General Information
Conflict of Interest
Request Clinical Trials
Associated Trial(s):- NCT01615029 - An Open Label, International, Multicenter, Dose Escalating Phase I/II Trial Investigating the Safety of Daratumumab in Combination With Lenalidomide and Dexamethasone in Patients With Relapsed or Relapsed and Refractory Multiple Myeloma
- NCT00473512 - A Phase I/II Open Label Study of the 17α-Hydroxylase/ C17,20 Lyase Inhibitor, Abiraterone Acetate in Patients With Prostate Cancer Who Have Failed Hormone Therapy
Request Clinical Trials
Data Request Status
Status: OngoingResearch Proposal
Project Title: Estimate MTD in Phase 1 Clinical Trials through Distributionally Robust Optimization and Machine Learning
Scientific Abstract:
Background: Phase 1 clinical trials often rely on rule-based methods like the 3+3 approach for dose escalation, which, while simple, can inaccurately estimate the Maximum Tolerated Dose (MTD) due to high variability and conservative dosing. Model-based methods such as the Continual Reassessment Method (CRM) and Bayesian Logistic Regression Model (BLRM) provide more complex modeling but face challenges with potential mis-specification and limited data.
Objective: To develop and validate a methodology that combines distributionally robust optimization with machine learning to improve MTD precision in Phase 1 clinical trials, overcoming the shortcomings of traditional methods.
Study Design: A retrospective analysis using existing clinical trial data to compare the new methodology against conventional dose-escalation methods. The study integrates robust optimization and machine learning, including transfer learning, to achieve more reliable MTD estimates.
Participants: Data will be analyzed from Phase 1 clinical trials with adequate dose-response and adverse event documentation, sourced from the YODA Project.
Primary and Secondary Outcome Measure(s): Primary outcome: occurrence of serious adverse effects at varying dose levels. Secondary outcome: incidence of all other adverse effects across doses.
Statistical Analysis: The analysis will compare the proposed method against traditional ones using bias and variance in MTD estimates. Robustness checks will assess performance across different dose-response curves to handle model mis-specification.
Brief Project Background and Statement of Project Significance:
The traditional rule-based approaches to dose escalation, such as the 3+3 method, are prevalently utilized in early clinical development. Their appeal largely stems from the simplicity of their execution. However, the variability in estimates produced using these methods is high, and their accuracy in targeting the true Maximum Tolerated Dose (MTD) is often questionable. Moreover, the conservative nature of the 3+3 rules, or their adaptations, may result in numerous subjects being exposed to subtherapeutic, lower doses.
In contrast, model-based approaches like the Continual Reassessment Method (CRM) and Bayesian Logistic Regression Model (BLRM) present alternatives for more efficiently determining the maximum tolerable dose of a new drug. These models typically assume a logistic function for the dose-response relationship. Nonetheless, there are two notable drawbacks to these methods. First, the assumed dose-response function may be mis-specified; real-world biological responses to new drugs can be significantly more complex than the logistic models predict. Additionally, subject heterogeneity means that responses might not conform uniformly to a single dose-response function. Deviations from the assumed model can lead to skewed MTD estimates, potentially resulting in either excessively aggressive dose escalations that compromise patient safety, or overly cautious dose levels that limit the drug's therapeutic efficacy.
Second, the limited participant numbers in Phase 1 trials can introduce substantial estimation errors in model-based approaches. Small cohorts provide limited data for shaping the dose-response curve, potentially resulting in unreliable MTD estimations.
Furthermore, existing research often overlooks the out-of-sample performance of these methods, including biases, consistency, and variance in MTD estimates. These measures are crucial for assessing the quality of MTD estimates.
In this research proposal, we aim to surmount these limitations by integrating robust optimization with recent advancements in machine learning.
Specific Aims of the Project: Our methodology is designed to refine MTD estimation, ensuring precision even when the underlying dose-response model is not accurately specified. We propose to validate our approach using real datasets from Phase 1 clinical trials. Our methodology aims to bridge current gaps in MTD estimation by incorporating cutting-edge machine-learning techniques within a robust optimization framework. This strategy is anticipated to improve the reliability of MTD estimations, thereby enhancing the overall efficacy and safety of Phase 1 clinical trials.
Study Design: Methodological research
What is the purpose of the analysis being proposed? Please select all that apply.: Develop or refine statistical methods Research on clinical trial methods
Software Used: Python
Data Source and Inclusion/Exclusion Criteria to be used to define the patient sample for your study:
I am requesting two clincial trials, whose NCT id is NCT01615029 and NCT00473512. The reason I request these two phase 1 trials is because they are both dose escalation studies and can provide a robustness check for our proposed method by applying it to multiple trials.
