Recommendations of the American Cancer Society workshop on design, conduct, analysis, and reporting of multicancer early detection trials with late‐stage incidence end points and post‐trial ongoing evaluation—The Peachtree Consensus

Abstract

Background

In a new era of multicancer early detection (MCED), late‐stage incidence has been proposed as an end point for the evaluation of test efficacy, marking a departure from the established end point of cancer mortality.

Methods

The American Cancer Society convened a panel of 15 academic researchers with expertise in cancer screening evaluation to develop principles assuring the validity and interpretability of MCED trials with late‐stage incidence end points and advancing implementation of tests with potential for meaningful clinical utility.

Results

The definition of late‐stage cancer is critical and affects study design, interpretation, and outcomes. The authors of the Peachtree Consensus recommend considering cancer‐type–specific definitions of late stage and paying attention to comparability of staging intensity in both trial arms. The duration of screening and the follow‐up interval after the last screen should be chosen based on the preclinical early stage and late‐stage durations of the target cancer types. The authors also recommend reporting aggregate results and results for individual cancer types as numbers permit and summarizing relative and absolute benefits. Predicted mortality reductions should be calculated to contextualize a late‐stage result and the inputs and assumptions involved should be reported. Given a significant late‐stage reduction suggestive of meaningful clinical benefit, the authors support launching consortium and demonstration studies while continuing to track trial mortality outcomes.

Conclusions

Trials with late‐stage incidence end points that are conducted using these principles should support the development of evidence‐based screening guidelines as well as the creation of accessible data resources for observational and modeling studies.

As the pace of technological development of new cancer screening tests quickens, it is imperative that research is designed to deliver findings faster. The resulting data can serve as the foundation for proof‐of‐principle demonstration projects (or in‐service evaluations) for further data gathering to guide implementation decisions. The route to consensus on the implementation of effective new screening technologies will be faster if investigations adhere to these recommended principles from the Peachtree Consensus for trial design, reporting results, and post‐trial considerations.

INTRODUCTION

For much of its history, the American Cancer Society (ACS) has played a leading role in promoting evidence‐based early cancer detection and both informing and developing population screening recommendations to reduce cancer mortality and morbidity. Screening efforts for breast, cervical, colorectal, lung, and prostate cancer have significantly reduced morbidity and mortality from these cancers, ¹ but a majority of cancer deaths each year still occur from cancers with no early detection strategy. The field of early cancer detection is now undergoing a fundamental shift. Many new blood‐based screening tests for cancer are in development, including multicancer early detection (MCED) tests that screen for multiple target cancers. The National Health Service (NHS)‐Galleri trial, the first randomized screening trial of an MCED test, recently reported primary results. ²

The established standard of randomized screening trials that use mortality as the primary end point leads to studies that are costly and lengthy, delaying implementation of new life‐saving technologies and even yielding results that are already outdated at trial completion. ³ , ⁴ Given the multiplicity and rapid development of new tests, there is substantial interest in identifying alternative, short‐term end points that could shorten the duration to first reporting and most likely reduce the cost of screening trials. ⁵ Several single‐cancer trials have been designed with late‐stage incidence as the primary end point, ⁶ , ⁷ and one breast cancer screening trial used predicted mortality (PM) based on stage and pathologic features at diagnosis. ⁸ The NHS‐Galleri trial was designed with late‐stage (stage III–IV) incidence as the primary end point, with plans for extended follow‐up to eventually report mortality outcomes. ⁴ , ⁹ This means that the field must now reckon with, validly interpret, and respond to findings from at least one MCED trial with an incidence‐based primary end point. New standards are needed to help ensure that these trials produce valid inferences concerning screening efficacy and that these results can be appropriately interpreted to quantify clinical utility. ¹⁰ The urgency of this need is evidenced by parallel efforts from the UK’s National Screening Committee to codify the evidence requirements in studies evaluating multicancer screening tests and to adjudicate the acceptability of surrogate end points in cancer screening trials. ¹¹ , ¹² The UK National Screening Committee’s position paper on surrogate end points focuses largely on whether end points such as late‐stage incidence might be acceptable and concludes that, as the as the definitive trial outcome, mortality cannot yet be replaced by surrogate outcomes to determine benefit. However, given that the NHS‐Galleri trial has already set a precedent by reporting on late‐stage incidence as a primary outcome, the time has come to formulate standards for the conduct, analysis, reporting, and interpretation of such trials.

