Risk factors for radiation-induced rib fractures following proton beam therapy for stage I non-small cell lung cancer: a retrospective study

Abstract

Background

Understanding and managing radiation-induced adverse events is becoming increasingly important in hypofractionated radiotherapy due to the use of higher doses per fraction compared with conventional radiotherapy. Specifically, toxicities of hypofractionated proton and carbon-ion beam therapy are still unclear. We investigated the clinical, anatomical, and dosimetric risk factors for radiation-induced rib fractures (RIRFs) following passive-scattering proton beam therapy (PBT) for stage I non-small cell lung cancer (NSCLC).

Methods

We retrospectively investigated patients with stage I NSCLC who underwent PBT with 66–70 Gy (relative biological effectiveness [RBE]) in 10 fractions, completing a minimum follow-up of 36 months. Rib fractures were detected by follow-up chest computed tomography (CT) examination, independent of symptoms of thoracic pain. Dose-volume histograms of separately contoured ribs on planning CT images were calculated by the treatment planning system in a retrospective manner. Kaplan–Meier and Cox proportional hazards analyses were performed on individual ribs to identify significant risk factors associated with RIRF.

Results

Among the 85 patients finally involved in this study, we identified 116 fractured ribs in 55 participants (64.7%). The 2- and 3-year frequencies of experiencing any RIRF were 36.5% and 52.9%, respectively. The median time-to-fracture was 23.5 months (range: 5–65). We used a total of 224 ribs irradiated over 50 Gy (RBE)—including all the detected fractured ribs—for statistical analysis. Univariate and multivariate analyses revealed the maximum-rib dose to a small volume, position of the maximum-dose point, bone mineral density, 1st rib number, and use of systemic corticosteroids to be related to the incidence of RIRFs.

Conclusions

In addition to dosimetric parameters, factors related to skeletal structure and bone strength are crucial predictors of proton RIRFs and should be considered for safer radiation therapy.

Background

A previous multi-institutional retrospective study reported that passive-scattering proton beam therapy (PBT) resulted in acceptable survival rates for stage I non-small cell lung cancer (NSCLC) in Japan [1]. This was the first nationwide study of its kind, mainly including patients with peripheral lung cancer who underwent treatment with 66 Gy (relative biological effectiveness [RBE]) in 10 fractions. Although a regimen involving 48 Gy in 4 fractions of X-ray-based stereotactic body radiotherapy (SBRT) is widely employed in Japan for early-stage inoperable lung cancers [2, 3], to the best of our knowledge, there are no Japanese institutions currently using hypofractionated PBT in < 10 fractions for the treatment of primary lung cancer. The physical advantages of the proton beam—for example, the depth-dose profile of the Bragg peak—mean that further hypofractionation will be tolerable with PBT compared with X-ray radiotherapy [4]. However, the current understanding of post-PBT toxicity poses a significant limitation. Adverse events associated with radiotherapy of the thoracic region include radiation pneumonitis, chest-wall pain, and radiation-induced rib fractures (RIRFs). The latter occur more frequently after SBRT than other methods of conventional radiotherapy which use a fraction dose of 2 Gy. Although the risk factors of RIRF after SBRT are well reported for lung cancer [5,6,7,8,9,10,11,12,13,14,15], the risk factors of particle therapy for lung, liver and breast cancer are considerably understudied [16,17,18,19,20] and the findings of previous studies as to whether sex and age are significant are inconsistent. These discrepancies may be attributed to differences in patient demographics, follow-up periods, endpoint definitions, and radiotherapy regimens. Thus, definitive analysis is essential in order to clarify the incidence and various risk factors of RIRF. The present study aimed to address this gap in the knowledge by investigating the occurrence of proton therapy-induced rib fractures among patients with NSCLC as comprehensively as possible.

