onc brain

Single-institution cohort from MD Anderson's INTERCEPT program, 2020-2024, N=110, with serial tumor-informed ctDNA during nonoperative management. Median follow-up 25 months (IQR 18-37).

Microsatellite stable stage I-III rectal adenocarcinoma achieving cCR or near-cCR after neoadjuvant therapy and managed nonoperatively. Median age 56; baseline cT3 in 72 and cT4 in 16; 69 received TNT, 41 CRT or chemotherapy alone.

Local regrowth and distant metastasis by longitudinal ctDNA status, by first post-NAT ctDNA (within 180 days), and per-sample accuracy for an event within ±90 days of each draw. Salvage-surgery pathology by ctDNA status at regrowth.

Twenty-three pts (21%) had local regrowth and 12 (11%) distant metastasis. Ever-positive ctDNA separated both curves (log rank p=0.0002 for regrowth, p<.0001 for metastasis). The per-sample table carries the operating characteristics.

Per-sample analysis pools 669 draws from 110 pts without accounting for repeated measures, so the confidence around 41% is softer than the denominator suggests. The first-post-NAT comparison rests on n=12 evaluable pts as reported in the source. No comparison against MRI or endoscopy, the tests ctDNA would have to beat.

The asymmetry between local (41%) and distant (74%) sensitivity is the informative result: intraluminal regrowth from a small residual burden sheds too little DNA to be caught reliably, while metastatic disease does. That biology argues for ctDNA as a distant-recurrence tool layered onto, not substituted for, luminal surveillance.

Retrospective single-center cohort; per-sample analysis treats 669 draws from 110 pts as independent. No head-to-head against MRI/endoscopy surveillance.

Post hoc intracranial analysis of the randomised DeLLphi-304 trial of second-line tarlatamab versus chemotherapy in SCLC. Two assessment frameworks: investigator RECIST in the ITT population and mRANO-BM by BICR in pts with baseline brain metastases. Data cutoff January 29, 2025; median follow-up 11.4 mo (tarlatamab) and 11.5 mo (chemotherapy).

ITT n=254 tarlatamab versus n=255 chemotherapy. The brain-metastasis cohort was n=67 versus n=56, defined as ≥1 brain metastasis at baseline per mRANO-BM by BICR plus ≥1 post-baseline scan. More than 70% had already received CNS-directed treatment, which is why the response categories were CR, non-CR/non-PD and PD rather than a conventional ORR.

This analysis reports time to CNS progression or death (CNS PFS) in the ITT population and in the brain-metastasis subset, plus best overall CNS response, CNS disease control rate, duration of CNS complete response and duration of CNS disease control. None of these was the trial's registered primary endpoint.

Both CNS PFS comparisons favor tarlatamab, and the effect is larger in the brain-metastasis subset (HR 0.40) than in the ITT population (HR 0.54). Intracranial complete response was 14.9% versus 5.4%, and CNS tumor shrinkage ≥30% occurred in 9/16 versus 5/13 pts with measurable lesions ≥10mm.

Brain metastases develop in the majority of SCLC pts, and second-line systemic options have historically been judged on extracranial disease with CNS control assumed to be the province of SRS, WBRT or prophylactic cranial irradiation. A bispecific T-cell engager showing an intracranial complete response rate of 14.9% is the notable part, because the working assumption for large-molecule agents has been limited intracranial penetration.

The ITT median CNS PFS was not estimable at 11.4 months of follow-up, so the ITT curve rests on the early portion of the data and the point estimate can still move. The two populations were also read with different tools and different models (stratified Cox for ITT, unstratified for the subset), so the ITT and subset HRs are not strictly comparable to each other.

The result establishes that intracranial disease does not progress faster under tarlatamab than under chemotherapy, and probably progresses more slowly. What it does not establish is whether that activity is sufficient to defer local therapy in a patient who would otherwise be referred for SRS, since the source reports no radiotherapy exposure data and no in-field versus out-of-field failure pattern.

Post hoc intracranial analysis of a randomised trial; CNS endpoints were not the registered primary, and the brain-met subset HR comes from an unstratified model.

Post-hoc genomic-classifier analysis of the ENZAMET ADT + enzalutamide cohort, N=320, testing whether Decipher (DPMC) predicts docetaxel benefit. Docetaxel receipt was investigator-elected, not randomised, so a propensity-score IPTW model carries the comparison.

mHSPC on an ADT + enzalutamide backbone with an evaluable Decipher score, split at the prespecified 0.85 cut. A separate panel restricts to low-volume disease.

Overall survival, read two ways: prognostic (Decipher stratum on ADT + enza) and predictive (docetaxel-by-Decipher interaction).

Prognostic signal is the cleanest number on the slide deck: HR 3.02 (1.50-5.76) for DPMC >0.85 on ADT + enza. The predictive claim rests on the interaction, p=0.02 unweighted and p=0.04 after IPTW, not on either stratum reaching significance.

Docetaxel was not randomised within this cohort, so IPTW is doing the causal work and the authors concede residual imbalance after propensity scoring. Low-volume panel numbers are small (n at risk 14 and 13 in the reported strata) with a CI of 1.70 (0.32, 8.06), wide enough to be uninformative.

Presented as level 1B evidence consistent with CHAARTED and STAMPEDE, where docetaxel benefit concentrated in high-volume disease. This analysis proposes a genomic rather than volumetric selector for the same decision.

The strongest defensible claim is the prognostic one: DPMC >0.85 marks worse OS on ADT + enza. The predictive claim (add docetaxel above 0.85, omit it below) is directionally consistent across unweighted and weighted models but is not established by a single non-randomised interaction.

