Citations

Abstract Background: For large language models (LLMs) to reach their potential as information technology tools that make medication use safer, clinically relevant benchmarks capable of automated grading and designed specifically to measure the performance of LLMs for medication tasks are required. The purpose of this study was to design a suite of benchmarking tests reflective of Comprehensive Medication Management (CMM; the standard of care for medication optimization) and quantify the baseline performance of the latest LLMs. Methods: We established six benchmarks representing critical stages of the CMM process: drug formulation matching, drug order (sig) generation, drug route matching, drug-drug interaction identification, renal dose identification, and drug-indication matching. For each benchmark, we curated a clinician-annotated dataset comprising 250 standardized input-output pairs including both inpatient and outpatient medications. We evaluated the clinical knowledge retrieval capabilities of three LLMs: GPT-4o-mini, MedGemma-27B, and LLaMA3-70B. We employed a zero-shot prompting strategy, excluding in-context examples, to assess the models’ internal clinical knowledge rather than their few-shot learning potential. To check reliability, each model was run three times using a temperature of 0.7 (a mid-range value of an LLM setting controlling text generation randomness). Performance was assessed using task-specific evaluation metrics including precision (positive predictive value), recall (sensitivity), F1-score, accuracy, and correctness consistency across trials. Results: Across six benchmarks, LLaMA3-70B demonstrated the highest performance in four tasks: drug-formulation matching (F1, 54.0% [95 CI: 50.1-58]), drug-order generation (accuracy, 88.0%), drug-route identification (F1, 74.3% [95 CI: 71-78]), and drug-indication identification (accuracy, 97.6% [95 CI: 95.6-99.2]). In the drug–drug interaction task, GPT-4o-mini achieved the highest overall accuracy (70.4% [95 CI: 64.8-75.7]). For renal dose–adjustment identification, GPT-4o-mini demonstrated the highest F1 score (83.3% [95 CI: 77.6-88]). Correctness-consistency scores ranged from 8.0% to 97.6% across benchmarks, with no model exhibiting uniformly superior consistency. Conclusions: Model performance varied substantially across medication-related tasks. LLaMA3-70B demonstrated promising baseline performance in tasks involving formulation, ordering, route, and indication. GPT-4o-mini showed potential advantages in drug–drug interaction detection and renal dose adjustment. These findings underscore the need for task-specific evaluation when deploying models for medication-focused clinical decision support.

Abstract Background The purpose of this study was to develop a series of benchmarks specifically for large language model (LLM) performance in the identification of drug-drug interactions (DDIs). Methods We evaluated six LLMs using a clinician-designed dataset of 750 DDI scenarios. Tasks included: (1) a pointwise two-drug classification task (2) a pairwise three-drug discrimination task and (3) a listwise 4–6 drug selection task. Zero-shot prompting was applied and performance was assessed using precision, recall, F1 score, accuracy, and self-consistency. Results Model performance varied by task structure and interaction severity. GPT-5-Chat demonstrated the highest recall and F1 score in the pointwise task (80.7% and 80.2%), whereas DrugGPT achieved superior accuracy in the pairwise (93.7%) and listwise task (92.6%). All models had improved performance with more severe compared to less severe DDIs (accuracy 72.0-99.2% vs. 58.3-86.7%). All models tended to have higher self-consistency in the pairwise and listwise experiments compared to the pointwise experiment (68.8-93.6% vs. 31.0-74.2%). No model exhibited uniformly high reliability across all tasks. Conclusions Current LLMs show inconsistent capabilities in identifying DDIs. Performance degrades as reasoning space expands with limited stability. These findings emphasize the need for multi-format evaluation frameworks and reliability-aware assessment when considering LLMs for medication-safety applications.

