- Email: [email protected]
Radiation dose to heart base linked with poorer survival in lung cancer patients
European Journal of Cancer 85 (2017) 106e113
Available online at www.sciencedirect.com
ScienceDirect journal homepage: www.ejcancer.com
Original Research
Radiation dose to heart base linked with poorer survival in lung cancer patients Alan McWilliam a,b,*, Jason Kennedy b, Clare Hodgson c, Eliana Vasquez Osorio a, Corinne Faivre-Finn a,b,1, Marcel van Herk a,b,d,1 a Division of Molecular and Clinical Cancer Science, School of Medical Sciences, Faculty of Biology, Medicine and Health, University of Manchester, Manchester Academic Health Science Centre, UK b Department of Radiotherapy Related Research, The Christie NHS Foundation Trust, Manchester, UK c Clinical Outcomes Unit, The Christie NHS Foundation Trust, Manchester, UK d NIHR Manchester Biomedical Research Centre, Central Manchester University Hospitals NHS Foundation Trust, Manchester Academic Health Science Centre, UK
Received 28 April 2017; received in revised form 21 July 2017; accepted 31 July 2017
KEYWORDS Lung cancer; Radiotherapy; Cardiotoxicity
Abstract Background: Advances in radiotherapy (RT) have allowed an increased proportion of lung cancer patients to be treated curatively. High doses delivered to critical thoracic organs can result in excess mortality with tolerance doses poorly defined. This work presents a novel method of identifying anatomical dose-sensitive regions within the thorax. Methods: A high-resolution, normal-tissue dosimetric analysis was performed to identify regions in the heart that correlate with poorer survival. A total of 1101 patients treated with curative-intent RT were selected and all computed tomography imaging and dose distributions were deformed to a reference. Mean dose distributions were created for patients who survived versus those who did not at a set time point. Statistical significance of dose differences was investigated with permutation testing. The dose received by the most statistically significant region of the thorax was collected in all patients and included in a multivariate survival analysis. Results: The permutation testing showed a highly significant region across the base of the heart, where higher doses were associated with worse patient survival (p < 0.001). Cox-regression multivariate analysis showed region dose, tumour volume, performance status and nodal stage were significant factors associated with survival, whereas cardiac mean dose, V5 and V30 showed no significance. Survival curves, controlling for these factors, were plotted with patients receiving doses greater than 8.5 Gy to the identified region showing worse survival (log-rank p < 0.001, hazard ratio 1.2).
* Corresponding author: The Christie NHS Foundation Trust, Radiotherapy Related Research (Dept 58), Wilmslow Road, Manchester, M20 4BK, UK. E-mail address: [email protected] (A. McWilliam). 1 Joint last authors. http://dx.doi.org/10.1016/j.ejca.2017.07.053 0959-8049/ª 2017 Elsevier Ltd. All rights reserved.
A. McWilliam et al. / European Journal of Cancer 85 (2017) 106e113
107
Conclusion: The application of this novel methodology in lung cancer patients identifies the base of the heart as a dose-sensitive region for the first time. ª 2017 Elsevier Ltd. All rights reserved.
1. Introduction Radiotherapy (RT) plays a major role in curative-intent treatment of lung cancer. Advances in radiation technology have enabled curative-intent doses to be delivered to a larger proportion of locally advanced patients while keeping doses to normal tissues within accepted safety limits. This increase in the proportion of patients treated with curative intent is expected to improve local control, which is known to be associated with better survival [1,2]. However, over the last 2 years, evidence relating irradiation of the heart to excess mortality has emerged. Bradley et al. [3] reported the outcomes from RTOG 0617, a multicentre phase III study comparing 60 Gy versus 74 Gy delivered in 2 Gy fractions. They showed that a higher treatment dose was associated with increased mortality and multivariate analysis of survival identified heart doses, volume receiving 5 Gy (V5) and volume receiving 30 Gy (V30), as associated with patient survival. Subsequently, a study showed that higher mean heart dose was significantly associated with higher cardiac event rates [4]. There is now great interest in investigating how dose-to-anatomical sub-structures of the heart links to survival. Despite this emerging evidence, in current practice, heart dose constraints remain poorly defined, generic and date back to articles published over 25 years ago [5,6]. One of the major limitations of previous studies is the evaluation of dose to the whole heart as one structure or divided into a small number of pre-defined sub-structures. There is, therefore, a clear unmet need to define heart sub-structures at risk and heart dose constraints for this group of patients. The analysis of large populationbased cohorts of lung cancer patients treated with curative RT is an attractive strategy to identify dosesensitive heart structures. However, this approach is limited by the fact that heart sub-structures are not delineated routinely as part of the RT planning process. The methodology in this article, applied to lung cancer patients for the first time, does not require any delineation structures. Instead, performs a high-resolution, voxel-by-voxel dosimetric analysis, identifying regions correlated with patient survival. 2. Methods A total of 1163 lung cancer patients treated between 2010 and 2013 at a single academic cancer centre, with
