Clinical Trial Dataset Research Articles

Prediction of cancer treatment response from histopathology images through imputed transcriptomics.

1551 Background: In recent years, the use of tumour molecular profiling within the clinic has allowed for more accurate cancer diagnostics, as well as the delivery of precision oncology. Rapid advances in digital histopathology have allowed the extraction of clinically relevant information embedded in tumor slides by applying machine learning methods, capitalizing on recent advancements in image analysis via deep learning. However, as in many supervised learning approaches, predicting response to therapy using whole slide images (WSI) of tissue stained with hematoxylin and eosin (H&E) requires large datasets comprising matched imaging and response data, severely restricting the applicability of this approach. Methods: To overcome this critical challenge, we introduce here for the first time a generic methodology for generating WSI-based predictors of patients’ response for a broad range of cancer types and therapies, which does not require matched WSI and response data for training. The approach, termed ENLIGHT-DeepPT, consists of: (1) DeepPT, a new deep-learning framework that predicts exome-wide tumor mRNA expression from slides, and (2) ENLIGHT, a recently published response prediction algorithm, applied here to predict response based on the DeepPT predicted expression values, instead of measured values. Results: First, we study the ability to predict tumor expression, showing that DeepPT successfully predicts transcriptomics in all 16 TCGA cohorts tested and generalizes well to two independent datasets. Second, aligned with our key aim, we analyze six independent clinical trial datasets of patients with different cancer types that were treated with various targeted and immune therapies, on which DeepPT was never trained. We show that ENLIGHT, without any adaptation, can successfully predict the true responders from the expression values imputed by DeepPT, using only H&E images: ENLIGHT-DeepPT successfully predicts true responders in these cohorts with an overall odds ratio of 2.19, increasing the baseline response rate by 38.8% on average among predicted responders. Conclusions: ENLIGHT-DeepPT is the first approach to successfully predict response to multiple targeted and immune cancer therapies directly from H&E slides, without requiring any drug-specific training data. ENLIGHT-DeepPT can provide rapid treatment recommendations without requiring tumor sequencing, bringing precision oncology to low-income countries and other situations where NGS is prohibitive.

A platform for comparing subgroup identification methodologies.

Since the advent of the phrase "subgroup identification," there has been an explosion of methodologies that seek to identify meaningful subgroups of patients with exceptional response in order to further the realization of personalized medicine. However, to perform fair comparison and understand what methods work best under different clinical trials situations, a common platform is needed for comparative effectiveness of these various approaches. In this paper, we describe a comprehensive project that created an extensive platform for evaluating subgroup identification methods as well as a publicly posted challenge that was used to elicit new approaches. We proposed a common data-generating model for creating virtual clinical trial datasets that contain subgroups of exceptional responders encompassing the many dimensions of the problem or null scenarios in which there are no such subgroups. Furthermore, we created a common scoring system for evaluating performance of purported methods for identifying subgroups. This makes it possible to benchmark methodologies in order to understand what methods work best under different clinical trial situations. The findings from this project produced considerable insights and allow us to make recommendations for how the statistical community can better compare and contrast old and new subgroup identificationmethodologies.

Use of Clinical Trial Simulations to Compare the Performance of Different Approaches for Population Analyses of Pediatric Pharmacokinetic Data.

The adequate characterization of the pharmacokinetics of a drug used in pediatrics is a mainstay of pediatric development programs and is critical for accurate dose selection in pediatrics. Analysis approaches can impact the estimation and characterization of pediatric pharmacokinetic parameters. Simulations were conducted to compare the performance of different approaches for analyzing pediatric pharmacokinetic data in the presence of extensive data from adult studies. Simulated clinical trial datasets were generated encompassing different scenarios that might be encountered in pediatric drug development. For each scenario, 250 clinical trials were simulated and analyzed using each of the following approaches: (1) estimating pediatric parameters using only pediatric data; (2) fixing specific parameters to adult estimates and estimating the remaining pediatric parameters using only pediatric data; (3) estimating pediatric parameters using adult parameters as informative Bayesian priors; (4) estimating pediatric parameters using combined adult and pediatric datasets with exponents for body weight effects estimated using adult and pediatric data; and (5) estimating pediatric parameters using combined adult and pediatric datasets with exponents for body weight effects estimated using pediatric data only. Each analysis approach was evaluated for its success in the estimation of true pediatric pharmacokinetic parameter values. Results demonstrated that analyzing pediatric data using a Bayesian approach generally performed best and had the lowest probability of significant bias in the estimated pediatric pharmacokinetic parameters among different scenarios evaluated. This clinical trial simulation framework can be used to inform the optimal approach for analyses of pediatric data for other pediatric drug development program scenarios beyond the cases evaluated in these analyses.

