Real-world Datasets Research Articles

Real world experience of trastuzumab deruxtecan for the treatment of metastatic breast cancer in the UK.

1024 Background: Trastuzumab deruxtecan (TDXd), an antibody drug conjugate, was first approved in the UK for the treatment of human epidermal growth factor receptor 2 positive (HER2+) metastatic breast cancer (mBC) in 2021. We aimed to review the efficacy and tolerability of TDXd in a real-world UK setting. Methods: We collected clinical and demographic data from medical records on HER2+ mBC patients treated with TDXd in the UK from Jan 2021 to Dec 2023 inclusive. This included: radiological response, the frequency of dose reductions and treatment delays, and reasons for discontinuation. Clinical trial patients were excluded. Descriptive statistical techniques were applied. Results: Data was collected from 26 UK cancer centres on 280 patients. The median age at TDXd initiation was 56 years (range 29-85). Median treatment duration was 9 cycles (range 1-40). 239 patients (85%) had visceral disease. 257 and 196 patients (92% and 70%, respectively) had received prior Trastuzumab and Pertuzumab, respectively. 233 patients (83%) had received prior TDM1. The median number of lines of treatment prior to TDXd was 3 (range 1 to 9). 20 patients (7%) were initiated on a reduced dose and 115 further patients (41%) required at least one dose reduction. 151 patients (54%) had at least one dose interruption due to: infection or neutropenia (n=61), other toxicity (n=85), interstitial lung disease (ILD) investigation (n=39), patient request (n=81), other concurrent illness (n=10), and cardiac investigation (n=6). Of these, 55 patients had more than one treatment break. 157 patients (56%) had discontinued TDXd treatment due to: disease progression or death (n=100), drug toxicity (n=39) of which 22 were due to ILD, and patient choice (n=8). 123 patients remained treatment at the time of analysis. The overall response rate of TDXd was 63% overall: stable disease 23%, partial response 56%, complete response 7%. Conclusions: The real-world data set is not directly comparable to either the DB02 or DB03 trials, as it included patients who had received TDXd in both the second line and later line settings. However, the majority (80%) of these patients were treated in the third line and beyond. The rates of drug interruption and reduction were higher in the real-world data set than in DB02, which may impact on efficacy in the real-world setting. Overall response rate is similar between trial data and our real-world data, although the latter was not measured by RECIST. Data collection is ongoing, including an analysis of real-world progression free survival. [Table: see text]

Real-world and clinical trial validation of a deep learning radiomic biomarker for PD-(L)1 immune checkpoint inhibitor response in stage IV NSCLC.

102 Background: Immune checkpoint inhibitor (ICI) therapy is standard-of-care for treatment of mutation-negative advanced non-small cell lung cancer (NSCLC). However, given known inaccuracy of PD-(L)1 markers, there is an unmet need to better identify patients most likely to derive clinical benefit from ICI. We developed and externally validated a generalizable CT imaging-based biomarker to predict response to ICI. Methods: We developed and validated a deep learning radiomic biomarker using an internally curated real-world dataset (RWD) of 2,010 stage IV NSCLC patients treated with PD-(L)1 ICIs in academic and community settings from US and Europe. Patients with missing baseline imaging, missing follow-up data, or EGFR/ALK oncogenic driver mutations were excluded, resulting in a total of 1,188 subjects. This RWD consisted of a discovery cohort (Dataset A, N=844) and a temporally distinct holdout cohort (Dataset B, N=344), which were used to generate performance metrics of the biomarker. To test generalizability, we validated our biomarker in a prospective clinical trial dataset evaluating Sasanlimab in PD-(L)1 therapy-naïve, advanced NSCLC patients (NCT02573259, Dataset C, N=54). We utilized a two-stage learning approach to model 6-month PFS. First, we used an independent multi-task deep-learning feature extractor trained on 19,184 whole chest CT scans. Second, we input the extracted features into a Cox proportional hazard (CoxPH) model along with age, sex, and baseline lesion measurements (sum of longest diameter and distant metastases counts), and the model generated a time-dependent PFS function and a response score. We performed 6-fold cross-validation on Dataset A to train and evaluate the models, which were subsequently ensembled and applied to independent Datasets B and C. To assess independence from PD-L1 status and key demographic covariates, we herein report multivariate adjusted hazard ratios (HR) for the group identified as low-risk based on the biomarker. Results: In Dataset A, the biomarker showed a cross-validation PFS adjusted HR of 0.49 (95% CI 0.38-0.63) in the all-comers cohort and 0.28 (0.17-0.46) in the first-line ICI monotherapy cohort (1LMono). In Dataset B, the PFS adjusted HRs were 0.54 (0.35-0.83) in all-comers and 0.18 (0.05-0.61) in 1LMono. In Dataset C, the adjusted HRs were 0.30 (0.14-0.68) for PFS, 0.29 (0.10-0.83) for OS, 0.31 (0.14-0.72) for TTP. Conclusions: In our validations in RWD and clinical trial cohorts, a deep-learning radiomic biomarker based on routine pre-treatment CT scans predicted response to ICI and stratified patients independently from PD-L1 status. This tool may inform clinical decision-making, such as to help guide whether concomitant chemotherapy may not be needed. In future work, we plan to further validate our approach in larger prospective datasets and expand its use to new indications.

