Low- and middle-income countries can reduce risks of subsequent neoplasms by referring pediatric craniospinal cases to centralized proton treatment centers

A subset of de-identified patient data [57, 58] of nine children—five girls and four boys—who were diagnosed with MB at The University of Texas MD Anderson Cancer Center (MD Anderson) were collected for this study. Separate CSI plans were created according to the standards of care of two leading cancer care facilities in their regions—the American University of Beirut Medical Center (AUBMC) using photon CSI and MD Anderson using proton CSI [59]. AUBMC is an academic hospital in a LMIC that delivers CSI with 3DCRT, and MD Anderson is an academic cancer hospital in a HIC that is capable of delivering proton therapy [60]. Patient data for this study were selected retrospectively using the systematic random sampling method [61], with the following inclusion criteria: between 2 and 14 years of age diagnosed with MB, treated on protocols or according to the current standards of care, availability of CT image data, and having treatment plans that were created between the dates of 1 August 2006 and 31 May 2010. The study was conducted under protocols approved by the Institutional Review Boards of AUBMC and MD Anderson. The cohort's age distribution was approximately uniformly distributed, ensuring a representative range of body sizes and probabilities of second cancers. The patients' characteristics are listed in table 1.

Simulation computed tomography (CT) images had been acquired as part of the routine proton therapy care for each patient and were collected from archived data. The CT image sets extended from the tops of the patients' heads to their mid-thighs, encompassing the majority of the patients' anatomy. Because the patients were from MD Anderson, unlike MB patients at AUBMC, they were simulated in the supine position and a digital couch was added to the images. For each patient, archived target and avoidance volumes were supplemented to include the contours for the 'whole body' (i.e., that captured in the CT image set) and each T, including, bladder, breast tissue, colon (including rectum), liver, lungs, ovaries, skin, stomach, thyroid, uterus, red bone marrow (RBM), and a 'remainder' volume that was created as a Boolean subtraction between the whole body and the union of all specific SMN sites. Because magnetic resonance imaging data were insufficient or missing, delineating the ovaries and RBM accurately was difficult. The size and position of the ovaries in the girls were approximated as 0.5 cm-diameter spheres located laterally to the body of the uterus. Bones were segmented within the body by identifying a range of Hounsfield Units between 150 and 3000, and the RBM volume was created automatically as a three-dimensional symmetric inner margin of 0.2 cm in the bones, excluding the teeth, with minor manual post-processing performed (e.g., small cavities filled and small volumes removed). The contours of the ovaries and RBM were confirmed by comparing the former's positions to those of the nominal locations identified by Pérez-Andújar et al [47] and the latter's regional anatomical percentages to those of the mathematical-model-based proportions of children derived by Cristy [62]. These images and contours were exported to RT Image and RT Structure Set objects in the Digital Imaging and Communications in Medicine-Radiation Therapy (DICOM-RT) format, which were de-identified to preserve anonymity of the patients and sent to AUBMC via password-protected and encrypted file transfer.

Treatment planning methods followed those of previous studies [52, 57, 63] and are briefly reviewed here. CSI fields were consistent with those of routine clinical practice of each institution—photon CSI at AUBMC and proton CSI at MD Anderson. Each CSI plan comprised left and right lateral or slightly posterior oblique cranial fields and one to three adjacent posterior-anterior spinal fields, depending on the length of spine. The photon and proton treatment plans had a common intent, specifically, to sterilize potential residual cancer cells in the clinical target, i.e., the tumor bed, cranial cavity, and spinal canal.

Photon CSIs were planned and therapeutic doses were calculated using a commercial treatment planning system (Panther, version 4.72, Prowess Inc., Concord, CA, USA). Fields of 6-Megavolt energy were designed to deliver a prescribed dose, D_Rx, of 23.4 Gy in 1.8 Gy fractions in the clinical target. Although normally junctions of abutting fields would be shifted longitudinally twice during the course of treatment and sub-fields would be added to improve the distribution of dose, for this study, these shifts and sub-fields were disregarded.

