Characteristics, Progression, and Output of Randomized Platform Trials

Key Points Question What are the characteristics, progression, and output of randomized platform trials? Findings In this systematic review of 127 platform trials with a total of 823 arms, primarily in the fields of oncology and COVID-19, the adpative features of the trials were often poorly reported and only used in 49.6% of all trials; results were available for only 65.2% of completed trial arms. Meaning The planning and reporting of platform features and the availability of results were insufficient in randomized platform trials.


Introduction
Randomized clinical trials (RCTs) are the criterion standard for evaluating health care interventions.

Key Points
Question What are the characteristics, progression, and output of randomized platform trials?Findings In this systematic review of 127 platform trials with a total of 823 arms, primarily in the fields of oncology and COVID-19, the adpative features of the trials were often poorly reported and only used in 49.6% of all trials; results were available for only 65.2% of completed trial arms.

Meaning
The planning and reporting of platform features and the availability of results were insufficient in randomized platform trials.

+ Supplemental content
8][19][20] Platform trial planning and conduct require consideration of their unique design features, methodological framework, and level of sophistication.This planning includes the potential use of a common control arm, nonconcurrent control data, the statistical framework (bayesian and/or frequentist), in silico trials (simulations), and the use of additional adaptive design features, such as response adaptive randomization (RAR; the change of the randomization ratio based on data collected during the trial), sample size reassessment, seamless design (seamless study phase transition), and adaptive enrichment (modification of eligibility criteria). 9,11,16Platform trials are stipulated to be more time efficient and cost efficient and are able to increase the output of the trial, benefiting both patients and researchers. 8,9,17Further potential benefits include the use of regulatory documentation (master protocol) and contracts beyond 1 trial and its respective duration, 8 quick initiation of new sites and intervention arms, 21 reuse of established infrastructure, 22 and quick study phase transition. 22pirical evidence about platform trials is needed to gain insight into the actual application of this design in clinical research practice and to learn about its benefits and pitfalls, so that the planning and conduct of platform trials can be further improved.Previous systematic reviews on platform trials are outdated 13,14 ; are restricted to the late-phase, multiarm, multistage design or COVID-19 trials 23,24 ; only investigated a small number of distinct platform trial features 23 ; or did not consider the output of platform trials in terms of completed, prematurely closed, and published trial arms. 25A comprehensive overview is currently lacking.We specifically wondered whether the incidence of platform trials continued to increase despite a fading pandemic, the extent to which distinctive features were actually used, whether recruitment failures were rare, and whether results from platform trials were consistently made available.We, therefore, conducted a systematic review of all available randomized platform trials to empirically determine (1) their incidence over time, (2) the actual frequencies of various distinctive platform trial characteristics (eg, common control arm, use of nonconcurrent control data, and RAR), (3) the incidence of added and dropped arms over time, (4)   the prevalence of discontinued trials due to poor participant recruitment, and (5) the availability of results for closed trial arms.

Methods
This systematic review is reported according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) reporting guideline. 26A detailed protocol was prospectively registered on Open Science Framework (OSF). 27

Systematic Search and Eligibility Criteria
The systematic search (including registries) was conducted on January 12, 2021, and was updated on Screening of titles and abstracts, trial registries, and full text were performed in duplicate.
Discrepancies were resolved by discussion or by involving a third reviewer (B.S. or M.B.).For each included report, we continued with forward and backward citation tracking (using Scopus).Citation C.M.S.).If multiple reports were available for 1 platform trial, these reports were organized and consolidated by registry numbers, acronyms, and the title of the trial.Once a platform trial was included, we determined if an official trial website was available (by screening the literature and registries and searching via Google).For each platform trial and each of their recorded arms, we searched in duplicate (registry, website, Google Scholar, and Google) for the master protocol, subprotocols, and results publications, if not previously found in the literature search.