Abstract 4287: The combination of statistical methods to compare observed and simulated data allowed to assess effectively the validity of mathematical model predictions in a context of a EGFR+ lung adenocarcinoma

Abstract Introduction In silico models proved to be a promising tool to complement and optimize clinical trials. These models should be validated to assess their capacity to reproduce real life behaviors. Opposed to real-life clinical trials where the amount of available data might be low, in silico approaches give us the possibility to simulate virtual populations of thousands of patients. This data size heterogeneity might be an issue. Moreover, in real life data, patients are seen at scheduled visits, leading to an observation time uncertainty (OTU), which is not the case in simulated data.In the context of the validation of a EGFR-mutant Lung Adenocarcinoma mathematical model, we depicted the interest of using combined validation methods to assess the capacity of the model to predict the time to tumor progression, from heterogeneous clinical trials datasets. Methods In this context, to overcome the limitation of using default log-rank test that could lead to biased results, 4 bootstrapped methods have been used: - 2 methods based on the log-rank test where the ratio of significant tests at a given alpha risk level is assessed, taking into account the OTU: The “default Log-rank test” and the “modified test” based on a combination of weighted log-rank tests (MaxCombo). - 2 methods based on prediction intervals. The “raw coverage”, which corresponds to the proportion of the observed time-to-event curve included in the prediction interval and the “juncture metric” which corresponds to the proportion of the observation period where the prediction interval overlaps with an interval bound between the observed data and the same data shifted by the OTU. Results As each one of the validation approaches has its own strengths and weaknesses and conditions of application, the validation process has been based on the combination of well selected methods. Validation metrics showed that the model reproduces successfully real life data depending on a given context. When all the metrics give the same result, we were able to easily conclude. But when it was not the case, this led us to investigate the origin of these discrepancies, taking into account method robustness. Conclusion The validation process is of the utmost importance to assess the level of credibility of the model. We showed that in order to fully evaluate the model, a combination of validation approaches is preferable. The use of multiple metrics highlights the eventual inconsistencies in model predictions and validation dataset’s content that would not have been detected by only a single approach. The drawback of combining metrics is that it complexifies the decision making, implying that the final decision, regarding whether the model is validated or not, is assumed by the user, without forgetting that the validation is not an end objective per se but a stepwise process. Citation Format: Evgueni Jacob, Laura Villain, Nicoletta Ceres, Jean-Louis Palgen, Adele L'Hostis, Claudio Monteiro, Riad Kahoul. The combination of statistical methods to compare observed and simulated data allowed to assess effectively the validity of mathematical model predictions in a context of a EGFR+ lung adenocarcinoma. [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2023; Part 1 (Regular and Invited Abstracts); 2023 Apr 14-19; Orlando, FL. Philadelphia (PA): AACR; Cancer Res 2023;83(7_Suppl):Abstract nr 4287.

Abstract 957: Prediction of patient response to targeted and immunotherapies from the tumor transcriptome in a wide set of indications and clinical trials

Abstract Background: Precision oncology is gradually advancing into mainstream clinical practice, demonstrating significant survival benefits. However, eligibility and response rates remain limited in many cases, calling for better predictive biomarkers. Methods: We present ENLIGHT, a transcriptomics-based computational approach that identifies clinically relevant genetic interactions and uses them to predict a patient’s response to a variety of therapies in multiple cancer types, importantly, without training on previous treatment response data. Consequently, in addition to its ability to predict patients' response to approved and well-studied therapies, ENLIGHT can predict the response to new treatments in early development, even before clinical data has accumulated. Accordingly, we study ENLIGHT in two translationally relevant scenarios: Personalized Oncology (PO), aimed at prioritizing approved treatments to a given patient, and Clinical Trial Design (CTD), selecting the subset of most likely responders in a patient cohort. Results: Evaluating ENLIGHT’s performance on 21 blinded clinical trial datasets spanning 11 indications and 15 different drugs in the PO setting, we show that it can effectively predict a patient’s treatment response across multiple therapies and cancer types, with an overall odds ratio of 2.59 (p=3.41e-8), and a 36% increase in response rate over the baseline (p=3.30e-13). Its prediction accuracy is better than other state-of-the-art transcriptomics-based signatures. Unlike most signatures that are prognostic or provide insights for only very few, specific treatments, ENLIGHT provides matching scores to a broad range of treatments. Quite strikingly, its performance is comparable to that of supervised predictors developed for specific indications and drugs. In combination with the IFN-γ signature, ENLIGHT achieves an odds ratio larger than 4 in predicting response to immune checkpoint therapy. In the CTD scenario, our results show that by excluding non-responders ENLIGHT can enhance clinical trial success for immunotherapies and other monoclonal antibodies, achieving > 90% of the response rate attainable under an optimal exclusion strategy. Conclusion: ENLIGHT is a powerful transcriptomics-based precision oncology pipeline developed by Pangea Biomed that broadly predicts response to both extant and novel targeted and immune therapies, going beyond context-specific biomarkers. Citation Format: Gal Dinstag, Eldad D. Shulman, Efrat Elis, Doreen S. Ben-Zvi, Omer Tirosh, Eden Maimon, Isaac Meilijson, Emmanuel Elalouf, Boris Temkin, Philipp Vitkovsky, Eyal Schiff, Danh-Tai Hoang, Sanju Sinha, Nishanth Ulhas Nair, Joo Sang-Lee, Alejandro A. Schäffer, Ze'ev Ronai, Dejan Juric, Andrea B. Apolo, William L. Dahut, Stanley Lipkowitz, Raanan Berger, Razelle Kurzrock, Antonios Papanicolau-Sengos, Fatima Karzai, Mark R. Gilbert, Kenneth Aldape, Padma S. Rajagopal, Tuvik Beker, Eytan Ruppin, Ranit Aharonov. Prediction of patient response to targeted and immunotherapies from the tumor transcriptome in a wide set of indications and clinical trials [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2023; Part 1 (Regular and Invited Abstracts); 2023 Apr 14-19; Orlando, FL. Philadelphia (PA): AACR; Cancer Res 2023;83(7_Suppl):Abstract nr 957.

