====== Glioblastoma prognosis ======

===== Overview =====

===== 🎯 General Prognosis (IDH-wildtype, classic histological GBM) =====
  * **[[Median Overall Survival]] (OS):** ~12–18 months  
  * **[[Median Progression-Free Surviva]]l (PFS):** ~6–9 months  
  * **2-year survival rate:** ~25%  
  * **5-year survival rate:** <5%

===== 🧬 Key Prognostic Factors =====
^ Factor ^ Better Prognosis ^ Worse Prognosis ^
| **[[Age]]** | < 50 years | > 65 years |
| **[[Performance Status]]** | KPS ≥ 70 | KPS < 70 |
| **[[Extent of Resection]]** | Gross total resection | Biopsy / subtotal |
| **[[MGMT]] Promoter Methylation** | Present | Absent |
| **[[IDH]] Status** | IDH-mutant (astrocytoma WHO 4) | IDH-wildtype |
| **Enhancement Pattern** | Non-enhancing (in some molGB) | Classic ring-enhancing lesion |
| **Treatment** | RT + Temozolomide | Palliative / incomplete RT |

===== 🔍 Special Subtype: Molecular Glioblastoma (molGB) =====
  * Defined by: **TERT mutation**, **EGFR amplification**, **+7/−10 chromosomal pattern**, **IDH-wildtype**
  * May lack contrast enhancement (mimicking LGG)
  * [[The Oncologist, 2025 | Zerbib et al.]]:
    * **molGB without CE:** median OS ~31.2 months
    * **molGB with CE or histGB:** median OS ~18–20 months
((Zerbib C, Robinet L, Ken S, Cavillon A, Roques M, Larrieu D, Siegfried A, Roux FE, Berjaoui A, Cohen-Jonathan Moyal E. Clinical [[outcome]] and [[deep learning]] imaging characteristics of patients treated by radio-chemotherapy for a "molecular" [[glioblastoma]]. Oncologist. 2025 Jun 4;30(6):oyaf127. doi: 10.1093/oncolo/oyaf127. PMID: 40542584.)).

===== 📉 Conclusion =====
Glioblastoma remains a highly lethal tumor. Prognosis depends on a combination of:

  * **Molecular profile** (especially IDH, MGMT, EGFR, TERT)
  * **Imaging phenotype**
  * **Extent of resection**
  * **Treatment completion**

Note: The term **"glioblastoma"** now refers strictly to **IDH-wildtype** tumors, per WHO 2021 classification. IDH-mutant high-grade tumors are classified as **astrocytoma, Grade 4**, with **better prognosis**.

===== Retrospective observational cohort studies =====

In a [[retrospective observational cohort study]], Zerbib et al., from the Department of Radiation Oncology, Institut Universitaire du Cancer de Toulouse Oncopole (IUCT-Oncopole), Claudius Regaud; INSERM UMR 1037, Cancer Research Center of Toulouse (CRCT); IRT Saint-Exupéry; Department of Engineering and Medical Physics, IUCT-Oncopole; Biostatistics & Health Data Science Unit, IUCT-Oncopole; Department of Neuroradiology, Hôpital Pierre-Paul Riquet, CHU Purpan; Department of Medical Oncology & Clinical Research Unit, IUCT-Oncopole; Pathology and Cytology Department, CHU Toulouse, IUCT-Oncopole; CerCo, Université de Toulouse, CNRS, UPS, CHU Purpan; Department of Neurosurgery, Hôpital Pierre-Paul Riquet, CHU Purpan; and University Toulouse III – Paul Sabatier,  
published in [[The Oncologist]], sought to evaluate and compare the **clinical outcomes** of patients with **molecular glioblastoma (molGB)** and **histological glioblastoma (histGB)** treated with standard radio-chemotherapy.  
They also assessed whether **artificial intelligence (AI)** models could accurately **distinguish molGB without contrast enhancement (CE)** from **low-grade gliomas (LGG)** using **MRI FLAIR** imaging features.

**Conclusion:**  
Patients with **molGB** and **histGB** showed **similar overall survival** under standard treatment.

  * However, **molGB without contrast enhancement (CE)** demonstrated a significantly **better median overall survival** (31.2 vs 18 months).  
  * **AI models** based on **FLAIR MRI features** were able to **differentiate non-enhancing molGB from LGG**, achieving a best-performing ROC AUC of **0.85**.

→ These findings support the **clinical relevance of non-enhancing molGB as a distinct subgroup** with better prognosis and highlight the **potential diagnostic utility of AI tools** in radiologically ambiguous cases.

----
This study presents itself as cutting-edge — mixing radiotherapy outcomes with artificial intelligence — but beneath the polished language and deep learning jargon lies a set of predictable flaws:

**❶ Retrospective and underpowered:**  
A 132-patient cohort — already heterogeneous — is further subdivided into **histGB**, **molGB with CE**, and **molGB without CE**. Statistical comparisons across these small subgroups are unreliable.

**❷ The AI angle?**  
Yes, a deep learning model differentiates molGB without CE from LGG with a **ROC AUC of 0.85**. Impressive? Only until you realize there is **no external validation**, **no real-world deployment**, and **no error analysis**. As usual, AI serves as a fashionable add-on — not a clinically deployable tool.

**❸ Overstated survival difference:**  
That non-enhancing molGB patients live longer (31.2 vs 18 months) is intriguing — but remains unexplained. There is **no analysis of methylation subclasses**, **no mention of MGMT promoter methylation**, and **no adjustment for diagnostic or therapeutic delays**. Are these findings rooted in biology or bias? The question isn’t asked.

**❹ Radiologic ambiguity ignored:**  
Although the paper acknowledges that molGB can mimic LGG radiologically, it fails to address the clinical consequence: these tumors may be **underdiagnosed and undertreated**. AI is referenced, but **no clinical workflow is proposed** to resolve this problem.

**❺ Conclusion inflation:**  
The abstract promises “diagnostic utility.” In reality, it delivers a **prototype model** with **unclear application**. Once again, **deep learning is praised**, but **no clinician can use it — not now, not soon**.

> In short: a visually attractive study, full of fashionable buzzwords and polite omissions.  
> Good for citations, weak for clinical change.  
> More [[radiogenomic theater]] than [[paradigm shift]].

((Zerbib C, Robinet L, Ken S, Cavillon A, Roques M, Larrieu D, Siegfried A, Roux FE, Berjaoui A, Cohen-Jonathan Moyal E. Clinical [[outcome]] and [[deep learning]] imaging characteristics of patients treated by radio-chemotherapy for a "molecular" [[glioblastoma]]. Oncologist. 2025 Jun 4;30(6):oyaf127. doi: 10.1093/oncolo/oyaf127. PMID: 40542584.)).