Abstract
Externally controlled trials have commonly been used when conducting a randomized controlled trial (RCT) is not feasible or ethical. By allowing the study of new treatments, use of external controls can lead to accelerated advances in the management of rare diseases. The use of external controls, however, introduces new challenges due to potential differences between the population the external controls are enrolled from and the population the patients on the new trial are enrolled from. Some differences include, but are not limited to, differences in how patients are diagnosed and treated, differences in the case mix of the underlying populations, differences in the ability to measure outcomes, and differences in data collection. We discuss the potential benefits and challenges of externally controlled trials, as well as strategies to mitigate bias, including the estimand and target-trial emulation framework. We also provide a brief overview of statistical methodology commonly used in these settings. We note that although the strategies presented may help mitigate some of these challenges, they cannot replace an RCT framework, and investigators should be aware of the potential limitations of externally controlled trials.
Introduction
Randomized controlled trials (RCTs) allow estimation of causal intent-to-treat treatment effects under mild assumptions.1 Because patients are randomly assigned to each arm, observed as well as unobserved variables are expected to be well balanced between the different study arms, and any differences in the outcomes are assumed to be caused by the treatment. It may not always be feasible to conduct an RCT, however. In rare disease settings, for example, low enrollment rates may make it difficult or impossible to adequately power a study that can be conducted within a reasonable time frame. In other settings, such as disease areas with a high unmet need, patients may not be willing to be randomly assigned.2 In externally controlled trials, outcomes from non-randomly assigned participants who received the standard of care treatment are compared with outcomes from the experimental treatment arm, either alone or in combination with concurrent randomized controls. Because external controls serve as a comparison group for the experimental treatment arm, it is crucial that the external control population is carefully selected.
By definition, external controls cannot be randomized, so causal intent-to-treat treatment effect estimation in this setting relies on stronger assumptions than when using concurrent randomized controls. In these settings, the selection of the external control group, appropriate study designs, and analytic methods are crucial to obtaining unbiased estimates of the causal treatment effect.3 Even in well-designed studies, it may be impossible to control all factors that can induce bias (eg, confounders or biases related to defining start of follow-up). Because of this, results from studies with external controls are more speculative than when a concurrent randomized control group is used, and governing agencies often recommend against the use of external controls unless no other viable option exists and a suitable external control population exists.4 In this article, we discuss the benefits and risks of externally controlled trials, as well as possible solutions to mitigate risk of bias. We also discuss considerations for the selection of the appropriate control group and briefly discuss designs that combine concurrent randomized controls with external controls.
What are external controls?
The guideline developed by the International Conference on Harmonisation of Technical Requirements for Registration of Pharmaceuticals for Human Use (ICH) defines external controls as “patients who are not part of the same randomized study as the group receiving the investigational agent.”5 External control data can come from a variety of sources, including previous trials, registries, or electronic health records.6 External controls coming from an earlier time period than study participants are referred to as historical controls.7 To differentiate from external controls, we will refer to patients randomized to the control group of the study that the treatment group is a part of as concurrent randomized controls. Here, patients enrolled to other studies taking place at the same time are also considered external controls.
Although the use of external controls may make otherwise impossible studies feasible, it also introduces many challenges to the interpretation of study results. Specifically, the use of external controls can lead to bias due to lack of randomization and blinding. Because of the potential pitfalls, the Food and Drug Administration (FDA) recommends the use of other study designs when possible. If other designs are not possible, they ask sponsors to consult with the FDA at early design stages to confirm appropriateness of an externally controlled trial.7 In general, the use of external controls has previously been limited to rare disease settings, diseases with no available treatments, and areas with a high unmet clinical need.8 Furthermore, existing guidance reserves the use of externally controlled trials for settings in which the effect size is expected to be large and in scenarios where the selected endpoint is objective, such as overall survival, so as not to introduce additional biases.5,7 A review of decisions by the FDA and the European Medicines Agency between January 1, 1999 and May 8, 2014 showed that 76 treatments were approved without an RCT, with some obtaining approval based on results from externally controlled trials, and most occurring for hematologic malignancies and oncology.9 External controls have also been used for various diseases within Children’s Oncology Group. In ACNS2031, for example, the efficacy of sodium thiosulfate for the reduction of hearing loss will be evaluated by comparing enrolled patients to a historical control group from a previous trial. ACNS1723, a phase II study of dabrafenib with trametinib in newly diagnosed BRAF-mutant high-grade glioma patients, will compare event-free survival using a historical control group as the control arm.