Proton CSIs were planned and therapeutic doses were calculated using a commercial treatment planning system (Eclipse, version 8.9, Varian Medical Systems, Inc., Palo Alto, CA, USA). Passive-scattering spread-out proton beams of nominal energies (i.e., energies prior to range shifting and modulation) between 140 and 225 MeV were designed to deliver a D_Rx of 23.4 Gy-RBE in 1.8 Gy-RBE fractions in the clinical target of each field. Following the recommendation of the International Commission on Radiation Units and Measurements [64], we applied an RBE value of 1.1 for protons, such that the D_Rx corresponded to an absorbed dose of 21.3 Gy.

Boost doses directed to the tumor resection bed were omitted in this research study because they bear much less importance for SMN risk than CSI fields [65]. All treatment plans were reviewed and approved by a board-certified radiation oncologist of the corresponding clinic (AM or FG).

For each patient, the mean equivalent dose, H_T, in each organ or tissue, T, was calculated as the mass-weighted mean of the equivalent dose in Sv, H_v, in each voxel, v, averaged over all v in T, or

$$\begin{eqnarray}&&{H}_{T}=\displaystyle \frac{\displaystyle {\sum }_{v{\rm{in}}T}{H}_{v}{m}_{v}}{\displaystyle {\sum }_{v{\rm{in}}T}{m}_{v}},\end{eqnarray} \tag{ 1 }$$

where m_v was the mass of the tissue in v, where

$$\begin{eqnarray}&&{H}_{v}=\overline{{w}_{R}}\cdot {D}_{v},\end{eqnarray} \tag{ 2 }$$

and where $\overline{{w}_{R}}$ was the mean radiation weighting factor and D_v was the absorbed dose in Gy in v. H_T, H_v, and D_v were estimated separately for therapeutic and stray radiations. For organs and tissues that were not captured completely in the CT image sets, i.e., RBM, skin, remainder, and whole body volumes, H_T values were reduced by the ratio of the mass of the captured portion of the patient to the mass of the patient recorded at treatment in his or her medical record. These values are listed in table 1.

Values of D_v from therapeutic radiation were calculated for each v throughout the modeled geometry of each patient using the TPS of each institution, with an isotropic dose calculation grid of 0.25 cm. The $\overline{{w}_{R}}$ value for all photons beams was taken as 1, and the $\overline{{w}_{R}}$ value of therapeutic protons was estimated as the approximate mean quality factor of 1.1 at any point within the fields based on linear-energy-transfer, microdosimetric, and radiobiological studies [66–72]. We applied $\overline{{w}_{R}}$ values from similar fields of previous studies, which found small variations between patients for neutrons produced inside the patients' bodies ('internal neutrons') and neutrons produced in the treatment unit ('external neutrons'). Specifically, the $\overline{{w}_{R}}$ values were 7.8 for external neutrons from cranial fields, 8.1 for external neutrons from spinal fields and, and 9.0 for internal neutrons [50, 65].

Stray radiation dose was estimated in the following manner. In photon CSI, because out-of-field dose has been shown to be inaccurately calculated by TPSs [73–75], the in-field H_v values were kept and the out-of-field H_v values were replaced. In particular, a measurement-based model for out-of-field dose as a function of distance to the field edge derived in a previous study [75] was used to replace the H_v values that were at least 1 cm away from the 50% isodose surface. Specifically, the following equation was applied to determine H_v in Sv:

$$\begin{eqnarray}&&{H}_{v}=\overline{{w}_{R}}\cdot {D}_{Rx}\cdot \displaystyle \frac{1\,{\rm{Gy}}}{100\,{\rm{cGy}}}\\ &&\,\,\cdot \,\left[\displaystyle \frac{{\alpha }_{1}}{\sqrt{2\pi {\sigma }_{1}^{2}}}{{\rm{e}}}^{-\tfrac{{(r-{\mu }_{1})}^{2}}{2{\sigma }_{1}^{2}}}+\displaystyle \frac{{\alpha }_{2}}{\sqrt{2\pi {\sigma }_{2}^{2}}}{{\rm{e}}}^{-\tfrac{{(r-{\mu }_{2})}^{2}}{2{\sigma }_{2}^{2}}}\right],\end{eqnarray} \tag{ 3 }$$

