Cardiac Risk After Heart-Sparing Breast Radiotherapy
ForLeft-sided breast cancer, 3D-CRT or IMRT, 2008-2018
TL;DRMax LAD ≥12 Gy EQD2: sHR 1.81 (1.04-3.16) for cardiac events; mean heart dose ≥2 Gy null (P=.99), 2223 left-sided pts.
The number that reaches the planning system is the physical-dose translation: 12 Gy EQD2 max LAD is about 10.5 Gy at 42.5 Gy/16 fx and 7 Gy at 26 Gy/5 fx. Discrimination was weak for both metrics (C index 0.58 vs 0.53), so this argues for adding an LAD max objective and motion management, not for retiring mean heart dose.
In left-sided breast cancer planned with 3D-CRT or IMRT, this supports carrying an LAD max objective alongside the usual heart constraint; it does not extend to right-sided disease, which sat outside the primary analysis.
The actionable number is the physical-dose translation: 12 Gy EQD2 max LAD is about 10.5 Gy at 42.5 Gy/16 fx and 7 Gy at 26 Gy/5 fx, both checkable at the workstation. Discrimination was weak for both metrics (C index 0.58 vs 0.53), so this adds an LAD max objective and breath-hold rather than retiring the heart constraint.
9 details 4 trials watching
Cross-sectional cohort of 4908 breast cancer pts treated with 3D-CRT or IMRT from 2008 to 2018 at one Canadian tertiary center, 2223 left-sided in the primary analysis. Median follow-up 10.8 years (IQR 8.4-13.1). Dosimetry auto-segmented from planning CT, converted to EQD2; competing-risks (Fine and Gray) regression adjusted for cardiovascular risk factors.
Breast cancer treated with 3-dimensional conformal or intensity-modulated RT, 2008 to 2018, with the primary analysis restricted to left-sided disease. Systemic cardiotoxic exposure (anthracycline, trastuzumab) is not reported in source.
3D-CRT or IMRT across the heart-sparing era; LAD and heart automatically segmented and dose converted to EQD2. The threshold is a max point dose to the LAD, not a mean, and the reference schedules are moderate hypofractionation (42.5 Gy in 16 fx) and ultrahypofractionation (26 Gy in 5 fx).
Adverse cardiac events: MI, or admission / ED visit for unstable angina, arrhythmia, heart failure, pericarditis, myocarditis. Coronary angiography and revascularization captured separately as CAD. Discrimination compared by ROC C index, adjusted association by competing-risks regression.
10-year cumulative incidence of cardiac event or CAD was 5.0% (95% CI 4.1-6.0). Metric-by-metric comparison is in the table above.
| Metric | Max LAD dose | Mean heart dose |
|---|---|---|
| Discrimination (C index) | 0.58 (95% CI 0.52-0.64) | 0.53 (95% CI 0.47-0.60) |
| Adjusted association | ≥12 Gy EQD2: sHR 1.81 (1.04-3.16), P=.04 | ≥2 Gy: not associated, P=.99 |
| Schedule | Physical max LAD dose |
|---|---|
| 42.5 Gy / 16 fx | approx 10.5 Gy |
| 26 Gy / 5 fx | approx 7 Gy |
Current whole-heart constraints descend from population dose-response work on cohorts irradiated when incidental cardiac exposure was far higher (Darby, NEJM 2013), the era in which mean heart dose had usable spread. This is the modern counterpart of those series, and the reversal it reports is what you would expect if heart-sparing planning compressed mean heart dose below its discriminating range. The referenced schedules, 42.5 Gy in 16 fractions and 26 Gy in 5 fractions (FAST-Forward), are current practice, so the dosimetric translation transfers.
The mean heart dose null is hard to separate from restricted range: a 2 Gy dichotomy inside a heart-sparing cohort may not span enough exposure for a gradient to show. The endpoint counts coronary angiography and revascularization, which track ascertainment and access as well as biology. Systemic cardiotoxic exposure is not reported in source, leaving an obvious confounder unaddressed.
