Crossing the First Span: Initial Feedback on Our Nature Medicine Collaboration

In my February column, I introduced my focus on ‘building bridges.’ This month, I want to introduce colleagues who have played a key role in launching an initiative that supports this focus. Through this collaboration, we recruited 33 statisticians specializing in oncology, metabolic diseases, and infectious diseases to serve as Nature Medicine’s first official statistical advisory panel.

To better understand how this initiative is shaping research quality and influencing the future direction of statistical methodologies in Nature Medicine, I spoke with Saheli Sadanand, the journal’s deputy editor.

Ji-Hyun: Reflecting on the memorandum of understanding (MOU) partnership, how will you measure success in the short term? Longer term, what is your vision for the collaboration?

Saheli: Nature Medicine is very excited to be partnering with the American Statistical Association on this pilot program. In the short term, we are measuring success through the level of participation of the statistical advisers, as well as the rigor and timeliness of their reports. After six months, we are pleased that most advisers across all tracks have reviewed at least one paper, and our final editorial team decisions have greatly benefited from the feedback we have received. In the long term, we hope to learn more about good practices for trial design and reporting from the statistical advisers and feature their views in the magazine section of our journal. In addition, given the very positive feedback we have received from our editorial team, we are looking to expand the program to include another track soon.

Ji-Hyun: In the past, statistical reviewers would be engaged as needed (ad hoc). How has having a standing group of statistical advisers contributed to manuscript quality?

Saheli: The statistical advisory panel has allowed us to improve the consistency of our review process, and importantly, to better ensure that we can offer timely feedback to our authors. In the past, it was often challenging to secure a statistical reviewer, particularly for certain areas covered at the journal or during high-volume submission periods, and this led to delayed decisions.

Ji-Hyun: How have the advisory groups helped to address the most common statistical issues and pitfalls in medical manuscripts? Could you share an example?

Saheli: A common issue that we encounter is the absence of signposting of efficacy endpoints that are exploratory or post hoc in nature, as well as the overstating of efficacy endpoints that a trial was not powered for. The statistical reviewers have offered helpful comments related to the fidelity of the reporting of prespecified endpoints, which is critical for the objective and transparent reporting of our clinical studies. As a member of the ICMJE [International Committee of Medical Journal Editors], we follow strict guidance for the consideration and reporting of clinical studies, and we appreciate the guidance of our statistical advisers in ensuring that these manuscripts are presented as transparently and responsibly as possible.

Ji-Hyun: Do you envision a future in which statistical reviews are integrated into the workflow for all manuscripts (beyond Nature Medicine, for example)?

Saheli: We are keen to expand the current pilot program to other topics covered at our journal, and we feel that statistical expertise is generally important for evaluation of clinical studies. We will continue to look for ways to ensure the rigor of the work that we publish.

Ji-Hyun: Reproducibility and transparency are key to sound science and public trust. How is Nature Medicine addressing these issues, and what can the statistical reviews contribute?

Saheli: We concur that these are critical issues for all of our manuscripts. As with other journals in the Nature portfolio, we require our published manuscripts to have a completed reporting summary, which is a standardized checklist that details the reagents, methodology, and statistical approaches used in our studies. This checklist must be submitted prior to initiating the peer-review process, and we invite all of our referees to look at this file in conjunction with the manuscript. Clinical trial manuscripts must also include the trial protocol and statistical analysis plan (when separate) for peer review, and we encourage our authors to publish these files with the final manuscript, if possible. In addition, our manuscripts must include a data availability statement, which details where data are located and any restrictions on data access, as well as a code availability statement if bespoke code was used in the study. These sections further ensure that our referees and readers can interrogate the reproducibility of the findings when data can be made publicly available or available under controlled access.

Statistical reviewers can further contribute to the reproducibility and transparency of our studies by offering constructive feedback on trial design (and elements that may influence the strength of the conclusions drawn in the study), as well as flagging data that are not signposted in line with prespecified criteria in the protocol and evaluating the presentation of clinical data sets.

Ji-Hyun: What advice would you give statisticians who aspire to contribute meaningfully to high-impact medical journals like Nature Medicine?

Saheli: We are always keen to meet new referees across all areas that we cover. If you would like to join our reviewer pool, you can email one of our deputy editors (Saheli Sadanand at s.sadanand@us.nature.com or Liam Messin at liam.messin@nature.com) to learn more. Statistical reviews form a crucial component of the evaluation process for our clinical studies, and statisticians’ granular feedback, as well as their input on the context for how a given study sits within the broader landscape in terms of trial design and execution, are valuable for us and for the medical research community.

