Meningioma recurrence risk estimates vary with era, classification, geography, healthcare

When a meningioma is removed, the question that haunts every follow-up scan is deceptively simple: will it come back? For decades, clinicians have answered with risk estimates drawn from single-center cohorts and historical comparisons, figures that shape how aggressively surgeons pursue resection, which patients are offered radiotherapy, and how closely survivors are monitored. A new study now warns that these estimates are far less stable than the field has assumed. Analyzing 4,111 patients with grade 1 and grade 2 meningiomas treated at 31 centers in 15 countries between 1990 and 2019, an international research team found that recurrence risk estimates shift substantially depending on the calendar period of diagnosis, the edition of the World Health Organization classification in use, the geographical setting, and even the way a nation’s healthcare system is financed. The findings, published on 28 August 2026 in the Journal of Neuro-Oncology, strike at the foundation of how brain-tumor research measures one of its most basic yardsticks.

Meningiomas arise from the meninges, the layered membranes that envelop the brain and spinal cord, and they rank among the most common primary intracranial tumors. Because most grow slowly, recurrence research demands follow-up stretching across years or decades, which is why single-center cohorts routinely accumulate patients over long stretches of calendar time. The customary shortcut—comparing contemporary results with historical cohorts that appear similar in design—has always been fraught, because the individual-level data needed to adjust for confounders are rarely available, limiting comparisons to qualitative rather than quantitative evaluation. The stakes reach well beyond academic bookkeeping: recurrence risk estimates underpin molecular classification schemes, risk stratification models, and the benchmark figures against which new meningioma therapies are judged. Led by neurosurgeon Christian Mirian of Copenhagen University Hospital, the research team therefore set out not to identify what drives recurrence, but to test whether the estimates themselves behave consistently when similar patients are evaluated across different eras, classification editions, geographical settings, and models of healthcare.

The evidence base is the PERNS database—short for PERsonalized NeuroSurgery—an international retrospective collaborative platform established to harmonize individual-level clinical, surgical, histopathological, and follow-up data from patients with primary meningiomas diagnosed, treated, and followed between 1990 and 2019. The analysis centered on 4,111 adults operated on for primary WHO grade 1 or grade 2 meningiomas: 3,008 carried grade 1 tumors and 1,103 grade 2 tumors, with 3,217 patients classified under the 2007 WHO edition and 894 under the 2016 edition. The cohort accrued 22,327 person-years of follow-up in total, with a median of 5.2 years and a range stretching from less than 0.1 to 28.7 years. Over that period, 645 patients—15.7 percent—experienced a recurrence, 276 died without recurrence, and 3,190 were censored alive and recurrence-free; only 235 patients, or 5.7 percent, received adjuvant fractionated radiotherapy. Crucially, recurrence was assessed locally by radiological evaluation without standardized criteria, because most cases predated the Response Assessment in Neuro-Oncology framework that now standardizes meningioma endpoints.

The study’s statistical centerpiece is regression standardization, also known as G-computation. Rather than comparing crude recurrence proportions, the researchers fitted multivariable logistic regression models—equipped with inverse probability of censoring weights to handle the vast variation in follow-up duration between centers and eras, and treating death without recurrence as a competing event—and used them to predict what the average five- and ten-year recurrence risk would be for a reference population if those same patients, carrying an identical distribution of ages, sexes, skull-base versus non-skull-base locations, WHO grades, classification editions, Ki-67 proliferation indices, Simpson resection grades, and radiotherapy exposures, had instead been diagnosed in another calendar period, graded under another WHO edition, treated at a different center, or managed in a different healthcare system. Four separate models covered the four comparisons, each adjusted for the full panel of clinical, surgical, and histopathological covariates, with interaction terms allowing the Ki-67 proliferation index to exert a different effect depending on the extent of resection. Confidence intervals came from bootstrap resampling, and any residual difference can only reflect factors the measured variables cannot capture.

The calendar-time comparison, which standardized the reference population of patients diagnosed between 2008 and 2012 to hypothetical diagnoses in other eras, produced the study’s headline result: more recent periods carried higher predicted recurrence risks. For grade 2 tumors, the predicted five-year risk was 1.60 times higher for patients diagnosed in 2013 or later than for identical patients diagnosed in 2007 or earlier (95 percent confidence interval 1.19 to 2.01, P = 0.004), and 1.51 times higher than for the same patients diagnosed in 2008–2012 (95 percent CI 1.09 to 1.92, P = 0.017); there was no significant difference between the two earlier periods. Among grade 1 tumors, where early recurrences are uncommon, the signal emerged later: at ten years, patients diagnosed in 2008–2012 had a predicted risk 1.35 times that of identical patients diagnosed by 2007 (95 percent CI 1.10 to 1.60, P = 0.007). Patients diagnosed from 2013 onward could not enter the ten-year analysis at all, since none had accrued a decade of follow-up when data collection closed in 2019.