Census tract–level social determinants of health and variability in stage at diagnosis of bladder, breast, and non-small cell lung cancers.

11170 Background: Social determinants of health (SDOH) can identify underserved populations, inform the patient journey, and identify opportunities to improve outcomes. In cancer, SDOH may influence stage at diagnosis through exposure to risk factors for aggressive cancer and decreased access to care. However, privacy restrictions can limit research with SDOH in real world datasets, which often only know a patient’s broad location. Methods: Geocoding was undertaken on addresses of patients diagnosed between 1/1/17 and 9/30/22 with bladder (n=6078), breast (n=28,958), and non-small cell lung cancer (NSCLC, n=14,957) in the Syapse Learning Health Network of U.S. community health systems. Location was linked to five indicators of SDOH at the census tract level: the Social Vulnerability Index (SVI); percent (%) of tract spending >30% of income on housing; % of tract with broadband internet; county designation as a primary care shortage area; and rural urban commuting area (RUCA). Nested multivariable ordinal logistic regression models estimated independent associations between demographic, clinical, and SDOH factors with stage at initial cancer diagnosis. The statistical significance of including SDOH variables was assessed using a likelihood ratio test (LRT) comparing the models before and after including SDOH. Results: After successful geocoding and linkage to census tracts, there were marginal differences in the distribution of the five SDOH across stages at diagnosis within the three tumor types. In multivariable models adjusted for year of diagnosis, age, sex, race, ethnicity, smoking, and primary payor, among the tested SDOH measures, county designated as primary care shortage (proportional odds ratios (OR): 1.31, 95% confidence interval (95% CI): 1.05-1.64), and non-metropolitan area (0.69, 0.50-0.94) were statistically significantly associated with stage at bladder cancer diagnosis. SVI (1.52, 1.31-1.76) was also statistically significantly associated and directly proportional with stage at breast cancer diagnosis. Among patients with NSCLC, broadband internet access (1.01, 1.00-1.01), primary care shortage (1.15, 1.03-1.27) and non-metropolitan area (0.85, 0.74-0.99) were statistically associated with stage at diagnosis. In all three tumors, LRTs found that models with SDOH were statistically significantly better at predicting stage at diagnosis than those without (p <0.001). Conclusions: Using granular, census-tract geospatial resolution for geocoding of SDOH, we find that patients who live in greater vulnerability are diagnosed at later stages than those who live in areas of less vulnerability, suggesting that consideration of SDOH variables is important in assessing stage at diagnosis and identify patients with an unmet need for outreach and screening.

A novel particle swarm optimization-based intelligence link prediction algorithm in real world networks

Link prediction in social network is an important topic due to its applications like finding collaborations and recommending friends. Among existing link prediction methods, similarity-based approaches are found to be most effective since they examine the number of common neighbours (CN). Current work presents a novel link prediction algorithm based on particle swarm optimization (PSO) and implemented on four real world datasets namely, Zachary’s karate club (ZKC), bottlenose dolphin network (BDN), college football network (CFN) and Krebs’ books on American politics (KBAP). It consists of three experiments: i) to find the measures on existing methods and compare them with our proposed algorithm; ii) to find the measured values of the existing methods along with our proposed one to determine future links among nodes that have no CN; and iii) to find the measures of the methods to determine future links among nodes having same number of CN. In experiment 1, our proposed approach achieved 75.88%, 78.34%, 82.63% and 78.36% accuracy for ZKC, BDN, CFN, and KBAP respectively. These results beat the performances of traditional algorithms. In experiment 2, the accuracies are found as 75.53%, 74.25%, 81.63% and 78.34% respectively. In experiment 3, accuracies are detected as 72.75%, 81.53%, 78.35% and 75.13% respectively.

Heterogeneous Interaction Modeling With Reduced Accumulated Error for Multiagent Trajectory Prediction.