where α₁ = 224.8 cm cGy Gy⁻¹, μ₁ = −2.16 cm, σ₁ = 1.57 cm, α₂ = 224.3 cm cGy Gy⁻¹, μ₂ = − 7.43 cm, and σ₂ = 10.28 cm. In-house codes were used to replace the out-of-field H_v values in the DICOM-RT files. In proton CSI, stray neutron contributions to H_v were considered separately from the TPS because of the increased RBE for carcinogenesis of neutrons compared to protons. Because neutron doses are not calculated separately by the TPS, the Monte Carlo Proton Radiotherapy Treatment Planning system [76] was used to calculate contributions to H_v from neutrons, and these values were added to contributions to H_v from protons. Data were extracted from each patient's DICOM-RT files to produce input files to its dose engine (Monte Carlo N-Particle eXtended code, version 2.7c, Los Alamos National Laboratory, Los Alamos, NM, USA) for simulating particle tracking and dosimetry through a field-specific proton source definition and geometry of the passive-scattering treatment unit (Probeat, Hitachi America, Ltd, Brisbane, CA, USA) and CT-based voxelized anatomy of the patient. A lattice tally [77] determined D_v per source particle throughout the patient anatomy from therapeutic protons, internal neutrons, and external neutrons. For every patient and particle type, D_v per source particle value was, first, scaled by the planned percentage weight of each field and, second, divided by the mean therapeutic proton D_v per source particle averaged in a central normalization sub-volume of the clinical target. This resulted in D_v in terms of Gy per therapeutic proton Gy for every patient and particle type.

Uncertainties were propagated for voxel and organ doses. For proton CSI, the Monte Carlo simulations conferred less than 0.3% and less than 4% statistical uncertainty in H_T from therapeutic protons and secondary neutrons, respectively. For photon CSI, uncertainties of less than 5% were applied for in-field H_v and the analytical model's root mean square deviation of 0.0167 Sv Gy⁻¹ was applied for out-of-field H_v.

The lifetime attributable risk LAR, of SMN incidence, LAR_I,T, for each radiogenic solid-tumor site that corresponded to T was determined using a linear-no-threshold model,

$$\begin{eqnarray}&&{L}{A}{{R}}_{I,T}=\displaystyle \frac{{I}_{T}}{{H}_{{\rm{ref}}}}\cdot {H}_{T},\end{eqnarray} \tag{ 4 }$$

in which I_T values were risk coefficients interpolated linearly between ages that were listed for H_ref = 0.1 Sv for each sex and SMN site in tables 12D-1 of the BEIR VII Report of the National Research Council of the National Academies (NRCNA) [78]. H_remainder was used to estimate the risk of solid cancers that may be located outside of the listed specific cancer sites, or 'other solid' tumors. Because incidence risk coefficients for NMSC were not provided in the BEIR VII Report, we instead applied values of I_NMSC that were derived from ICRP Publication 60 [68], adjusting for age and sex, as described previously [65].

Unlike solid tumors, because the NRCNA recommended a no-threshold linear-quadratic model of risk as a function of dose for LAR_I,leukemia, it was necessary to give the projected risks of leukemia special consideration. For low to moderate doses, this could be done by applying the recommended quadratic rather than linear multiplicative term, D + θD², where θ was related to the degree of curvature of the function. However, this model was based on the data of Preston et al [79], who fit a linear-quadratic model to updated dosimetric data in the range of 0–2 Gy. Their data was for atomic bomb survivors, and the recommended model did not account for cell sterilization, post-therapy repopulation, and circulation effects that may occur from fractionated exposures in a childhood radiotherapy cohort [80, 81]. Notwithstanding this omission, excess relative and absolute risk estimates of leukemia derived from medical radiation studies, with average doses ranging up to 2 Gy, were similar to those of the model [78 pp 13, 183, 188]. However, in the case of CSI of young children, the mean equivalent dose in the RBM can exceed 2 Sv and are neither acute nor chronic. For these reasons, we were motivated to not overestimate the LAR_I,leukemia for children in the same way as Berrington de Gonzelez et al, who cited the 'limited information about the shape of the dose response curve above about 2 Gy,' and lack of statistical power at high doses [82]. They reserved the linear-quadratic model for acute doses only, applied a linear model for chronic exposures, and restricted the application of their algorithm to 'lower than about 1 Gy' for childhood exposures. Similarly, we applied the full linear-quadratic model to 2 Sv and removed the quadratic term above 2 Sv. In other words,