Context makes those numbers intelligible. When Donald Simpson defined recurrence in 1957, he meant the reappearance of symptoms caused directly by tumor growth after a period of relief—a definition bounded by what patients could feel. Contemporary imaging is far sharper: high-resolution gallium-68 DOTA-TOC positron emission tomography can flag meningioma lesions as small as 0.1 cubic centimeters. Rising estimates over time, the authors argue, therefore track intensifying surveillance and more sensitive technology rather than deteriorating surgery, and the grade-specific timing supports that reading. The excess among grade 1 tumors appeared at ten years but not five, consistent with these slow-growing lesions continuing to recur long after operation, while the grade 2 excess surfaced within five years, the window in which atypical meningiomas typically relapse. The patterns, the team emphasizes, should not be interpreted as evidence that treatment efficacy has worsened; they are consistent with evolving follow-up strategies built on more frequent and more sensitive imaging.

The comparison between WHO classification editions delivered a quieter verdict. The pivotal change between the 2007 and 2016 editions was the elevation of brain invasion to a standalone criterion for grade 2, and among grade 2 patients the five-year risk was indeed nominally higher under the newer edition (risk ratio 1.28, 95 percent CI 0.94 to 1.62), though the difference fell short of statistical significance (P = 0.10). At ten years the ratio was 1.13 (95 percent CI 0.90 to 1.36). Grade 1 estimates were statistically indistinguishable across editions at both horizons, although the five-year comparison carried enormous uncertainty (risk ratio 0.69, 95 percent CI 0.00 to 1.51) because early recurrences are rare among these tumors. Prior evidence on the prognostic weight of brain invasion has been mixed, and while the finding hints at a short-term fingerprint of the 2016 criteria on recorded recurrence within this cohort, the data stop short of proof.

Geography told a more disquieting story. For this comparison the team confined itself to seven European cohorts from tax- or social-insurance-funded systems, each with a median follow-up of at least five years and at least twenty grade 2 patients, and standardized every center’s estimate to the distribution of patient characteristics in a 410-patient Geneva cohort serving as the reference. Even after aligning age, sex, tumor location, grade, proliferation index, and resection extent, predicted recurrence risks swung widely between centers. The explanation surfaced through the study’s methodological innovation: “Cohort-Event” plots that trace every individual patient from the year of diagnosis to the end of observation, color-coded for recurrence, recurrence-free death, or censoring. The plots exposed stark irregularities—one center enrolled patients unsystematically between 1990 and 2000, and during that window recruited exclusively patients who had already recurred, before shifting to consecutive enrollment, while another recorded patients consecutively early in the study period and then drifted into sporadic additions consisting almost entirely of recurrent cases, most detected early. Non-uniform data accrual, in other words, can warp recurrence estimates even after rigorous adjustment, quietly sabotaging comparability between cohorts that look methodologically identical on paper.

Healthcare systems left their own signature. Taking patients treated in social-insurance-funded systems—mandatory multi-payer arrangements such as those of Germany, France, Switzerland, Hungary, Japan, and South Korea—as the reference, the models predicted higher recurrence risks for identical patients if treated in mixed or privately funded systems such as those of the United States, India, and China (risk ratio 1.47, 95 percent CI 1.20 to 1.74, P < 0.001), while estimates for tax-funded systems such as those of Spain, Italy, Norway, Sweden, and Canada ran generally comparable or lower, with a ten-year ratio of 0.69 (95 percent CI 0.48 to 0.90, P < 0.001). The authors are unequivocal that these figures do not crown any system superior or inferior. What they expose is machinery: follow-up routines, access to imaging, documentation practices, and financial incentives that shape surveillance intensity and thresholds for reintervention all feed into whether, and when, a recurrence is detected and recorded—and therefore into what the statistics ultimately say.

The consequences ripple directly into clinical trials. Single-arm meningioma studies routinely benchmark efficacy against historical progression-free survival rates such as PFS-6 or PFS-12, thresholds inherited from cohorts diagnosed, scanned, and recorded under earlier regimes; if recurrence detection is sensitive to follow-up intensity, imaging modality, and data-accrual practices, then historical benchmarks are not stable yardsticks, and observed treatment effects may partly reflect differences in outcome ascertainment rather than genuine therapeutic benefit. The findings cast an equally pointed shadow over molecular research, since modern classifiers are often built by linking molecular profiles from retrospective specimens to outcomes recorded under heterogeneous historical conditions—meaning that associations celebrated as biology may partly be artifacts of how events were found and logged. Recurrence events, the authors conclude, remain inherently “locked” to the historical context in which they were detected. The team acknowledges its retrospective design cannot capture unmeasured factors such as selective enrollment, molecular markers absent from older datasets, or the unstandardized follow-up schedules that predated modern response criteria. Its prescription, however, is procedural: transparent visualization of cohort composition and individual-level event timing, alongside explicit documentation of surveillance practices, recurrence definitions, and data accrual, so that the numbers oncology relies upon can finally be compared on honest terms.