, "consistent adherence" (49.8%), "low start and increasing" (20.8%), "gradually decreasing" (23.5%), and "low and decreasing" (5.9%). Adolescents experiencing inconsistent adherence trajectories were more likely to be older, live in rural areas, and have sexually acquired HIV. Compared with the consistent adherence trajectory, the odds of durable viral suppression were lower among adolescents in the low start and increasing (adjusted odds ratio [aOR]: 0.62, 95% CI: 0.41 to 0.95), gradually decreasing (aOR: 0.40, 95% CI: 0.27 to 0.59), and the low and decreasing adherence (aOR: 0.25, 95% CI: 0.10 to 0.62) trajectories. Adherence to antiretroviral treatment remains a challenge among ALHIV in South Africa. Identifying adolescents at risk of nonadherence, based on their adherence trajectories may inform the tailoring of adolescent-friendly support strategies.