Ji-Hyun: Are there areas where you believe statisticians should focus their efforts to align with the needs of modern medical research?

Saheli: We and other journals are increasingly seeing observational studies and trials involving AI, and guidance from statisticians on how to improve the design of these studies would be welcome. In addition, we continue to see traditional 3+3 phase 1 trials across many fields, and we are interested in learning more about new and innovative approaches for phase 1 trial design.

Natalie Dean, who leads the infectious diseases cohort, is an associate professor in the department of biostatistics and bioinformatics at Emory University’s Rollins School of Public Health.Michele Guindani, who leads the metabolic diseases cohort, is a professor in the department of biostatistics at the University of California at Los Angeles’ Fielding School of Public Health.

I also invited the leaders of the infectious and metabolic diseases cohort to share their experiences: Natalie Dean, who leads the infectious diseases cohort, and Michele Guindani, who leads the metabolic diseases cohort.

Reflection on the MOU Partnership

Natalie: Building this partnership between ASA and Nature Medicine will hopefully provide more avenues to elevate statistical voices to a wider audience. Furthermore, there may be opportunities for ASA members to gain an inside look into the publishing process, learning from the Nature Medicine editorial team.

Advice for the Statistical Community

Natalie: The most interesting and impactful science is from interdisciplinary teams, and statisticians can contribute more than just their expertise in statistics. Some statisticians, like me, have dedicated their careers to one application area of interest (for me, infectious diseases and vaccines), rather than working broadly across many domains. By sticking to fewer topics, we can develop a deeper understanding of the science, yielding better analyses. I’d encourage statisticians to find long-term partnerships with collaborators who value statistical expertise.

When writing for a high-impact audience, effective communication is paramount. We have to be able to clearly and succinctly [communicate] difficult concepts, and we need to distinguish the most critical kernels from the wealth of information (Why does it matter?).

Addressing AI and Big Data Challenges

Michele: I believe that the deployment of AI and large-scale data sets provides both challenges and opportunities for journal editors and—more in general—statistics.

There are certainly challenges, which many of us statisticians can likely immediately recognize. On the one hand, a good study should ensure transparency and be reproducible. Many AI-driven studies rely on complex algorithms that are difficult to replicate without access to full data sets and model parameters. The lack of standardized reporting guidelines for AI-based research makes it even more challenging for reviewers and editors to evaluate the validity of the studies, including the quality of the data. It’s essential that statistical editors make sure that the studies’ results are robust. This often requires a careful consideration of the data distribution, the modeling assumptions, and potential sources of bias or heterogeneity—all things that may not be immediately apparent when using AI methods.

An additional challenge is represented by the fact that AI models tend to prioritize prediction accuracy over interpretability, making it difficult to derive actionable medical practices. Furthermore, by their very nature, AI-based methods focus on identifying associations between features and outcomes rather than establishing causal relationships. Large data sets are high-dimensional, with many variables of different types and origins. As a result, it may be hard to identify confounding factors that could influence the association between a study’s outcome and predictor variables. All of these points underscore the importance for statistical editors to keep informed about the latest methodologies, their strengths, and potential pitfalls, and to encourage comparisons with traditional statistical approaches that could provide more interpretability to a study results.

Ultimately, the goal is to use AI-driven analyses to inform clinical decisions. In radiology, for example, AI has been increasingly used to identify abnormalities like tumors, fractures, or lesions with high accuracy. While its use can transform the field and alleviates the burden on medical practitioners—reducing the likelihood of missed diagnoses due to human fatigue or oversight—it may be affected by implicit biases, and it may do little to achieve the goal of improving, rather than merely detecting, health conditions.

As I mentioned, the rise of AI and large-scale data sets in medical research also presents many opportunities for journal editors. AI tools are here to stay, and their integration offers a chance to rethink and improve reporting guidelines to ensure transparency and reproducibility across the board. The potential of AI could also extend to improving the peer-review process. While we’re not fully there yet, automated tools for statistical checking—such as code audits—could help detect errors before publication and encourage the sharing of code, data, and models. For example, the responsibilities of associate editors of reproducibility at JASA [Journal of the American Statistical Association] might become partially automated in the future. AI tools could also be incorporated into peer review to assist with fraud detection, statistical validation, and even matching reviewers to submissions. Of course, these applications require careful oversight and discussion. Still, given the rapid advancements in recent years, it’s exciting to consider how we might leverage AI in the editorial process moving forward.