Dynamical complex systems composed of interactive heterogeneous agents are prevalent in the world, including urban traffic systems and social networks. Modeling the interactions among agents is the key to understanding and predicting the dynamics of the complex system, e.g., predicting the trajectories of traffic participants in the city. Compared with interaction modeling in homogeneous systems such as pedestrians in a crowded scene, heterogeneous interaction modeling is less explored. Worse still, the error accumulation problem becomes more severe since the interactions are more complex. To tackle the two problems, this article proposes heterogeneous interaction modeling with reduced accumulated error (HIMRAE) for multiagent trajectory prediction. Based on the historical trajectories, our method infers the dynamic interaction graphs among agents, featured by directed interacting relations and interacting effects. A heterogeneous attention mechanism (HAM) is defined on the interaction graphs for aggregating the influence from heterogeneous neighbors to the target agent. To alleviate the error accumulation problem, this article analyzes the error sources from the spatial and temporal perspectives, and proposes to introduce the graph entropy and the mixup training strategy for reducing the two types of errors, respectively. Our method is examined on three real-world datasets containing heterogeneous agents, and the experimental results validate the superiority of our method.

Symbolic regression as a feature engineering method for machine and deep learning regression tasks

In the realm of machine and deep learning (DL) regression tasks, the role of effective feature engineering (FE) is pivotal in enhancing model performance. Traditional approaches of FE often rely on domain expertise to manually design features for machine learning (ML) models. In the context of DL models, the FE is embedded in the neural network’s architecture, making it hard for interpretation. In this study, we propose to integrate symbolic regression (SR) as an FE process before a ML model to improve its performance. We show, through extensive experimentation on synthetic and 21 real-world datasets, that the incorporation of SR-derived features significantly enhances the predictive capabilities of both machine and DL regression models with 34%–86% root mean square error (RMSE) improvement in synthetic datasets and 4%–11.5% improvement in real-world datasets. In an additional realistic use case, we show the proposed method improves the ML performance in predicting superconducting critical temperatures based on Eliashberg theory by more than 20% in terms of RMSE. These results outline the potential of SR as an FE component in data-driven models, improving them in terms of performance and interpretability.

Autobalanced Multitask Node Embedding Framework for Intelligent Education.

Recently, online education has become popular. Many e-learning platforms have been launched with various intelligent services aimed at improving the learning efficiency and effectiveness of learners. Graphs are used to describe the pairwise relations between entities, and the node embedding technique is the foundation of many intelligent services, which have received increasing attention from researchers. However, the graph in the intelligent education scenario has three noteworthy properties, namely, heterogeneity, evolution, and lopsidedness, which makes it challenging to implement ecumenical node embedding methods on it. In this article, an autobalanced multitask node embedding model is proposed, named MNE, and applied to the interaction graph, settling a few actual tasks in intelligent education. More specifically, MNE builds two purpose-built self-supervised node embedding learning tasks for heterogeneous evolutive graphs. Edge-specific reconstruction tasks are built according to the semantic information and properties of the heterogeneous edges, and an evolutive weight regression task is designed, aiding the model to perceive the evolution of learners' implicit cognitive states. Then, both aleatoric and epistemic uncertainty quantification techniques are introduced, achieving both task- and node-level weight estimation and instructing subtask autobalancing. Experimental results on real-world datasets indicate that the proposed model outperforms the state-of-the-art graph embedding methods on two assessment tasks and demonstrates the validity of the proposed multitask framework and subtask balancing mechanism. Our implementations are available at https://github.com/ccnu-mathits/MNE4HEN.

Development and validation of MOSPROC, an easily applicable predictive score of mortality for patients with prostate cancer.

e13598 Background: Predicting life expectancy in prostate cancer patients is particularly difficult as different clinical factors significantly influence clinical outcome. Here we have developed MOSPROC, a simplified mortality predictive score to identify prostate cancer patients at high risk of mortality. Methods: The training set (TRS) (n=2,035) included metastatic Castration Resistant PC (mCRPC) patients from 4 randomized phase 3 trials (NCT00273338, NCT00988208, NCT00617669, and NCT00519285). The validation set included patients from NCT00554229 (mCRPC; n=256) (TS1) and NCT00626548 (M0 CRPC; n=661) (TS2) trials. Clinical trial datasets were obtained from www.projectdatasphere.org . All datasets contained more than 100 clinical variables per patient at baseline. Synthetic representations of every patient were generated and input into a deep learning framework to identify subgroup of patients based on their similarities. The resultant subgroups were correlated with overall survival (OS). Differential variables between subgroups were identified and included in a multivariable logistic regression model. Independent predictive factors found to be statistically significant were used to generate a predictive score of survival that was validated in the validation sets. Results: The deep learning framework identified two different subpopulations in the TRS: LL (n=824) and HH (n=1,204) with significantly different survival outcome. Patients in HH had a higher risk of death compared to LL (median OS 17.6 months vs. 27.3 months, respectively; HR 2.09 95%CI 1.84-2.39; p<.0001). Feature contribution analysis identified 36 differential features between both subpopulations including anthropometrics, performance status, vital signs, biochemistry, previous and current cancer therapies, and concomitant medications. From these, PSA, ALP, and AST were the best independent predictive factors that were then selected to create a simplified predictive score for survival. A score ≥34 points was considered a high score (HS) while if < 34, a low score (LS). When applied to TRS, the predictive score yielded an area under the curve of 0.91. Patients with HS presented a significantly shorter OS (median OS 18.4 vs. 23.9 months; HR 1.49 95%CI 1.31-1.71; p<.005). The predictive score was then validated in two additional trials with similar results. In both cases, patients with a HS presented a significantly shorter OS (TS1: HR 2.56, 95%CI 1.79-3.66, p<.01; TS2: HR: 3.46, 95%CI 1.89-6.33, <.001). Conclusions: By combining deep learning and classical biostatistical tools, we have developed MOSPROC, an easily applicable predictive score that, based on baseline values of PSA, ALP and AST, can identify prostate cancer patients with a higher risk of death. Further work is required to validate this approach in prospective cohorts of PC patients in additional clinical trials and/or real-world datasets.

