Precision Medicine · global
Which Children With Brain Tumors Could Receive Less Radiotherapy? MANTIS Seeks Clues in Molecular Data
An AI model built using data from 2,786 patients attempts to turn childhood brain tumor risk grouping into personalized survival predictions. A retrospective analysis of reduced radiotherapy intensity offers clues, but the findings are not yet sufficient to guide treatment changes.
For children with malignant brain tumors, the costs of treatment may extend into adulthood: alongside controlling the tumor, treatment must protect learning, development, and future quality of life as much as possible. Which children need to maintain treatment intensity, and which might receive less radiation? A study published in Nature Genetics seeks more detailed clues in tumor molecular data.
An international team led by researchers at Dana-Farber Cancer Institute developed an artificial intelligence model called MANTIS, studying medulloblastoma and ependymoma. According to the paper abstract listed by the Broad Institute, the team built the model using genome-wide DNA methylation array data from 2,786 patients to estimate each patient's probability of survival over the next 10 years, rather than simply assigning patients to a few broad risk groups.
The model uses both DNA methylation patterns and copy number changes; the former reflect chemical marks on DNA, while the latter provide information about gains and losses of genomic segments. Such methylation testing is already used to help diagnose brain tumors, so the study's practical appeal lies in whether existing test data can also support prognosis. Children with the same diagnosis, even those in the same risk group, may still have very different disease courses.
According to Dana-Farber, MANTIS outperformed current risk assessment methods for many patients across multiple large, independent validation cohorts. The highest concordance index reported in the paper abstract was 0.80 ± 0.07. This metric measures the model's ability to distinguish the order of patients' survival times; it cannot be directly interpreted as meaning that “predictions for every patient are 80% accurate,” nor does it mean the model has been shown to improve treatment outcomes.
The part more closely connected to clinical decisions involved extending the model to estimate the effects of different radiotherapy intensities. A retrospective analysis found that, among the approximately 60% of average-risk medulloblastoma patients identified by the model, those who received lower-dose craniospinal irradiation showed no statistically significant survival disadvantage (P = 0.683). Reports on the same study from Dana-Farber and The ASCO Post also noted that lower-dose irradiation was associated with better neurocognitive performance.
However, “no significant difference was found” does not establish that the two doses are equally effective. This retrospective finding proposes an approach to patient selection that warrants testing in prospective trials. It cannot yet confirm that reducing doses based on the model can preserve both tumor control and cognitive function. The radiotherapy analysis also focused on medulloblastoma and cannot be directly extended to support dose reduction in ependymoma.
In institutional coverage and reports on the same study, senior author Volker Hovestadt and first author Sabina Stefan described personalized treatment intensity as a future direction and emphasized the need for further clinical research. MANTIS currently offers a path that can be tested: tumor testing could potentially help answer not only “What disease is this?” but also “How much treatment is needed?” Whether it can actually reduce long-term harm still requires confirmation in prospective trials.