For these reasons, the ACS convened an international workshop of screening experts and thought leaders in the field with the objective of developing specifications for MCED trials with late‐stage incidence end points. Workshop participants also considered ways in which such trials might inform investment in companion population studies or screening programs to investigate how promising tests might be implemented. This document reports on the consensus recommendations from the ACS workshop.

Mortality and late‐stage incidence as end points in cancer screening trials

A central premise of the ACS workshop was that the primary objective of cancer screening is to reduce cancer‐specific mortality. However, there was also a consensus that the primary pathway to reduced mortality is likely to be largely through a reduction in the extent of disease at diagnosis, which also is likely to reduce other adverse outcomes, including treatment morbidity. Stage‐specific incidence represents a summary of extent of disease at diagnosis; therefore, we considered whether the use of late‐stage incidence as a primary trial end point could be justified by available evidence that it represents a surrogate end point for cancer‐specific mortality in screening trials. A key criterion for surrogacy of an intermediate end point is that it should be on the causal pathway between the intervention and the primary end point. In the case of screening and late‐stage incidence, this means that the effect of screening on late‐stage incidence, together with stage‐specific survival, would determine the effect of screening on cancer deaths (Figure 1). Whether reductions in mortality caused by screening are predicted by effects of screening on late‐stage incidence has been investigated empirically in meta‐analyses of previously conducted single‐cancer screening trials. ¹³ , ¹⁴ , ¹⁵

FIGURE 1.

Conceptual model of screening reducing late‐stage diagnoses, which, when combined with stage‐specific cumulative incidence of cancer death, translates into reducing cancer‐specific deaths. Source : Core 8 catchment areas of the Surveillance, Epidemiology, and End Results program for cases of known and staged cancer types (i.e., excluding brain and spinal cord, leukemia, and plasma cell myeloma) diagnosed in 2018. A hypothetical 20% reduction in late‐stage diagnoses because of screening was assumed.

Despite limited numbers of trials for each cancer site, the meta‐analyses generally demonstrated a positive correlation between the observed late‐stage incidence reduction and the observed mortality reduction across trials. The association between late‐stage incidence and mortality varied, depending on the type of cancer and the follow‐up time at which the correlation was estimated. The estimated associations were more modest when late stage was defined as stage IV rather than combined stages III and IV, ¹³ , ¹⁴ , ¹⁵ suggesting that the strength of a potential surrogate relationship could depend on how late stage is defined. Although the empirical meta‐analyses were highly informative, a recent simulation‐based study showed that it is possible that such analyses may fail to identify a surrogate relationship even if one exists. ¹⁶ The study concluded that causal reasoning backing a surrogate relationship may be more reliable than empirical meta‐analysis.

We concluded that both empirical evidence and mechanistic reasoning support considering late‐stage incidence as the primary outcome in a screening trial; however, it is premature to assert that it could serve as a surrogate for cancer mortality for every cancer type and every screening modality. Although we considered alternatives to late‐stage incidence as the primary end point, including PM, we concluded that late‐stage incidence was currently the most transparent and acceptable option. At the same time, we acknowledged the limitations of a simple binary classification of stage and its inability to reflect within‐stage advances in diagnosis or within‐stage shifts that might also affect mortality.