Incomplete Data Meets Uncoupled Case: A Challenging Task of Multiview Clustering.

Incomplete multiview clustering (IMC) methods have achieved remarkable progress by exploring the complementary information and consensus representation of incomplete multiview data. However, to our best knowledge, none of the existing methods attempts to handle the uncoupled and incomplete data simultaneously, which affects their generalization ability in real-world scenarios. For uncoupled incomplete data, the unclear and partial cross-view correlation introduces the difficulty to explore the complementary information between views, which results in the unpromising clustering performance for the existing multiview clustering methods. Besides, the presence of hyperparameters limits their applications. To fill these gaps, a novel uncoupled IMC (UIMC) method is proposed in this article. Specifically, UIMC develops a joint framework for feature inferring and recoupling. The high-order correlations of all views are explored by performing a tensor singular value decomposition (t-SVD)-based tensor nuclear norm (TNN) on recoupled and inferred self-representation matrices. Moreover, all hyperparameters of the UIMC method are updated in an exploratory manner. Extensive experiments on six widely used real-world datasets have confirmed the superiority of the proposed method in handling the uncoupled incomplete multiview data compared with the state-of-the-art methods.

Face mask classification using convolutional neural networks with facial image regions and super resolution

<p>Face mask classification is relevant to public health and safety, so an approach for face mask classification using Multi-Task Cascaded Convolutional Networks (MTCNN) for face detection on image data, ResNet152 architecture for feature extraction, and super-resolution method, BSRGAN, for enhanced image quality was proposed. The classification model was trained by a fully connected layer of neural networks. The goal is to classify each facial image into three classes: the image with a mask, without a mask, or with an incorrectly worn mask. The performance of each classification model on two real-world datasets was evaluated by Accuracy, Precision, Recall, and F1 score for different sets of input patterns which were features extracted from the facial image regions including their combinations. Using multiple image regions, i.e. face, nose, and mouth, as resources for preparing input features showed the improved classification performance compared to using single image regions. In addition, the super-resolution technique applied to medium or large-sized images can improve the performance of the face mask classification model. Our findings may further guide the development for greater effective models and techniques on face mask classification contributing to practical scenarios.</p>

Impact of carboplatin and cisplatin shortages on treatment patterns in patients with metastatic solid tumors.

11149 Background: The United States experienced cisplatin (CP) and carboplatin (CB) shortages in 2023, leading to potential rationing or switching to alternative therapies in multiple cancers. Using a US nationwide, real-world oncology dataset, we assessed changes in the use of CP and CB in 7 cancers during the shortage period. Methods: All patients (pts) from the Flatiron Health electronic health record-derived de-identified database (~280 cancer clinics, ~800 sites of care) with a machine learning-extracted metastatic (met) diagnosis of a cancer of interest between March 2022 and June 2023 and evidence of first-line (1L) therapy, were selected. Pts with multiple primaries were excluded. Platinum of interest was identified based on frequency of pre-period 1L use for a given cancer. The pre-period was June 2022 through first month of reported shortage (February 2023 for CP; April 2023 for CB); data cutoff was June 2023. Clinical characteristics of platinum-treated pts stratified by pre- vs post-period by disease were assessed. 1L platinum utilization rates (UR), calculated as the proportion of 1L-treated pts initiating CB/CP within 30 days of met diagnosis, were plotted by month. Cohort-level impact was evaluated with an interrupted time series analysis—overall and stratified by practice type, practice size, and socioeconomic status (SES). Multiple testing was controlled using the Benjamini-Hochberg procedure. Results: 10,983 pts received 1L therapy in the study period. Monthly platinum UR were stable across diseases in the pre-period. However, there was a significant post-period decrease in CB usage for non-small cell lung (NSCLC) and endometrial cancers (EC), and a decreasing trend in UR for CB in bladder (BC), ovarian (OC), and small cell lung cancers (SCLC). We also observed a decreasing trend in CP usage for cholangiocarcinoma (CC). Post-period monthly UR were stable for CB and CP in head and neck cancer (HNC), and for CP in BC. Odds ratios (OR) representing the month-over-month change in the odds of receiving platinum therapy during both pre- and post-periods are shown (table). OR trends varied by practice type and disease; no notable trends were seen by practice size or SES. UR with additional follow-up times will be presented. Conclusions: We observed decreased platinum use across multiple cancers during the shortage period, most notably for met NSCLC and EC, suggesting shortages led oncologists to seek alternatives to standard 1L platinum-based regimens. Further study of how treatment changes due to platinum shortages impacted pt outcomes is warranted. [Table: see text]