In some settings, the use of external controls may prove to be more challenging, and in some cases not appropriate. The National Cancer Institute Community Oncology Research Program (NCORP) aims to bring cancer clinical trials to people in their communities with an emphasis on representation of participants from minority/underserved communities who are often underrepresented in clinical trials. Because of this, the populations enrolled to NCORP studies are often different from those in other trials, affecting the ability to find a suitable external control group (eg, they might be more compatible with data from electronic health records than trial data). Additionally, trials within the NCORP focus on cancer screening and prevention, health-related quality of life, and cancer care delivery, resulting in endpoints that can be more affected by the lack of blinding and randomization in externally controlled trials. In scenarios where the use of external controls is not appropriate, study designs incorporating external control groups can limit advances. For example, for diffuse midline glioma, a rare fatal disease with limited advances in therapy, several single-arm trials with a historic control group have been conducted to evaluate the effect of ONC201. Hansford et al. criticized the reliability of the historical control comparisons and advocated for a trial with a concurrent randomized control group.10,11 The following sections focus on design considerations for the proper use of external controls.
Estimand and target-trial emulation frameworks
If an externally controlled study is deemed necessary, the appropriate study design and selection of external controls is crucial to reducing bias. Two frameworks that are particularly useful when designing externally controlled trials are the estimand and the target-trial emulation frameworks.12-15 The estimand framework requires investigators to define a scientific question that can be translated to a statistical estimand (ie, the quantity that is the target of estimation and answers the scientific question, commonly some version of a treatment effect in treatment trials). With these in mind, one then determines the population of interest, treatments, the primary endpoint(s), intercurrent events, and the estimator to be used. The target-trial emulation framework aims to minimize the limitations of observational studies by designing a study that mimics an RCT as closely as possible with available data.12 Careful use of the two frameworks together can allow investigators to reduce bias of estimation of the estimand of interest in externally controlled trials.
Selection of external controls
One important consideration in both the estimand and the target-trial emulation frameworks is the selection of the external control group. Below, we discuss the considerations introduced by Pocock (1976) for the selection of external controls.
Eligibility
The first consideration is the selection and assessment of eligibility criteria. In an RCT, eligibility criteria determine the target population for the trial. In externally controlled trials, however, inconsistent eligibility criteria may lead to incomparable groups and, therefore, biased results.3,13 External controls are often participants from other studies with potentially different eligibility criteria. If individual participant-level data are available for external controls, these can be used to ensure that participants included satisfy the current criteria. In some scenarios, however, the required data may not have been collected. Consider a historical control trial where a new biomarker has been identified and is necessary to confirm patient diagnosis. If the biomarker was not discovered until after the historical trial, biomarker information will not be available, making it impossible to assess whether historical controls meet the current study’s criteria. In these settings, the study committee must decide whether the eligibility criteria for the study should be modified, or if a new control group must be selected. Although it may be tempting to use all available information, it is also important to ensure that only data that would have been available at diagnosis are used to assess eligibility. Otherwise, this too, may induce bias.16 Additionally, when individual patient-level data are not available, it can be impossible to confirm that the patients on the previous study meet the new study’s eligibility criteria.
Treatment assignment, endpoint selection, and confounding
Other aspects of study design are the selection of the treatment for the control arm, the outcome, and evaluation of treatment response. If an RCT design were to be selected, participants in the control arm would typically receive the current standard of care. To emulate an RCT, the selected external control group must have also received the protocol-defined standard of care. Other considerations related to treatment include changes in the availability and experience administering the treatment, as well as adherence. To the latter point, intention-to-treat analyses are common in RCTs; however, in most cases external controls will have been selected because they received the standard of care and therefore the intention-to-treat framework differs from that of typical RCTs.