Data Extraction
The variables for this systematic review were chosen based on discussions with methodologists and statisticians of platform trials, previous reviews on the topic, and the critical appraisal checklists by Park et al. 20,28 All relevant data were extracted in duplicate (by different researchers).Differences were consolidated by a third reviewer.All authors worked in teams of 2 from trial protocols (master and subprotocols), result publications, trial registries, and the official trial websites into a REDCap data sheet. 29,30We documented the different labels used in study records (eg, "platform trial," "trial platform," "platform study," "platform design," or "platform protocol") to explore the general use of the term platform trial.We extracted baseline characteristics for each included platform trial and each of their individual arms (see list of all baseline characteristics in eAppendix 2 in Supplement 1).Furthermore, distinct platform trial features were recorded.These features included the use of a common control arm and, if the common control arm could be updated during the trial, the use of nonconcurrent control data, adaptive design elements (eg, RAR, adaptive enrichment, seamless design, sample size readjustment), a statistical framework (bayesian, frequentist, or both), multiplicity adjustments (to multiple arms and for interim analyses), and feasibility studies (in silico trials or simulations or pilot trials).We determined the progression and output of the platform trial by the starting number of arms, the total number of arms, the number of arms added, the number of arms dropped (including the reason), and the status and availability of the results for each intervention arm (output of platform trial).Further features of interest included the use of biomarker stratification or subpopulations, integration of nonrandomized arms, interim analysis (reporting of frequency, outcome, and trigger), or the use of a factorial design.The format of the master protocol and the results publications were also recorded (as peer-reviewed publication, preprint, and full protocol on website or registry).Furthermore, we calculated the ratio of available results publications to the number of closed arms.The ratio was calculated twice, once including and once excluding results available as abstracts only.We contacted all principal investigators with a report detailing the most important information extracted from their platform trial.Principal investigators were asked to approve the accuracy of extracted data and to clarify missing or unclear information (eAppendix 3 in Supplement 1).

Statistical Analysis
We summarized the characteristics of the included platform trials using the median and IQR for continuous variables and numbers and percentages for categorical variables.Baseline characteristics were stratified by sponsorship (industry vs not industry sponsored) and COVID-19 indication.
Previous research has identified differences in the discontinuation rate, reporting quality, and transparency between industry-sponsored and non-industry-sponsored traditional RCTs 31,32 ; as such, we stratified platform trial characteristics by sponsorship.Because it was expected that platform trial features are often recorded in the master protocol, we conducted a sensitivity analysis including only trials with an available master protocol.Data cleaning and analysis were conducted with R, version 1.4.1103(R Project for Statistical Computing).

Results
A total of 9155 records were identified.We determined 431 eligible records, resulting in 127 unique randomized platform trials included in our sample (the list of all included platform trials can be found    4).The

Discussion
Existing platform trials predominantly focus on evaluating drugs and tend to cluster in medical areas, such as oncology, COVID-19, and other infectious diseases.After the peak in 2020 with the arrival of the COVID-19 pandemic, the initiation of new platform trials has decreased.However, there has been a noticeable diversification of medical fields and interventions of platform trials over the past 5 years.
This diversification encompasses areas such as neurology, dermatology, and general surgery, as well as the testing of behavioral, surgical, or dietary interventions.
Among the observed platform trials, 49.6% incorporated at least 1 additional adaptive design feature.A total of 58.3% of platform trials added at least 1 arm, and 62.2% dropped at least 1 arm  (21.3% did neither, although planned).Consequently, the approximately 40% of trials that never added an arm may have incurred higher planning and setup costs compared with traditional RCTs without benefiting from the cost savings of additional arms. 33A common control arm was used in only 73.2% of platform trials, which is lower than one would expect for a major platform trial advantage (increased efficiency) and is below the percentage previously reported. 23This finding may underline the belief of many stakeholders that the establishment of collective trial infrastructures (including communication networks, overall data management and monitoring plans, and standardized documents across arms) is reason enough to justify the use of the platform trial design. 226][37] A bayesian design was frequently used because this statistical framework fits well with the adaptive nature of platform trials 25,35 ; however, bayesian trial designs may be less commonly understood by a general medical and scientific readership, posing challenges for interpretation and uptake of results.
In addition, the use of features such as RAR and nonconcurrent controls should be considered carefully.Response adaptive randomization, for instance, requires a well-planned run-in phase, may inflate type I error, typically requires a higher sample size, and can be associated with slow accrual of outcome data. 38About 8% of platform trials considered nonconcurrent control data in an attempt to further increase statistical power; however, this approach carries a high risk for bias. 22,37,39gulators criticize the use of nonconcurrent controls in confirmatory trials because statistical modeling can only partially address the potential bias. 37,38most 80% of platform trial protocols were publicly available in some format, much higher than previously determined for traditional RCTs. 24,25However, reporting of essential features, such as adjustment for multiplicity, use of nonconcurrent control data, and criteria for dropping and adding new arms, was often unsatisfactory.Full results publications were available for 47.9% of   poor recruitment in RCTs, 10%-15%). 31,32However, it is possible that this proportion will increase due to recruitment hurdles and the increasing scarcity of eligible patients for COVID-19 trials toward the end of the pandemic.Publication of full results for closed arms (47.9%) was lower than what is generally seen for traditional RCTs (78.5% at 10-year follow-up). 32Availability of full results publications and overall transparency were generally better in non-industry-sponsored platform trials.
Overall, industry-sponsored platform trials accounted for approximately one-third of the total and predominantly focused on early-phase investigations, while late-phase trials were mostly not sponsored by industry.Seamless designs, combining early-and late-phase trials, although still a minority (18.1%), are becoming increasingly more common. 14