Dynamic effects of prognostic factors and individual survival prediction for amyotrophic lateral sclerosis disease.

Amyotrophic lateral sclerosis (ALS) is a neurodegenerative disease affecting motor neurons, with broad heterogeneity in disease progression and survival in different patients. Therefore, an accurate prediction model will be crucial to implement timely interventions and prolong patient survival time. A total of 1260 ALS patients from the PRO-ACT database were included in the analysis. Their demographics, clinical variables, and death reports were included. We constructed an ALS dynamic Cox model through the landmarking approach. The predictive performance of the model at different landmark time points was evaluated by calculating the area under the curve (AUC) and Brier score. Three baseline covariates and seven time-dependent covariates were selected to construct the ALS dynamic Cox model. For better prognostic analysis, this model identified dynamic effects of treatment, albumin, creatinine, calcium, hematocrit, and hemoglobin. Its prediction performance (at all landmark time points, AUC ≥ 0.70 and Brier score ≤ 0.12) was better than that of the traditional Cox model, and it predicted the dynamic 6-month survival probability according to the longitudinal information of individual patients. We developed an ALS dynamic Cox model with ALS longitudinal clinical trial datasets as the inputs. This model can not only capture the dynamic prognostic effect of both baseline and longitudinal covariates but also make individual survival predictions in real time, which are valuable for improving the prognosis of ALS patients and providing a reference for clinicians to make clinical decisions.

PD24-11 HOT CLEAR CELL RENAL CELL CARCINOMA IMMUNE STATUS: ILLUSION OR REALITY?

You have accessJournal of UrologyCME1 Apr 2023PD24-11 HOT CLEAR CELL RENAL CELL CARCINOMA IMMUNE STATUS: ILLUSION OR REALITY? Yi Lu and Hongjun Li Yi LuYi Lu More articles by this author and Hongjun LiHongjun Li More articles by this author View All Author Informationhttps://doi.org/10.1097/JU.0000000000003302.11AboutPDF ToolsAdd to favoritesDownload CitationsTrack CitationsPermissionsReprints ShareFacebookLinked InTwitterEmail Abstract INTRODUCTION AND OBJECTIVE: The drivers and resistors of the anti-tumor response remain incompletely elucidated. High tumor mutation burden (TMB) and increased infiltration by T cells are associated with the answer to immune checkpoint inhibitors ICIs. We conducted the bioinfomatic study to better understand immunotherapy resistance in clear cell renal cell carcinoma (ccRCC). METHODS: The patients and sequencing data were selected from The Cancer Genome Atlas KIRC (TCGA-KIRC), and a clinical trial dataset. To classify worse-survival ccRCC resistant to immunotherapy, immune prognosis model (IPM) risk score classifier was constructed with prognostic costimulatory molecule genes through survival analysis and validated as well. For function analysis, ESTIMATE, ImmuCellAI, TIDE, and Gene Oncology (GO) were used. RESULTS: The 366 Co-expressed genes (CEGs) correlated with IPM risk score were enriched with p<0.05 and correlation |R-value|>0.5. GO analysis indicated CEGs were enriched within immune-associated gene sets, such as T cell homeostasis. TIDE algorithm predicted all the patients' response to ICIs, and non-responders were enriched in higher IPM risk scores and a real-world ICIs-treated ccRCC cohort validated it with a consistent finding with p=0.028 in chi-square test. ESTIMATE algorithm showed the ESTIMATE, and immune scores of high-risk patients, markedly increased compared with low-risk patients. Nevertheless, tumor purity was thinner than low-risk patients. Besides, TMB and cytolytic activity were higher in high-risk patients. ImmuCellAI algorithm revealed the levels of CD8+ T cells, CD4+ T cells, dendritic cells, and MAIT, together with macrophages, increased in the tumor microenvironment of high-risk patients. The TIDE T cell function estimate was used to test the T cell dysfunction of high-risk patients, which was markedly worse compared with low-risk patients. Meanwhile, immune inhibiting cells, namely exhausted T cells, nTreg cells, iTreg cells, and cancer associated fibroblast (CAF), had an increased fraction in high-risk ccRCC. Remarkably, the microsatellite instability (MSI) score was lower in high-risk ccRCC. Interestingly, IPM risk level was independent from MSI level in prognostic influence. CONCLUSIONS: We found a subtype of refractory ccRCC (IPM high risk), and subsequent function analysis indicated the reason for its insensitivity to immune blockades was T cell dysfunction. In IPM high-risk ccRCC, T cells' high infiltration, hot immune status, is an illusion. The reality is the increased infiltration of dysfunctional CD8+ and CD4+ T cells, immunosuppressive cells (i.e., iT-regs, nT-regs, and CAFs) as well as low MSI. Source of Funding: This work is supported by the grant from National Population Health Science Data Sharing Service Platform Clinical Medical Science Data Center (NCMI-ABD02-201906) to Hongjun Li © 2023 by American Urological Association Education and Research, Inc.FiguresReferencesRelatedDetails Volume 209Issue Supplement 4April 2023Page: e727 Advertisement Copyright & Permissions© 2023 by American Urological Association Education and Research, Inc.MetricsAuthor Information Yi Lu More articles by this author Hongjun Li More articles by this author Expand All Advertisement PDF downloadLoading ...