$$\begin{eqnarray}&&\begin{array}{l}{L}{A}{{R}}_{{I},{\rm{leukemia}}}=\left\{\begin{array}{c}{{I}}_{{\rm{leukemia}}}\left(\displaystyle \frac{{{H}}_{{\rm{RBM}}}+{\theta }{{{H}}_{{\rm{RBM}}}}^{2}}{{{H}}_{{\rm{ref}}}+{\theta }{{{H}}_{{\rm{ref}}}}^{2}}\right),\,{\rm{for}}\,{{H}}_{{\rm{RBM}}}\leqslant 2\,{\rm{Sv}}\\ {{I}}_{{\rm{leukemia}}}\left[\displaystyle \frac{(2\,{\rm{Sv}})+{\theta }{(2{\rm{Sv}})}^{2}}{{{H}}_{{\rm{ref}}}+{\theta }{{{H}}_{{\rm{ref}}}}^{2}}\right]+{{I}}_{{\rm{leukemia}}}\left(\displaystyle \frac{{{H}}_{{\rm{RBM}}}-2\,{\rm{Sv}}}{{{H}}_{{\rm{ref}}}}\right),\,{\rm{for}}\,{{H}}_{{\rm{RBM}}}\gt 2\,{\rm{Sv}}\end{array}\right.,\end{array}\end{eqnarray} \tag{ 5 }$$

where θ was 0.88 Sv⁻¹. By combining like terms and inserting the values for θ and H_ref, equation (2) was simplified to the following:

$$\begin{eqnarray}&&LA{R}_{I,\mathrm{leukemia}}=\left\{\begin{array}{c}{I}_{{\rm{leukemia}}}\left(\displaystyle \frac{{H}_{{\rm{RBM}}}+\theta {{H}_{{\rm{RBM}}}}^{2}}{{H}_{{\rm{ref}}}+\theta {{H}_{{\rm{ref}}}}^{2}}\right),\,{\rm{for}}{H}_{{\rm{RBM}}}\leqslant 2\,{\rm{Sv}}\\ {I}_{{\rm{leukemia}}}\left(50.7+\displaystyle \frac{{H}_{{\rm{RBM}}}-2\,{\rm{Sv}}}{{H}_{{\rm{ref}}}}\right),\,{\rm{for}}\,{H}_{{\rm{RBM}}}\gt 2\,{\rm{Sv}}\end{array}\right..\end{eqnarray} \tag{ 6 }$$

As an example, this LAR_I,leukemia model is shown in figure 1 for a 5-year-old boy and a 5-year-old girl. Compared to applying the linear-quadratic model directly, this approach was intended to result in lower LAR_I,leukemia values at high H_RBM, than if we applied the recommended linear-quadratic model directly.

Zoom In Zoom Out Reset image size

The BEIR VII Report and ICRP Publication 60 contained recommendations for dose- and dose-rate effectiveness factors (DDREFs) to be applied to reduce the risk estimates of solid tumors for low-dose and low-dose-rate exposures. For solid tumors in which the exposures are of high dose but the dose rates are moderate or fractionated, DDREFs should be applied as 1.5 for all solid tumors except NMSC and 2.0 for NMSC. For leukemia, the rationale for DDREF had already been accounted in the curvature of the linear-quadratic portion of the BEIR VII model, so we did not correct additionally our risk values for leukemia.

The above methods were used in the same way to project the LAR of mortality for each cancer site, LAR_M,T, except substituting M_T for I_T, in which the values of M_T were risk coefficients interpolated between ages listed for each sex and SMN site in Table 12D-2 of the BEIR VII Report. Because mortality risk coefficients were not provided in the BEIR VII Report for thyroid cancer and NMSC, for M_thyroid and M_NMSC we multiplied the values of I_thyroid and I_NMSC, respectively, by the proposed lethality fractions for thyroid cancer (0.1) and NMSC (0.002), respectively, taken from ICRP Publication 60. Values of I_T and M_T derived for each patient are listed in table 1.

We did not consider a possible leveling (i.e., plateau) of the slope in the dose-effect model at high doses. This decision was based on the very limited evidence of this effect in cancer survivors, especially when considering individual SMN sites [46]. The exceptions of this are breast cancer, which may (or may not) plateau at high doses, and thyroid cancer, which appears to downturn at doses beyond those of this study [83, 84].