The inclusion/exclusion criteria of the trial NCT01615029 is listed: 18 Years and older (Adult, Older Adult ); All Sex Group. All other criteria are the same with the Inclusion/Exclusion Criteria shown in the clinicaltrial.gov.
The inclusion/exclusion criteria of the trial NCT00473512 is listed: 18 Years and older (Adult, Older Adult ); Male. All other criteria are the same with the Inclusion/Exclusion Criteria shown in the clinicaltrial.gov.
Primary and Secondary Outcome Measure(s) and how they will be categorized/defined for your study: Our the study design requires the individual patient level data regarding each patients' adverse effect occurrence. Thus, the primary outcome measures include all serious adverse effect indicators, and time to response. The secondary outcome measures: other adverse effect indicators.
Main Predictor/Independent Variable and how it will be categorized/defined for your study: Again, we want to use machine learning and transfer learning to predict the patients outcome by learning patients with similar characteristics. Thus, the main Predictor/Independent Variable include in our study is: patient age, weight, height, gender, dose of treatment exposed to, and all other patient characteristics.
Other Variables of Interest that will be used in your analysis and how they will be categorized/defined for your study: not applicable.
Statistical Analysis Plan:
We aim to estimate the average dose-response function across the population to derive the MTD. In an ideal scenario with sufficient patient subjects, we could gradually test each dose level on a large cohort of patients to obtain an empirical dose-response function, providing an accurate estimate of the population dose-response function. However, with a small sample (or to use the sample more efficiently), clinical researchers typically rely on a pre-assumed dose-response function (e.g., a logit function), estimating only its parameters to reduce the required sample size. If the assumed dose-response function is misspecified, the MTD will be biased. Therefore, we propose a distributionally robust approach that considers a family of dose-response functions close to the empirical dose-response function. This family provides a conservative estimate of the MTD, which we will use to decide whether to escalate to the next dose level.
Additionally, each patient has unique characteristics such as gender, BMI, and race. A small sample may be imbalanced in some of these dimensions, resulting in a sample that is not representative of the population. For example, if a sample consists of 30% women while 50% of the patient population are women, and gender affects the dose response, then the estimated average dose-response function will be biased. To address this issue, we will use machine learning to estimate the heterogeneous dose-response function and then apply inverse probability weighting to assign different weights to our sample, ensuring that the estimates are representative of the population.
Narrative Summary: Traditional dose escalation methods like the 3+3 approach or model-based approach may inaccurately estimate the Maximum Tolerated Dose (MTD) in Phase 1 trials, potentially compromising patient safety and drug efficacy. We propose a new methodology that integrates robust optimization with advanced machine learning techniques, such as transfer learning, to enhance MTD estimation accuracy. Our approach will be validated using real Phase 1 trial data, aiming to significantly improve both safety and efficacy by addressing the limitations of current dosing models.
Project Timeline:
Anticipated project start date: 2024 June
Analysis completion date: 2024 Dec
Date manuscript drafted and first submitted for publication: 2025 Jun
Date results reported back to the YODA Project: 2025 Jun
Dissemination Plan:
Anticipated Products: The primary product will be a detailed research paper presenting the methodologies, findings, and implications of our study. This paper will detail the development and validation of our hybrid methodology for MTD estimation, comparisons with traditional methods, and an analysis of the robustness and efficiency of our approach.
Conference Presentations: We will prepare presentations for delivery at major clinical and pharmacological conferences. These presentations will focus on the innovative aspects of our methodology, particularly the integration of robust optimization with machine learning in the context of Phase 1 clinical trials.
Code Availability: To foster transparency and enable further research, we will make the developed code available, subject to privacy and ethical considerations. This will allow other researchers to replicate our study or extend our methodology.
Target Audiences: Clinical Researchers and Biopharma and those involved in pharmacological research, and clinical trial design.
Suitable Journals: Journal of Clinical Oncology, operations research, management science, Biometrics, Statistics in Medicine
Bibliography:
Neuenschwander, B., Branson, M., and Gsponer, T. (2008). Critical Aspects of the Bayesian Approach to Phase I Cancer Trials. Statistics in Medicine, 27(13), 2420–2439. doi: 10.1002/sim.3230.
Zhou, H., Yuan, Y. and Nie, L., 2018. Accuracy, safety, and reliability of novel phase I trial designs. Clinical Cancer Research, 24(18), pp.4357-4364.