Methods

Eligibility criteria

We retrospectively reviewed all 197 patients with stage I NSCLC who underwent passive-scattering PBT at the Medipolis Proton Therapy and Research Center (MPTRC) between April 2011 and March 2018. Diagnoses of NSCLC were accepted according to the 7th edition of the Union for International Cancer Control. Inclusion criteria were as follows: (1) treatment with a prescribed dose of 66–70 Gy (RBE) at the isocenter in 10 fractions, (2) minimum radiographic follow-up of 36 months, (3) no previous radiation or surgical intervention on ribs adjacent to the tumors treated with PBT. Exclusion criteria were as follows: (1) local and/or bone metastatic recurrence after PBT, to avoid difficulty in distinguishing RIRFs from pathological fractures, (2) any additional radiotherapy overlapping the field of proton-radiation.

Treatment procedure

Patients were immobilized in the supine position with a thermoplastic body shell (ESFORM; Engineering System, Matsumoto, Japan). Nonhelical simulation computed tomography (CT) (Aquilion LB; Canon Medical Systems, Ohtawara, Japan) with a slice thickness of 2 mm was performed at the end-exhale phase using a respiratory gating system (AZ-733V; Anzai Medical, Shinagawa, Japan). Internal margins were evaluated using four-dimensional CT (4DCT) to measure motion of lung tumors during breathing. A treatment planning system with a pencil-beam algorithm for proton dose calculation (XiO-M; Elekta, Stockholm, Sweden/Hitachi, Tokyo, Japan) was used for contouring and dose-distribution calculation. Gross tumor volume (GTV) was determined from treatment planning CT images using pulmonary window settings. Clinical target volume (CTV) was defined as the 5-mm isotropic expansion of the GTV, excluding bony structures or the chest wall. Planning target volume (PTV) was defined as the CTV plus a set-up margin of 5 mm and an anisotropic internal margin, which was determined using 4DCT. We applied a generic RBE value of 1.1 for PBT planning. The majority of the treatment plans consisted of two (range: 1–3) gantry angles (Fig. 1a). For patient positioning, we performed bony-structure-based manual registration between reconstructed radiographs and daily orthogonal kilo-voltage X-ray images prior to proton irradiation, corrected according to visible tumors or the diaphragm. A synchrotron was used to accelerate protons to kinetic energies of 150 or 210 MeV, and the distal end of the proton beam was finely adjusted to match the PTV through range shifters and patient-specific compensating filters. Bar-ridge filters spread the Bragg peak into 30–140-mm-water-equivalent thickness in 10-mm steps. Respiratory-gated proton beam delivery [21] was conducted during the expiratory phase, except in patients who exhibited negligibly small tumor motion.

Fig. 1

Screenshots of an illustrative proton treatment plan for a tumor in the right lower lobe. (a) Proton dose distribution consisted of two gantry angles on the simulation computed tomography image in the axial plane. (b) A three-dimensional model of the body and right 4th–7th ribs contoured after final follow-up. Fractures of the 5th and 6th ribs (illustrated in red) were detected at 17 and 23 months after completion of proton beam therapy, respectively. The 4th and 7th ribs adjacent to the fractured ribs (illustrated in green) showed no evidence of fracture at the time of the last follow-up (59 months)

Follow-up and detection of radiation-induced rib factures

The endpoint of this study was the occurrence of an RIRF, which was defined as cortical discontinuity of the rib, detected by CT. Follow-up chest CT was performed every 3 months in the first year, then every 6–12 months for at least 4 years after completion of PBT, in accordance with the recommended interval [22]. The standard follow-up length of 5 years was extended depending on the patients’ condition from a clinical perspective. For the purposes of this study, the total follow-up time was defined as the time from the final treatment date to the last CT scan date. All follow-up CT images were reviewed, regardless of the presence or absence of clinical symptoms, to identify the rib number, fracture location, and appearance time of RIRF.