The asymmetry that matters is cost: an LAD max objective plus breath-hold usually costs optimization time, not target coverage, so a weak association is enough to justify it, while it would not justify trading away chest wall or nodal coverage. The measurement problem cuts the other way, since a max point dose to a small mobile auto-segmented vessel is among the least reproducible quantities to write into a protocol. Motion management carries the least methodological baggage of the two recommendations: it lowers LAD dose and heart dose together.
Cross-sectional single-center cohort with a cut point derived in the same data; C index 0.58 barely above chance and its interval overlaps mean heart dose's.
- External validation of the 12 Gy EQD2 LAD cut point
- Whether LAD-directed planning prospectively lowers cardiac events n=400 · primary completion 2026-04 · DIBH vs free-breathing cardiac dose, paired plansn=750 · primary completion 2027-12 · IMPT vs IMRT/VMAT, cardiac toxicity endpoint
- Generalizability to regional nodal irradiation and 5-fraction schedules active Postmastecomy Internal Mammary Nodal Irradiation for High-risk Breast Cancer Patients Phase 3n=2400 · primary completion 2025-11 · phase 3 IMN irradiation vs none, n=2400recruiting Ultra-Hypofractionated vs. Hypofractionated Radiation for Node-Positive Breast Cancer Phase 2n=220 · primary completion 2034-04 · randomised ultra-hypofx vs hypofx with nodal RT
📚 Sources · 📄 1 paper
Abstract
The longer read
The claim worth arguing about is not that LAD dose predicted events, it is that mean heart dose did not, in a cohort planned the way people plan now. The dose-response work anchoring current whole-heart constraints was built on women treated when incidental cardiac exposure was much higher and much more variable, and a metric's discriminating power depends on the spread of exposure in the sample where it is measured. Dichotomizing mean heart dose at 2 Gy inside a heart-sparing cohort leaves little room for a gradient to appear, so P=.99 is at least as consistent with compressed range as with the metric being biologically uninformative. The defensible conclusion is that whole-heart mean dose has run out of dynamic range in modern planning, not that whole-heart dose stopped mattering.
The discrimination numbers should temper the other direction just as hard. A C index of 0.58 (0.52-0.64) sits close to chance, and its interval overlaps the mean heart dose estimate (0.53, 0.47-0.60) across most of its width, so 'LAD discriminates better' is a direction rather than a demonstrated separation. The 12 Gy EQD2 threshold was found by cut point search in the same data that then produced the hazard estimate attached to it, which inflates both the apparent knee and the sHR of 1.81 whose lower bound is 1.04. Read as a hypothesis about where a threshold might sit, it is useful. Read as a constraint, it is untested outside one center.
What makes it reasonable to act on anyway is the asymmetry of cost. The physical-dose translation the authors supply, near 10.5 Gy at 42.5 Gy in 16 fractions and 7 Gy at 26 Gy in 5 fractions, is usable at the workstation without further arithmetic, and in most left-sided plans holding LAD max beneath those values costs a breath-hold and some optimization time rather than target coverage. A weak association is enough to justify a nearly free intervention. It would not justify one that traded away chest wall or internal mammary coverage, and that boundary is where an LAD objective should stop.
The real counterweight is measurement. A maximum point dose to a small, mobile, automatically segmented vessel is close to the least reproducible quantity a planning protocol can specify: it is sensitive to contour propagation error, to cardiac and respiratory motion during delivery, and to whether the planning CT caught the vessel in a representative position. Mean heart dose earned its place partly by being insensitive to all of that. Writing an LAD max constraint into practice without first standardizing delineation risks importing that noise into QA, which is why the motion-management half of the recommendation rests on firmer ground than the constraint half: it lowers LAD dose and heart dose together.
Two things would make this wrong. If higher LAD dose tracks unfavorable chest wall and cardiac anatomy, and that anatomy also tracks baseline cardiovascular risk, residual confounding survives adjustment for recorded risk factors. And nothing here is randomized, so whether lowering LAD dose lowers events remains an extrapolation, not a finding.