Effect of Mailed Fecal Immunochemical Test Outreach for Patients Newly Eligible for Colorectal Cancer Screening

IntroductionLimited data exists on the effectiveness of organized outreach campaigns on CRC screening completion for patients who are newly eligible for such screening.MethodsWe conducted an analysis of an existing clinical trial dataset of a publicly funded safety-net health system serving low-income populations.ResultsA total of 619 patients aged 50–51 received the outreach intervention and 3108 patients aged greater than 51 years old who had no prior history of FIT testing similarly received the outreach intervention. Patients newly eligible for FIT were more likely to complete a FIT test compared with older patients who had yet to complete a FIT test (58.3% vs 40.5%, p < 0.001).ConclusionPatients who are newly eligible for colorectal cancer screening are more likely to respond to outreach interventions than older patients without a prior history of FIT, indicating newly eligible patients across diverse populations may benefit from targeted outreach intervention.

Healthcare systems data in the context of clinical trials – A comparison of cardiovascular data from a clinical trial dataset with routinely collected data

BackgroundRoutinely-collected healthcare systems data (HSD) are proposed to improve the efficiency of clinical trials. A comparison was undertaken between cardiovascular (CVS) data from a clinical trial database with two HSD resources. MethodsProtocol-defined and clinically reviewed CVS events (heart failure (HF), acute coronary syndrome (ACS), thromboembolic stroke, venous and arterial thromboembolism) were identified within the trial data. Data (using pre-specified codes) was obtained from NHS Hospital Episode Statistics (HES) and National Institute for Cardiovascular Outcomes Research (NICOR) HF and myocardial ischaemia audits for trial participants recruited in England between 2010 and 2018 who had provided consent. The primary comparison was trial data versus HES inpatient (APC) main diagnosis (Box-1). Correlations are presented with descriptive statistics and Venn diagrams. Reasons for non-correlation were explored. ResultsFrom 1200 eligible participants, 71 protocol-defined clinically reviewed CVS events were recorded in the trial database. 45 resulted in a hospital admission and therefore could have been recorded by either HES APC/ NICOR. Of these, 27/45 (60%) were recorded by HES inpatient (Box-1) with an additional 30 potential events also identified. HF and ACS were potentially recorded in all 3 datasets; trial data recorded 18, HES APC 29 and NICOR 24 events respectively. 12/18 (67%) of the HF/ACS events in the trial dataset were recorded by NICOR. ConclusionConcordance between datasets was lower than anticipated and the HSD used could not straightforwardly replace current trial practices, nor directly identify protocol-defined CVS events. Further work is required to improve the quality of HSD and consider event definitions when designing clinical trials incorporating HSD.

Impact of dataset size and long-term ECoG-based BCI usage on deep learning decoders performance.

