AbstractBackgroundWhile many have speculated on the reasons for gamma comparison insensitivity for patient‐specific quality assurance analysis, the true reasons for insensitivity have not yet been elucidated. Failing to understand the reasons for this technique's insensitivity limits our ability to either improve the gamma metric to increase sensitivity of the comparison or the capacity to develop new comparison techniques that circumvent the limitations of the gamma comparison.PurposeTo understand the underlying cause(s) for gamma comparison insensitivity and determine if simple plan characteristics can quantitatively predict for gamma comparison sensitivity.MethodsKnown MLC and MU errors of varying magnitudes were induced on simple test fields to preliminarily investigate where gamma failures first begin to appear as error magnitude is increased. Gamma value maps between error‐induced plan calculations and error‐free plan calculations were created for 20 IMRT and 20 VMAT cases, each on three different detector geometries—ArcCHECK, MapCHECK, and Delta4. Gamma value maps were qualitatively compared to dose‐gradient maps, and quantitative comparisons were performed between various plan descriptors and the computed gamma sensitivity for five different classes of induced errors were utilized to determine if any plan descriptor could predict the gamma sensitivity on a case‐by‐case basis. All comparisons were performed in a calculation‐only scenario to remove uncertainties introduced by comparisons made with real patient specific QA measurements.ResultsGamma value maps with increasing induced error magnitude illustrated that gamma comparisons fail first in high‐dose, low‐gradient regions of the field. Conversely, in areas of high gradient, gamma values typically remain low, even in the presence of large errors, regardless of detector geometry and gamma normalization setting. Thus, the complex, and often overlapping, high dose gradients in plans appear to be a limiting factor in gamma comparison sensitivity as the number of points along these gradients may often outnumber the points available for failing the comparison in lower gradient regions of the field. None of the simple plan descriptors studied were able to quantitively predict gamma comparison sensitivity, suggesting that quantitatively predicting the sensitivity of gamma comparisons on a case‐by‐case basis may require a combination of multiple factors or metrics not studied here.ConclusionsSimple plan descriptors and the number of points in high‐dose, low‐gradient regions of the field did not quantitively predict for gamma comparison sensitivity. However, it is clear from gradient and gamma value maps that gamma comparisons fail first in high‐dose, low‐gradient regions of the field in the presence of known induced errors, which we have shown to be independent of detector geometry and gamma comparison normalization setting. Gamma comparison sensitivity is thus limited by the ever‐increasing complexity of plans and is particularly important to consider as treatment volumes become smaller and the complexity of overlapping plan gradients increases. This suggests that new methods for patient‐specific QA comparisons are required to circumvent this limitation.