In silico clinical trial comparing predicted subsequent malignant neoplasms in photon versus proton therapy of a pediatric cohort with intracranial tumors

Graduation date: 2018Access restricted to the OSU Community, at author's request, from 2017-08-09 to 2019-08-09Background: The leading cause of death 20 years after treatment for children surviving a cancer of the central nervous system was from a subsequent malignant neoplasm (SMN) (1). Although it was been shown that proton therapy considerably reduces the risk of a fatal SMN in children receiving craniospinal irradiation compared to photon therapy, it has not been studied for intracranial tumors. We hypothesized that proton therapy in a high-income country would provide no benefit in reducing the risk of a fatal SMN in children with intracranial tumors compared to photon therapy in a low- to middle-income country. \ud Methods: We performed an in silico clinical trial comparing photon and proton treatments, in which we tested whether a cohort of 7 pediatric patients with intracranial tumors would not have a statistically-significant difference in the lifetime attributable risk, LARTotal, of a fatal SMN. These pediatric patients, ages 3-12, were treated at the American University of Beirut Medical Center (AUBMC) using photon 3DCRT and were retrospectively selected for this study. The selection criteria included the diagnosis of a low grade localized brain tumor, age of 2-14 years old, availability of computed tomography image sets, and treatment plans constructed between January 1, 2009 to September 30, 2011. For the photon therapy arm, plans were adjusted slightly to conform to the current standard of care and boost fields were removed. For the proton therapy arm, treatment planning was performed at the University of Texas MD Anderson Cancer Center according to the standard of care for passive-scattering proton therapy. The respective clinically-commissioned treatment planning systems (TPSs) were used to calculate the therapeutic dose from photon and proton therapies. Due to missing anatomy in the patients’ computed tomography (CT) simulations, supplemental anatomy using a computational phantom was fused to each patient for calculation of stray radiation dose in out-of-field organs and tissues. To estimate stray radiation dose in photon therapy, we made measurements in an anthropomorphic phantom and derived a relationship between absorbed dose and distance from the field edge. For proton therapy, the stray radiation of greatest concern was neutrons, and only neutron doses were considered. To account for neutrons produced in the patient, an analytical model was trained and validated using previous Monte Carlo simulations of two children who received intracranial proton therapy fields. To account for neutrons generated in the treatment unit, an analytical model from the literature was adjusted for clinical realism and validated using data from previously-published Monte Carlo simulations of the same two children. Equivalent doses were estimated using the respective radiation weighting factors for photons, protons, and neutrons. The equivalent doses from stray and therapeutic radiation were summed, and the mean equivalent dose in each organ and tissue, T, at risk for SMN was calculated. Lifetime attributable risk of mortality, LART, from a SMN was determined for each cancer site using a widely applied model from the literature. The ratio (LARproton/LARphoton) of the combined LART of all cancer sites, LARtotal, was used to compare the two modalities.\ud Results: We observed that supplement phantoms combined with analytical models were sufficient for estimating stray radiation in missing out-of-field anatomy. The analytical model used to estimate stray radiation in photon therapy reproduced the 12-year-old boy’s in- anthropomorphic phantom measurements with a RMSD of 0.75 cGy Gy-1, i.e., 6.6% of the dose at the field edge. Additionally, the equivalent doses estimated by the internal neutron model that we developed were within 13.5% of the Monte Carlo for distances between 3 cm to 10 cm and were within a factor of two for distances between 10 cm to 20 cm. The equivalent doses estimated by the external neutron analytical model that we adjusted for clinical realism deviated by less than a factor of two from the Monte Carlo results of two children with intracranial tumors. We observed that both the internal and external neutron analytical models were of sufficient accuracy for estimating dose from stray neutrons. For photon therapy, the largest mean organ equivalent doses were found in in the red bone marrow, remainder, skin, and thyroid and were greater than 0.8 Sv. This was true for proton therapy as well except for the thyroid for which the average equivalent dose across all patients was 0.337 ± 0.154 Sv. The main result of these studies was that we found that proton therapy reduced the risk of developing a fatal SMN in these pediatric patients for which the ratio of LARtotal was 0.75 ± 0.22. This led to the rejection of the null hypothesis (H0: ratio of LARtotal =1) with a p-value of 0.011. The primary contributors to LARtotal were leukemia and other solid tumors for which the ratio of LART were 0.80 ± 0.27 and 0.76 ± 0.26, respectively. In general, the ratio of LART was less than one except for bladder cancer in all the children, ovarian and uterus cancers in the girls, and prostate cancer in the boys. However, the LART for each of these cancer sites was small, i.e., less than 0.02%. Other observations included that the LARtotal decreased as the age of the children increased.\ud Conclusions: We conducted an in silico clinical trial that disproved our hypothesis and implicated that children treated with intracranial photon fields in low- to middle-income countries would have a statistically-significant reduced risk of a fatal SMN if they were instead treated with intracranial proton fields. Additionally, our findings from these studies suggest that applying supplemental anatomy and contours from generic computational phantoms and models of out-of-field dose may be used to determine organ doses in clinical and research radiotherapy studies when the actual patients’ anatomies are not available. Furthermore, our methods demonstrated the feasibility of using commercial TPSs combined with analytical models to quantify therapeutic and stray radiation in proton and photon therapy. This ability introduces the possibility of quickly comparing different modalities or optimizing treatment plans to minimize long-term morbidities, such as SMNs. The internal neutron model that we developed as part of this study was most accurate within 10 cm of the treatment fields where the internal neutron dose contributes the most to overall exposures. This model combined with the external neutron model, can be used to estimate the equivalent dose from stray neutrons produce in proton therapy with sufficient accuracy. Future work includes programming the analytical models used in this study as add-ons for commercial TPSs