In brain-computer interfaces (BCI) research, recording data is time-consuming and expensive, which limits access to big datasets. This may influence the BCI system performance as machine learning methods depend strongly on the training dataset size. Important questions arise: taking into account neuronal signal characteristics (e.g., non-stationarity), can we achieve higher decoding performance with more data to train decoders? What is the perspective for further improvement with time in the case of long-term BCI studies? In this study, we investigated the impact of long-term recordings on motor imagery decoding from two main perspectives: model requirements regarding dataset size and potential for patient adaptation. We evaluated the multilinear model and two deep learning (DL) models on a long-term BCI & Tetraplegia (ClinicalTrials.gov identifier: NCT02550522) clinical trial dataset containing 43 sessions of ECoG recordings performed with a tetraplegic patient. In the experiment, a participant executed 3D virtual hand translation using motor imagery patterns. We designed multiple computational experiments in which training datasets were increased or translated to investigate the relationship between models' performance and different factors influencing recordings. Our results showed that DL decoders showed similar requirements regarding the dataset size compared to the multilinear model while demonstrating higher decoding performance. Moreover, high decoding performance was obtained with relatively small datasets recorded later in the experiment, suggesting motor imagery patterns improvement and patient adaptation during the long-term experiment. Finally, we proposed UMAP embeddings and local intrinsic dimensionality as a way to visualize the data and potentially evaluate data quality. DL-based decoding is a prospective approach in BCI which may be efficiently applied with real-life dataset size. Patient-decoder co-adaptation is an important factor to consider in long-term clinical BCI.

Multi-arm U-Net with dense input and skip connectivity for T2 lesion segmentation in clinical trials of multiple sclerosis

T2 lesion quantification plays a crucial role in monitoring disease progression and evaluating treatment response in multiple sclerosis (MS). We developed a 3D, multi-arm U-Net for T2 lesion segmentation, which was trained on a large, multicenter clinical trial dataset of relapsing MS. We investigated its generalization to other relapsing and primary progressive MS clinical trial datasets, and to an external dataset from the MICCAI 2016 MS lesion segmentation challenge. Additionally, we assessed the model’s ability to reproduce the separation of T2 lesion volumes between treatment and control arms; and the association of baseline T2 lesion volumes with clinical disability scores compared with manual lesion annotations. The trained model achieved a mean dice coefficient of ≥ 0.66 and a lesion detection sensitivity of ≥ 0.72 across the internal test datasets. On the external test dataset, the model achieved a mean dice coefficient of 0.62, which is comparable to 0.59 from the best model in the challenge, and a lesion detection sensitivity of 0.68. Lesion detection performance was reduced for smaller lesions (≤ 30 μL, 3–10 voxels). The model successfully maintained the separation of the longitudinal changes in T2 lesion volumes between the treatment and control arms. Such tools could facilitate semi-automated MS lesion quantification; and reduce rater burden in clinical trials.

Leveraging process mining for modeling progression trajectories in amyotrophic lateral sclerosis

BackgroundAmyotrophic Lateral Sclerosis (ALS) is a neurodegenerative disease whose spreading and progression mechanisms are still unclear. The ability to predict ALS prognosis would improve the patients’ quality of life and support clinicians in planning treatments. In this paper, we investigate ALS evolution trajectories using Process Mining (PM) techniques enriched to both easily mine processes and automatically reveal how the pathways differentiate according to patients’ characteristics.MethodsWe consider data collected in two distinct data sources, namely the Pooled Resource Open-Access ALS Clinical Trials (PRO-ACT) dataset and a real-world clinical register (ALS–BS) including data of patients followed up in two tertiary clinical centers of Brescia (Italy). With a focus on the functional abilities progressively impaired as the disease progresses, we use two Process Discovery methods, namely the Directly-Follows Graph and the CareFlow Miner, to mine the population disease trajectories on the PRO-ACT dataset. We characterize the impairment trajectories in terms of patterns, timing, and probabilities, and investigate the effect of some patients’ characteristics at onset on the followed paths. Finally, we perform a comparative study of the impairment trajectories mined in PRO-ACT versus ALS–BS.ResultsWe delineate the progression pathways on PRO-ACT, identifying the predominant disabilities at different stages of the disease: for instance, 85% of patients enter the trials without disabilities, and 48% of them experience the impairment of Walking/Self-care abilities first. We then test how a spinal onset increases the risk of experiencing the loss of Walking/Self-care ability as first impairment (52% vs. 27% of patients develop it as the first impairment in the spinal vs. the bulbar cohorts, respectively), as well as how an older age at onset corresponds to a more rapid progression to death. When compared, the PRO-ACT and the ALS–BS patient populations present some similarities in terms of natural progression of the disease, as well as some differences in terms of observed trajectories plausibly due to the trial scheduling and recruitment criteria.ConclusionsWe exploited PM to provide an overview of the evolution scenarios of an ALS trial population and to preliminary compare it to the progression observed in a clinical cohort. Future work will focus on further improving the understanding of the disease progression mechanisms, by including additional real-world subjects as well as by extending the set of events considered in the impairment trajectories.

