The Evolution of Radiotherapy in Children’s Oncology Group Trials, from 1998–2025

Abstract

Purpose:

Significant advancements in external beam radiotherapy (RT) over the past 20+ years have occurred with evolution from 2D techniques based on plain film x-rays to three-dimensional (3D) conformal and intensity modulated photon techniques. Likewise, proton radiotherapy (PRT) was also introduced and has become more widely available. Each advance in radiation technique improves targeting and has the potential to mitigate late effects associated with RT. In this study, we use metadata from past and present Children’s Oncology Group (COG) trials that included RT to describe the evolution of different RT modalities across time and disease sites.

Methods:

Imaging and Radiation Oncology Core (IROC) database was queried for protocols with at least one patient receiving RT during 1998–2025. Protocol metadata including primary disease site, study timeline and quality assurance (QA) was extracted, as was non-identifiable patient-level variables including age, institution, year of enrollment, and RT modality utilized. Descriptive statistics were used for downstream analysis.

Results:

The study population comprised 8,898 patients, including 7,881 from the USA and 1,017 from other countries. The median age at enrollment was 9 years; California, Texas, Florida and New York were among the top contributing states. Of the 83 included COG trials, 48 included at least one patient receiving PRT. The number of trials allowing PRT peaked in 2010 with 17 protocols. Trials involving central nervous system (CNS) tumors and rhabdomyosarcoma soft tissue sarcomas (RST) dominated both in the number of trials and enrollment. They were also among the earliest to incorporate PRT into treatment protocols. From 1998 to 2025, 3D photon-based treatments declined from 73% to 15%, while IMRT/VMAT usage rose from 26% to 89% over the same interval. The use of PRT increased from 2% pre-2000 to nearly 40% in 2021, which has since stabilized. QA review demonstrated >90% protocol compliance across all RT modalities both nationally and internationally.

Conclusions:

Over 25 years of COG experience, photon therapy has transitioned from 3D conformal to IMRT/VMAT, while PRT use has continued to expand, particularly in CNS and RST trials. These trends highlight how pediatric cooperative-group studies continue to shape national standards and underscore the need for future work linking modality adoption with toxicity, survivorship, and access across diverse treatment settings.

Introduction:

Over the last several decades, the field of pediatric oncology has evolved dramatically, resulting in markedly improved outcomes for nearly all subtypes of pediatric cancers.¹ Although 5-year survival has served as a key marker for treatment success, long-term pediatric cancer survivors are at significant risk of late toxicities including secondary cancers, cardiopulmonary disease, and other chronic health conditions.^2,3 Radiotherapy (RT) is an important therapeutic component in children with cancer, and is frequently used in multimodal therapy strategies for a multitude of tumors including those of the central nervous system (CNS), bone and soft tissue. In the last two decades, advancements in RT techniques have allowed more conformal targeting of tumors,^4,5 exemplified by the development of intensity-modulated radiation therapy (IMRT) / volumetric-modulated arc therapy (VMAT)⁶ and proton radiotherapy (PRT).^7,8

The Children’s Oncology Group (COG) is the world’s largest organization devoted exclusively to childhood and adolescent cancer research. More than 90% of the 16,000 children and adolescents diagnosed with cancer each year in the United States are cared for at COG member institutions; with more than 100 active clinical trials, there are approximately 12,000 patients registered on COG trials each year.⁹ Therefore, the metadata from the COG trials are uniquely informative for understanding nationwide trends and patterns of care. The objectives of this study were to (1) quantify temporal trends in all RT modalities usage, (2) identify patient-specific factors and protocol characteristics associated with type of RT modality, and (3) evaluate longitudinal QA performance for each modality, using enrollment data from all COG trials between 1998 and 2025. As implementation of PRT becomes more widespread, we sought to evaluate the trends in all modalities of RT use on COG clinical trials. To our knowledge, no studies to date have reported on the implementation and evolution of external beam RT techniques in pediatric cooperative group trials, including across the COG portfolio.