Strengths and Limitations
Our study has some strengths.b Full results as peer-reviewed results publications, master protocol (n = 76 [59.8%]) were considered (eTable 5 in Supplement 1).Second, the reporting was not always consistent across different sources.We handled these discrepancies by creating an information hierarchy, giving priority to peer-reviewed manuscripts and the feedback received by investigators (followed by preprints, websites, and then other sources).Third, although highly desirable, we did not consider resource use and costs of platform trials in this review.Evidence from a hypothetical costing study suggested increased costs associated with the planning and setup of platform trials compared with traditional RCTs are due to the complex protocols and longer setup times. 33These increased costs were mitigated when more arms were added to the trial, which was less time intensive and reduced costs long term. 40,41Fourth, a comparison of platform trials with traditional parallel-arm RCTs was possible only on an indirect level.However, a direct comparison of platform trials with traditional RCTs with the same research question is planned in a future project, as described in our study protocol. 27Fifth, this systematic review provides only a snapshot of the current platform trial landscape.Two-thirds of identified platform trials are still ongoing, and the COVID-19 pandemic may have had an influence on the progression and output of our sample.
Furthermore, methodological background and reporting guidelines for platform trials were lacking at the start of this project and are currently still evolving.Therefore, regular updates of this systematic review are necessary to gain further insights into progression patterns and output from randomized platform trials and to determine the most appropriate application of this design in the future.

Conclusions
In this systematic review, we found that platform trials were initiated most frequently during the beginning of the COVID-19 pandemic and appeared to decrease thereafter, with a trend toward more diversified medical fields and interventions.Despite the potential for complexity, most made use of only 1 adaptive feature, or none.Forty percent of platform trials did not add an arm and, thereby, may have missed efficiency gains and incurred higher planning and setup costs compared with traditional RCTs. 33Premature arm closure for poor recruitment was rare.The reporting of platform features, the status of trial arms, and the results of closed arms needs to be improved.Guidance and infrastructure are needed so that the status and results of individual trial arms can be reported in a timely manner (eg, adaptations of trial registries for platform trials) and so that decisions about the need for a platform design and its planning is optimized.

July 28 ,
2022.Data were extracted until December 2022.Investigators were contacted for verification of the data in February 2023.We performed a systematic search of Medline (OVID), Embase (OVID), Scopus, and several trial registries (Clinicaltrials.gov,European Union Drug Regulating Authorities Clinical Trials Database, and International Standard Randomized Controlled Trial Number registry).To increase the sensitivity of the search, we included gray literature servers (OSF and Zenodo) and preprint servers (Europe PubMed Central) (search date: July 21, 2022).The detailed search strategy is available in eAppendix 1 in Supplement 1.An information specialist helped us design and review our search strategy.Trials were included if they were RCTs and planned to add or drop arms.

Figure
Figure.Frequency of Initiation of Platform Trials

a
Common control planned to be updated (n = 31) or has been updated (n = 3) and cannot be updated (n = 8).b Multiple adaptive designs possible for platform trials.c Platform trials with 1 additional design (41 [65.1%]),

a
Reasons for discontinuation: change in treatment landscape (n = 3), low event rate (n = 3), insufficient funding (n = 2), safety (n = 1), and unclear (n = 1).b Includes the 4 planned platform trials.c Proportion calculated based on trials that added and dropped arms.d Proportion calculated based on trials with available master protocol.