PURPOSE Previous studies suggest that SPOP mutations result in increased sensitivity to androgen receptor pathway inhibitors (ARPIs) but are limited by lack of granular data. We hypothesize that SPOP mutations are associated with improved outcomes across the spectrum of advanced prostate cancer (PC). METHODS Using the real-world clinicogenomic Prostate Cancer Precision Medicine Multi-Institutional Collaborative Effort database, we analyzed outcomes based on SPOP mutation status. The primary end point was overall survival (OS) from the diagnosis of metastatic disease. Secondary end points included real-world progression-free survival (PFS) and PSA90 response. RESULTS Among 2,097 patients with metastatic PC, 5.5% (N = 115) had SPOP -mutated tumors. Compared with SPOP wild-type, patients with SPOP mutations were older at diagnosis (median 65 v 63 years; P = .001) and had higher (≥8) Gleason sum (68% v 54%; P = .02), higher incidence of lung metastasis (17% v 6%; P = .001), and increased frequency of intraductal features (13% v 5%; P < .001). SPOP mutations were associated with improved PSA90 response with first ARPI exposure in the castrate-resistant setting (60% v 40.6%; OR, 2.20 [95% CI, 1.15 to 4.19]; P = .02) but not in the castrate-sensitive setting (78.9% v 71.5%; OR, 1.49 [95% CI, 0.66 to 3.37]; P = .43). Median PFS with first ARPI did not differ in SPOP- mutated versus wild-type group in both the castrate-sensitive (24.2 v 19.2 months; hazard ratio [HR], 0.80 [95% CI, 0.49 to 1.32]; P = .39) and castrate-resistant settings (15.0 v 12.4 months; HR, 0.81 [95% CI, 0.58 to 1.12]; P = .20). In multivariable analysis, SPOP mutations were associated with improved OS (HR, 0.65, P = .02). CONCLUSION SPOP mutations were associated with longer OS despite the presence of aggressive clinical features. The distinct clinical and molecular features of SPOP -mutated PC support its consideration as a unique molecular subtype with prognostic implications.

Abstract Introduction: Primary mediastinal B-Cell lymphoma (PMBCL) is a rare form of large B-Cell lymphoma (LBCL) with around 90% of patients (pts) cured with first line (1L) treatment. For pts with relapsed or refractory (RR) PMBCL, the optimal treatment is unclear. Prior retrospective analysis in the rituximab (R) era have shown high response rates to platinum salvage and high-dose chemotherapy followed by autologous stem cell transplant (HDT/ASCT) in second line (2L) RR PMBCL (Bakos et al. ASH 2023 and Dumke et al. ASCO 2024). The pivotal phase III LBCL trials in early relapsing pts that compared CD19 chimeric antigen receptor-T cell (CAR-T) with HDT/ASCT as 2L treatment had few PMBCL pts: none on ZUMA-7 and 17 PMBCL on TRANSFORM, of whom only 8 were randomized to 2L CAR-T. This study aims to evaluate outcomes of 2L therapies for RR PMBCL, including pts treated with 2L CAR-T. Methods: We performed a multicenter retrospective study in pts with RR PMBCL treated between 2001 and 2025. Clinical and pathologic characteristics, treatment regimens, and efficacy outcomes were collected and analyzed. Results: 139 pts were included with 43% male, 69% Caucasian, 7.9% Black, 8.6% Asian and 79% Non-Hispanic. R-EPOCH (64%) and (32%) were the most common 1L treatments and all pts received R. Consolidative radiation was used in 21 pts (15%) as part of 1L therapy, 28% after RCHOP. Primary refractory disease with relapse ≤6 month (mo), or early relapse ≤12 mo, occurred in 83% (N = 115) and 93% (N = 129 pts), respectively. At relapse, median age was 34 years (yrs), 44% had advanced stage disease, 31% had bulky disease (≥ 7.5cm). CNS relapse occurred in 8.6% of pts. Overall and complete response rate (ORR and CR) to any 2L treatment was 44% and 27%, respectively. At median follow-up of 40 months (mo) from the start of 2L treatment, median PFS of the entire cohort was 3.2 mo with a 3 yr OS of 70%. For 2L treatment, N=93 (67%) pts were treated with platinum salvage with intention to proceed with HDT/ASCT. Most commonly used platinum salvage regimens were ICE± R (n= 53, 38%), R-DHAP (n=23, 17%), R-GemOx (n= 5, 3.6%), and GDP ± R (n= 10, 7.2%), and the majority (95%) of pts received R with 2L therapy. ORR and CR to platinum salvage was 46% and 15% respectively. With a median follow up of 30 mo from start of 2L therapy, median PFS