This is a PDF file of an article that has undergone enhancements after acceptance, such as the addition of a cover page and metadata, and formatting for readability, but it is not yet the definitive version of record. This version will undergo additional copyediting, typesetting and review before it is published in its final form, but we are providing this version to give early visibility of the article. Please note that, during the production process, errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

Background For proton therapy, a relative biological effectiveness (RBE) of 1.1 has broadly been applied clinically. However, as unexpected toxicities have been observed by the end of the proton tracks, variable RBE models have been proposed. Typically, the dose‐averaged linear energy transfer (LETd) has been used as an input variable for these models but the way the LETd was defined, calculated, or determined was not always consistent, potentially impacting the corresponding RBE value. Purpose This study compares consistently calculated LETd with other quantities as input variables for a phenomenological RBE model and attempts to determine which quantity that can best predicts proton RBE. The comparison was performed within the frame of introducing a new model for the proton RBE. Methods High‐throughput experimental setups of in vitro cell survival studies for proton RBE determination are simulated using the SHIELD‐HIT12A Monte Carlo particle transport code. Together with LET, z∗2/β2$z^{*2}/\beta ^2$, here called effective Q (Qeff), and Q are scored. Each quantity is calculated using the dose and track averaging methods, because the scoring includes all hadronic particles, all protons or only primaries. A phenomenological linear‐quadratic‐based RBE model is subsequently applied to the in vitro data with the various beam quality descriptors used as input variables and the goodness of fit is determined and compared using a bootstrapping approach. Both linear and nonlinear fit functions were tested. Results Versions of Qeff and Q outperform LET with a statistically significant margin, with the best nonlinear and linear fit having a relative root mean square error (RMSE) for RBE2Gy ± one standard error of 1.55 ± 0.04 (Qeff, t, primary) and 2.84 ± 0.07 (Qeff, d, primary), respectively. For comparison, the corresponding best nonlinear and linear fits for LETd, all protons had a relative RMSE of 2.07 ± 0.06 and 3.39 ± 0.08, respectively. Applying Welch's t‐test for comparing the calculated RMSE of RBE2Gy resulted in two‐tailed p‐values of <0.002 for all Q and Qeff quantities compared to LETd, all protons. Conclusions The study shows that Q or Qeff could be better RBE descriptors that dose averaged LET.