Defining late stage and documenting diagnostic intensity

Both epidemiologic and clinical considerations factor into the decision about how to define late stage. There should be a marked difference in survival between late‐stage and early stage disease. Also, late‐stage diagnoses should account for a non‐negligible proportion of cancer deaths. From a clinical perspective, the definition of late stage should reflect cancers for which treatment generally is either ineffective (or substantially less effective) or significantly less tolerable, leading to poorer survival and/or higher morbidity. Other considerations in defining late stage may include whether treatment with curative intent is an option, and this definition would likely include both interstage and intrastage tumor characteristics. One could define late stage as IV or as combined III and IV as has been considered in previous studies ¹⁷ , ¹⁸ , ¹⁹ ; for example, a specific reduction in stage IV will have different clinical implications if it is accompanied by an increase or a decrease in stage III. However, neither of these definitions reflects the clinical heterogeneity of how cancers are diagnosed and managed. For instance, late‐stage pancreatic cancer might best be defined as unresectable disease (typically, American Joint Committee on Cancer stage ≥IIB; Figure 2), whereas late‐stage ovarian cancer might best be defined as combined stages III and IV because the clinical management is often indistinguishable and survival is poor for both of these stages. In contrast, colon cancer outcomes justify classifying only stage IV as late stage because stage III colon cancer is typically treated with curative intent; in contrast, stage IV colon cancer is typically incurable. Because of this heterogeneity, we recommend cancer‐type–specific definitions of late stage for each cancer targeted by an MCED test. Furthermore, because stage‐based outcomes are not well defined for cancer types that are not stageable (e.g., brain and spinal cord, plasma cell myeloma, most types of leukemia, and some lymphomas), we recommend excluding these cancer types from the definition of the primary end point. We recognize that within stage, survival may be quite heterogeneous because of other clinicopathologic features like disease subtype, tumor size, and nodal involvement that are associated with outcomes. ²⁰ For example, in a study of the effect of prostate‐specific antigen screening on late‐stage incidence in the European Randomized Study of Screening for Prostate Cancer, metastatic disease was defined by either M1 disease on imaging or by a prostate‐specific antigen value >100 ng/mL. ²¹ In cancers for which these features are important, alternative definitions of late stage that accommodate them should be considered.

FIGURE 2.

Pancreatic cancer‐specific survival stratified by (left) American Joint Committee on Cancer (AJCC) stages I–IV and (right) resectable status. Source: Core 8 catchment areas of the Surveillance, Epidemiology, and End Results program for cases diagnosed in 2018–2022. Survival for AJCC stage II may be higher than for stage I because of treatment selection, pathologic upstaging (e.g., finding positive nodes) on surgery, or other artifacts. Potentially resectable was defined as AJCC stages ≤IIA, and likely unresectable was defined as AJCC stages ≥IIB.

In general, we expect clinical staging to be more reliably consistent across patients and institutions despite being subject to misclassification compared with pathologic staging. Pathologic staging may be more predictive of mortality than clinical staging, but it is not always recorded; furthermore, neoadjuvant therapies can complicate the interpretation of pathologic stage at diagnosis. Ideally, both clinical and pathologic (where available) stages should be recorded in addition to information about neoadjuvant immunotherapy, chemotherapy, or radiation before surgical resection. Rules for defining stage in patients receiving neoadjuvant therapy should be applied equally in both arms.

Although it might seem desirable to establish rigid staging standards to promote comparability across trial arms, we do not anticipate this will be feasible given variations in patterns of care and availability of diagnostic technologies. We instead recommend pursuing a de‐minimus staging standard for all cases: for example, including a physical examination, standard tumor markers, and body imaging of the chest, abdomen, and pelvis. However, we expect the modalities used for staging to vary across sites based on access to imaging facilities and insurance coverage policies, which may vary by country/region. Therefore, documentation of tests done for staging (including imaging examinations, biomarkers, procedures, and biopsies) and their results will be needed to assess the comparability of staging intensity between trial arms. Summaries of these data could include distributions of numbers and types of diagnostic tests for screen‐detected and interval‐detected cancers in the screen arm and for cancers in the control arm. In addition, we recommend specifying a target upper limit on the interval between diagnosis and staging (e.g., 3 months) and documenting the time to staging by mode of diagnosis (screen vs. interval cases).