was 2.6 mos, and 3 yr PFS and OS was 15% and 64%, respectively. Of the 93 pts treated with platinum salvage, only 25 pts (27%) proceeded with HDT/ASCT- of whom, N = 17 (68%) achieved a CR after ASCT. In patients that proceeded to ASCT, relapses occurred in 10 pts (40%) with a 3 yr PFS and OS 56% and 83%, respectively. CAR-T was used as 2L in 20 pts (14%)- most commonly either axi-cel (N =7) or liso-cel (N = 8). ORR and CR to CART was 89% and 55% respectively, median PFS and OS was NR after CAR-T infusion and 3 yr PFS and OS was 55% and 95%, respectively. Other 2L treatments included pembrolizumab (N =3), brentuximab vedotin (BV) + nivolumab (N= 4) or single agent BV (N=2). Radiation was given as part of 2L treatment in 25.9% of pts. 2L CAR-T therapy was associated with improved ORR (OR 7.60, 95% CI 1.68 – 34.55, p = 0.002), improved CR rate (OR 4.15, 95% CI 1.49 – 11.53, p = 0.008) and improved PFS (HR 0.18, 95% CI 0.06 – 0.57, p = 0.004) compared to pts treated with 2L platinum salvage. There was no difference in OS between pts treated with platinum salvage or 2L CAR-T. A total of 96 patients progressed on 2L therapy, and 92 pts (95.8%) received third-line (3L) treatment. Third line treatments included chemoimmunotherapy +/- HDT/ASCT (N= 23), CD19 CAR-T (N = 35), clinical trial (N=7), check point inhibitor-based regimen (N= 11), BV based regimen (N= 8), allogeneic stem cell transplant (N = 3) or other (N = 5), with 7 pts receiving radiation as part of 3L therapy. Median PFS and OS from start of 3L therapy was 6.7 mo and 6.4 mo, respectively. Conclusions: We report the largest cohort of pts with RR PMBCL treated in the rituximab era and the largest number of PMBCL pts treated with 2L CAR-T. This is the first analysis, to our knowledge, to show a statistically significant benefit in PFS, ORR and CR rate of CAR-T compared to platinum salvage followed by HDT/ASCT as 2L treatment in a study of exclusively RR PMBCL pts. Our dataset was largely (93%) compromised of pts with primary refractory or early relapsing disease and provides support for using CAR-T as 2L treatment in this high-risk subset of RR PMBCL patients.

Abstract Introduction: The 5th edition of the World Health Organization (WHO-5) and the International Consensus Classification (ICC) recognize TP53-mutated myeloid neoplasms (MN) as a distinct disease category, reflecting their uniformly dismal outcomes and the urgent need for novel therapeutic strategies. However, both classifications do not include TP53-non-mutatedcases with 17p (TP53 locus) loss, despite emerging evidence that such lesions may confer biologic and clinical behaviour similar to TP53-mutated disease. Our preliminary work demonstrated that 17p loss without TP53 mutation in therapy-related MN (t-MN) is associated with poor survival, approximating that of TP53-mutated cases (Hiwase et al., Blood 2023). Yet, the true frequency and prognostic impact of 17p loss in TP53-non-mutatedMN remains largely undefined, representing a critical gap in current disease classification and risk stratification. To address this critical knowledge gap, we aim to define the frequency and prognostic impact of 17p loss across the TP53 locus in TP53-non-mutatedMN, including both myelodysplastic syndromes (MDS) and acute myeloid leukemia (AML). Finally, we compared the survival outcomes of this cohort with those of TP53-mutated MN. Method This international collaborative study was designed to evaluate the frequency and prognostic significance of 17p loss in TP53-non-mutatedMN. We included MDS and AML patients who had conventional cytogenetic analysis and did not have TP53-mutation(TP53-non-mutated)by next-generation sequencing (NGS) with sensitivity of ≥2% variant allele frequency (VAF). These cases were systematically reviewed for 17p loss using karyotype data, and where available, fluorescence in situ hybridization (FISH) and SNP array were used to confirm the 17p/TP53 locus deletion. To assess the clinical impact, we compared the overall survival (OS) of TP53-non-mutatedcases with 17p loss against a reference cohort of TP53-mutated MNs with VAF ≥2% (n=603) (South Australian Myeloid Neoplasm Registry (SA-MN) and Mayo Clinic, Rochester, USA). Results We screened 8876 cases of TP53-non-mutatedMN cases across an international cohort comprising the SA-MN (Australia; n=1046), the Düsseldorf MDS Registry (Germany; n=545), Munich Leukaemia Laboratory (Germany; n=5785), and King's College Hospital, (United Kingdom; n=1500). We identified 85 (1%) cases with 17p loss across the TP53 locus, representing the largest international cohort to define the frequency of 17p loss in TP53-non-mutatedMN. Complex karyotype was present in 40% (n=34) of cases, frequently co-occurring with del(5q) and del(7q). Clinically, most cases were MDS (n=69; 81.2%), including MDS-IB1 (n=19), MDS-IB2 (n=15), MDS-LB (n=16), t-MDS (n=8), MDS/MPN overlap (n=3), CCUS (n=1), or unspecified MDS (n=7), while AML accounted for 16 cases (18.8%). Among evaluable cases, the majority were male (78.6%) and had a median age of 66.5 years at diagnosis. The most frequent cytogenetic abnormality causing 17p loss across the TP53 locus was monosomy 17 (39%), followed by i(17q) (29%), add (17p) (14%), derivative chromosome 17 from unbalanced translocations (11%), and del(17p) (7%). The majority represented de-novo MN (60%), with t-MN in 40% of cases. Median OS of evaluable 17p loss in TP53-non-mutatedcases (n=28; 10.2 months) was poor and comparable to TP53-mutated MN (n=603, 8.7 months; P=0.42). Similarly, survival censored for a stem cell transplant was also equally poor (10.4 vs. 8.7 months; P=0.34) in both groups. Complex vs. non-complex karyotype (7.7 vs. 10.7 months; P=0.68) did not significantly influence the OS of 17p loss. Median OS of MDS cases tended to be longer compared to AML (16.7 vs. 9.1 months; P=0.11), although not statistically significant. Conclusion: In this largest international study, 1% of TP53-non-mutatedMNs harbor 17p loss across the TP53 locus, displaying poor survival equivalent to TP53-mutated disease. Exclusive reliance on NGS to identify TP53 mutations misses these high-risk structural TP53 abnormalities, which are readily detectable by conventional karyotyping. These findings provide compelling evidence to include TP53-non-mutatedcases with 17p loss in the next revision of the classifications, enabling accurate prognostication and equitable access to curative and clinical trials of novel therapies.

Abstract Background: Sickle cell crisis (SCC) remains a leading cause of emergency hospitalization and acute morbidity in patients with sickle cell disease (SCD). With improved survival in SCD, long-term complications such as cardiovascular disease are emerging as critical determinants of morbidity and mortality. While right-sided heart failure has been adequately explored in SCD, the specific contribution of the other cardiovascular conditions to in-hospital mortality during sickle cell crisis remains unclear. This study evaluates the independent impact of cardiovascular comorbidities on in-hospital mortality among adults hospitalized for SCC using a large, nationally representative database. Methods: This is a retrospective analysis involving adult hospitalizations (≥ 18 years old) for SCC in the United States (U.S) using 2016-2020 National Inpatient Sample (NIS). The study was exempt from institutional review board approval as the NIS database contains deidentified patient information. ICD-10 codes used for SCC were: D570, D5721, D5741, D5743, D5745, D5781. Regarding cardiovascular co-morbidities, the study considered age, race (African American, Hispanic and Caucasian), dyslipidemia, hypertension, diabetes mellitus (both type 1 and type 2), systolic and diastolic heart failure, rheumatic valvular heart disease, atrial fibrillation/flutter, pulmonary hypertension, chronic coronary artery disease and ischemic stroke. Univariable Cox regression was performed to identify independent predictors associated with in-hospital mortality among patients admitted for SCC. All factors from the univariable analysis with p&lt;0.20 were included in a multivariable Cox regression model. All the outcomes were adjusted for age, sex, and Charlson Comorbidity Index. P-value of ≤0.05 was used to determine statistical significance. Results: The current study analyzed 424,540 SCD patients admitted for SCC (mean age 32.4 years, 55.4% females), from 2016 to 2020. Each additional year of age beyond 18 was associated with a linearly increasing in-hospital mortality among study patients hospitalized for SCC. This observation was confirmed in both