CancerX Data Sprint: Advancing national oncology data standards through a unique public-private partnership.

e23283 Background: In a concerted effort to enhance the quality, accessibility, and effectiveness of comprehensive real-world cancer datasets, CancerX, a public-private partnership announced by the White House recently conducted a demonstration project, the CancerX Data Sprint. CancerX Data Sprint attempted a novel approach to quickly inform collective efforts of the Center for Medicare and Medicaid Innovation (CMMI), the National Cancer Institute (NCI), the US Food and Drug Administration (FDA), and the Office of the National Coordinator for Health Information Technology (ONC) to advance cancer-related data standards. Specifically, at the intersection of CMMI’s Enhancing Oncology Model (EOM) and the development of ONC’s United States Core Data for Interoperability (USCDI+) – Oncology extension, CancerX leveraged its unique model and broad membership to report novel insights to advance data-driven innovations in oncology research, care, and policy. Methods: From Sept to Dec 2023, the Data Sprint convened 150+ member organizations from industry and government partners to 1) Identify the high-value research questions that could be asked about the standardized real-world data (RWD) generated through EOM 2) Supplement the existing data elements with new ones to extend and enhance the ability to conduct scientific, clinical, and research inquiries with comprehensive datasets. The CancerX community opined via surveys, written feedback, and live workshops to identify key data elements, discuss their clinical value and technical feasibility, and consider their use in real-world settings as part of the comprehensive datasets. Results: Building on the current EOM clinical data elements, we identified and reported 15 new clinical data elements for considerations with high clinical utility and ease of technical abstraction. This comprehensive RWD encompasses a diverse range of patient demographics, clinical details, treatment histories, and outcomes, creating a more accurate and inclusive portrayal of cancer treatment effectiveness and patient journeys. This depth of information is crucial for optimizing the personalization of treatment pathways, conducting impact assessments, and addressing healthcare disparities. Essentially, the new datasets serve as a valuable repository of insights, driving advancements in cancer research, treatment and elevating care quality. Conclusions: CancerX remains committed to supporting federal data initiatives like USCDI+ Cancer standards development and others by providing recommendations for essential data elements implementation to enhance clinical utility, technical feasibility, and exchange of standardized cancer data. By adopting a unique approach and assembling an interdisciplinary group, we can ensure that data standardization efforts prioritize high-utility data elements, ultimately maximizing benefits for our patients.

A real-world longitudinal assessment of the treatment-related symptoms and associated burden among patients receiving chemotherapy.