Design, analysis, and reporting considerations for late‐stage incidence trials

Design and analysis considerations

Screening duration and length of follow‐up after the final screen

The goal to complete trials quickly by using the shorter‐term end point of late‐stage diagnosis is at odds with the known dynamics of early cancer detection tests, which, by their nature, advance the timing of diagnosis and inflate the incidence (including the late‐stage incidence) observed in the screening arm during the screening period. ²² This inflationary effect is more pronounced when the preclinical late‐stage duration is longer and may manifest as a surge because of the detection of prevalent disease that paradoxically increases late‐stage diagnoses in the screen arm. Later screens may also be affected in a similar manner but to a lesser extent than the initial screen. Thus a sufficient number of screens and sufficient follow‐up after the last screen are required to overcome this inflationary effect. At the same time, trial designs should not make the follow‐up so long that the late‐stage reduction becomes heavily diluted by cases in both arms diagnosed after the end of the screening rounds.

Decisions about the screening interval, number of screening rounds, and follow‐up after the last screening round will be a critical determinant of the estimated screening efficacy and should be based on what is known about the detectable preclinical durations of the target cancer types. This may be informed by analyses of banked prediagnostic blood samples if available. These durations will vary across cancer types, often substantially, making determination of the optimal screening interval and number of screens to conduct a challenge. A modeling study ²³ demonstrated that, under longer detectable early stage durations, reductions in late‐stage diagnoses were more pronounced but took longer to manifest. In this case, too few screening rounds may underestimate late‐stage incidence reductions that would only become evident after more screening rounds. In the absence of prior knowledge of the range of detectable early stage durations for the target cancers, we recommend being conservative and erring on the side of more rather than fewer screening rounds to improve the ability to identify a potentially beneficial effect.

Analyses to increase statistical power

We recommend that MCED trials collect and store blood samples from both control and screen arm participants to increase trial power and reduce required sample sizes. ²⁴ Because the effects of screening are concentrated among individuals who trigger a positive signal, comparing outcomes between ever‐positives across both arms—using either intended‐effect ²⁵ or targeted‐analysis ²⁶ frameworks—can yield greater efficiency than analyses comparing outcomes among all participants on each arm. These frameworks can be applied to both late‐stage incidence and mortality end points. The key assumptions for both analyses are that there is (1) no effect of a negative screening test on the outcome of interest, (2) no loss of signal from storage at the time of analysis of control arm specimens, and (3) no differential noncompliance with specimen collections between the arms. The last of these assumptions requires that participants be blinded regarding their assigned arm and unblinded only if they have a positive result in the screening arm.

Reporting results of late‐stage incidence trials

Aggregate versus per‐cancer reporting

We support the use of aggregate end points for MCED trials, but we also recommend reporting disaggregated, site‐specific results to identify which cancer types contribute most to any observed benefit. Reporting of histology‐specific or etiology‐specific results may also be appropriate when sample size and data permit. The results will be descriptive rather than definitive because we expect low statistical power for individual cancer types or subtypes in MCED trials designed to identify a specified aggregate result ²⁷ ; therefore, the single‐cancer results should be interpreted with caution. Reporting results for cancer types with versus without existing screening tests may also be appropriate.

Subgroup analyses

To explore possible moderating effects of structural disparities on late‐stage incidence, we recommend a descriptive summary of outcomes stratified by prespecified socioeconomic and demographic factors, such as insurance coverage (in US studies), level of education, and a validated instrument for deprivation. Ideally, key factors of interest will be prespecified and used in stratified randomization. This will facilitate consideration of whether the efficacy of an MCED test is consistent across diverse populations or whether its performance is modulated by systemic variations in access to care in addition to staging intensity.

Contextualizing reported late‐stage incidence reductions for assessment of potential clinical utility

The clinical implications of an observed reduction in late‐stage incidence in an MCED trial will rest on the targeted cancers and the definition of late stage. Although there may be a desire to define a single threshold for adequacy of a reported late‐stage incidence reduction to conclude that a test is likely to be clinically useful, this presents a challenge in the multicancer setting. These considerations motivate us to consider ways to contextualize the reported result to enhance interpretation of the implications for clinical utility.