Methods:

We queried the Imaging and Radiation Oncology Core (IROC) national database for all COG trials active between 1998 and 2025. The study population encompassed trials conducted across seven countries and 751 institutions, including the United States (and Puerto Rico), Australia, Canada, Israel, New Zealand, Switzerland, and the Netherlands. Patients were included in this analysis regardless of age, provided all other clinical inclusion criteria were met. Eligible trials included those where at least one patient receives any type of RT. For trials allowing PRT, IROC was queried and at least one patient must have received PRT to be considered for “allowing protons” as only the later protocols made it explicit whether PRT could be used. Trials primarily focusing on acute myeloid leukemia (AML), acute lymphocytic leukemia (ALL) and stem cell transplantation (SCL) were excluded from this analysis since total body irradiation is used as a conditioning regimen prior to stem cell transplantation. Protocol metadata including primary disease site and study phase was extracted, as was non-identifiable patient-level variables including age, location, year of enrollment, and RT modality (proton versus photon/electron). Central QA review data was also collected for all available patients to determine RT protocol compliance. Institutional IRB exemption was confirmed for the use of this de-identified dataset. Descriptive statistics were used to quantify the proportion of patients receiving RT on protocol based on primary disease site, as well as temporal trends in RT utilization/protocol compliance over time.

Because the RTQA process evolved over the study interval, QA outcomes were interpreted within the context of the contemporaneous central review workflow. In the earlier portion of the study period, RT objects were frequently submitted in hard-copy or mixed hard-copy/digital formats and were reviewed retrospectively or, when feasible, during the first several days of treatment. With maturation of secure digital submission platforms and IROC workflows, central review has increasingly shifted toward real-time, pre-treatment review when required by protocol, particularly for studies in which eligibility, response assessment, target definition, or RT initiation depends on central review.

Results:

Characteristics of Enrolled Patients

From 1998 to 2025, 83 COG trials met the inclusion criteria (Supplementary Table. 1), enrolling 8,898 patients. Of these, 7,881 patients were accrued within the USA, while 1,017 were from international sites. Most enrolled patients were between 1 and 17 years of age (Supplementary Fig. 1A). Across the study period (1998–2025), the annual median age ranged from 3 to 17 years (Supplementary Fig. 1B). Specifically, the median age was 9 years for all RT modalities combined, 9 years for photon therapy, and 7 years for PRT. In a majority of years, median age of those receiving PRT was lower than those enrolled in all RT and photon groups (Supplementary Fig. 1B).

In terms of state-level accrual across all RT modalities, California contributed the highest number of cases between 1998 and 2020, with Texas and Florida ranking second and third. In contrast, New York emerged as the leading contributor during the 2021–2025 period (Fig. 1A–C); When analyzing at the city-specific level, Memphis, Los Angeles, Boston, Houston, and Montreal were the five largest contributors (Fig. 1D).

Figure 1.

Geographic distribution of the enrolled patients across all RT modalities in binned eras. A) 1998–2010; B) 2011–2020; C) 2021–2025. D) City-level accrual of all RT modalities, showing cities contributing over 50 cases of all time (1998–2025).

Characteristics of Enrolled COG Trials

The total enrollment is driven largely by trials for CNS (No. of trials = 35, No. of enrollment = 3,213) and RST (No. of trials = 16, No. of enrollment = 2,126) tumors (Fig. 2A). Total trial enrollment and the number of patients receiving RT closely paralleled the increase in number of COG trials over the years, reaching a plateau between 2004 and 2012 (Fig. 2B). Notably, the Hodgkin lymphoma (HOD) trial AHOD0031 (N = 1,008; 2002–2010) contributed significantly to total enrollment during this period.

Figure 2.

Timeline of Initiation/Closure of COG Trials Including RT. A) Each bar represents the COG protocol enrollment period, organized and color-coded by primary disease site. B) Line plot outlining total number of patients enrolled on COG trial each year (regardless of receipt of RT; orange), patients receiving any types of RT (green), and those receiving PRT (blue).