P277 Predicting disease relevant features in Crohn’s Disease and Ulcerative Colitis from Haematoxylin &amp; Eosin stained whole slide images using self-supervised deep learning

Abstract Background Histopathological endpoints are evolving as a treatment target in Inflammatory Bowel Disease (IBD). Use of histology to screen entrants could add value in IBD clinical trials; for example, by refining eligibility criteria to ensure studies recruit patients with definitive active inflammation at the microscopic level. Several histopathological indices have been developed, but the relative complexity of available scores hinders the development of an AI algorithm without large-scale labour-intensive annotation by a pathologist. We aimed to develop computer vision tools to assist decoding the complex clinical disease features at the histological level for both Crohn’s Disease (CD) and Ulcerative Colitis (UC). This will inform understanding of disease pathology and patient stratification to support clinical trial development strategies. Methods A total of 1397 clinically annotated Haematoxylin &amp; Eosin (H&amp;E) images were included from 418 CD and 218 UC patients enrolled in a multicentred longitudinal Study of a Prospective Adult Research Cohort with IBD (SPARC IBD) obtained from the IBD Plexus program of the Crohn’s &amp; Colitis Foundation. We developed an image quality control (QC) algorithm to automatically identify image and tissue processing/staining artefacts negatively impacting analysis (e.g., out-of-focus, tissue folds, overstained regions) and excluded these regions (Fig. 1). Next, a self-supervised learning (SSL) deep learning computer vison model was developed and trained to predict disease relevant features including disease diagnosis and lesional macroscopic appearance (inflammation, erosions and ulcers). Finally, to better understand the model’s predictions, we generated heatmap overlays on the tissue that show regions which the model considers to be most predictive (Fig. 2) and shared these with pathologists for qualitative evaluation. Results We find that the SSL model performs well on different downstream classification tasks such as UC vs CD (area under curve (AUC) = 0.79) and normal vs lesional tissue (AUC = 0.76). Specialist pathologist collaboration further confirmed that the heatmap overlays identified clinically relevant tissue features, including inflammatory cell infiltrates (Fig. 2). Conclusion These encouraging results support further exploration of this deep-learning algorithm to distinguish disease specific characteristics in this set of images from CD and UC patients. Further work is ongoing to validate the heatmap approach on endoscopic scores, we also plan to validate our model on IBD clinical trial datasets.

Natural cubic splines for the analysis of Alzheimer's clinical trials.

Mixed model repeated measures (MMRM) is the most common analysis approach used in clinical trials for Alzheimer's disease and other progressive diseases measured with continuous outcomes over time. The model treats time as a categorical variable, which allows an unconstrained estimate of the mean for each study visit in each randomized group. Categorizing time in this way can be problematic when assessments occur off-schedule, as including off-schedule visits can induce bias, and excluding them ignores valuable information and violates the intention to treat principle. This problem has been exacerbated by clinical trial visits which have been delayed due to the COVID19 pandemic. As an alternative to MMRM, we propose a constrained longitudinal data analysis with natural cubic splines that treats time as continuous and uses test version effects to model the mean over time. Compared to categorical-time models like MMRM and models that assume a proportional treatment effect, the spline model is shown to be more parsimonious and precise in real clinical trial datasets, and has better power and Type I error in a variety of simulation scenarios.

Developing Technical Specifications for Submitting Clinical Trial Data Sets for Treatment of Noncirrhotic Nonalcoholic Steatohepatitis (NASH) Liver Fibrosis

Nonalcoholic steatohepatitis (NASH) can progress to cirrhosis and liver failure and is associated with an increase incidence of liver cancer. Currently, there are no approved therapies for treatment of NASH fibrosis.&amp;nbsp; During evaluation of a new drug application, it can be challenging to differentiate between progression of liver disease and potential drug-induced liver injury (DILI) in subjects with NASH fibrosis and DILI.&amp;nbsp; Therefore, clinical trials for treatment of NASH fibrosis require standardized disease-specific metadata to adequately support evaluation of efficacy and safety, including potentially DILI assessment. To improve reviewability and quality of data submission, it is helpful for all stakeholders to understand specifications needed for subject level data submission to the FDA. Development of technical specifications requires a multidisciplinary approach. This report describes the rationale, process and methods used in developing Technical Specifications for Submitting Clinical Trial Data Sets for Treatment of Noncirrhotic Nonalcoholic Steatohepatitis (NASH). The recommendations outlined in the NASH technical specifications pertain to submission of the sponsor’s tabulated and analysis data sets in a standardized manner based on CDISC standards to improve reviewability. These specifications also provide an opportunity for dialogue between the sponsor and regulatory agency to discuss issues related to trial design or conduct that may affect the content of these data sets. These specifications are intended to support the draft guidance for industry Noncirrhotic Nonalcoholic Steatohepatitis with Liver Fibrosis: Developing Drugs for Treatment (NASH Guidance) and reflect the data standards and processes described in the FDA Study Data Technical Conformance Guide and the FDA Data Standards Catalog.

