The modelled benefits of individualizing radiotherapy patients’ dose using cellular radiosensitivity assays with inherent variability

Abstract Objective: To model the increases in local tumour control that may be achieved, without increasing normal tissue complications, by prescribing a patient's dose based on cellular radiosensitivity measured using an assay possessing inherent variability. Method: Patient populations with varying radiosensitivity were simulated, based on measured distributions among cancer patients of the surviving fraction of their fibroblasts given a dose of 2 Gy in vitro (SF 2 ). The dose-response curve for complications in the population was assessed using a formula relating SF 2 to normal tissue complication probability (NTCP), by summing the data for the individuals. This curve was similar to clinically-derived dose-response curves. The effect of individualizing the patients' doses was explored, based on individual radiosensitivities measured by SF 2 , so that every patient had the same low (5%) value of NTCP. Results: It was found that a significant gain (up to around 30%) in tumour control probability (TCP) was predicted for the population when the doses were individualized using a predictive assay result strongly correlated with NTCP. A greater gain in TCP was predicted when each of the individuals were assumed to have a higher sensitivity and the distribution of radiosensitivity in the population was widened to compensate. The gain in TCP was less (around 20%) when considering less-sensitive patients and a narrower distribution of radiosensitivities. The effect of assay variability and other factors that could affect the predictive power of the assay was simulated. Assay variability and an imperfect correlation between in vitro cell survival and tissue complications, rapidly increased the NTCP for the population when treated with individualized doses. However the individualized doses could be reduced so that NTCP declined to an acceptable level, but in this case the TCP for the population also declined. For example, when the assay variability was half the true variability in SF 2 , the gain in TCP was reduced to around 6%. Also, the predicted gains in population TCP were higher if tumour and normal tissue radiosensitivity were assumed to be correlated. In this case, and in the absence of assay variability, increases in population TCP of about 50% and 30% were predicted, depending on the assumed relative sensitivities of the individual patients compared with that of the population average. For practical application, the division of the patient population simply into three groups of high, average and low radiosensitivity was also examined. The three groups were treated with different doses and the NTCP for the population was kept below 5%. Although the gain in population TCP was less than that predicted with the full individualization, considerable gains of up to 20% were still predicted. This method of dividing the population was more resilient to assay variability and other factors that may affect complications in patients. The modelling suggests that small improvements in TCP (5–10%) may still be achievable even if the correlation between SF 2 and late complications is lower at around −0.4 to −0.6, as reported in some clinical series. Conclusion: Modelling based on measured distributions of fibroblast radiosensitivity shows that improvements in tumour control rates may be achievable through the individualization of radiotherapy dose prescriptions of cancer patients, when assay variability is less than about 50% of the true variability in radiosensitivity, and with greater benefits if tumour and normal tissue radiosensitivity are correlated. Tripartite stratification of the population proved to be less sensitive to assay uncertainty, and can provide most of the benefits of the full individualization.