When reporting results from MCED trials with late‐stage incidence end points, we recommend reporting absolute and relative reductions in late‐stage incidence because the same relative reduction may have very different implications for the absolute number of individuals benefitting. A common measure reflecting absolute benefit is the number needed to screen to avert one late‐stage diagnosis. We acknowledge that these figures represent conservative estimates of long‐term benefits under continued screening. The use of absolute measures is important for MCED screening because the absolute benefit is the sum of the benefit across cancer sites, whereas the relative benefit is an average of the relative benefits at each site included in the test. Absolute measures are more meaningfully comparable across different tests but are not generalizable across populations with different incidence and mortality rates for the targeted cancers. We also recommend reporting results by screening round (and mode of detection), not only cumulatively over the study duration.

A clinical benefit expected from reduced late‐stage incidence reduction is a decline in intensive systemic treatments, which carry toxicity and implications for quality of life. Therefore, one metric for interpreting the results of a late‐stage incidence trial is the frequency of intensive systemic treatments in the screen and control arms. We recommend that the specific treatments provided be recorded similarly across trial arms and summarized.

Perhaps the ultimate metric for contextualizing a late‐stage reduction is the corresponding effect on cancer mortality. In published meta‐analyses, ¹³ , ¹⁴ , ¹⁵ cancer mortality reductions were generally more modest than the corresponding late‐stage incidence reductions. While waiting for mortality results, predicting mortality reduction based on the observed stage distribution could help to assess whether the observed late‐stage reduction is likely to be clinically meaningful. PM combines observed stage distributions with and without screening with predicted stage‐specific survival. The predicted stage‐specific survival is usually based on survival in the absence of screening either from registries, historical data, or the trial’s control group and can be formulated to incorporate dependence on clinicopathologic features other than stage. We caution that this may not reliably represent survival by stage in the presence of screening, particularly if the screening test preferentially identifies more aggressive disease, as has been reported, for example, in some MCED studies. ²⁸ By its nature, PM involves a predictive model; different models may produce different results. Consequently, we recommend that PM be used to contextualize a reported late‐stage incidence result, but care should be taken before interpreting it as a measure of clinical utility. In addition, any reported PM should explicitly include documentation of its methodology, inputs, and assumptions, supplemented by robust sensitivity analyses to quantify uncertainty.

Population studies to inform clinical utility after trials with late‐stage incidence end points

To move screening technologies more swiftly from discovery to implementation, the ACS workshop participants emphasized the role of demonstration projects, observational cohort studies, and modeling studies, which collectively inform stakeholders on the real‐world performance, logistics, and outcomes of new screening programs.

Demonstration projects and observational cohorts

In the late 1970s, the Breast Cancer Detection Demonstration Project provided critical age‐specific performance data on mammography after early results from the Health Insurance Plan of Greater New York trial. ²⁹ , ³⁰ Similarly, the National Cancer Institute’s Breast Cancer Surveillance Consortium, established in 1973, continues to serve as an authoritative resource for evaluating breast cancer screening performance and diagnostic outcomes within routine clinical practice. ³¹ , ³² Also of interest are the randomized implementation projects, such as the AgeX trial in England ³³ and the randomized expansion of colorectal cancer screening in Finland. ³⁴ Such programs, conducted alongside or after publication of MCED trials, can provide essential data on patient, provider, and system‐level factors that influence screening effectiveness in diverse populations. These projects should ideally include public–private partnerships for data sharing and transparency; demographic, economic, and health systems diversity; timely and high‐quality results; and opportunities for large‐scale collaboration. ²⁵

Modeling and cost‐effectiveness studies

Modeling studies extend trial findings by simulating long‐term outcomes across broader hypothetical populations. These studies first learn rates of cancer development and progression in the absence of screening based on incidence patterns in the screen and control arms. They then use the estimated rates to simulate screening scenarios and time horizons that differ from the trial as conducted. Inputs, such as the costs of screening, diagnostic confirmation, treatment, and anticipated harms, may be added to the simulations for cost‐effectiveness analyses. Screening trials represent a uniquely valuable resource for learning and validating models ³⁵ ; making these data accessible for modeling purposes will greatly enhance the opportunity to learn about disease natural history and screening approaches that are likely to produce advantageous benefit‐harm tradeoffs. Independent validation of models is critical because model projections depend on assumptions and inputs; therefore, we recommend developing pathways for sharing trial data and modeling software to support independent modeling efforts.