The work of two groups within ASA can be very important in this regard. For example, The Committee on Data Science and Artificial Intelligence can guide the ASA in developing strategic initiatives that promote transparency, reproducibility, and ethical standards in AI-driven research, directly influencing how journals handle AI-based submissions. Similarly, the new ASA interest group Stats Up AI complements this by empowering statisticians to take leadership roles in AI research, ensuring that statistical rigor is maintained in developing and applying AI methods. Together, these groups can help improve the quality and integrity of medical research and publishing in the era of AI.

Ji-Hyun: Reflecting on your experience as a panel member for Nature Medicine over the past 6–7 months, what do you consider the most valuable insight gained from reviewing manuscripts? How do you see this experience contributing to your professional growth? If it did not, could you explain why?

Michele: Serving as a panel member has been very instructive, offering a distinct perspective compared to my role as an editor for statistical journals. In the statistical field, the focus typically centers on the modeling framework and data analysis after data collection, with the experimental setup often considered a fixed aspect of the study. In contrast, in Nature Medicine, the attention extends to the entire study—including experimental design, protocol development, clinical trial setup, and subsequent analysis.

Adhering to the original analytical plan outlined in the study protocol is essential. It minimizes biases, enhances reproducibility, and ensures the validity of the study’s conclusions. The most compelling studies are grounded in carefully designed protocols that clearly define objectives, hypotheses, inclusion and exclusion criteria, and analysis plans. A well-prepared statistical analysis plan should be developed in tandem with the study design, rather than [come] as an afterthought. Manuscripts that lack a transparent study protocol often introduce ambiguity in analytical decisions, raising concerns about reproducibility and selective reporting. Deviating from this plan without strong justification can compromise the study’s integrity and introduce unintended biases. Upholding scientific integrity, ensuring valid statistical inference, maintaining transparency, avoiding selective reporting, and complying with regulatory standards are fundamental to producing credible research.

A further insight from my panel experience is the critical role of clear communication in scientific writing. Some submissions—although they may be technically sound, they may lack coherence, making it challenging for reviewers to fully grasp the implications of the findings. High-profile submissions are often well-crafted and ambitious, but those that truly stand out exhibit methodological rigor, clear reasoning, and reproducibility. Reviewing manuscripts for Nature Medicine has highlighted the elements that make a study compelling: a well-structured narrative; clearly articulated research questions; and a balanced discussion that contextualizes findings without overstatement. This process also keeps us statisticians at the forefront of current research, deepening our understanding of the fields in which we work. I feel like I am improving my ability to position both my own work and that of others within the broader scientific landscape.

Finally, I asked four oncology reviewers to share their perspectives. What follows are insights from Chen Hu, Cara Joyce, Mithat Gönen, and Nolan Wages.

Ji-Hyun: With the rise of AI and large-scale data sets in medical research, what statistical challenges are most pressing for journal editors?

Mithat: Large-scale data sets that AI uses are not always well curated. It can be difficult to know what population a given data set is supposed to represent, how the variables are defined and ascertained or if they are defined consistently across individual units, and if common problematic issues like missingness needs special handling. Another statistical challenge is validation. While it is now widely accepted that AI methods overfit and independent validation is absolutely necessary before even the simplest claims can be made, creative data re-use might result in so-called “data leakage,” and it is difficult to realize this from the descriptions given in a manuscript. Finally, there is the issue of small data sets and AI. I encounter many people using sophisticated AI methods on 200-patient data sets (a common explanation is that it is a large data set for their disease) and expect them to work. I am not aware of many rules of thumb or guidelines of sample sizes to be used with AI methods but it would be helpful if one can authoritatively say you should not use AI in this data set and stick to regression.

Cara: Because big data resources are often readily available and easily accessible, there may be a tendency for researchers to jump right into analysis without taking care in defining their research question, fully considering biases and confounders, and drafting a detailed analysis plan. John Tukey long ago warned against precise answers to the wrong question, and, more recently, Xiao-Li Meng coined the phrase “Big Data Paradox,” asking us to weigh data quality, quantity, and problem difficulty to draw better population-level inferences from data. We require aims, endpoints, and a protocol for prospective interventions available via the Clinicaltrials.gov record, for example. Without a similar mechanism for documenting design and analysis considerations for retrospective observational studies, journal editors may find that the factors that contribute to the reproducibility crisis—p-hacking, hypothesizing after results known, etc.—are hard to detect or quantify in many papers using AI. Preregistration is one underutilized and maybe imperfect potential solution.