e23158 Background: For patients (pts) with cancer, treatment with cytotoxic chemotherapy is associated with a variety of treatment-related symptoms including nausea, vomiting, diarrhea and constipation. Often the ability to manage these symptoms when they occur is the difference between completion of chemotherapy and early discontinuation. The first weeks of treatment are critical for understanding the pts’ acute experience with chemotherapy. The objective of this study is to longitudinally evaluate treatment-related symptoms and associated treatment burden among chemotherapy pts through a real-world electronic remote symptom monitoring system. Methods: Pts treated with chemotherapy in U.S. academic cancer centers were enrolled in Carevive’s Patient Reported Outcomes Mobile Platform (Carevive PROmpt®) between September 2020 and Dec 2023, with 3 month median follow-up. A total of 476 pts were enrolled, across early (81%) and late (19%) stage disease. Weekly PRO-CTCAE-derived surveys assessed severity, frequency, and interference for 16 symptoms often associated with cancer treatment: anxiety, constipation, cough, decreased appetite, diarrhea, fatigue, general pain, insomnia, mouth/throat sores, muscle pain, nausea, numbness and tingling, rash, sadness, shortness of breath, and vomiting. The GP5 assessed weekly overall bother with the side effects of treatment. Questionnaires were administered weekly starting at treatment initiation. Results: The mean age of chemotherapy pts was 67.3 (SD = 13.2) years, with 75% female, and 20% African American. At treatment initiation, 302 pts reported symptoms by PRO-CTCAE, and 100 completed the GP5. At week 1, constipation, general pain (GP), nausea, numbness/tingling (N/T), and shortness of breath were experienced by 20-25% of pts. Over the first 4 weeks of treatment, pts reported an increase in constipation, decreased appetite, diarrhea, and nausea at week 2, with a return to week 1 frequency over the remaining weeks. Diarrhea, however, remained elevated. While week 1 capture rates were lower, patterns for treatment bother were similar. At week 1, 34% of pts reported being bothered at least “Somewhat” by the side effects of treatment. This increased to 48% at week 2, but returned to near the week 1 frequency over the remaining weeks. Conclusions: For remote-monitored pts in this real-world data set treated with chemotherapy both alone and in combination, the most frequent symptoms at the start of treatment were nausea, GP, and N/T. The frequency of pts with GP and N/T did not change over the course of treatment suggesting their possible presence prior to the observed regimen. Increases in chemotherapy-related symptoms show the importance of ongoing assessment over the first 4 weeks of treatment, ideally with feedback to treating physicians and concomitant symptom management to optimize successful chemotherapy completion.

Pan-tumor circulating tumor DNA testing in the community oncology setting.

e14556 Background: Circulating tumor DNA (ctDNA) technology allows for monitoring the genomic profile of cancer over time non-invasively, and in a way that is more representative of existing tumor clones than tissue sampling. The clinical and research applications of this new technology are numerous, and they are potentiated through serial ctDNA testing of individual patients. It is important to identify and characterize real-world data sets that will permit the development of both ctDNA’s clinical and research applications. Methods: This is an observational study of breast, non-small cell lung cancer (NSCLC), colorectal and prostate cancer patients who had ctDNA testing in The US Oncology Network from August 2014 to August 2023. All unique ctDNA reports per patient were collected, along with the report dates and patient characteristics. Rate of change in number of unique reports per month through time was evaluated as well as frequencies of serial testing and median time lapse between diagnosis and date of first ctDNA test by cancer type. Results: Number of new unique ctDNA testing reports per month increased for The US Oncology Network from 2014 through 2018 at a rate of 0.86 (95% CI [0.70, 1.02]). From 2018 to 2023 this rate increased to 5.87 (95% CI [5.16, 6.59]) new reports per month. The total number of unique patients with one of the 4 cancer indications and at least one ctDNA test during the study time period was 10,852. The cancer indication with greatest number of patients with ctDNA testing is NSCLC with 5,503 patients, followed by breast with 2,408, and colorectal and prostate with 1,544 and 1,397 respectively. Black patients were equally represented across all 4 cancers at rates of 13.6% for colorectal, 12.7% for breast, and 12% for NSCLC and prostate. Patients over 65 made up 66% for prostate and NSCLC but only 38% and 32% for colorectal and breast respectively. Most patients presented with a single ctDNA test (n = 9482), but an important proportion presented with 2 serial ctDNA tests or more (n = 1370, 12.6%).The median time between either cancer diagnosis date or date of first visit in The Network and the date of the first ctDNA test report was 39 days for NSCLC, 1,690 days for breast, 377 days for colorectal and 772 days for prostate, and presented with a skewed, long right tailed distribution. Median times were similar for Black and White patients across all cancers. Conclusions: The number of patients with ctDNA testing data available is increasing through time without evidence of plateauing. Although the total number of patients represented is still a very small fraction of the total patient population, the rate of increase in number of new reports through time is equivalent to an exponential increase in the total number of reports. 12.6% of patients presented with two tests or more, potentially allowing for pre and post treatment assessments. There’s an opportunity to better align cancer research and cancer treatment paradigms with this new technology.

Exploratory data analysis for electric vehicle driving range prediction: insights and evaluation

One of the biggest challenges of electric vehicle (EV) users has been predicting the amount of driving time their vehicles will have on one battery charge. Planning a trip and reducing range anxiety depends on an accurate range estimate. This study aims to anticipate the EV driving range using machine learning methods. In this research, several regression models for predicting EV driving range will be developed and compared. A real-world dataset comprising various factors affecting EV range, such as power, trip distance, energy consumption, driving style, and environmental factors, is used for analysis. The dataset is preprocessed using exploratory data analysis methods to manage missing values, outliers, and categorical variables. The findings of this study contribute to the expanding area of EV range prediction and provide EV buyers, producers, and regulators with insightful information. The user experience can be improved, EV adoption can be boosted, and effective charging infrastructure design is made possible with accurate range prediction. The study also highlights the importance of model selection and data pretreatment in making accurate predictions.