univariate (HR: 1.06, 95% CI: 1.05–1.07, p&lt;0.001) and multivariate (HR: 1.05, 95% CI: 1.04–1.06, p&lt;0.001) analyses. On univariable analysis of cardiovascular co-morbidities in patients admitted for SCC, a higher in-hospital mortality was conferred by: hypertension (HR: 2.40; 1.95 – 2.94, p &lt;0.01) dyslipidemia (HR: 2.38; 1.54 – 3.65, p &lt;0.01), diabetes mellitus (HR: 2.15; 1.51 – 3.04, p &lt;0.01, systolic (HR: 5.20; 3.70 – 7.31, p &lt;0.01) and diastolic heart failure (HR: 2.11; 1.44 – 3.08, p &lt;0.01), atrial fibrillation/flutter (HR: 5.43; 4.04 – 7.32, p &lt;0.01), pulmonary hypertension (HR: 2.34; 1.75 – 3.13, p &lt;0.01) and ischemic stroke (HR: 5.75; 3.30 – 10.03, &lt;0.01); a lower in-hospital mortality was associated with the African American race (HR: 0.43; 0.20 – 0.91, p = 0.027). In multivariable analysis of cardiovascular comorbidities that demonstrated nominal significance (p &lt; 0.20) in univariable analysis, only the following remained significantly associated with in-hospital mortality: systolic heart failure (HR: 2.65; 1.77 – 3.98, p &lt;0.01), atrial fibrillation/flutter (HR: 2.01; 1.41 – 2.8), p &lt;0.01) and ischemic stroke (HR: 3.25; 1.85 – 5.71, p &lt;0.01). Although not statistically significant in univariable analysis (p = 0.065), chronic coronary artery disease (CAD) emerged as a significant predictor of reduced in-hospital mortality in the multivariable analysis (HR: 0.52; 0.29–0.91, p = 0.021). Conclusions: Among adults hospitalized for sickle cell crises, select cardiovascular comorbidities, including systolic heart failure, atrial fibrillation/flutter, and ischemic stroke, were independently associated with significantly increased in-hospital mortality. Interestingly, CAD was linked to reduced mortality in adjusted models. This may be attributed to closer medical surveillance and management in patients with established CAD, along with the potential protective mechanism(s) in SCD that limit atherosclerotic progression — warranting further investigation

Abstract Introduction Elevated levels of fetal hemoglobin (HbF) ameliorate sickle cell disease (SCD) manifestations. BEAM-101 is an investigational cell therapy comprising ex vivo base-edited autologous CD34+ hematopoietic stem and progenitor cells (HSPCs) that directly target the HBG1/2 promoters to disrupt BCL11A binding, switching production from sickle (HbS) to majority anti-sickling HbF. We present ongoing data from the BEACON study (NCT05456880), a Phase 1/2, single-arm, open-label study evaluating a single dose of BEAM-101 in individuals with SCD and severe vaso-occlusive crises (sVOCs). Methods Eligible patients were aged 12–35 years with SCD and ≥4 sVOCs in the 2-year period prior to informed consent. After mobilization with weight-based or tiered fixed-dose plerixafor, autologous CD34+ HSPCs were collected by leukapheresis and genetically modified with an adenine base editor. Following conditioning with pharmacokinetically adjusted myeloablative busulfan (target cumulative AUC 80 hr*mg/L), patients received a single dose of BEAM-101 (≥3.0×106 viable CD34+ cells/kg). Patients were then monitored for neutrophil and platelet engraftment, adverse events (AEs), total Hb, Hb fractions, % F-cells, hemolysis markers, sVOCs, and additional protocol endpoints for up to 24 months. Results As of June 16, 2025, enrollment has been completed and 46 patients have initiated mobilization. Of the 26 patients who have received BEAM-101, mean duration of follow-up was 6.2 (range 1.0–18.7) months. Baseline demographics were: age 18–34 years, 24 βS/βS genotype, 21 self-reported Black/African American, and 11 female. One patient, who successfully engrafted and obtained stable HbF levels &gt;60%, withdrew consent 5 months post-infusion. For the 26 treated patients, the median number of stem cell collection cycles was 1 (range 1–4); 18 (69.2%) required a single cycle, 4 required 2 cycles, and 4 required &gt;2 cycles. Target cell dose was achieved with a median of 3 (1–12) collection days. Patients received a mean BEAM-101 dose of 8.4 (range 3.2–23.4) ×106 viable CD34+ cells/kg. By the data cut, patients achieved neutrophil (n=26) and platelet (n=24) engraftment at a median of 18 (12–30) days and 19 (11–50) days, respectively, with a median duration of 7 (4–17) days of neutropenia. Of