Regulatory and coverage mechanisms providing structured pathways to ongoing evidence generation

In the United States, screening and diagnostic tests are classified as medical devices and are subject to regulatory requirements for safety and effectiveness as judged by the US Food and Drug Administration (FDA). In the case of diagnostic tests, the FDA has stated that analytical and clinical validity are in their purview, whereas clinical utility is not. For any diagnostic test that the FDA regulates, there is potential for mandated post‐approval studies. ³⁶ , ³⁷ Ideally, such studies could also provide an opportunity for public–private partnerships that can further contribute to our understanding of MCED test performance and outcomes.

In the United Kingdom, before implementation, screening tests require evaluation by the UK National Screening Committee, which requires both trial outcomes and health economic modelling. Furthermore, there is the potential in the United Kingdom to create studies that examine screening implementation issues through large‐scale, in‐service evaluations of screening.

In the United States, there is a separate evaluation of screening and diagnostic tests to obtain coverage by the Centers for Medicare and Medicaid Services, which also offers mechanisms for studies of new technologies under the national coverage decision process. One potential result of a national coverage decision review could be the application of coverage with evidence development. This mechanism links reimbursement to prospective data collection, addressing uncertainties regarding real‐world utility and implementation. In the United Kingdom, a large lung cancer screening implementation program invited over 2 million people for chest computed tomography screening and confirmed population effectiveness within 5 years, particularly in regions with high rates of socioeconomic deprivation. ³⁷ This initiative has provided not only tools for implementation of lung cancer screening but also valuable data for modeling and cost‐effectiveness studies.

The ACS workshop participants supported beginning robustly designed adjunct and complementary studies after findings for late‐stage incidence that are strongly indicative of potential clinical utility. Such studies could involve regulatory agencies and/or large health systems and should aim to collect high‐quality data on test performance and outcomes, identify implementation challenges and best practices, and make the data collected accessible to the broader research community.

CONCLUSION

In the past decade, the prospect of novel multicancer screening tests has evolved rapidly. What has not moved rapidly is the development of a consensus on how we can best realize the potential of these novel technologies, which, if proven to be effective, could reduce the burden of cancer beyond what is attainable by current screening strategies.

One goal of the ACS workshop was to propose principles to strengthen the validity of trials with late‐stage incidence end points given concerns about the time frame typically required for trials with mortality end points. A second goal was to highlight the value of complementary studies focused on clinical utility, best practices, and implementation challenges so long as the late‐stage incidence data from trials provide confidence that the test is likely to lead to meaningful clinical utility and the safety data indicate acceptably low concerns and harms.

Workshop participants agreed that there is justification for considering late‐stage incidence as a primary aggregate outcome of an MCED trial; however, at this time, results based solely on reductions in late‐stage incidence are insufficient to serve as a basis for population screening recommendations. Conversely, in some settings, lack of an observed stage shift may be meaningful from a policy perspective. We emphasized that late stage is an incomplete concept—how late stage is defined not only affects trial duration and design but also has major implications for interpretation and quantification of clinical utility. Contextualizing an observed late‐stage reduction to better understand its implications for clinical utility will be an important part of interpreting and reporting the results of trials with late‐stage incidence as the primary end point. We agreed that clinically important and statistically significant late‐stage incidence reduction could be sufficient to launch large complementary studies of test performance, best clinical practices, and implementation strategies while continuing to follow up the trial for mortality to confirm that benefits outweigh harms.