Development and validation of RSClin N+ tool for hormone receptor-positive (HR+), HER2-negative (HER2-), node-positive breast cancer.

508 Background: The RSClin tool was developed to provide estimates of recurrence risk and absolute chemotherapy benefit for patients with HR+/HER2- node-negative breast cancer using the 21-gene recurrence score (RS) and clinicopathologic factors. For patients with node-positive disease, we developed a new tool (RSClin N+) to increase the prognostic and predictive utility of risk estimates by combining RS with clinicopathologic factors and menopausal status. Methods: We used Cox regression to estimate 5-year risk of invasive disease or death (iDFS) and likelihood ratio (LR) tests to compare fit of RSClin N+ with RS alone and clinicopathologic factors (tumor grade, tumor size, positive lymph nodes [N1/N2], age) alone in 5283 node-positive patients treated with chemoendocrine therapy (CET) vs endocrine therapy alone (ET) in the S8814 (n=367: N1=227, N2=140, all postmenopausal) and RxPONDER (n/N1=4916) trials. Data from both studies were pooled and stratified by study for postmenopausal models; premenopausal models were fit only on RxPONDER data. Absolute CET benefit across combination of covariate values was estimated as the difference between ET and CET risk estimates. Validation of RSClin N+ was performed in 592 Nmic/N1 patients in the Clalit registry real-world dataset. Results: For pre- and postmenopausal patients, RSClin N+ provided significantly more prognostic information for iDFS than RS or clinicopathologic factors alone (LR Chi-square p<0.05). In postmenopausal patients, RS and clinicopathologic factors were independently prognostic, but only RS showed interaction with CET benefit (p=0.016). Absolute CET benefit for iDFS ranged from <0.1% to 21.5% as the RS increased from 0 to 50. Table shows examples of risk stratification and predicted benefit by RS within clinical low and high risk patients. In premenopausal patients, both RS and clinicopathologic factors remained prognostic but no interaction was observed between RS and CET.In external validation, RSClin N+ risk estimates were concordant with (Lin’s concordance=0.92) and prognostic for observed risk (HR 1.75; 95% CI 1.38 to 2.20). Conclusions: The RSClin N+ model provides improved estimates of prognostic risk and absolute CET benefit than clinical or genomic data alone in node-positive, HR+/HER2- breast cancer and could be used in patient counseling. [Table: see text]

Beyond Homophily and Homogeneity Assumption: Relation-Based Frequency Adaptive Graph Neural Networks.

Graph neural networks (GNNs) have been playing important roles in various graph-related tasks. However, most existing GNNs are based on the assumption of homophily, so they cannot be directly generalized to heterophily settings where connected nodes may have different features and class labels. Moreover, real-world graphs often arise from highly entangled latent factors, but the existing GNNs tend to ignore this and simply denote the heterogeneous relations between nodes as binary-valued homogeneous edges. In this article, we propose a novel relation-based frequency adaptive GNN (RFA-GNN) to handle both heterophily and heterogeneity in a unified framework. RFA-GNN first decomposes an input graph into multiple relation graphs, each representing a latent relation. More importantly, we provide detailed theoretical analysis from the perspective of spectral signal processing. Based on this, we propose a relation-based frequency adaptive mechanism that adaptively picks up signals of different frequencies in each corresponding relation space in the message-passing process. Extensive experiments on synthetic and real-world datasets show qualitatively and quantitatively that RFA-GNN yields truly encouraging results for both the heterophily and heterogeneity settings. Codes are publicly available at: https://github.com/LirongWu/RFA-GNN.

Hierarchical Neighbors Embedding.

Manifold learning now plays an important role in machine learning and many relevant applications. In spite of the superior performance of manifold learning techniques in dealing with nonlinear data distribution, their performance would drop when facing the problem of data sparsity. It is hard to obtain satisfactory embeddings when sparsely sampled high-dimensional data are mapped into the observation space. To address this issue, in this article, we propose hierarchical neighbors embedding (HNE), which enhances the local connections through hierarchical combination of neighbors. And three different HNE-based implementations are derived by further analyzing the topological connection and reconstruction performance. The experimental results on both the synthetic and real-world datasets illustrate that our HNE-based methods could obtain more faithful embeddings with better topological and geometrical properties. From the view of embedding quality, HNE develops the outstanding advantages in dealing with data of general distributions. Furthermore, comparing with other state-of-the-art manifold learning methods, HNE shows its superiority in dealing with sparsely sampled data and weak-connected manifolds.