Chen: One of the biggest gaps in AI-driven medical research is the disconnect between statistical performance and clinical impact. As a statistical adviser for Nature Medicine and editorial board member of top clinical oncology journals, I frequently encounter studies boasting high accuracy or AUC yet providing little evidence of how these models improve patient care. A model with a slight predictive edge may still be impractical if it relies on hard-to-collect data or produces results that clinicians struggle to interpret. I often encourage authors to compare their models to existing clinical decision tools, justify their real-world feasibility, and demonstrate how the model will meaningfully influence medical decision-making.

Another major issue is over-reliance on performance metrics without proper validation. Many models perform well on a single data set but fail to generalize due to biases in training data or lack of external validation. In my reviews, I emphasize the need for domain-specific sensitivity analyses—such as stratifying performance by demographics or comorbidities—to identify potential biases. I also push for multi-institutional validation to ensure models hold up beyond their original data set.

A third challenge is missing data and biases in large-scale data sets. AI models are highly sensitive to data abnormalities, yet many studies fail to transparently report how missing data is handled. A model trained on incomplete data sets may lead to biased predictions, particularly if missing data patterns correlate with clinical outcomes. In my reviews, I stress the importance of transparent reporting on missing data handling, whether through imputation, weighting, or sensitivity analyses.

Nolan: One of the most pressing statistical challenges for journal editors is ensuring the rigor, transparency, and reproducibility of AI-driven analyses. Many machine learning models used in medical research lack interpretability, making it difficult to assess their validity and generalizability. Additionally, biases in training data can lead to misleading conclusions, particularly in clinical applications where demographic representation is critical. Editors must also navigate the complexities of evaluating predictive model performance, addressing issues such as overfitting and proper validation techniques. Statistical rigor in AI applications extends to ensuring appropriate documentation of methods and code sharing. Reproducibility remains a concern, as many studies do not provide enough detail for independent verification.

Ji-Hyun: How do you foresee the role of statisticians evolving in handling these challenges in the publication process?

Mithat: The best way to deal with these challenges is to convince the individual research teams to include a statistician in all aspects of their data analysis. It is a tall order.

In some ways, it is a failure of our profession that we are unable to communicate our value to biomedical researchers, although we have been working with them for more than half a century. I am not suggesting a requirement by journals to have statistical co-authors (which would only lead to superficial inclusion as co-authors) but a systematic exposure of PIs [principal investigators] to the idea that statistical skills are crucial to projects using AI in medical data sets.

It is an uphill battle because inclusion of a statistician is likely to reduce the number of claims that can be made by a paper and might be perceived as a short-term loss. It is the job of statisticians and perhaps journal editors to convince the scientific community at large that it will be a long-term gain. Besides this, we statisticians should continue to say and do what we think is right. We should also find ways to say it without being perceived as ‘Negative Nellies,’ or ‘Dr. Nos.’

And, finally, continue to contribute our expertise as editors and reviewers to filter out the poorly executed studies and reward the good ones.

Cara: I don’t think statisticians’ roles have changed drastically given new algorithms or large data sets; we still need to request sufficient methodological detail to make studies replicable, offer suggestions to improve results reporting, and ensure conclusions are appropriately tempered based on design and causal assumptions. We also need to join the conversation when we notice that lower-quality work is published—it could be PubPeer post-publication discussion, journal replies to the authors, or original articles aimed at establishing better practices. With respect to the latter, the work of the STRATOS initiative (STRengthening analytical thinking for observational studies) comes to mind as a model for this type of methods work.

Chen: As a statistical advisory board member, I help editors identify common pitfalls—overfitting, data leakage, and lack of external validation. I push for better calibration, transparency, and reproducibility, ensuring that AI research is not just statistically sound but also clinically meaningful. Without strong statistical oversight, flawed AI studies risk misleading clinicians and affecting patient care.

Moving forward, statisticians need to be more than just gatekeepers. AI research involves interdisciplinary teams with varying statistical expertise, and our role should be to guide authors toward better practices—helping them refine their methods rather than just pointing out flaws.

Journals can also play a role by pushing for real-world impact in AI studies. Accuracy and AUC alone don’t tell the full story. I’d like to see journals require authors to compare AI models to existing clinical tools, validate them in diverse populations, and provide a clear roadmap for clinical implementation.