Post-hoc biomarker subgroup with non-randomised docetaxel use; both stratum HRs non-significant after weighting, residual imbalance conceded. Read rests on an interaction p=0.04.

Clinico-transcriptomic analysis of NRG cohorts modeling the 22-gene GC as a continuous variable (per 0.1 unit) alongside clinical covariates, with treatment carried as a covariate. Contributing trials, N, and follow-up duration are not reported in the source.

NCCN ≥ high-risk localized prostate, spanning NCCN HR and VHR. Baseline age, PSA, and grade distributions are not reported in the source.

Every branch of the framework keeps RT + ADT. No dose, fractionation, target volume, or ADT duration is reported, so RT enters as a fixed backbone rather than a variable under test.

Prognostic performance assessed on MFS, distant metastasis, and OS. No single primary endpoint is stated, and the framework itself is not compared against a randomized alternative.

GC was independently prognostic for all three endpoints (p<0.001); the row-level hazard ratios are not legible in the source slide. Approximately ¼ of the ≥HR population carries discordant clinical vs biomarker risk.

STAMPEDE M0 (Attard, Lancet 2022) is the external yardstick: the CT HR and CT VHR curves are overlaid on that trial's RT + ADT arm rather than on a randomized internal control. Eligibility, staging era, and ADT duration differ between cohorts, so the vertical gap carries trial effects alongside risk-group effects.

The GC term's hazard ratios and confidence intervals are not legible in the source, so the size of the increment over clinical variables cannot be checked. The 9% clinically-high / biomarker-low cell is where the framework withholds intensification, and its outcomes are the ones a reader most needs before acting.

The contribution is a parsimonious integer score computable at the consult desk, extending Spratt et al. (JCO 2018) to the modern VHR population. What it does not settle is whether the signal separating these curves also identifies who benefits from AAP: a prognostic split widens absolute benefit for a uniformly effective drug without any interaction, and none is reported here.

Retrospective pooled analysis; GC shown prognostic, not predictive. Intensification threshold benchmarked against STAMPEDE across trials, not randomized within cohort; no treatment-by-GC interaction in source.

Post-hoc analysis of the pivotal single-arm phase 2 celiac plexus radiosurgery trial (NCT03323489), n=90 evaluable. Candidate baseline predictors screened by univariate then multivariate logistic regression, with only multivariate-significant variables carried into the score. Internal validation by bootstrap resampling, 500 iterations, for an optimism-corrected AUC.

Pain response per parent protocol: a ≥2-point reduction in average pain on the Brief Pain Inventory-Short Form, baseline to three weeks. Parent-trial response was 53% (95% CI 42-64%).

Only neurotoxic chemotherapy exposure (OR 5.1, p=0.009) and baseline pain intensity (OR 1.8, p=0.003) survived multivariate analysis; age and therapy line dropped out to collinearity. The resulting 2-point score separated response into 32% / 53% / 89%, with AUC 0.716 apparent, 0.714 optimism-corrected.

The score was derived and internally validated in the same 90 pts, so the bootstrap corrects optimism from resampling but not from the variable-selection step that preceded it. The 2-point stratum is n=19, and the dichotomy at pain >6 was picked from the same data that shows steeper response above 7 and 8.

The neurotoxic-chemo term is the interesting one because it is a timing variable a clinician controls, not a fixed patient trait: it implies referral for celiac SRS competes with the systemic sequence rather than following it. Whether that reflects true nerve injury blunting an ablative pain response, or confounding by later-line disease biology, the post-hoc design cannot separate.

Post-hoc derivation on the parent trial's own 90 pts; internal bootstrap only, no external cohort. Design dominates the read regardless of effect size.

Multicenter retrospective cohort, Spain, 2010 to 2022, N=369 treated with definitive trimodality therapy (maximal TURBT then concurrent chemoradiotherapy). Predictors of response identified by multivariable logistic regression. Follow-up duration not reported in source.

cT2-T4aN0M0 MIBC selected for bladder preservation. Median age 76, 85.1% male. Fitness for cystectomy and completeness of TURBT are not reported in source.

Concurrent chemoradiotherapy, regimen at each center's discretion. 5-FU-based CRT predicted higher complete local response (OR 4.9, 95% CI 1.1-22.1, p=0.038). The comparator regimens are not named in source.

Dose, fractionation and target volume are not reported in source. The only technique signals reported are verification frequency (weekly portal imaging, OR 0.35 for CLR) and a non-significant trend favoring VMAT.

Primary: complete local response (CLR). Secondary: OS, CSS, recurrence patterns, salvage cystectomy. No OS or CSS estimate appears in the source, so the survival half of the conclusion cannot be checked.

CLR 63.7%, salvage cystectomy 9.7%. Progression 28.8%, with systemic failure (10.7%) running at or above local-only failure (10.1%).

BC2001 established the locoregional-control gain from adding chemotherapy to bladder radiotherapy, and the pooled RTOG bladder-preservation experience set the complete-response benchmark; both were protocol populations. This adds European real-world multicenter data at a median age of 76, with no internal cystectomy comparator.

No follow-up duration, OS or CSS estimate is reported, so the durability behind the preservation claim cannot be judged. The 5-FU odds ratio spans 1.1 to 22.1, compatible with a marginal or a large effect, and in a retrospective series regimen choice tracks renal function and performance status.

Systemic failure at 10.7% running at or above local-only failure at 10.1% argues the ceiling in this population is micrometastatic disease rather than the bladder, which caps what further local intensification can buy. Read conservatively, the predictors favor modern image guidance and an active concurrent backbone, not any specific verification schedule.

Retrospective multicenter cohort with no cystectomy comparator; imaging and chemo predictors come from multivariable regression across a 2010-2022 era shift, so a causal reading is unsupported.