Treatment plan-related data

After evaluating RIRF during the period up to the final follow-up, ribs that were irradiated with a dose of 30 Gy (RBE) or more, ribs in which fractures were observed, and nonfractured ribs adjacent to the fractured ribs were retrospectively contoured by hand on simulation CT images (Fig. 1b). The minimum distance between GTV and ribs (GTV–rib distance) was obtained using dummy contours generated from the GTV with three-dimensional symmetrical expansion. Using the standard functions of the XiO-M, structural resolution was set to 1 mm and dosimetric parameters were extracted from the dose-volume histograms of the separately contoured ribs. Dosimetric data that were used for statistical analyses included the maximum-point dose (D_max), maximum doses to small absolute volume (D_0.5cc, D_1cc, D_2cc), and volumes irradiated with over 50, 55, 60, and 65 Gy (RBE) (V₅₀, V₅₅, V₆₀, V₆₅).

Bone mineral density measurement

Bone mineral density (BMD) of the lumbar vertebrae and hips is a major component of bone strength and is associated with fracture risk. Although this parameter is typically assessed using dual-energy X-ray absorptiometry, the MPTRC facility lacked the necessary equipment for its implementation. For the purposes of this study, the pretreatment simulation-CT-derived mean Hounsfield unit values of the 10th thoracic vertebral cancellous bone (HU_Th10) were taken to represent BMD [23]. This point was selected for retrospective data collection because the Th11–12 vertebrae were outside the field of view for some participants (Fig. 2).

Fig. 2

Measurement of the bone mineral density. Mean computed tomography values of the 10th thoracic vertebral cancellous bone in the rectangular region of interest (yellow rectangle) on the midsagittal plane were evaluated as a substitute for bone mineral density measurements.

Statistical analysis

Fracture locations were defined using a previously described methodology for categorizing tumor locations into three groups [6]. The time-to-fracture was analyzed for each fracture location using the Wilcoxon rank-sum test, with p values adjusted using Holm’s method. We compared CT-derived HU_Th10 values between male and female patients using the Wilcoxon rank-sum test, and the association with age was evaluated using Pearson’s correlation coefficient. More accurate prediction of RIRF risk requires borderline doses that may contribute to fracture development to be determined, and data from ribs exposed to low doses that are unlikely to result in RIRF were excluded from statistical analysis. We took 50 Gy (RBE) to be the threshold dose, and only ribs with D_max exceeding this value were included in most statistical analyses. Receiver operating characteristic (ROC) curves were generated to extract the most relevant dosimetric factor for the prediction of RIRF and optimal cut-off values. Comprehensive risk assessment was carried out by dividing high-dose-exposed ribs into two or three groups according to clinical (i.e., which patient they belong to), anatomical, and dosimetric factors. Factors used for classification included age, sex, body mass index, hemoglobin A1c (HbA1c), administration of systemic corticosteroids, smoking history, laterality, tumor location, anatomically classified rib number, HU_Th10, position of maximum-dose point in the rib, and D_1cc. Data truncated at a follow-up time of 5 years were used for Kaplan–Meier and Cox proportional hazards model analyses to avoid loss of reliability due to the limited sample size. All statistical analyses were performed using R, version 4.2.1 (R Foundation for Statistical Computing, Vienna, Austria). A p value of < 0.05 was considered statistically significant.

Results

Subjects

Eighty-five patients with 87 tumors met the criteria. Two participants had two tumors in the ipsilateral lung; one underwent PBT twice sequentially and the other received a common radiation field covering two GTVs. A total of 13 participants received systemic corticosteroids (prednisolone): 11 for radiation pneumonitis occurring several months after PBT and 2 for rheumatic diseases. The characteristics of the study population and tumors are summarized in Table 1.