RCTs may be blinded such that participants do not know what treatment arm they were assigned to until after participating in the study, reducing the potential for Hawthorne effects, whereby knowledge of the treatment assigned affects subsequent behavior in ways that influence the outcome. In externally controlled trials, participants may have known what treatment they were receiving, particularly if they were not part of an RCT. To reduce the potential for bias, endpoints should ideally be objective. Furthermore, the way that the endpoint is measured or defined must be the same for trial participants and external controls.3,12 Oftentimes, for example, the criteria for a relapse may change based on newly available information, such as the discovery of new biomarkers or newly available technology. This information may change the rates of the observed outcome. If at the time of the new trial, available information allows for better identification of the event, participants in the experimental arm may look different from the external control group.
Additionally, because participant characteristics may not be well balanced between the treatment arms, the trial design must specify a priori selected potential confounders and statistical methods for confounder adjustments. This will ensure that required information is collected in the current trial, and that it is readily available for external controls. As with eligibility criteria, one challenge encountered is availability of data. Even if data are collected for all known confounders, the existence of unmeasured confounders would make adjustment for all confounders impossible.12 It is not possible to determine whether all confounders have been accounted for using the available data, and sensitivity analyses are often conducted to assess the potential impact of unobserved confounding. When considering a single-arm study comparing to summary-level information from prior studies, comparing characteristics of the study samples and adjusting for confounding may be difficult or impossible to do.
Timing and follow-up
There are several considerations related to time that are important for the design of externally controlled trials. One such decision is the selection of “baseline” for external controls. It is important that at this timepoint, external controls satisfy the study’s eligibility criteria. A common baseline timepoint is initial diagnosis. If information regarding initial diagnosis is available and eligibility criteria are satisfied, defining baseline for external controls is straightforward. In some cases, however, eligibility criteria may allow for more flexibility in the timing of enrollment. This also increases flexibility for external controls, making the selection of a baseline timepoint unclear. As discussed by Hernan and Robins (2016), under some scenarios patients from external control populations may be eligible at multiple timepoints. Consider, for example, a trial that allows the enrollment of patients at any timepoint after diagnosis. In this scenario, multiple timepoints may satisfy the definition of baseline, or time zero, for external controls. Various options have been used for the selection of time zero.9,12,16,17
Another consideration when designing externally controlled trials is frequency of follow-up. The follow-up times should be similar for the external control group and for participants in the experimental arm. Although this study design aspect can be accounted for, there is still potential for surveillance bias, in which patients who are more sick are seen more often. One should be aware that if this is differential by arm, it may also lead to bias.13
Environment
The criteria discussed by Pocock (1976) also require that patients were treated at the same institution and with largely the same clinicians as those participating in the current trial.3 This is done to mitigate differences due to care provider and site; however, this criterion may be difficult, if not impossible, to satisfy, particularly when using historical controls. Similarly, current events may affect patient behavior, including study accrual and retention. These may be impossible to replicate and account for in externally controlled trials, leading to potential biases.
Incorporating information from external controls
Up to this point, we have discussed considerations for the selection of external controls in clinical trials. Statistical methodology has been developed for use in these settings, with a focus on alleviating some of the challenges related to bias induced by differences in the population underlying the treatment and the control arms. In this section, we provide a brief overview of existing methodology. For more detailed information, we refer the reader to the references provided in this text.