Comparing literature-based cell-free miRNA panels to whole genome miRNA screening for cancer detection.

e13587 Background: Cell-free miRNAs (cf-miRNAs) found in bodily fluids like plasma or serum are adept at detecting cancer. When constructing a cancer detection model with a reasonably sized training sample, it seems logical that employing whole genome screening could enhance prediction performance and unveil novel miRNAs. However, screening all cf-miRNAs could be costly and its potential improvement compared to a curated panel remains uncertain. To evaluate the trade-off between accuracy and cost-effectiveness, our study aims to compare the performance of two approaches: a literature-based cf-miRNA panel and a whole genome screening. We utilized real-world datasets to assess the effectiveness of these two methods in cancer detection. Methods: To curate a cf-miRNA panel, we conducted a comprehensive review of published miRNA-related databases, identifying 27 databases, including dbDEMC 2.0, miRCancer, miRStress, HMDD, and others. Of these, 11 databases were both available and downloadable, yielding 196,839 records (database entries). Focusing on cf-miRNAs, we filtered the records to only include those clearly identified as originating from plasma or serum, resulting in 13,595 cf-miRNA records. Next, to ensure robustness, we employed two strategies: requiring multiple records for a given miRNA and confirming its mention in multiple databases to finalize the cf-miRNA panel. For comparison of prediction performance, we obtained a large-scale dataset from the Gene Expression Omnibus, specifically the GPL21263 platform, for modeling experiments. This dataset comprised 8,174 subjects with 2,565 miRNA targets across seven cancer types derived from various cf-miRNA-based cancer studies. We then trained machine learning models on both the curated panel and the full miRNAs from this dataset to facilitate a comparative analysis. Results: Out of the 13,595 cf-miRNA records, 499 miRNAs had over 4 records, and 612 miRNAs were mentioned in multiple databases. There was an overlap of 339 miRNAs, forming our final cf-miRNA panel. The model accuracy using this panel to distinguish between seven cancer types was 67.62%. In contrast, directly employing machine-learning to model whole genome screening resulted in an accuracy of 93.00%, demonstrating a significant improvement compared to the panel-based results. Conclusions: Our study curated a cost-effective cf-miRNA panel of 395 miRNAs, demonstrating the capability to distinguish multiple cancers. However, while the curated panel exhibited reasonable efficacy in cancer detection, the whole genome screening model displayed superior discriminatory power. Hence, for optimizing cancer detection methods, especially in large-scale datasets, employing whole genome screening appears to be a more promising strategy.

Dynamic machine learning model to forecast patient availability for clinical trials.

e13641 Background: Patient eligibility is a routine concept in clinical research, comprised of simple or complex set of inclusion or exclusion criteria applied in a specific sequence. Our methodology introduces the concept of ‘availability’ as a narrower term that is closer to the actual number that will be considered for a specific study and who may decide to participate. This specific research develops Machine Learning (ML) model that dynamically predicts likelihood of Multiple Myeloma progression and departure from targeted outcomes to sustain therapy and to indicate potential availability for clinical trials – to reduce screening efforts and improve screening precision of clinical research organizations. Methods: The cohort for training these models was drawn from the ConcertAI Oncology Research database, further enriched by key variables derived from unstructured data sources. Eligible patients for the analysis had a diagnosis of Multiple Myeloma made after Jan 1, 2020. Progressions were labelled using International Myeloma Working Group Criteria. Patients that did not satisfy the criteria were censored. Labels were assigned to patients after initial diagnosis of MM, and the duration between diagnosis to label was computed to build time to event (TTE) models. Over 2000 features were created to build this model, e.g., laboratory values, vitals, biomarkers, medications, ECOG, stage, etc. Variable creation preserved temporality of lab variables and vitals and noted presence of other variables. Several ML methods were employed to build the TTE models, with gradient boosting resulting in the highest c-index (0.8) based on 12 clinically relevant features. A temporal validation of the model was performed on a simulated index date. Results: Cohort used for training the model comprised of N = 4315 patients with average age at diagnosis of 72.3 years, and ~11% of the patients labelled as progressing. Temporal validation was performed on N = 1874 patients, with ~9% progression rate. The area under the curve (AUC) and weighted F1 score for this cohort was 87% and 89% respectively. Conclusions: Patient availability assessments are complex, not supported effectively by existing solutions and often require 100’s of hours from highly trained clinical personnel. An AI model predicting Multiple Myeloma progression built using a large-scale real-world dataset demonstrated 2-4x improvement in forecasting patient availability over other approaches. Integrating these solutions into clinical workflows can improve the effectiveness of research teams; enable more patients to have clinical trials as viable options at the right time; and allow care teams to have forward visibility to choices of standard-of-care and clinical trial alternatives in the best interest of patient outcomes. [Table: see text]