Table 1 Characteristics of the study population and tumors

Statistical summary of the occurrence of radiation-induced rib fracture

The median radiographic follow-up time was 59 months (range: 36–113 months). The total number of follow-up CT examinations was 979, and the mean acquisition interval was 5.03 months. During follow-up, 116 fractured ribs at 134 fracture sites were detected in 55 participants (64.7%). Multiple fracture sites in one rib (maximum number of sites of two) were identified for 18 ribs of 13 participants. The 2- and 3-year frequencies of experiencing at least one RIRF were 36.5% and 52.9%, respectively. The median time-to-fracture was 23.5 months (range: 5–65 months; derived from the fractured ribs, not from fracture sites). Of the 116 fractures, 93 (80.2%) developed within 36 months of PBT (Fig. 3).

Fig. 3

Histogram of time-to-fracture development

The floating ribs (11th–12th), located in the lower part of the thoracic cage, were outside of the field of radiation, and no RIRFs were detected in the 9th–12th ribs. Of the 116 fractures, 82 developed in the right side of the body and 34 in the left. The laterality of all fracture sites was consistent with the irradiation field involving the tumor. Most of the fractures occurred in the true ribs (1st–7th), which are directly connected to the sternum via the costal cartilage (Fig. 4).

Fig. 4

Butterfly chart of rib fractures illustrating rib number and laterality

Of the 134 fracture sites, 15 were on the anterior side, 79 on the lateral side, and 40 were on the posterior side of the individual ipsilateral lungs (Fig. 5a). One case of multiple fractures in the same rib is shown in Fig. 5 (which is the same individual shown in Fig. 1). As in the illustrative case, the initial fracture developed distal to the vertebral body, with the subsequent fracture developing proximal to the vertebral body in 13 cases. The opposite pattern was observed in three cases, and two were unidentifiable because the fracture sites were observed simultaneously in the same follow-up CT images.

Fig. 5

Follow-up computed tomography images showing a case of multiple fracture sites in the same rib. (a) Illustration of the boundaries used to define fracture location in three fields (anterior/lateral/posterior), created in steps of 120° around the center of the ipsilateral lung in the axial plane. No evidence of fracture in the right 6th rib was observed at 20 months after therapy. (b) The initial fracture site of the right 6th rib appeared in the lateral field (white arrow). (c) The subsequent fracture site appeared in the same rib, in the posterior field (red arrow).

Time-to-fracture, categorized by fracture location, was statistically significant in pairwise comparisons (Fig. 6). Fractures tended to occur sooner after PBT as the region shifted from anterior to lateral to posterior, i.e., more distal from the vertebral body.

Fig. 6

Box plot showing the time-to-fracture for the 134 fracture sites. All combinations of pairwise comparisons by fracture location were statistically significant (* p < 0.05, *** p < 0.005).

Dosimetric parameters

A negative cross-correlation between the GTV–rib distance and D_max was observed (Fig. 7). The minimum value of D_max among the fractured ribs was 53.2 Gy (RBE); hence, the 224 ribs with D_max >50 Gy (RBE) were selected as input for subsequent analyses.

Fig. 7

Scatter plot showing the gross tumor volume–rib distance versus the maximum-point dose of ribs. Data points of ribs with fracture (red circle) were concentrated in the upper-left area.

ROC curves were generated for dosimetric parameters including, D_max, D_0.5cc, D_1cc, D_2cc, V₅₀, V₅₅, V_60, and V₆₅ as well as GTV–rib distance to identify the best predictor for RIRFs and determine the corresponding cut-off values. Among these, the D_1cc was found to be the most effective parameter for predicting RIRF, achieving the highest Youden’s index, whereas the GTV–rib distance exhibited the lowest area under the curve (AUC) and Youden’s index (Table 2). Under the null hypothesis that the AUC is equal to 0.5, the p values for all parameters were < 0.0001.

Table 2 Receiver operating characteristic analysis for the dosimetric parameters and gross tumor volume–rib distance

Bone mineral density

The median CT value of the Th10 vertebral cancellous bone was 138.1. The median HU_Th10 for males was 151.0 (range: 56.3–374.9), which was significantly higher than that for females of 127.6 (range: 33.6–200.1) (p < 0.005). Figure 8 shows the scatter plots and regression lines of HU_Th10 versus age for male and female participants. The correlation coefficient between HU_Th10 and age was − 0.35 (p < 0.05) for males and − 0.59 (p < 0.001) for females, which implies the effect of aging.