External controls can be included alone or in combination with concurrent randomized controls. Combining external and concurrent randomized controls can be done only when concurrent controls are available—that is, there exists the ability to randomly assign patients to different treatment arms so that a group of concurrent controls is available for comparison with the external control group. Pocock was among the first to discuss the latter and provided a list of criteria to be met before considering the combination of external and concurrent randomized controls.3 We note that inclusion of concurrent randomized controls allows for comparisons between the two control groups, providing some level of assurance of the comparability of external and concurrent randomized controls. In both cases, standard methods developed for observational studies can be used to account for differences between the treatment and control groups such as propensity scores or doubly robust methods.18-22 Other statistical methods have been proposed to account for the use of various external control groups in the same study23 and to allow for differential weighting, or even exclusion, of external controls based on their comparability to concurrent randomized controls. One method allows us to “test-then-pool” controls. That is, a formal hypothesis test is conducted to compare the external and concurrent randomized controls. If the two are found to be comparable, all controls are weighted the same.24 Power priors and hierarchical modeling allow for differential weighting of external controls based on their similarity to concurrent randomized controls.25 Although the use of external controls may help reduce the required sample size, if the control group is not found to be appropriate, some statistical methods can result in increased type I error rates.24 A recent critical appraisal of this approach in the precision medicine setting, which is similar to the setting of rare diseases, identified numerous issues that limit the utility of the augmented design. In fact, none of the scenarios presented supported the use of an augmented design.26
Although we do not discuss the use of summary data from external controls (rather than individual-level data) in great length, we note that this use of external controls is common and is often used to come up with a benchmark for comparison in single-arm studies.27 However, lack of individual patient-level data substantially complicates bias adjustments because standard methods rely on access to individual patient-level data and the FDA draft guidance on externally controlled trials does not discuss the use of summary-level information.7
Type I error
An important consideration when incorporating external controls is control of the type I error rate. The uncertainty in the information from external controls and for point estimates derived from external control data should be accounted for. In the case of historical controls, for example, outcomes are fixed and known at the start of the study, and designs commonly used to incorporate external control data may lead to increased type I error rates. Polley et al. (2024) investigated the performance of the externally augmented design in which patients are randomly assigned to the treatment and control arms at a fixed ratio, such as 1:1, until an interim analysis determines whether concurrent and external controls are comparable. If they are found to be comparable, newly enrolled participants are randomly assigned using a different randomization ratio such that the randomization favors the experimental arm. Otherwise, the trial proceeds without the use of external controls. The authors found that the type I error rate is increased even in settings where concurrent and external controls are not comparable, and an RCT is ultimately conducted.24
Other groups have investigated the use of historical controls in the group sequential framework. In the standard group sequential setting, interim analyses are conducted after a certain amount of information has been collected, with outcome data accumulating for all arms throughout the trial. When incorporating historical controls, the interim analyses largely depend on the accumulated events in the experimental arm, as outcomes in the historical control group are known. At each interim analysis time point, the data collected up that time are used for the experimental arm, whereas all available data are included for the external control group. These designs have been shown to maintain the type I error rates when all the required assumptions are met.28,29
When designing an externally controlled trial, it is important to consider the operating characteristics of the study design under various scenarios at the design stage. This will help understand the potential benefits and challenges introduced by the use of external controls and can inform whether the use of external controls is appropriate, or whether another design may be preferred.
Recommendations
External controls can be a useful tool when an RCT is not possible. As discussed in this article, however, the use of external controls can lead to many challenges and may result in biased treatment effect estimates. The FDA limits the use of external controls to disease areas where there are no other viable study designs, treatment effects are expected to be large, and a suitable external control population exists.4 We note that improvements in cancer treatments are often small, so detecting a large treatment effect with the use of external controls may be especially challenging for cancer trials.
When an external control study is deemed appropriate, the external control population and study design should be carefully selected with these challenges in mind. If treatment-emergent adverse events are not well understood, we recommend the use of concurrent randomized controls to compare the prevalence of these adverse events between the control and experimental arm with a lower risk of bias. Regulatory bodies, statisticians, and other stakeholders should be consulted in early stages to ensure the appropriateness of the proposed study and design. It should also be noted that although appropriate design and use of statistical methods help mitigate some of the challenges associated with the use of external control data, the results from studies with an external control group rely on stronger assumptions and are more speculative than results from RCTs. The assumptions underlying the conclusions of externally controlled trials should be clearly stated and the plausibility of the assumptions discussed. Because of the reliance on stronger assumptions, it is recommended that externally controlled trials are used only when a fully randomized trial is infeasible or unethical and there is a reasonable chance that an externally controlled trial will be able to demonstrate the effectiveness of the treatment.
Disclaimer
The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.
Contributor Information
Michelle M Nuño, Children’s Oncology Group, Monrovia, CA 91016, United States; Department of Population and Public Health Sciences, University of Southern California, Los Angeles, CA 90032, United States.
Stephanie L Pugh, NRG Oncology Statistics and Data Management Center, American College of Radiology, Philadelphia, PA 19103, United States.
Lingyun Ji, Children’s Oncology Group, Monrovia, CA 91016, United States; Department of Population and Public Health Sciences, University of Southern California, Los Angeles, CA 90032, United States.
Jin Piao, Children’s Oncology Group, Monrovia, CA 91016, United States; Department of Population and Public Health Sciences, University of Southern California, Los Angeles, CA 90032, United States.
