{"doi":"10.1093/neuonc/noaf211","title":"Bayesian reappraisal of ACNS0332 and ACNS0334 strengthens subgroup treatment effects in high-risk pediatric Group 3 medulloblastoma","abstract":"Bayesian reappraisal of ACNS0332 and ACNS0334 strengthens subgroup treatment effects in high-risk pediatric Group 3 medulloblastoma We were excited by the results of the Children’s Oncology Group phase 3 ACNS0334 trial,1 particularly the subgroup analysis suggesting an event-free survival (EFS) benefit of high-dose methotrexate (HD-MTX) in young children (<3 years) with high-risk Group 3 medulloblastoma (hr-MB-G3), the most aggressive medulloblastoma subtype. However, our excitement was tempered by the challenge of achieving sufficient statistical power in trials of rare pediatric cancers, which often necessitates accepting greater uncertainty. Similarly, an earlier promising subgroup signal from the phase 3 ACNS0332 trial,2 showing improved EFS with carboplatin intensification in older children (ages 3–21) with hr-MB-G3, was not powered for subgroup analyses. As a result, hr-MB-G3 subgroup analyses in both trials were underpowered: our Cox regression using reconstructed individual patient data yielded non-significant hazard ratios (HRs) for EFS of 0.33 (95% confidence interval [CI], 0.09-1.22; two-tailed P = .1) for HD-MTX in ACNS0334 and 0.54 (95% CI, 0.26-1.13; two-tailed P = .1) for carboplatin intensification in ACNS0332 (Figure 1). These limitations—which both trials acknowledge—may hinder clinical adoption of otherwise promising therapies. Therefore, we conducted a rigorous Bayesian reanalysis using historical data to reassess evidence for (1) HD-MTX arm of ACNS0334 in younger children and (2) carboplatin-intensified arm of ACNS0332 in older children with hr-MB-G3. We reanalyzed the trials using individual patient-level data reconstructed from published Kaplan-Meier curves, incorporating historical control data from patient populations of similar age groups treated in SJYC073 and PBTC-0264 (younger; <4 years), and SJMB03,5 MET-HIT-2000-AB4,6 and PNET-HR + 57 (older). Using the BayesFBHborrow R package, we implemented a model that dynamically borrows from historical data, reducing the degree of borrowing from the historical data when they differ from the current trial to avoid inflating treatment effects due to time trends or changing standards of care. We pre-specified similarity thresholds for the ACNS0334 and ACNS0332 analyses using the empirically determined log-hazard range (reflecting the estimated risk of EFS events over time) in each trial’s control group, resulting in prior borrowing weights of 49.6% and 69.2% for the historical data matched to the control groups of ACNS0334 (younger children) and ACNS0332 (older children), respectively—reflecting a cautious, data-driven stance. Treatment effects were estimated with full uncertainty quantification using 10 000 Markov Chain Monte Carlo simulations, yielding HRs and credible intervals (CrI; the Bayesian equivalent of CIs) that reflect both within-trial and historical variation. Our Bayesian reanalysis yielded more precise estimates: mean HR of 0.24 (95% CrI, 0.05-0.64) for the HD-MTX arm of ACNS0334 in younger and 0.44 (95% CrI, 0.23-0.76) for the carboplatin-intensified arm of ACNS0332 in older children with hr-MB-G3 (Figure 1). The probability of observing these HRs under the null hypothesis of no treatment effect was 1% and 0.3%, respectively. Adjusting for up to 5 multiple subgroup comparisons and assuming a modest 20% prior chance of true benefit, the posterior probability, which reflects the updated likelihood of a real treatment effect given the data, was 92% and 98%, respectively. We also developed a Bayesian analog to the Fragility Index8—the minimum number of events that must be erased (censored) to make the 95% CrI include 1. This number exceeded the total observed events in both analyses, reflecting highly stable results. By cautiously borrowing from historical trial data, we addressed a key limitation of the original hr-MB-G3 subgroup analysis conducted in ACNS0334 and ACNS0332. We implemented a Bayesian approach to generate intuitive, probability-bas","journal":"Neuro-Oncology","year":2025,"id":581851,"datarank":0.0,"base_score":0.0,"endowment":0.0,"self_citation_contribution":0.0,"citation_network_contribution":0.0,"self_endowment_contribution":0.0,"citer_contribution":0.0,"corpus_percentile":null,"corpus_rank":null,"citation_count":0,"citer_count":0,"citers_with_citation_signal":0,"citers_with_endowment":0,"datacite_reuse_total":0,"is_dataset":false,"is_dataset_confidence":0.9524,"is_data_producer":false,"deposit_databanks":null,"is_oa":true,"file_count":0,"downloads":0,"has_version_chain":false,"published_date":"2025-01-01","fair_score":null,"fair_percentile":null,"algorithm_id":"datarank_citation_only_1hop_v6","ranking_scope":"data_only","authors":[{"id":1493748,"name":"Michael A Huang","orcid":null,"position":1,"is_corresponding":false},{"id":1493749,"name":"J. A. Beall","orcid":null,"position":2,"is_corresponding":false},{"id":353771,"name":"Akshitkumar M. Mistry","orcid":"0000-0002-7918-5153","position":3,"is_corresponding":false},{"id":1198638,"name":"Rahim Abo Kasem","orcid":"0000-0002-9635-7471","position":0,"is_corresponding":true}],"reference_count":8,"raw_metadata":null,"created_at":"2026-07-19T02:58:51.328454Z","pmid":"41252268","pmcid":null,"fwci":null,"citation_percentile":null,"influential_citations":0,"oa_status":null,"license":null,"views":0,"total_file_size_bytes":0,"version_count":0,"fair_f":null,"fair_a":null,"fair_i":null,"fair_r":null,"fair_zscore":null,"fair_rationale":null,"fair_model":null,"fair_agent_version":null,"fair_fulltext_source":null,"fair_has_llm":null,"fair_computed_at":null,"clinical_trials":[],"software_tools":[],"db_accessions":[],"linked_datasets":[],"topics":[]}