SRS for Vestibular Schwannomas: TCP Analyses (HyTEC)
ForSporadic vestibular schwannoma, SRS or fSRS, non-NF2
TL;DRPooled model of 35 series and 5162 pts: 12 Gy single-fraction predicts 3-5y TCP of 91.2% (LQ) or 95.6% (LQ-L). No outcome data below 11 Gy.
The two models disagree on α/β (12.4 vs 2.97 Gy), which changes how fSRS ranks: LQ puts 18 Gy/3fx (93.6%) above 12 Gy x1 (91.2%), while LQ-L puts it below (94.0% vs 95.6%). Calling fractionation 'equivalent' therefore depends on the model chosen. Single-fraction dose reduction below 11 Gy has no outcome data behind it.
In a sporadic VS being planned for SRS or fSRS, this supports the common 12-13 Gy/1fx, 18-21 Gy/3fx and 25 Gy/5fx prescriptions as all sitting high on the control curve. It does not extend to NF2-related tumors or salvage after prior resection.
The LQ vs LQ-L split on α/β (12.4 vs 2.97 Gy) changes the fSRS ranking: 18 Gy/3fx predicts 93.6% vs 91.2% for 12 Gy x1 under LQ, but 94.0% vs 95.6% under LQ-L. Single-fraction prescriptions below 11 Gy have no outcome data behind them.
12 details
HyTEC literature review and pooled TCP model: 35 publications, 5162 pts. A 3-parameter Poisson model with α/β as a free parameter, fit by maximum likelihood under both LQ and LQ-L dose conversion.
Sporadic VS treated with SRS or fSRS. NF2 pts excluded, and the authors tried to exclude progression after prior resection. Pediatric pts and repeat SRS were not analyzed.
The most common prescriptions were 12 to 13 Gy in 1 fraction, 18 to 21 Gy in 3 fractions, or 25 Gy in 5 fractions. Gamma Knife, CyberKnife and linac series were pooled. The prescription was treated as the minimum GTV dose and the sole determinant of TCP.
3- to 5-year tumor control, using 5y when reported, otherwise 3y. Where both were given they were typically within 0%-3% of each other.
LQ: EQD2 of 13, 19, 25 and 33 Gy predicted TCP of 80%, 90%, 95% and 98%. LQ-L: EQD2 of 18, 26, 35 and 45 Gy for the same levels. Per-regimen TCP is in the table above.
| Regimen | LQ (α/β 12.4) | LQ-L (α/β 2.97) |
|---|---|---|
| 10 Gy x1 | 85.0% | 89.7% |
| 11 Gy x1 | 88.4% | 93.1% |
| 12 Gy x1 | 91.2% | 95.6% |
| 13 Gy x1 | 93.5% | 97.33% |
| 18 Gy/3 fx | 93.6% | 94.0% |
| 25 Gy/5 fx | 97.2% | 96.9% |
The source cites prior comparative work that found SRS and fractionated stereotactic RT "equally effective", with no NTCP differences in hearing preservation, trigeminal toxicity or facial nerve toxicity. This model is consistent with that: every fSRS regimen modeled sits high on the curve.
Only 26% of reports gave conformity indices, and none reported Dxx/Vxx, so dosimetry was reduced to the prescription. TCP at 0 Gy was assumed at 30%. There were too few data to model TCP by tumor volume.
The curve is flat across standard prescriptions, so it cannot separate 12 from 13 Gy, or single-fraction from fSRS, on control alone. Toxicity and logistics should drive that choice. Whether control holds beyond 10y is still open.
Dose-response model fit to pooled, mostly retrospective series with heterogeneous control definitions. It supports current prescriptions but adds no randomized or prospective dose-comparative data.
- Tumor control durability beyond 10 years after SRS
- Control at single-fraction doses below 11 Gy
- Volume-dependent TCP for larger schwannomas
📚 Sources · 📄 1 paper
The longer read
The value of this report is less in any single number than in what it says about the dose range where VS radiosurgery is actually practiced. Every common prescription sits on the shoulder of the curve. Under either model, moving from 12 to 13 Gy in one fraction, or from single-fraction to 18 Gy in 3 or 25 Gy in 5, moves predicted 3-5y control by a few percentage points at most. A curve this flat says that tumor control should not be what decides between regimens. Cranial nerve outcomes, tumor volume, brainstem proximity and patient logistics are the real levers, and this paper cannot address any of them.
The disagreement between the two dose-conversion models matters more than its placement in the methods suggests. LQ returns an α/β of 12.4 Gy, which is implausibly tumor-like for a slow-growing schwannoma. LQ-L returns 2.97 Gy, which matches radiobiological intuition and fits marginally better. The choice changes which regimen looks stronger. Under LQ, 18 Gy in 3 fractions outperforms 12 Gy single-fraction. Under LQ-L it slightly underperforms. A reader who cites this paper to argue that fSRS matches or beats SRS on control should know that the ranking depends on a model assumption, not on observed data that discriminate between regimens.
Several methodological choices bias the predictions, and the reader should know which way each one pushes. Assuming 30% nonprogression at 0 Gy is presented as conservative: a lower value would make SRS look more effective. Using crude control at median follow-up where Kaplan-Meier estimates were absent, which was the case in more than half the articles, tends to overstate control, because late failures beyond the median go uncounted. Treating the prescription as the minimum GTV dose discards real coverage differences across Gamma Knife, CyberKnife and linac series. The heterogeneity of failure definitions (radiologic progression, symptomatic progression, need for salvage) means the pooled endpoint is not one endpoint. Most of these push the apparent control rate up, not down.
The point most relevant to current practice is the lower bound. Dose de-escalation in VS radiosurgery has been driven by efforts to preserve hearing and facial nerve function. This model has no outcome data below the equivalent of 11 Gy in one fraction. The predicted 85.0% (LQ) or 89.7% (LQ-L) at 10 Gy is extrapolation with wide confidence bands, not evidence that 10 Gy is adequate. Anyone considering going below 11 Gy for hearing preservation is outside the data this model was built on.
What would make the conclusions wrong is late failure. Few series extend beyond 10 years, and the untreated natural history the authors cite shows growth in more than 95% of tumors by 10 years. If a meaningful fraction of controlled tumors regrow in the second decade, a 3-5y TCP overstates long-term control at every dose. The reporting standards in Table 2 are the most durable contribution here. Without KM curves, at-risk numbers and complete dosimetry from future series, the next version of this model will have the same limitations.