James J Dignam, NRG Oncology Statistics and Data Management Center, American College of Radiology, Philadelphia, PA 19103, United States.
Jon A Steingrimsson, Department of Biostatistics, Brown University, Providence, RI 02903, United States.
Author contributions
Michelle M. Nuño, PhD, Stephanie L. Pugh, PhD, Lingyun Ji, PhD, Jin Piao, PhD, James J. Dignam, PhD, Jon Steingrimsson, PhD. All authors contributed to the conceptualization, project administration, and writing of the paper (writing, editing, and reviewing).
Funding
Research reported in this article was supported by the Children’s Oncology Group, the National Cancer Institute of the National Institutes of Health under award numbers U10CA180886, U10CA180899, U10CA180820, U10CA180794, and UG1CA189828.
Monograph sponsorship
This article appears as part of the monograph “Statistical and Practical Considerations in Design and Analysis of Clinical Trials with Patient-Centered Outcomes,” sponsored by the National Cancer Institute (NCI) and the following NCI Community Oncology Research Program (NCORP) Research bases: NRG Oncology (UG1CA189867), Alliance (UG1CA189823), SWOG (UG1CA189974), ECOG-ACRIN (UG1CA189828), Wake Forest University (UG1CA189824), University of Rochester (UG1CA189961), and Children’s Oncology Group (5UG1CA189955-11).
Conflicts of interest
None declared.
References
- 1. Bhide A, Shah PS, Acharya G. A simplified guide to randomized controlled trials. Acta Obstet Gynecol Scand. 2018;97:380-387. 10.1111/aogs.13309 [DOI] [PubMed] [Google Scholar]
- 2. Burcu M, Dreyer NA, Franklin JM, et al. Real-world evidence to support regulatory decision-making for medicines: considerations for external control arms. Pharmacoepidemiol Drug. 2020;29:1228-1235. 10.1002/pds.4975 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3. Pocock SJ. The combination of randomized and historical controls in clinical trials. J Chronic Dis. 1976;29:175-188. 10.1016/0021-9681(76)90044-8 [DOI] [PubMed] [Google Scholar]
- 4. FDA Guidance for Industry. Considerations for the design and conduct of externally controlled trials for drug and biological products. 2023. https://www.fda.gov/media/164960/download
- 5.International Conference on Harmonization (ICH) E10. Choice of control group and related issues in clinical trials. 2000. . https://database.ich.org/sites/default/files/E10_Guideline.pdf [PubMed]
- 6. Marion JD, Althouse AD. The use of historical controls in clinical trials. JAMA. 2023;330:1484-1485. 10.1001/jama.2023.16182 [DOI] [PubMed] [Google Scholar]
- 7. Center for Drug Evaluation and Research. Considerations for the design and conduct of externally controlled trials for drug and biological products. January 31, 2023. https://www.fda.gov/regulatory-information/search-fda-guidance-documents/considerations-design-and-conduct-externally-controlled-trials-drug-and-biological-products. Accessed December 21, 2023.