Table 1 .
General Platform Trial Characteristics

Table 3 )
Supplement 1.A total of 38 platform trials (29.9%) were initiated in 2020, the highest reported incidence of newly started platform trials in 1 year thus far.This number has since decreased (25 of 127[19.7%]in2021)(Figure).A common control arm was reported to be used in 73.2% of all platform trials (93 of 127); 7.9% trials (10 of 127) planned to use nonconcurrent control data for their statistical analysis (not reported for 61 of 127 trials [48.0%]) (Table2).Adaptive design elements were integrated in approximately .Reasons for discontinuation included change in treatment landscape (3 of 10), low event rates (3 of 10), insufficient funding (2 of 10), and safety concerns (1 of 10), and, for 1 platform trial, the reason for discontinuation remained unclear.The number of arms at the start of the platform trial and the total number of arms was typically higher in industry-sponsored trials (median number of arms at start, 4 [IQR,2-5]; median total number of arms, 6 [IQR,4-8]) than in non-industry-sponsored trials (median number of arms at start, 3 [IQR,2-4]; median total number of arms, 5 [IQR, 4-7]) (Table3).Overall, 58.3% platform trials (74 of 127) added at least 1 arm, and 62.2% (79 of 127) dropped at least 1 arm during their progression; although planned, 21.3% of platform trials (27 of 127) neither added nor dropped an arm.Of the 85 platform trials that added or dropped an arm during the trial, the corresponding registry entry was not updated for 19 trials (22.4%).Half of all platform trials (64 of 127 [50.4%]) made results available from at least 1 comparison.Data on progression and output stratified by COVID-19 vs non-COVID-19 trials can be found in eTable 6 in Supplement 1.The 127 platform trials had a total of 823 arms, including 206 control arms (Table4).Of the 823 arms, 385 (46.8%) were ongoing, 34 (4.1%) were in the planning phase, and 353 (42.9%) were closed.Of the 353 closed arms, 189 (53.5%) were completed, 56 (15.9%) were stopped for futility, 20(5.7%) were stopped due to new external evidence, 9 (2.5%) were stopped for safety concerns, and 26 (7.4%) were stopped for practical reasons, including poor recruitment (5[1.4%]).Less than half of the closed arms (169 of 353 [47.9%]) made full results available.Making results available was more common and faster for non-industry-sponsored trials compared with industry-sponsored trials (150 of 277 [54.2%] vs 19 of 76 [25.0%]); however, there is evidence for confounding because COVID-19 trial results were available substantially faster than results for non-COVID-19 trials (Table

JAMA Network Open | Statistics and Research Methods
platform trial arms stratified by COVID-19 vs non-COVID-19 trials can be found in eTable 7 in Supplement 1.The form of results availability (as peer review, preprint, abstract, and on registry) is available in eTable 8 in Supplement 1.We contacted investigators of platform trials to verify the extracted data and achieved a high response rate (active agreement, 46.5% [59 of 127]; taciturn agreement, 15.7%[20 of 127]; no response, 37.8% [48 of 127]) (eTable 9 in Supplement 1).

Table 2 .
Specific Platform Trial Characteristics

Table 3 .
Platform Trial Progression and Output

JAMA Network Open | Statistics and Research Methods
Downloaded from jamanetwork.comby guest on 03/23/2024 closed arms.Premature closure of platform trial arms due to recruitment problems was infrequent, occurring in only 1.4% of trials, which is in contrast to traditional RCTs (discontinuation rate due to

Table 4 .
To our knowledge, it is the first study investigating key platform trial features, protocol and results availability, and the status of individual arms.An additional strength of our study was that we contacted investigators of platform trials to verify the extracted data and achieved a high response rate (active agreement, 46.5% [59 of 127]; taciturn agreement, 15.7% [20 of 127]; no response, 37.8% [48 of 127]) (eTable 9 in Supplement 1); responses typically confirmed the accuracy of gathered data, and only minor adjustments were necessary.Our study has the following limitations.First, available information was sometimes limited, especially if only a registry entry was available.We have, therefore, conducted sensitivity analyses showing how the proportion of certain variables changed if only platform trials with an available Status of Platform Trial Arms and Trial Arm Results a Research question answered.
Wilson N, Biggs K, Bowden S, et al.Costs and staffing resource requirements for adaptive clinical trials: quantitative and qualitative results from the Costing Adaptive Trials project.BMC Med.2021;19(1):251.doi:10.1186/s12916-021-02124-z 41.Wason JMS, Dimairo M, Biggs K, et al.Practical guidance for planning resources required to support publiclyfunded adaptive clinical trials.BMC Med.2022;20(1):254.doi:10.1186/s12916-022-02445-7Detailed Flow Chart and Reasons for Exclusion eTable 1. Report Labels and Reasons for Exclusion of Reports in Literature and Registry Screening eTable 2. Other Baseline Characteristics eTable 3. Baseline Characteristics by COVID and Non-COVID Platform Trials eTable 4. Specific Platform Trial Characteristics in COVID and Non-COVID Trials eTable 5. Specific Platform Trial Characteristics for Platform Trials With Full Available Master Protocol eTable 6. Platform Trial Progression and Output of COVID and Non-COVID Trials eTable 7. Status of Platform Trial Arms and Trial Arm Results in COVID and Non-COVID Trials eTable 8. How Were Results Made Available for Arms? eTable 9. Survey Response Rates eAppendix 3. Example of eMail Template and Report Sent to Platform Trial Teams eTable 10.List of Randomized Platform Trials 40.