routine curative-intent RT (55 Gy in 20 fractions), with or without induction chemotherapy, were randomly selected for analysis. Institutional approval had been gained to use these patients. Three-dimensional conformal and intensity-modulated radiation therapy (IMRT) plans were both used in this work. Patient images were deformable registered to a reference patient using the Nifty Registration package, (NiftyReg, UCL, UK [7]). To avoid potential effects from sliding tissue between the ribs and the lung, bone was excluded from the registration process by truncating Hounsfield Units at 100. The deformable registration is based on a Bspline parameterisation approach. The planned dose distribution was normalised to the reference by directly applying the derived deformable vector field. A visual validation of the registration was performed to ensure that all patients were successfully normalised into the same spatial reference. This approach allows large numbers of patients to be included in the analysis without the requirement for additional contouring. A sub-set of 386 patients, in whom the heart had been contoured by a clinical oncologist, were used to evaluate the accuracy of deformable registration. The uncertainty was estimated using the standard deviation of the centre of mass coordinates of deformed heart contours. To assess the influence of this uncertainty on the data mining results, dose distributions for each patient were blurred by a Gaussian filter with corresponding width. Results using original dose distributions and blurred distributions were compared. The difference in mean dose distributions for patients who survived versus those who did not survive at a given time point from end of treatment (3e36 months) was calculated, with patients censored for follow-up. To test if the dose difference between the two groups was statistically significant, permutation testing was used, with 1000 permutations performed. The maximum t-value was used over the average dose distribution to test for significance [8]. The test statistic, maximum t-value, is calculated from the difference in mean dose in a voxel between the two groups divided by the standard deviation of the voxels, with the maximum t-value selected. Permutations then generate random samples to determine the distribution of maximum t-values, this tests the null hypothesis that there is no difference between the two groups. This approach indicates areas of the anatomy, where the observed dose difference is related to a statistically significant difference in patient survival. To
108
A. McWilliam et al. / European Journal of Cancer 85 (2017) 106e113
ensure that the permutation testing was not influenced by variability in dose, profiles were plotted through the region of highest statistical significance for the dose difference, standard deviation of dose and the t-value. The area of highest statistical significance was used to define a small region of interest, defined at 90% of the maximum t-value. The mean of the individual doses received to this region was calculated for every patient. Using univariate analysis the parameters (patient and tumour characteristics) significantly associated with patient survival were determined (selected as those with p < 0.05). Factors that remained significant were included in a Cox-regression multivariate survival model. Tumour volume was collected for all patients, as volume will influence dose delivered to any anatomical region. Mean lung dose (total lung volume minus overlapping gross tumour volume) was also calculated, as this is a known predictor of lung toxicity. For comparison with published results the mean dose, the V5 and V30 were collected for the 386 patients with heart contours. Hazard ratios (HR) were calculated for all variables with 95% confidence intervals (CIs). KaplaneMeier curves were plotted as quartiles of dose received to the identified anatomical region, allowing an appropriate dose to be selected with which to group patients. Log-rank tests were performed to show any significance in overall survival. All statistical analysis was performed using the Statistical Package for Social Sciences, version 22 (SPSS, Chicago, IL). Validation was performed on a set of 89 sterotactic ablative radiotherapy (SABR) patients with stage I nonsmall cell lung cancer (NSCLC) treated with 60 Gy in five fractions (100% dose to cover 95% of the planning target volume). Patients were processed using the same methodology and reference patient. The identified region was transferred onto these patients, the dose to this area was collected and KaplaneMeier curves plotted. To calculate the appropriate dose to group the validation cohort, a biological effective dose (BED) calculation was performed. The dose selected to group the patients treated with 55 Gy in 20 fractions was used to calculate a BED (assuming an alpha/beta ratio of 2 Gy [9,10]). This BED was then used to calculate the BED with which to group the patients treated with SABR. 3. Results Visual inspection of the deformable image registration identified 62 registration failures, patients who displayed significant atelectasis or had previous surgery and were removed from subsequent analysis. Patient demographics for the 1101 patients who remained are included in Table 1. No discrimination of the patients was made on these characteristics. The standard deviation in centre of mass coordinates, for the 386 patient heart contours available, was