- 8. Jahanshahi M, Gregg K, Davis G, et al. The use of external controls in FDA regulatory decision making. Ther Innov Regul Sci. 2021;55:1019-1035. 10.1007/s43441-021-00302-y [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9. Hatswell AJ, Baio G, Berlin JA, Irs A, Freemantle N. Regulatory approval of pharmaceuticals without a randomised controlled study: analysis of EMA and FDA approvals 1999–2014. BMJ Open. 2016;6:e011666. 10.1136/bmjopen-2016-011666 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10. Hansford JR, Bouche G, Ramaswamy V, et al. Comments and controversies in oncology: the tribulations of trials developing ONC201. J Clin Oncol 2024;JCO2400709. 10.1200/JCO.24.00709 [DOI] [PubMed] [Google Scholar]
- 11. Venneti S, Kawakibi AR, Ji S, et al. Clinical efficacy of ONC201 in H3K27M-mutant diffuse midline gliomas is driven by disruption of integrated metabolic and epigenetic pathways. Cancer Discov. 2023;13:2370-2393. 10.1158/2159-8290.CD-23-0131 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12. Hernán MA, Robins JM. Using big data to emulate a target trial when a randomized trial is not available: Table 1. Am J Epidemiol. 2016;183:758-764. 10.1093/aje/kwv254 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13. Fu EL. Target trial emulation to improve causal inference from observational data: what, why, and how? JASN. 2023;34:1305-1314. 10.1681/ASN.0000000000000152 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14. Arnold K, Antunes L, Coles B, Lee H. Application of the target trial emulation framework to external comparator studies. Front Drug Saf Regul. 2024;4:1380568. 10.3389/fdsfr.2024.1380568 [DOI] [Google Scholar]
- 15. Polito L, Liang Q, Pal N, et al. Applying the estimand and target trial frameworks to external control analyses using observational data: a case study in the solid tumor setting. Front Pharmacol. 2024;15:1223858. 10.3389/fphar.2024.1223858 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16. Shin JI, Grams ME. Trial emulation methods. Am J Kidney Dis. 2024;83:264-267. 10.1053/j.ajkd.2023.07.024 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17. Hernán MA, Alonso A, Logan R, et al. Observational studies analyzed like randomized experiments: an application to postmenopausal hormone therapy and coronary heart disease. Epidemiology. 2008;19:766-779. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18. Rosenbaum PR, Rubin DB. The central role of the propensity score in observational studies for causal effects. Biometrika. 1983;70:41-55. 10.1093/biomet/70.1.41 [DOI] [Google Scholar]
- 19. Li X, Miao W, Lu F, Zhou XH. Improving efficiency of inference in clinical trials with external control data. Biometrics. 2023;79:394-403. 10.1111/biom.13583 [DOI] [PubMed] [Google Scholar]
- 20. Robins JM, Rotnitzky A, Zhao LP. Estimation of regression coefficients when some regressors are not always observed. J Am Stat Assoc. 1994;89:846-866. 10.1080/01621459.1994.10476818 [DOI] [Google Scholar]
- 21. Liu J, Zhang J, Mitchell A, Fang M, Tian L. Causal inference for longitudinal data based on historical controls. J Biopharm Stat. 2023;33:289-306. https://www.tandfonline.com/doi/full/10.1080/10543406.2022.2148164 [DOI] [PubMed] [Google Scholar]
- 22. Cheng Y, Wu L, Yang S. Enhancing treatment effect estimation: a model robust approach integrating randomized experiments and external controls using the double penalty integration estimator. In: Proceedings of the Thirty-Ninth Conference on Uncertainty in Artificial Intelligence. PMLR; 2023:381-390. https://proceedings.mlr.press/v216/cheng23a.html. Accessed February 26, 2024.
- 23. Lim J, Walley R, Yuan J, et al. Minimizing patient burden through the use of historical subject-level data in innovative confirmatory clinical trials: review of methods and opportunities. Ther Innov Regul Sci. 2018;52:546-559. 10.1177/2168479018778282 [DOI] [PubMed] [Google Scholar]
- 24. Polley MYC, Schwartz D, Karrison T, Dignam JJ. Leveraging external control data in the design and analysis of neuro-oncology trials: pearls and perils. Neuro-Oncol. 2024;26:796-810. 10.1093/neuonc/noae005 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25. Viele K, Berry S, Neuenschwander B, et al. Use of historical control data for assessing treatment effects in clinical trials. Pharm Stat. 2014;13:41-54. 10.1002/pst.1589 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26. Freidlin B, Korn EL. Augmenting randomized clinical trial data with historical control data: precision medicine applications. JNCI J Natl Cancer Inst. 2023;115:14-20. 10.1093/jnci/djac185 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27. Yom SS, Torres-Saavedra P, Caudell JJ, et al. Reduced-dose radiation therapy for HPV-associated oropharyngeal carcinoma (NRG Oncology HN002). J Clin Oncol. 2021;39:956-965. 10.1200/JCO.24.00709 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28. Wu J, Xiong X. Survival trial design and monitoring using historical controls. Pharm Stat. 2016;15:405-411. 10.1002/pst.1756 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29. Wu J, Li Y. Group sequential design for historical control trials using error spending functions. J Biopharm Stat. 2020;30:351-363. 10.1080/10543406.2019.1684305 [DOI] [PMC free article] [PubMed] [Google Scholar]