Fig. 8

Scatter plots and regression lines of Hounsfield unit values of the 10th thoracic vertebral cancellous bone versus age for male and female participants

Univariate and multivariate analyses

Kaplan–Meier curves were generated to evaluate and compare the fracture-free probability among groups according to clinical, anatomical, and dosimetric factors. The analysis of HU_Th10 indicated a significantly lower fracture frequency in the group with high BMD (Fig. 9).

Fig. 9

Kaplan–Meier curves of the fracture-free probability for high and low bone mineral density. The shaded area indicates the 95% confidence intervals. ^|Censorship for lost to follow-up

The univariate analysis revealed sex, use of systemic corticosteroids, rib number, HU_Th10, position of the maximum-dose point, and D_1cc to be statistically significant. Although patients with diabetes mellitus and high HbA1c levels are generally considered to have an increased risk of fractures, HbA1c did not show a significant difference (Table 3).

Table 3 Results of univariate analysis

With the exception of sex (to avoid multicollinearity), five of the above six factors were applied to the Cox proportional hazards model. The following five factors were found to be statistically significant: use of systemic corticosteroids, 1st rib, HU_Th10, maximum-dose point in posterior ribs, and D_1cc (Table 4).

Table 4 Results of multivariate analysis

Discussion

In addition to exposure of small-rib volumes to high-dose radiation, our study suggests that multiple factors, including use of systemic corticosteroids, 1st rib, HU_Th10 value, and position of the maximum-dose point, are associated with frequencies of RIRF. Apart from the radiation dose, these factors are all related to the structural strength of the bone, thus indicating that comorbidities strongly associated with bone strength—such as osteoporosis, collagen diseases requiring steroid therapy, and breast cancer in elderly women—significantly increase the fractures when starting radiotherapy.

The most delayed case in the present study occurred at 65 months after PBT, highlighting the fact that irradiated ribs can remain fragile and at risk of fracture for a very long time. The median time-to-fracture development from completion of proton therapy was 23.5 months, which was not significantly different from the most frequently reported value of 22 months for X-ray-based SBRT [7,8,9]. Therefore, studies that carry out ROC curve analysis using participants who were followed up for less than 2 years should be considered with the caveat that dependence on the follow-up period needs to be carefully validated. The present study imposed a sufficiently long minimum follow-up period of 36 months. The 3-year cumulative incidences of Grade 1 and Grade 2 RIRFs after SBRT are reportedly 45% and 3%, respectively, from a study involving 126 patients with stage I primary lung cancer [11]. In comparison, the present study did not distinguish between Grades, and identified a slightly higher 3-year frequency of RIRF. Increasing the number of ports or introducing the pencil-beam scanning technique to reduce the dosage to the ribs proximal to the lung tumor could help to reduce the incidence of RIRF.

The use of a small number of beams and certain gantry angles can expose even ribs distant from the tumor to high doses, potentially leading to fractures. In cases where treatment plans achieve high dose conformity, the distance between the tumor and ribs is strongly correlated with the rib dose, and is a risk factor for RIRF [7, 10]. While the distance between GTV and ribs can be measured on CT images without dose-distribution calculations, the GTV-CTV and CTV-PTV margin settings and prescribed doses differ between institutions. Therefore, rib dose is a more universal and direct risk factor for RIRF than the distance.

Small volumes of ribs exposed to high doses such as D_max, D_0.5cc, and D_1cc can be a trigger for inducing fractures occurring in a narrow region comparable to the width of the ribs. Considering this, ribs exhibit characteristics of so-called serial organs. It is generally known that D_max, which is defined for infinitesimal volumes, is dependent on the accuracy of the contour. However, CT images based on electron density of body tissues provide a clear contrast between soft tissue and bone cortex, enabling the precise contouring of the ribs. Furthermore, it should be noted that bony-structure-based patient positioning and respiratory-gated irradiation allows the actual rib dose to be estimated accurately, even using a general dose-volume histogram that does not account for respiratory organ motion.