Finally, statisticians can help shape editorial policies by establishing clearer reporting standards for AI research. Stronger guidelines on methodology transparency and external validation will improve reproducibility and elevate the impact of published studies.

Nolan: Statisticians will play a crucial role in shaping editorial policies and improving the peer-review process for AI-driven medical research. Their expertise is essential in developing guidelines for evaluating AI models, ensuring appropriate validation strategies, and enforcing transparency in reporting. Journals will increasingly rely on statistical reviewers with expertise in ML, causal inference, and large-scale data analytics to assess methodological rigor and detect potential pitfalls. Additionally, statisticians can contribute by educating the broader research community about best practices. As AI becomes more integrated into clinical decision-making, statisticians will be essential in bridging the gap between methodological advancements and real-world applicability.

Ji-Hyun: Reflecting on your experience as a panel member for Nature Medicine over the past 6–7 months, what do you consider the most valuable insight gained from reviewing manuscripts? How do you see this experience contributing to your professional growth? If it did not, could you explain why?

Mithat: Too early to say since I only reviewed a few papers, but I think I will enjoy the exposure to the variety of topics and methods used in the papers I am reviewing. I also learn quite a bit from reading other referee reports, so to the extent they are available to me, it will help me learn and grow.

Cara: I enjoy peer review to gain exposure to new developments in medicine and also to keep up with innovations in clinical trial design, implementation, and analysis. It makes me a better and more well-rounded statistician to learn from my peers. It’s great to see how experienced trialists handle and present their study-specific challenges of missing data, intercurrent events, endpoint measurement, multiplicity, and so on. Even small components of the peer-review process are helpful. For example, I saw some creative data visualization in one paper that I’ll keep in mind for my own future work. I must weigh these personal benefits against the time necessary to do things right—a proper review of a protocol, analysis plan, manuscript, and supporting documents may take the better part of a work day.

Outside of Nature Medicine, I get review requests several times a week, and I have to be selective with what I sign up for to maintain balance with my ‘day job.’ I hope our panel is able to regroup on ways to make this program sustainable for Nature Medicine and perhaps emulate it at other prominent journals. While I know we are providing a valuable service to science, I suspect our group of busy statisticians may not be able to maintain a high level of participation long term.

Chen: Reviewing for Nature Medicine has reinforced the importance of balancing statistical rigor with real-world relevance. A technically advanced model isn’t useful if it doesn’t change decision-making or integrate into clinical workflows. Many AI-driven studies focus on incremental performance gains without addressing their practical applicability. As statisticians, we must push beyond accuracy metrics and ask: Does this model make sense in a clinical setting? Is the data it requires routinely available?

Another key lesson has been bridging the gap between AI developers and clinicians. AI researchers often optimize for performance, while clinicians prioritize interpretability and usability. I frequently see studies where analytic methods are strong but the clinical rationale is weak. Our role is to ensure models are designed with meaningful performance metrics, robust validation strategies, and transparent reporting.

Finally, peer review has shown me how different disciplines evaluate research, improving how I communicate statistical insights. Seeing how clinicians, domain experts, and statisticians critique the same study has made me more mindful of how to frame statistical findings for broad audiences—an essential skill for research, teaching, and policy.

Nolan: One of the most valuable insights I have gained is a stronger appreciation for the critical role of statistical and methodological rigor in shaping high-impact medical research. Reviewing manuscripts for Nature Medicine has reinforced the importance of clear study design, appropriate statistical analysis, and transparent reporting. I have seen firsthand how small lapses in methodology can lead to misleading conclusions with potential clinical implications. This experience has also deepened my appreciation for interdisciplinary collaboration. The best-reviewed papers often involve close coordination between statisticians, clinicians, and data scientists, demonstrating that robust statistical design is a central component of high-quality research.

Professionally, serving on this panel has enhanced my ability to critically assess cutting-edge research, stay current with emerging statistical challenges in medical research, and refine my approach to mentoring students and collaborators on best statistical practices.

Ji-Hyun: In your experience, are early-phase oncology trials increasingly moving beyond the traditional 3+3 design for dose determination? While innovative approaches are gaining traction, challenges and pitfalls remain. What trends have you observed, and what recommendations would you offer to improve these methodologies? Even though there are several new approaches (assumed to be better) available, the classic methods are widely and persistently used. Why?