PRT Utilization Trends

Among all trials meeting the inclusion criteria, 48 included at least one patient who received PRT. Regarding trials allowing PRT, primary central nervous system (CNS) and rhabdomyosarcoma soft tissue sarcomas (RST) studies predominated between 1998–2025 (40.8% (N=20) and 24.5% (N=12) for CNS and RST, respectively), with active protocols nearly every year (Fig. 3A). The number of open trials reached a peak in 2010–2012, with a gradual decline in open studies until regaining an upward trend beginning in 2021. Notably, several renal tumors (REN) and rare tumors (RAR) protocols have more recently and explicitly allowed PRT usage starting in 2020, contributing to this overall recent rise in eligible COG trials. During 1998–2025, the annual percentage of open trials incorporating PRT delivery showed an upward trend, reflecting the overall growing inclusion of PRT within trial protocols (Fig. 3B).

Figure 3.

Overall Trends of RT Use in COG Trials Involving Proton Therapy. A) Bar plot showing number of enrolling COG trials each year (allowing PRT), color-coded by primary disease site. B) Line plot showing the annual percentage of COG trials permitting PRT relative to all trials permitting RT. C) Line plot showing the proportion of patients receiving PRT relative to all those receiving RT on trial annually. Abbreviations: CNS, central nervous system; RST, rhabdomyosarcoma soft tissue sarcomas; NBL, neuroblastoma; EWS, Ewing Sarcoma; HOD, Hodgkin lymphoma; REN, renal tumors; RET, retinoblastoma; RAR, rare tumors.

The number of patients treated with PRT has shown an ongoing upward trend, peaking in 2018 (Fig. 2B). A subsequent decrease in the number of patients receiving PRT during 2021–2025 may reflect the closure of several CNS, Ewing Sarcoma (EWS) and HOD trials. The overall use of PRT makes up a small proportion of total RT initially, but starting from 2021, PRT use comprised 38.5% on average of all patients receiving RT (Fig. 3C) despite the decreasing number of open trials.

Photon Utilization Trends

To further compare the PRT with different photon techniques, we evaluated patients either receiving three-dimensional (3D) conformal or IMRT/VMAT photon therapy. 3D conformal was the predominant technique used in photon modalities, with utilization plateauing 2002–2009, while IMRT/VMAT use increased steadily but remained approximately 50% lower than 3D conformal during that period. The intersection occurred in 2012, when the use of IMRT/VMAT surpassed 3D conformal and became the more prevalent photon-based RT technique (Fig. 4A). Similar trends were observed when the data were stratified by US and international cohorts (Supplementary Fig. 2A). Since some patients received multiple modalities of RT (multiple sites or mixed modalities), patients receiving both 3D conformal and IMRT/VMAT were counted as receiving each modality in Fig. 4A-B.

Figure 4.

Trends in Photon-Based RT Techniques over Time. A) Line plot showing the percentage of patients receiving either 3D or IMRT/VMAT photon RT relative to patients receiving photon-based RT. B) Bar plot showing the proportion of patients receiving 3D and IMRT/VMAT photon RT relative to all those receiving RT on trial, grouped by era.

PRT and Photon Utilization by Primary Disease Sites

Consistent with the predominance of COG trials targeting CNS and RST tumors shown in Fig. 2A, the number of patients receiving RT for these two sites are also the highest (CNS, N = 2,787; and RST, N = 1,962). The number of patients receiving PRT on CNS (N = 361), RST (N = 248), neuroblastoma (NBL, N = 138), EWS (N = 104), and HOD (N = 72) protocols was higher than other sites across 1998–2025. The percentage of PRT use in CNS and RST trials fluctuated between 20–50% in the last 10 years (Fig. 5A-B, Supplementary Fig. 3A-B); however, PRT use in RST increased to 88% in 2025 (Supplementary Fig. 3B). CNS and RST trials were also among the earliest to integrate PRT into treatment protocols, considering both enrollment volume and timing.

Figure 5.

Trends in PRT and Photon Use by Primary Disease Site Over Time. A-H) Line plots showing the second-order polynomial trendline fitted to the raw data to illustrate the longitudinal trajectory of IMRT/VMAT, PRT and 3D photon utilization across various disease sites. The coefficients of determination (R²) of each group were shown, denoting the fit of the model to the observed data. Solid lines represent the chronological shifts of IMRT/VMAT among patients receiving photon therapy by disease site each year, dotted lines represent the chronological shifts of 3D conformal photon therapy among patients receiving photon therapy, and dashed lines represent the chronological shifts of PRT relative to all radiotherapy use.