We calculated total lifetime risks of SMN incidence and mortality in the following manner. The total LAR of developing any cancer, LAR_I,total, was the sum of the LAR values for each cancer site, or

$$\begin{eqnarray}&&{L}{A}{{R}}_{I,\mathrm{total}}=\displaystyle \sum _{T}{L}{A}{{R}}_{I,T}.\end{eqnarray} \tag{ 7 }$$

We assumed that the risk for development of a radiogenic cancer was independent of developing any other radiogenic cancers and that it is possible to develop multiple SMNs [2]. In other words, because of the mathematical possibility of attributing an incidence to multiple cancers, the sum was used and correlated terms were not subtracted. However, because of the mathematical impossibility of attributing a fatality due to multiple cancers, we calculated the total LAR of any fatal cancer, LAR_M,total, differently than LAR_I,total. Specifically, unlike the NRCNA, who calculated LAR_M,total as the sum of each LAR_M,T, we considered fatal cancers to be mutually exclusive events in the model, i.e., a person cannot die of one radiogenic cancer and then subsequently die of other radiogenic cancers. In this case, the correlated terms—LAR_M,bladder × LAR_M,breast, LAR_M,bladder × LAR_M,breast × LAR_M,colon, and so on—would need to be omitted. However, rather than taking the sum of all LAR_M,T and then subtracting all correlated terms, to account for this, we used the following methods to achieve the equivalent result. The probability of surviving of any single radiogenic cancer, S_T, was defined as the complement of LAR_M,T, so that S_T was one minus LAR_M,T. Similarly, the probability of surviving all SMNs, S_total, was one minus LAR_M,total. Assuming that each S_T was independent of surviving all other cancers, S_total was the product of the probabilities of surviving each SMN. As a result, the following was true:

$$\begin{eqnarray}{L}{A}{{R}}_{M,\mathrm{total}} & = & 1-{S}_{{\rm{total}}}\\ & = & 1-\displaystyle \prod _{T}{S}_{T}\\ & = & 1-\displaystyle \prod _{T}(1-{L}{A}{{R}}_{M,T}).\end{eqnarray} \tag{ 8 }$$

In summary, like the NRCNA, we used equation (7) to determine LAR_I,total, and, unlike the NRCNA, we used equation (8) to determine LAR_M,total.

When comparing two treatment modalities in terms of SMN risk, it has been shown that the ratios of LAR carry less uncertainty than absolute LAR values [85–88]. Therefore, as a relative figure of merit, we defined the ratio of the LAR from proton CSI to the LAR from photon CSI, RLAR, or

$$\begin{eqnarray}&&{R}{L}{A}{R}=\displaystyle \frac{{L}{A}{R}({\rm{protonCSI}})}{{L}{A}{R}({\rm{photonCSI}})}.\end{eqnarray} \tag{ 9 }$$

In both incidence and mortality, the risk coefficients of each cancer site cancel each other in the numerator and the denominator of RLAR, resulting in identical values for RLAR_I,T and RLAR_M,T. Therefore, we reported RLAR_T values, which were applicable to incidence and mortality. However, the same was not true for RLAR_I,total and RLAR_M,total, so these ratios were calculated separately for incidence and mortality. We reported mean RLAR values averaged across all patients for each T, $\overline{{R}{L}{A}{{R}}_{{T}}},$ any T for incidence, $\overline{{R}{L}{A}{{R}}_{I,\text{total}}},$ and any T for mortality, $\overline{{R}{L}{A}{{R}}_{M,\text{total}}}.$ Unlike dose values, variances in these mean RLAR values were reported at approximately the 95% confidence level (i.e., as two sample standard deviations). One-sided t-tests (α = 0.05) were performed separately to determine whether $\overline{{R}{L}{A}{{R}}_{I,\text{total}}}$ and $\overline{{R}{L}{A}{{R}}_{M,\text{total}}}$ were significantly less than 1 (H₀: $\overline{{R}{L}{A}{R}}$ = 1, H_A: $\overline{{R}{L}{A}{R}}$ < 1), which would indicate that proton CSI confers less LAR than photon CSI with significance.