Mithat: Single institutional trials or trials using very expensive treatments like CAR T therapy sometimes have only a few dose levels and deficiencies of 3+3 are not as pronounced in those situations. Some institutions do not have the infrastructure for the continual updating of the dose toxicity relationship that some of the newer methods require. Some of the papers presenting new methods present unreasonable claims, use questionable simulation strategies, etc., that make the users question the authenticity of their findings. Finally, some of us (both the statisticians and the clinicians) are old dogs who are reluctant to learn new tricks. I think these are the reasons why these novel methods did not gain as much traction as they could have.

Cara: It’s the experience every collaborative statistician faces where a PI enters a consultation saying, “but we’ve done it this way for years” or “my colleagues did it this way and published in this great journal.” I have not yet successfully convinced teams to level up their early-phase designs in spite of the evidence I’ve shared about superior alternatives. Things may change slowly with time, patience, building deeper collaborations … and with journals publishing trials using these methods. Perhaps Nature Medicine or others could highlight design features when trials using these methodologies published. The JAMA Guide to Statistics and Methods, for example, pairs a short (usually two pages) statistical explainer with new original research. It’s great to see methods reach a broader audience in this way.

Chen: From my reviews at Nature Medicine and my own work, I’ve seen more early-phase oncology trials move away from the traditional 3+3 design toward adaptive, model-based approaches like BOIN [Bayesian optimal interval] and CRM [continual reassessment method]. These methods are more efficient and precise, making them a better fit for the increasing complexity of oncology drug development. But 3+3 still dominates—not because it’s better, but because it’s simple, familiar, and easy to use. Many trialists see it as “good enough” for small trials, even though it often leads to suboptimal dose selection. Now, this outdated approach is under scrutiny. The US Food and Drug Administration’s Project Optimus has highlighted major flaws in dose selection. Traditional methods tend to push doses too high, focusing on dose-limiting toxicities (DLTs) while ignoring whether lower doses might work just as well. This makes the case for adaptive designs that consider both safety and effectiveness, not just toxicity. Statisticians are key to making this shift happen. We need better models that balance risk and benefit, moving beyond DLT-based decision-making.

To speed up adoption, journals and funding agencies must step up. Journals should set clearer reporting rules, requiring authors to justify their dose-finding method and consider alternatives. Funding agencies should tie grants to better methodological choices, prioritizing trials that use modern designs. More funding for statistical expertise in early-phase trials would ensure these methods aren’t just talked about but actually used. By pushing for better trial designs, statisticians, journals, and funders can help improve oncology research and, ultimately, patient outcomes.

Nolan: Yes, early-phase oncology trials are gradually shifting away from the traditional 3+3 algorithm, but progress has been slow. More adaptive, model-guided methods—such as the Bayesian optimal interval (BOIN) method and the continual reassessment method (CRM)—are increasingly recognized for their statistical efficiency, superior estimation of the maximum tolerated dose (MTD), and flexibility in accommodating complex design challenges, including combination therapies, patient heterogeneity, late-onset toxicities, and multiple endpoints for dose optimization. However, despite the availability of these more advanced approaches, the 3+3 design remains widely used due to several factors:

1. Familiarity and Simplicity: Many clinical investigators and review entities are accustomed to 3+3, and its straightforward implementation makes it appealing.
2. Institutional Barriers: Many clinical trial review committees are hesitant to disapprove studies that use the 3+3 design.
3. Operational Challenges: More advanced dose-finding designs are often viewed as being harder to interpret and implement. They require real-time data monitoring, specialized statistical expertise, and infrastructure for computations, which may not always be available.
4. Historical Precedent: Since many oncology drugs have historically been approved using 3+3-based trials, some sponsors and investigators perceive it as an adequate approach.

To improve dose-finding methodologies, I recommend continued education and advocacy for model-guided designs, particularly through workshops, regulatory guidance updates, and improved software tools that make implementation more accessible. Increased collaboration between statisticians, clinical investigators, and review entities to tailor trial designs to specific research questions will be essential in overcoming these longstanding barriers. Lastly, securing increased funding for statisticians engaged in clinical trial design and implementation can help overcome resource limitations, enabling broader access to novel trial designs and fostering innovation in early-phase oncology studies.

Dear fellow statisticians and data scientists, I cannot end this column without acknowledging the increasingly complex landscape we face. Yet, despite these challenges, I deeply appreciate the resilience, creativity, and dedication of the ASA community. By advocating for the essential role of research in addressing global challenges, we can navigate this landscape together.

Thank you for being part of our ASA community.