We concluded that, by paying attention to the principles outlined for conducting, analyzing, and interpreting trials with late‐stage incidence end points and by investing in efforts to conceptualize, prepare for, and support complementary studies, we can establish a new pipeline for evaluating novel screening tests that will expedite the time from development of an effective screening test to its clinical implementation (see Table 1).

TABLE 1.

Recommendations of the American Cancer Society Peachtree Consensus workshop.

1. The definition of late‐stage cancer as a trial end point has critical implications for study design and interpretation. We recommend cancer‐specific definitions of late‐stage incidence based on epidemiologic and clinical considerations that are practical and can be implemented in the context of large‐scale clinical studies. To maintain the integrity of the late‐stage incidence end point, we recommend excluding non‐stageable malignancies.

2. Efforts to ensure comparability of staging practices in both trial arms will be important to limit bias in the reported late‐stage reduction. We recommend specifying a minimal staging protocol and a target maximum time between diagnosis and stage determination. We also recommend summarizing staging pathways and timing in both arms.

3. The interval between screens and the follow‐up interval after the last screen should be chosen based on the preclinical early stage and late‐stage durations of the target cancer types; this choice is complicated in the multicancer early detection setting, in which these intervals may vary across cancer types. We recommend erring on the side of doing more screening rounds and allowing an appropriate follow‐up interval after the last screen to facilitate identifying a potentially beneficial effect.

4. We recommend reporting late‐stage reductions for aggregated and disaggregated cancer types, recognizing that per‐cancer estimates will be descriptive and should be interpreted with caution.

5. We recommend summarizing both relative and absolute reductions in late‐stage incidence, recognizing that these estimates, computed over a short‐term screening program, will be conservative for longer term programs.

6. We recommend calculating predicted mortality reductions to contextualize a late‐stage incidence result, reporting inputs and assumptions made in the calculation, and conducting informative sensitivity analyses.

7. If a significant late‐stage reduction suggestive of meaningful clinical benefit is observed, we support launching consortium studies and demonstration projects to guide potential implementation while continuing to follow for mortality outcomes in the trial. We also recommend developing accessible data resources for modeling and cost‐effectiveness studies.

AUTHOR CONTRIBUTIONS

Ruth Etzioni: Conceptualization; writing—original draft; writing—review and editing. Larry Kessler: Conceptualization; writing—original draft; writing—review and editing. Deb Schrag: Writing—original draft; writing—review and editing. Peter Sasieni: Writing—original draft; writing—review and editing. Hilary A. Robbins: Writing—original draft; writing—review and editing. Hormuzd A. Katki: Writing—original draft; writing—review and editing. Stephen W. Duffy: Writing—original draft; writing—review and editing. Roman Gulati: Writing—original draft; writing—review and editing. Diana Buist: Writing—original draft; writing—review and editing. Sean Tunis: Writing—original draft; writing—review and editing. Rebecca Landy: Writing—original draft; writing—review and editing. Jane Lange: Writing—original draft; writing—review and editing. Alpa V. Patel: Writing—original draft; writing—review and editing. William L. Dahut: Writing—original draft; writing—review and editing. Robert A. Smith: Conceptualization; writing—original draft; writing—review and editing; funding acquisition; project administration.

CONFLICT OF INTEREST STATEMENT

Ruth Etzioni reports personal/consulting fees from Guidepoint Global and Oregon Health Sciences University and stock ownership in Seno Medical outside the submitted work. Deb Schrag reports grants/contracts from GRAIL Inc. outside the submitted work. Peter Sasieni reports personal/consulting or advisory fees from GRAIL Inc. and MEDIAN Technologies outside the submitted work. Stephen W. Duffy reports personal/consulting fees from GRAIL Inc. outside the submitted work. Diana Buist reports personal/consulting fees from Delfi Diagnostics and Natera outside the submitted work. Sean Tunis reports grants/contracts from the Laura and John Arnold Foundation and the MCED Consortium; and personal/consulting fees from GRAIL Inc. outside the submitted work. Jane Lange reports personal/consulting fees from GRAIL Inc. outside the submitted work. The remaining authors disclosed no conflicts of interest.