Values of H_T are listed in table 2 for photon CSI at AUBMC and proton CSI at MD Anderson for the five girls and four boys in this study. Most H_T values followed a general trend of being larger for younger children than for older children, indicating H_T is the strongly affected by the size of the patient. The most notable exception to this was H_liver in proton CSI, which had the opposite trend of larger values for older patients. For H_RBM and H_prostate in photon CSI and H_RBM, H_breasts, H_liver, H_ovaries, and H_uterus in proton CSI, we did not observe a dependency on age. For each patient, in photon CSI, the H_T values were greater than 1 Sv in all T, except for the prostate. However, in proton CSI, only H_RBM, H_remainder, H_skin, and H_lungs were greater than 1 Sv. In some cases, high H_T values indicated that the photon CSI therapeutic fields crossed organs at risk that are not usually considered to be within the CSI target, for example, the bladder of the 3-year-old boy. The overall statistical uncertainties in H_T ranged from less than 1% for most T to less than 10% for small organs and tissues (i.e., prostate, uterus, and ovaries).

In general, even after accounting for stray radiation, mean H_T values were higher for photon CSI than for proton CSI. These values and their 95% confidence intervals are listed in table 2. The average H_T was higher for photon CSI than for proton CSI for all T, except the skin, which had approximately equal H_T. Only a slight reduction in average H_T for proton CSI was observed for the lungs, remainder, and RBM. The variances in H_T were due to the trends in H_T with age as well as the degree of partial irradiation of some organs and tissues, not because of statistical uncertainties. Apart from the whole body volume, the organs and tissues that received the highest average H_T values in photon CSI were the thyroid, RBM, colon, and remainder, and those for proton CSI were the RBM, remainder, skin, and lung.

In proton CSI, neutrons accounted for large portions of H_T for several organs and tissues. Contributions to H_T from external and internal neutrons along with average values are listed for each patient in table 3. Neutrons accounted for more than 40% of the average H_T in the prostate, breast tissue, ovaries, uterus, bladder, thyroid, stomach, liver, and colon. In these organs and tissues, more than two-thirds of the neutron equivalent doses were from external neutrons, except for the thyroid, in which there was an even contribution between external and internal neutrons.

The average projected LAR_I,T values (figure 2(a)) were higher in photon CSI than in proton CSI for every radiogenic cancer category, with the exception of NMSC, and these risks varied greatly between cancer sites. Average projected lifetime risks of incidence greater than 10% were found in other solid tumors, NMSC, lung cancer, and leukemia in both photon and proton CSI and in thyroid cancer, breast cancer, and colon cancer in photon CSI. Average projected absolute reductions in risk of incidence by more than 11% by applying proton CSI rather than photon CSI were observed in thyroid cancer, breast cancer, and colon cancer. Compared to boys, girls were at more than twice the lifetime risk for thyroid cancer, lung cancer, and NMSC in both modalities and stomach cancer in photon CSI. Compared to girls, boys were at elevated lifetime risk of developing a liver cancer and bladder cancer in both modalities and colon cancer in proton CSI. Children of ages at exposure less than 8 years had higher lifetime risk of incidence than children of ages at exposure greater than 8 years in every cancer site for both modalities, and this was true by more than a factor of 2 for bladder cancer and other solid tumors in both modalities, prostate cancer in proton CSI, and breast cancer and lung cancer in photon CSI.

Zoom In Zoom Out Reset image size

The average projected LAR_M,T values (figure 2(b)) were higher in photon CSI than in proton CSI for every radiogenic cancer site having non-negligible LAR_M,T, and these risks varied greatly between cancer sites. The highest average lifetime risks of mortality (i.e., greater than 4%) were for lung cancer, other solid tumors, and leukemia in both modalities and for colon cancer and breast cancer in photon CSI. Average projected reductions in risk of mortality by more than 4% by applying proton CSI rather than photon CSI were observed in colon cancer, lung cancer, and breast cancer in girls. Compared to boys, girls were at more than twice the lifetime risk of a fatal thyroid or lung cancer for both modalities and a fatal stomach cancer for photon CSI. In these T, they also had much larger risk coefficients than boys. Compared to girls, boys were at almost twice the lifetime risk of developing a fatal colon cancer in photon CSI. With the exception of leukemia, children of ages at exposure less than 8 years had higher lifetime risk of mortality than children of ages at exposure greater than 8 years for every cancer site and both modalities, and this was true by more than a factor of 2 for bladder cancer and other solid tumors in both modalities, prostate cancer in proton CSI, and breast cancer and lung cancer in photon CSI. Unlike with incidence, we did not observe a relationship between the lifetime risk of a fatal leukemia and age at exposure in photon CSI or proton CSI.