Similar analyses were performed evaluating the use of 3D and IMRT/VMAT photon-based techniques. We observed a dramatic decrease in 3D techniques in photon therapy beginning in 2011, especially among CNS, RST and NBL trials (Fig. 5A-C, Supplementary Fig. 3A-C). IMRT/VMAT has become the dominant photon technique for nearly all tumor sites, in terms of both the number and percentage of photon treatments since 2011 (Supplementary Fig. 3A-H).

PRT and Photon Utilization by States/Cities

In terms of state-level accrual for photon radiotherapy, California consistently contributed the highest number of cases between 1998 and 2020, while Texas and New York ranked second and third. Like what was found in all RT modalities, New York also emerged as the leading contributor for photon RT during the 2021–2025 period (Supplementary Fig. 4A–C). Regarding PRT, Massachusetts and California led contributions from 1998–2010; Texas contributed the most from 2011–2020; Florida, Texas, and Tennessee were the top contributors from 2021–2025 (Supplementary Fig. 4D-F), corresponding to the increase in number of proton facilities in Florida (n=8), Texas (n=3), and Tennessee (n=3) in the 2021–2025 era compared to the pre-2010 era where Florida and Texas each had one proton facility (Tennessee had zero).

When analyzing specific cities, Los Angeles, Memphis, Montreal, Atlanta and Chicago were the five largest contributors for photon utilization (Supplementary Fig. 5A), while Boston, Jacksonville, Loma Linda, Memphis and Houston were the top five for PRT utilization (Supplementary Fig. 5B).

QA Review Outcomes

IROC performance evaluation data showed a relatively low number of violation cases when considering all RT modalities collectively (Fig. 6A) or PRT independently (Fig. 6B). Among cases with definitive QA outcomes, the combined rate of “Appropriate” and “Acceptable Deviation” exceeded 90%, showing a consistent longitudinal improvement across the three eras (Fig. 6A). However, a transiently lower QA concordance rate was observed for the PRT subgroup during the 1998–2010 period. A significant proportion of evaluations from 2022 onward remain pending finalization, which is expected for recently activated, actively enrolling, or recently closed protocols. Final QA status may lag enrollment while complete imaging and RT objects, treatment records, and committee adjudication are submitted, reconciled, and finalized. Therefore, the pending category in the most recent era should be interpreted primarily as data immaturity at the time of extraction rather than evidence of protocol noncompliance or inferior QA performance. Comparative analysis based on geographic data revealed that QA performance in the US, which accounts for the majority of cases, is comparable to that of international cohorts (Fig. 6C-D). Corresponding PRT analysis was unable to be generated due to extremely limited number of cases of PRT internationally (N = 1).

Figure 6.

Quality Assurance (QA) Performance Evaluation of RT on COG Trials Over Time. A-B) Combo plot showing the performance evaluation of patients receiving any form of RT (A) or PRT (B) over time. C-D) Combo plot showing the performance evaluation of patients receiving any form of RT in USA (C) and other countries (D). The left Y axis corresponds to the bar plot showing the number of cases of each performance evaluation category. The right Y-axis corresponds to an overlaid line plot representing the cumulative proportion of cases rated as either “Appropriate” or “Deviation Acceptable” among all reviewed cases (excluding those marked as “Unevaluable” or “Pending Final Evaluation”), connecting the percentages shown above each era in all panels.