Uncertainties can occur in dose-scaling methods associated with variations in the number of fractions, such as the Linear-Quadratic model; therefore, this study only included patients who underwent 10-fraction regimens, and the lowest D_max among the fractured ribs was 53.2 Gy (RBE). This value is in good agreement with the report on X-ray-based SBRT in 7–9 fractions [12], suggesting that the frequency of rib fractures is remarkably low if the maximum dose equivalent to 50 Gy (RBE) or less is delivered in 10 fractions. Further investigation to objectively determine the threshold dose for the risk of “radiation-induced” rib fractures, irrespective of observer bias, is warranted. It should be noted that including ribs with no risk of fracture in ROC curve analysis increases the number of true negatives, although no improvement in prediction accuracy can be achieved. The AUC values in the present study indicated moderate predictive power, highlighting the limitations of predicting fracture incidence by dosimetric factors alone.

Age and sex have been reported to contribute to the rate of RIRF, along with D_0.5cc as a representative dose parameter [5], and BMD is known to correlate with these two factors. The CT-derived HU_Th10 values, which were taken to represent BMD in the present study, exhibited consistent correlation and were significant explanatory variables for RIRFs. If a study included a certain population of elderly females—who tend to have low BMD—age and sex would more likely be identified as factors relevant to RIRFs. However, it would be more direct and appropriate to focus on BMD itself as a risk factor.

Consideration of anatomical differences between individual ribs is important when assessing the risk of fracture. Previous studies that specified anatomical rib numbers reported very low rates of fracture in the first rib (1 out of 178 when reviewing all previous studies: Taremi et al.: 0 of 41 [5], Miura et al.: 0 of 77 [11], Carducci et al.: 1 of 60 [13]). The present study identified two first-rib fractures out of 116, and multivariate analysis revealed this rib, which is thicker and shorter, to be significantly more resistant to fractures compared with the typical 3rd–9th ribs. Furthermore, the lower risk that was observed when the maximum-dose point was located posteriorly rather than laterally, appears to be linked to the significantly delayed fracture development when the fracture occurred closer to the vertebral body. We postulate that the low fracture risk proximal to the vertebral body is due to the thicker bone and restricted mobility due to the surrounding tissue.

The risk of RIRFs was increased among individuals receiving steroid therapy in the present study. The majority of participants who received steroids used them for a limited period, ranging from several weeks to months, depending on the severity of their radiation pneumonitis. A previous meta-analysis revealed that the risk of fractures increases rapidly within the first 3–6 months after starting oral corticosteroids, and decreases after cessation [24]. Thus, preventive measures against fractures should be taken during corticosteroid treatment for radiation-induced pneumonitis. However, given the limited number of cases in this study, further investigations involving detailed analyses of dosage and duration of steroid use are warranted.

The present study has some limitations which should be acknowledged. This was a single-institution retrospective study. Thus, future prospective studies should obtain the BMD using a universal and quantitative manner to avoid introducing measurement errors across multiple institutions. We estimated BMD using simulation CT images obtained with a single scanner under the same image-acquisition protocols.

Conclusions

We identified the following risk factors for RIRF related to skeletal structure, bone strength, and radiation dose among patients with NSCLC undergoing PBT: position of the maximum-dose point, BMD, rib number, use of systemic corticosteroids, and maximum-rib dose to a small volume. These findings not only contribute to accurate risk communication with patients but also emphasize the potential drawbacks of considering only dose during treatment planning. Furthermore, the method used in this study, which involves individual rib analysis, is expected to provide valuable insights when applied to other regimens, specifically in the introduction of hypofractionated radiotherapy while maintaining the patients’ quality of life.