Personalizing Radiotherapy Prescription Dose Using Genomic Markers of Radiosensitivity and Normal Tissue Toxicity in NSCLC

Abstract

Introduction: Cancer sequencing efforts have revealed that cancer is the most complex and heterogeneous disease that affects humans. However, radiation therapy (RT), one of the most common cancer treatments, is prescribed on the basis of an empirical one-size-fits-all approach. We propose that the field of radiation oncology is operating under an outdated null hypothesis: that all patients are biologically similar and should uniformly respond to the same dose of radiation.

Methods: We have previously developed the genomic-adjusted radiation dose, a method that accounts for biological heterogeneity and can be used to predict optimal RT dose for an individual patient. In this article, we use genomic-adjusted radiation dose to characterize the biological imprecision of one-size-fits-all RT dosing schemes that result in both over- and under-dosing for most patients treated with RT. To elucidate this inefficiency, and therefore the opportunity for improvement using a personalized dosing scheme, we develop a patient-specific competing hazards style mathematical model combining the canonical equations for tumor control probability and normal tissue complication probability. This model simultaneously optimizes tumor control and toxicity by personalizing RT dose using patient-specific genomics.

Results: Using data from two prospectively collected cohorts of patients with NSCLC, we validate the competing hazards model by revealing that it predicts the results of RTOG 0617. We report how the failure of RTOG 0617 can be explained by the biological imprecision of empirical uniform dose escalation which results in 80% of patients being overexposed to normal tissue toxicity without potential tumor control benefit.

Conclusions: Our data reveal a tapestry of radiosensitivity heterogeneity, provide a biological framework that explains the failure of empirical RT dose escalation, and quantify the opportunity to improve clinical outcomes in lung cancer by incorporating genomics into RT.

Keywords: Mathematical modeling; Non–small cell lung cancer; Personalized medicine; Radiation oncology.

Figure 1.. We identify wide heterogeneity in RxRSI which can be simplified into three distinct groups of patients from genomics and RT dosing schedules.

(A) Distribution of RSI in a cohort of 1,747 patients with NSCLC (the TCC cohort). (B) Calculating RxRSI (the physical dose required to achieve an optimized biological outcome) for each patient in a clinical cohort of 60 patients with known clinical outcome, dose received and RSI reveals three groups: patients who require less than SOC dose (50Gy), patients who require a dose within the SOC range (50–70 Gy) and patients who require more than the SOC dose (> 70 Gy). (C) Translating to primary radiation doses and a larger (TCC) cohort, we see that there is a subset of patients who are optimized by 60Gy (blue), a small subset of patients would benefit from moderate (up to 74Gy - grey) and a large cohort (red) who would need greater than 74Gy.

Figure 2.

Combined TCP and NTCP model). (A) The cumulative distribution function of the (bimodal) RxRSI is a TCP curve (and approximates a sigmoid). (B) Probability of grade 3 or greater toxicity with dose for each of esophagus (purple), lung (yellow) and heart (blue). (C) TCP (blue) corrected by NTCP (yellow) as a function of dose.

Figure 3.. An in silico trial of dose escalation using the competing outcomes model in NSCLC matches the outcomes of a recent cooperative group trial.

(A) Schematic of our in silico trial designed to match RTOG 0617, with patients drawn uniformly at random from the TCC cohort. (B) A Kaplan-Meier curve depicting penalised local control (pLC). The 60 and 74Gy arms are predicted to have statistically indistinguishable outcomes (penalized local control) through 5 years. (C) Using the combined model accurately predicts the results of RTOG 0617.

Figure 4.. Empiric dose escalation reached a local optima at 60Gy, but personalized dosing offers significant benefits with current technology.

(A) A radiation dose of 60Gy will provide optimal tumor control for approximately 40% of the population, and escalation to 74Gy will only optimize a further 18.6%, while exposing all to additional toxicity. (B) A Kaplan-Meier curve depicting an in silico trial of 60Gy vs. 74Gy, with escalation only for those the RSI based model predicts who would benefit. (C) A Kaplan-Meier curve depicting an in silico trial of predicted optimal dose in the range 45–80Gy.