Discussion:

To our knowledge, this is the first study to leverage cooperative group metadata to explore the evolution of RT techniques in pediatric cancer patients. In this study, we utilized COG enrollment data from nearly a quarter century, spanning the eras of significant RT advancements including IMRT/VMAT and PRT adaptation. Based on this retrospective analysis, IMRT/VMAT gained significant traction around 2012 in pediatric patients, consistent with prior reports in adult and elderly cancer populations where IMRT replaced 3D as the most common modality by 2007.^10–12 Shen et al.¹² reported that IMRT usage has been increasing annually beginning in 2002 for Medicare patients, but the rate of growth began to decline, especially after 2007. Authors postulated the slowing growth rate likely reflected the saturation of IMRT adoption by radiation facilities. Unlike photon-based treatments, which deliver both entrance and exit doses to normal tissue, PRT exposes normal tissue only to an entrance dose. Because of the physical properties of the Bragg peak, the radiation stops completely within / adjacent to the intended treatment volume. Numerous dosimetric comparisons and retrospective cohort studies have shown the superiority of the dose distribution of PRT when compared to photons (with robust evidence in pediatric cancers^13–15); This is especially important in pediatric patients considering the radiosensitivity of developing tissue, late effects, and long life expectancy for pediatric survivors.^16–18 However, a major hurdle for the widespread implementation of PRT has been that proton centers are significantly more expensive than photon therapy centers and are therefore much less accessible. In our study, PRT users first accounted for over 20% of all RT modalities in 2014, representing nearly a threefold increase from 5.4% in 2008, which is significantly higher than the 83% increase in the proportional contribution of PRT in a previous analysis of private insurance claims for nonelderly adult cancer patients during the same time period.¹¹ As of 2025, there are 125 proton radiation centers worldwide in operation (45 located in USA), of which 107 were commissioned after 2010.¹⁹ This expansion underscores the increasing availability of PRT. Our analysis of COG patients revealed that PRT has been stabilizing at around 35% of all modalities used in RT since 2018, whereas IMRT/VMAT continues to be used more frequently, largely due to lower costs,⁶ broader accessibility, and reduced infrastructure requirements.

In our analysis, the number of total patient enrollment and patients receiving RT plateaued between 2004 and 2011, followed by a gradual decline. Although the number of trials began to increase again after 2020, the number of patients enrolled as well as those receiving RT has continued to decrease. This trend appears to be driven primarily by a sharp decline in the number of trials and patient enrollment targeting CNS, RST, REN and HOD disease sites. An apparent paradigm shift towards treatment de-escalation, driven largely by the advent of new targeted /immunotherapies in both pediatric hematological malignancies, and solid tumors may be contributing to this decline. Many patients live far away from proton centers, placing them and their families under financial and social burdens.^12,20,21 The COVID-19 pandemic and related travel restrictions may have further limited trial enrollment. A report from multicenter New York Area Institutions²² showed that 42% of patients receiving radiotherapy were significantly affected by COVID-19 pandemic either on workup or available modalities, treatment delay or cancelling, or changes in radiation timing and number of fractions.

Given that CNS tumors are the second most frequent pediatric cancer after leukemia²³, this explained the high representation in COG trials contributing to greater enrollment and thus a larger proportion of patients receiving RT. In this analysis, the initiation of PRT implementation has varied significantly among different disease sites. Trials targeting CNS and RST cancers were among the earliest to incorporate this modality. The reasons for these observations are likely multifactorial. First, the high incidence rate, and potential for dissemination throughout neuroaxis of CNS tumors⁵ demand the dosimetric advantages of PRT (especially in the setting of craniospinal irradiation), as well as the long-term benefits of reducing particular adverse events such as neurological deficits, endocrine dysfunction, growth retardation, hearing impairment, vascular disorders, as well as secondary cancer incidence.²⁴ Second, since many adverse events are dose-dependent, PRT is also commonly used in soft tissue sarcomas and Ewing sarcoma, where primary lesions are often located near critical anatomical structures (i.e., head and neck, spine, or pelvis).²⁴ Third, the relative radiosensitivity of rhabdomyosarcoma and Ewing sarcoma in combination with the high conformality of advanced RT techniques including PRT, have led to the adaptation of less extensive/morbid surgical resections in favor of these modalities.²⁴ PRT is able to maximize dose to the target volume while minimizing dose to surrounding tissue.