Clinically oriented prediction of patient response to targeted and immunotherapies from the tumor transcriptome

Precision oncology is gradually advancing into mainstream clinical practice, demonstrating significant survival benefits. However, eligibility and response rates remain limited in many cases, calling for better predictive biomarkers. We present ENLIGHT, a transcriptomics-based computational approach that identifies clinically relevant genetic interactions and uses them to predict a patient's response to a variety of therapies in multiple cancer types without training on previous treatment response data. We study ENLIGHT in two translationally oriented scenarios: personalized oncology (PO), aimed at prioritizing treatments for a single patient, and clinical trial design (CTD), selecting the most likely responders in a patient cohort. Evaluating ENLIGHT's performance on 21 blinded clinical trial datasets in the PO setting, we show that it can effectively predict a patient's treatment response across multiple therapies and cancer types. Its prediction accuracy is better than previously published transcriptomics-based signatures and is comparable with that of supervised predictors developed for specific indications and drugs. In combination with the interferon-γ signature, ENLIGHT achieves an odds ratio larger than 4 in predicting response to immune checkpoint therapy. In the CTD scenario, ENLIGHT can potentially enhance clinical trial success for immunotherapies and other monoclonal antibodies by excluding non-responders while overall achieving more than 90% of the response rate attainable under an optimal exclusion strategy. ENLIGHT demonstrably enhances the ability to predict therapeutic response across multiple cancer types from the bulk tumor transcriptome. This research was supported in part by the Intramural Research Program, NIH and by the Israeli Innovation Authority.

External validation of prediction models for patient-reported outcome measurements collected using the selfBACK mobile app

BackgroundExternal validation is essential in examining the disparities in the training and validation cohorts during the development of prediction models, especially when the application domain is healthcare-oriented. Currently, the use of prediction models in healthcare research aimed at utilising the under-explored potential of patient-reported outcome measurements (PROMs) is limited, and few are validated using external datasets. ObjectiveTo validate the machine learning prediction models developed in our previous work [29] for predicting four pain-related patient-reported outcomes from the selfBACK clinical trial datasets. MethodsWe evaluate the validity of three pre-trained prediction models based on three methods— Case-Based Reasoning, Support Vector Regression, and XGBoost Regression—using an external dataset that contains PROMs collected from patients with non-specific neck and or low back pain using the selfBACK mobile application. ResultsOverall, the predictive power was low, except for prediction of one of the outcomes. The results indicate that while the predictions are far from immaculate in either case, the models show ability to generalise and predict outcomes for a new dataset. ConclusionExternal validation of the prediction models presents modest results and highlights the individual differences and need for external validation of prediction models in clinical settings. There is need for further development in this area of machine learning application and patient-centred care.

A Sensitive Machine Learning-Based Approach to Assess Multiple Myeloma t(11;14) Genetic Subtype from Histopathology Images