For each treatment, total risks of any SMN incidence and SMN fatality were computed as LAR_I,total and LAR_M,total, respectively, while excluding the unrealistic risk of a fatality from multiple cancer sites (see section 2.4). The variations in the average values below were reported as two sample standard deviations of each mean value and, therefore, reflect variability across ages rather than the small uncertainties in physical dose calculations or large uncertainties in the risk model.

The mean LAR_I,total values averaged across patients of all ages and both sexes were 168% ± 111% and 88% ± 44% for photon CSI and proton CSI, respectively. The girls were at higher risk of any SMN incidence than the boys by factors of 1.98 ± 1.53 and 1.47 ± 1.07 for photon CSI and proton CSI, respectively. Children less than 8 years of age at exposure were at approximately twice the LAR_I,total than children greater than 8 years of age for either treatment modality. NMSC accounted for approximately one-sixth and one-third of LAR_I,total of each patient for photon CSI and proton CSI, respectively. In proton CSI, the contributions to LAR_I,total from external neutrons and internal neutrons were small compared to those of therapeutic protons, specifically, they were 5.0% ± 2.1% and 3.6% ± 1.7%, respectively, or, together, one-tenth of the overall risk.

The mean LAR_M,total values averaged across patients of all ages and both sexes were 41% ± 18% and 26% ± 10% for photon CSI and proton CSI, respectively. The girls were at higher risk of any fatal SMN than the boys by factors of 1.43 ± 0.78 and 1.34 ± 0.77 for photon CSI and proton CSI, respectively. Children less than 8 years of age at exposure about 1.6 times higher LAR_M,total than children greater than 8 years of age at exposure for either treatment modality. In proton CSI, the contributions to LAR_M,total from external neutrons and internal neutrons were small compared to those of primary protons, specifically, they were 1.8% ± 0.6% and 1.1% ± 0.4%, respectively, or, together, one-ninth of the overall risk.

The values of $\overline{{R}{L}{A}{{R}}_{T}}$ for the nine patients are shown in figure 3 for each T. The $\overline{{R}{L}{A}{{R}}_{T}}$ values were much less than 1 for subsequent bladder, breast, liver, ovarian, prostate, stomach, thyroid, uterine, and colon cancers and less than 1 for other solid tumors, leukemia, and lung cancer. Among these, the statistical significance of $\overline{{R}{L}{A}{{R}}_{T}}$ < 1 was weakest for subsequent lung cancer, having a p-value of 0.012 for its one-sided t-test. In two cases, the 6- and 9-year-old girls, RLAR_lung was not less than 1. RLAR_NMSC was greater than 1 for each patient except the 9-year-old boy, which suggests that proton CSI is of no benefit over photon CSI for reducing the risk of NMSC. The only cancer site in which a relationship was observed clearly between RLAR_T and age was breast cancer, for which the RLAR_T values increased with age, indicating that a larger reduction in breast cancer risk by proton CSI may be achieved for younger girls than for older girls.

Zoom In Zoom Out Reset image size

Values of $\overline{{R}{L}{A}{{R}}_{I,\text{total}}}$ and $\overline{{R}{L}{A}{{R}}_{M,\text{total}}}$ for the cohort were 0.56 (95% CI, 0.37–0.75) and 0.64 (95% CI, 0.45–0.82), respectively. This result indicated with statistical significance the potential of reducing the risk of SMN in future pediatric MB patients in LMICs by replacing photon CSI with proton CSI. Specifically, the projected lifetime risks of SMN incidence and mortality were reduced by 44% and 36%, respectively. The value of $\overline{{R}{L}{A}{{R}}_{I,\text{total}}}$ was slightly lower for girls than for boys, but $\overline{{R}{L}{A}{{R}}_{M,\text{total}}}$ values were similar between girls and boys. This could be interpreted as a slightly stronger potential reduction in SMN incidence for girls compared to boys by using proton CSI rather than photon CSI. Similar values of $\overline{{R}{L}{A}{{R}}_{I,\text{total}}}$ and $\overline{{R}{L}{A}{{R}}_{M,\text{total}}}$ were observed in younger children and in older children.