The QA process itself also evolved substantially across the same interval. Early cooperative group RTQA was largely retrospective because treatment plans and images were submitted in hard-copy format and reviewed after treatment completion or, when feasible, during the first days of treatment. As digital transfer and centralized review platforms matured, COG/IROC progressively moved toward prospective, real-time review of imaging and RT treatment objects before treatment delivery. This transition was driven by cooperative group experience demonstrating that RT deviations can influence trial outcomes and that pre-treatment review decreases deviation rates and improves interpretability of study results. Modern COG RTQA commonly includes protocol-specific credentialing, digital submission of diagnostic and simulation imaging, RT structure sets, treatment plans, and dose files, and central review by disease-site and RTQA experts before treatment start when mandated by protocol. This evolution is important for the present study: increasing use of IMRT/VMAT and PRT occurred in parallel with a more rigorous QA infrastructure capable of supporting complex conformal treatment across many institutions while maintaining high protocol compliance.²⁵

Selection bias should be considered when interpreting the abovementioned findings. Enrollment in cooperative group trials often requires access to specialized pediatric oncology centers, typically in urban or academic settings, and may disproportionately include patients with higher socioeconomic status and better access to advanced technologies, multidisciplinary care, and supportive services. Consequently, outcomes from cooperative group cohorts may not fully reflect the broader pediatric cancer population. Another limitation of this study is the restricted access to patient specific data within the IROC dataset due to Protected Health Information (PHI) constraints. Consequently, while we were able to characterize the overall age and geographic distribution of the enrolled cohort, we could not evaluate potential disparities in RT access across different racial, ethnic, or gender demographics. Follow-up studies will be extremely important to answer these questions. It was beyond the scope of our study to include analysis of clinical outcomes or adverse events, and we are therefore unable to draw conclusions about the risks/benefits based on RT modality. However, individual COG trial studies will continue to analyze the effects of RT modality as is being reported currently.^26,27 None of the results thus far show any difference in cure rates between IMRT and PRT. Also, this study did not distinguish between passive scatter proton therapy versus pencil beam scanning (PBS) and therefore, we are unable to determine if there are differences between these techniques within PRT treated patients.

Conclusion:

This is the first systematic study to evaluate the evolving patterns of all modalities of RT within COG trials over the past quarter century. Our findings indicate that around 2010, IMRT/VMAT gained significant traction, replacing 3D as the predominant modality, while PRT use began to increase steadily. This trend aligns with the remarkable technological advancements in RT during that period. The highest adoption of PRT was observed in CNS and RST protocols, which also dominate the number of included trials and total patient enrollment. The absolute number of patients treated with PRT remains lower than IMRT/VMAT, but now stable; approximately 35–40% of all patients receiving RT on COG trials receive PRT. These trends illustrate how RT techniques have evolved within COG trials and highlight the need for future work that situates these patterns within national/international practice and links modality selection to toxicity, survivorship, and access to care.

Supplementary Material

1

Suppl Fig 1. Age distribution in this study. A) Bar plot showing the age distribution of all patients included in the analysis. B) Bar plot showing the annual median age in all RT modalities, photon and PRT from 1998–2025.

Suppl Fig 2. Trends in Photon-Based RT Techniques over Time. A) Line plot showing the percentage of patients receiving either 3D or IMRT/VMAT photon RT relative to patients receiving photon-based RT, grouped by the US and international countries.

Suppl Fig 3. Trends in PRT and Photon Use by Primary Disease Site Over Time. Solid lines represent the raw data of IMRT/VMAT among patients receiving photon therapy by disease site each year, dotted lines represent the raw data of 3D conformal photon therapy among patients receiving photon therapy, and dashed lines represent the raw data of PRT relative to all radiotherapy use.

Suppl Fig 4. Geographic distribution of cases across photon and PRT in binned eras. Photon utilization in A) 1998–2010; B) 2011–2020; C) 2021–2025. PRT utilization in D) 1998–2010; E) 2011–2020; F) 2021–2025.

Suppl Fig 5. City-level accrual of photon and PRT. A) Cities contributing over 50 cases of photon all time (1998–2025); A) Cities contributing over 10 cases of PRT all time (1998–2025);

Data Sharing Statement:

The data underlying this article will be shared on reasonable request to the corresponding author.