KEY POINTS
- This review revisits the classic question of how accurately radiotherapy must be delivered, four decades after the influential 1987 analysis that proposed target absorbed-dose variability of approximately ≤3.5% relative standard deviation. The authors argue that biological dose-response evidence still supports requirements of roughly the same order today.
- Small systematic errors can matter clinically because modern dose-response curves remain steep. Using a representative normal-tissue model, the authors show that a 2% systematic dose increase could raise a 30% toxicity probability to approximately 39%; existing IAEA recommendations therefore aim to keep systematic dose biases below approximately 1–2%.
- Reference dosimetry has improved substantially. Best-case ionization-chamber-based uncertainty is approximately 1.0% for MV photons, 1.7% for protons and 2.7% for light-ion beams under current international reference-dosimetry frameworks.
- External audits suggest that this level of dosimetric performance is achievable in mature systems. Contemporary photon audits generally show measured-to-planned variability around 1.5–2.5% at reference-dose points, 2–3.5% within target-like regions and 3–4% across all measured locations.
- The dominant challenge has therefore shifted from delivering the correct dose at a single reference point to delivering the correct three-dimensional spatial dose distribution. Modern linac mechanical uncertainties can generally be maintained within 1 mm/1° and often around 0.5 mm/0.5°, while image-to-treatment isocenter coincidence can also be held below approximately 1 mm with appropriate QA.
- Hypofractionation makes accuracy increasingly consequential. Interfraction random error averages out with many fractions, whereas few-fraction stereotactic treatments provide little opportunity for this averaging; steep high-dose gradients simultaneously increase sensitivity to geometric and dosimetric deviations.
- Particle therapy introduces additional uncertainty through range, anatomy along the beam path, LET and variable RBE. Despite these differences, the authors conclude that current clinical and biological data are insufficient to justify a fundamentally different numerical accuracy requirement for particle therapy, making robust optimization, Monte Carlo dose calculation, imaging and external QA particularly important.
CLINICAL TAKEAWAY
Modern radiotherapy has become much better at delivering the right absolute dose; the harder problem is now making sure that dose lands in exactly the right place throughout treatment. For highly conformal photon and particle therapy, accuracy should be viewed as a full chain—from imaging and contouring through calculation, geometry, adaptation and biological uncertainty—not merely machine output calibration.