the 24 evaluable patients, 9 (37.5%) achieved platelet engraftment without requiring platelet transfusion through the transplant process. No patients had ≥Grade 3 AEs or serious AEs related to BEAM-101. One patient died due to respiratory failure, assessed as likely related to busulfan conditioning, 4 months post-infusion as previously reported. Mean total Hb increased to 12.1 (range 9.0–14.7) g/dL by Month (M) 1 (n=24), 15.6 (11.9–18.6) g/dL by M6 (n=11), and was sustained through follow-up. Four patients had total Hb levels above the normal range beyond M6 without any associated clinical manifestations or need for therapeutic intervention. Erythropoietin levels trended towards normal in all patients, including those with elevated total Hb. Mean endogenous HbF (HbF/[HbF+HbS]) was &gt;60% and HbS was &lt;40% by M1, which was sustained through last follow-up. At M6, mean % F-cells were 99.2% (97.0–99.9; n=10) and mean HbF/F-cell was 20.8 (18.0–23.9; n=9) pg, exceeding the protective threshold of 10 pg against sickling. Peripheral blood editing was at a mean of 68.5% by M3 (n=16), 69.6% by M6 (n=10), and 72.8% by M12 (n=5). Hemolysis markers and sickling parameters normalized or improved for all patients. No patients experienced any investigator-reported sVOCs post-engraftment. Conclusions The BEAM-101 treatment process is characterized by a low number of mobilization and collection cycles and high editing efficiency, resulting in rapid neutrophil and platelet engraftment and robust pancellular HbF expression, with durability observed up to 18 months. Additionally, there is a corresponding reduction in HbS levels to &lt;40%, and improvements or resolution of hemolysis parameters and anemia, along with resolution of investigator-reported sVOCs post-engraftment. The safety profile of the BEAM-101 treatment regimen is consistent with busulfan conditioning, autologous HSCT, and underlying SCD. These results demonstrate the potential for base editing of the HBG1/2 promoters as a safe and effective therapeutic modality for the treatment of SCD that also has the potential to minimize hospitalization and treatment process burden for patients, families, and sites.

PURPOSE STELLAR (ClinicalTrials.gov identifier: NCT02796261 ) was a phase III, randomized, open-label trial of eflornithine + lomustine versus lomustine monotherapy in patients with recurrent grade 3 astrocytoma. METHODS At trial initiation, eligibility criteria included: age ≥18 years, anaplastic astrocytoma (2016 WHO CNS Tumor classification [WHO CNS4]), first recurrence ≥6 months after radiation and temozolomide (TMZ), Karnofsky performance status ≥70, and no imaging findings consistent with grade 4 glioblastoma. Random assignment (1:1) was stratified by isocitrate dehydrogenase ( IDH ) mutation, age, resection extent, and geography. Patients received eflornithine (2.8 g/m 2 orally, every 8 hours [2 weeks on, 1 week off]) + lomustine (90 mg/m 2 orally, once every 6 weeks), or lomustine monotherapy (110 mg/m 2 once every 6 weeks). The primary end point was overall survival (OS). RESULTS Among 343 patients randomly assigned across 74 sites in eight countries, there was no difference in survival between eflornithine + lomustine and lomustine monotherapy (median OS 23.4 v 20.3 months, hazard ratio [HR], 0.94). Following changes in classification and grading in the 2021 WHO CNS5, a subset analysis of patients with IDH- mutant, grade 3 astrocytoma (n = 196), defined in 2024, before unblinding, showed clinically meaningful improvements in median OS with eflornithine + lomustine versus lomustine monotherapy (34.9 v 23.5 months, HR, 0.64) and median progression-free survival (PFS, 15.8 v 7.2 months, HR, 0.57). No differences were observed among patients with CNS grade 4 disease. Grade ≥3 treatment-emergent adverse events of relevance were related to reversible myelosuppression (eflornithine + lomustine 42% v lomustine monotherapy 29% of patients) and hearing impairment (24% v 0%). No new safety signals were identified. CONCLUSION Clinically meaningful improvements were observed; eflornithine + lomustine doubled PFS and improved OS in patients with recurrent IDH -mutant, grade 3 astrocytoma, but not grade 4 tumors, after prior radiotherapy and TMZ, consistent with its cytostatic mechanism of action.