Background: Multiple myeloma (MM) encompasses several cytogenetically distinct plasma cell neoplasms, in which the overexpression of anti-apoptotic BCL-2 family proteins facilitate the survival and accumulation of myeloma cells in the bone marrow. Venetoclax (Ven) is a highly selective, potent oral BCL-2 inhibitor that is currently being evaluated as a targeted therapy for the treatment of t(11;14)-positive relapsed/refractory MM (RRMM). t(11;14) is present in 15%-20% of individuals with MM and has recently been shown to be associated with lower overall survival and response rates to current therapies compared with other standard risk markers, indicating a need for additional therapeutic options (Bal et al. Br J Haematol. 2021;195:e113-e116). t(11;14) MM has been shown to retain unique features of B-cell biology; however, pathologist evaluation of t(11;14) has indicated that only 50% of t(11;14)-positive cases have distinctive lymphoplasmacytic morphology (Hoyer et al. Am J Clin Path. 2000;113:831-837; Garand et al. Leukemia. 2003;17:2032-2035). Here, we report a machine learning (ML)-based approach that is highly sensitive in determining t(11;14) status in digitized biopsy slides from a Ven clinical trial dataset. Methods: Bone marrow core biopsies from patients with RRMM who had known t(11;14) status were stained with hematoxylin and eosin (H&E) and the immunohistochemistry (IHC) marker CD138, a protein expressed on the plasma cell membrane. Slides were digitized and scanned at 40× (Aperio). For IHC model development, 213 slides (61 positive, 152 negative) passed quality control for sufficient tumor regions free of artifacts. First, a color-thresholding model identified regions of CD138 staining; these regions were then used to train a minimally supervised end-to-end (E2E) model for prediction of t(11;14) status. For H&E model development, 218 slides (64 positive, 154 negative) passed quality control. An initial tissue segmentation model was trained from pathologist annotations of regions of myeloma; these regions were used in E2E training for status prediction. The distribution of positive and negative cases was balanced across training, validation, and test sets. To reduce the number of false- negative cases, a combination approach was used in which cases that were predicted by both the IHC and H&E model as negative for t(11;14) were recorded as negative, and cases predicted as positive by either model were recorded as positive. Model performance was assessed using F1 score, which is a weighted average of the precision and recall values; Cohen's Kappa statistic, which measures agreement between actual and assigned classes, taking chance agreement into account; and area under the ROC (receiver operating characteristics) curve (AUC), which measures separability of classification. Results: The sensitivity for predicting t(11;14)-positive cases was robust for both models, with values of 0.75 for the IHC model, 0.67 for the H&E model, and 0.75 for the combination of IHC and H&E models (Table). Both the validation of IHC and H&E models showed a high predictive value for detecting t(11;14)-negative samples. In the validation and test sets, respectively, the IHC model yielded values of 0.87 and 0.88, and the H&E model yielded values of 0.86 and 0.84. Combination of IHC and H&E model predictions resulted in the best performance in the validation set, a predictive value 0.96 for t(11;14)-negative cases. Positive predictive values were not as high, with values of 0.56 and 0.53 for the IHC and H&E models, respectively. Conclusions: This ML-based model shows potential in classifying t(11;14) MM, with high predictive value for t(11;14)-negative cases and high sensitivity for detecting t(11;14)-positive cases. Both the IHC and H&E models can accurately predict the majority of positive and negative t(11;14) cases. These results suggest that this ML algorithm has high sensitivity for use as a screening tool. Additional optimization to improve specificity of the ML-based model could enable use as a diagnostic for t(11;14) status. This approach further demonstrates that t(11;14) MM possesses unique biological and histopathological features and warrants further development to potentially identify patients for targeted therapies like Ven that has been shown to be effective in this subgroup of patients. Clinical trials evaluating Ven in t(11:14)-positive RRMM are ongoing. Figure 1View largeDownload PPTFigure 1View largeDownload PPT Close modal

Association of artificial intelligence-powered and manual quantification of programmed death-ligand 1 (PD-L1) expression with outcomes in patients treated with nivolumab ± ipilimumab

AbstractAssessment of programmed death ligand 1 (PD-L1) expression by immunohistochemistry (IHC) has emerged as an important predictive biomarker across multiple tumor types. However, manual quantitation of PD-L1 positivity can be difficult and leads to substantial inter-observer variability. Although the development of artificial intelligence (AI) algorithms may mitigate some of the challenges associated with manual assessment and improve the accuracy of PD-L1 expression scoring, use of AI-based approaches to oncology biomarker scoring and drug development has been sparse, primarily due to the lack of large-scale clinical validation studies across multiple cohorts and tumor types. We developed AI-powered algorithms to evaluate PD-L1 expression on tumor cells by IHC and compared it with manual IHC scoring in urothelial carcinoma, non-small cell lung cancer, melanoma, and squamous cell carcinoma of the head and neck (prospectively determined during the phase II and III CheckMate clinical trials). 1,746 slides were retrospectively analyzed, the largest investigation of digital pathology algorithms on clinical trial datasets performed to date. AI-powered quantification of PD-L1 expression on tumor cells identified more PD-L1–positive samples compared with manual scoring at cutoffs of ≥1% and ≥5% in most tumor types. Additionally, similar improvements in response and survival were observed in patients identified as PD-L1–positive compared with PD-L1–negative using both AI-powered and manual methods, while improved associations with survival were observed in patients with certain tumor types identified as PD-L1–positive using AI-powered scoring only. Our study demonstrates the potential for implementation of digital pathology-based methods in future clinical practice to identify more patients who would benefit from treatment with immuno-oncology therapy compared with current guidelines using manual assessment.

Applying interpretable machine learning workflow to evaluate exposure-response relationships for large-molecule oncology drugs.

The application of logistic regression (LR) and Cox Proportional Hazard (CoxPH) models are well-established for evaluating exposure-response (E-R) relationship in large molecule oncology drugs. However, applying machine learning (ML) models on evaluating E-R relationships has not been widely explored. We developed a workflow to train regularized LR/CoxPH and tree-based XGboost (XGB) models, and derive the odds ratios for best overall response and hazard ratios for overall survival, across exposure quantiles to evaluate the E-R relationship using clinical trial datasets. The E-R conclusions between LR/CoxPH and XGB models are overall consistent, and largely aligned with historical pharmacometric analyses findings. Overall, applying this interpretable ML workflow provides a promising alternative method to assess E-R relationships for impacting key dosing decisions in drug development.