Table 1 Patient demographics from the 1101 patients selected for analysis. Not all patients had complete sets of records available; the ‘Total in group’ column shows the number of data points available for each variable. Variable
Sub-variable Sub-total
Total in group
Gender
Male Female
1101
Age (median) Tumour size (cm3) (median) Smoking history Current Ex-smoker Life-long non-smoker T-stage T1 T2 T3 T4 N-stage N0 N1 N2 N3 Induction chemotherapy Yes No
593 508 73 (38e95) 34.5 (0.5e528) 153 197 8 159 434 238 169 546 137 257 66 266 835
1101 1101 359
1000
1006
1101
found to be 3.9 mm in the lefteright, 4.9 mm anterioreposterior and 7.4 mm cranialecaudal direction after deformable registration. All time points investigated from 6 months post RT showed a statistically significant correlation between the mean dose distribution and survival (p < 0.001). Results from the original and blurred distributions were almost identical. The most significant dose difference was located across the base of the heart, where excess dose was associated with poorer patient survival. A representative result from the permutation testing using the blurred dose distribution is shown in Fig. 1, where patient survival at 12 months has been set. The region of highest significance (p-value uncorrected for multiple testing) overlaps with the aorta and includes the origin of the coronary arteries and sinoatrial node. The coronal and sagittal views show that the significant area is contained across the base of the heart (Fig. 1A). Profiles of dose difference (between patients who survived versus those who did not at 12 months), standard deviation of dose across all patients and the tstatistic were plotted (Fig. 1B). Profiles were taken through the origin of the left coronary artery. Note that the standard deviation is fairly homogenous over the relevant heart region, therefore the location of significance is defined by the dose difference between dead and living patients. The area of highest significance, defined at 90% of the maximum t-value, was used to define a region of interest. This was located in the base of the heart and limited at the boundary with the lung. The volume of the identified anatomical region was 12 cm3 with a median average dose, across all patients of 16.3 Gy. Dose to this
A. McWilliam et al. / European Journal of Cancer 85 (2017) 106e113
109
Fig. 1. In the top row (A) axial, sagittal and coronal views from the computed tomography scan of the reference patient. The statistical significance from the permutation testing is overlaid (p < 0.001). This shows that excess dose to the base of the heart was highly correlated with poorer patient survival. The lower row (B) shows profiles taken through the left coronary artery (as indicated on the patient anatomy). Dose difference, standard deviation of dose and the t-value from the permutation testing are plotted.
region for all patients was collected, and the Coxregression univariate analysis showed that survival was correlated with dose to this area (p Z 0.04), Table 2. However, the mean dose to the heart, V5 and V30 were not predictive of survival for the sub-group of patients (386 patients with heart contours analysed). These findings remained in the Cox-regression multivariate analysis, where the defined region in the base of the heart provided a HR of 1.25 (95% CI 1.01e1.56, p Z 0.04). This analysis indicates the importance of this region in patient survival. Nodal stage (categorical), performance status (categorical), age (continuous, HR 1.01; 95% CI 1.00e1.02, p Z 0.04) and tumour volume (continuous, HR 1.01; 95% CI 1.01e1.01, p < 0.001) were also strongly correlated in the multivariate analysis. T-stage, gender and mean lung dose showed significance in univariate analysis but did not maintain significance in multivariate analysis. Survival curves were plotted on the quartiles of dose to the identified heart area (1st quartile, 0e8.5 Gy, 2nd Quartile 8.5e16.3 Gy, 3rd quartile 16.3e27.3 Gy and 4th quartile 27.3e49), quartiles defined with equal numbers in each, Fig. 2A. Mean survival times of the quartiles were 28, 24, 21 and 17 months, respectively. The 1st quartile dose, 8.5 Gy, was selected as the dose to group the patients. The KaplaneMeier survival curves included with a clear split between the patient groups was seen, with those receiving a dose greater than
8.5 Gy, showing significantly worse survival, log-rank p < 0.001, Fig. 2B. The survival curves with nodal stage, set as a categorical variable, showed a similar response between those patients receiving a low dose to the identified anatomical area and those receiving a high dose, Fig. 3. Log-rank tests between the patients receiving less or greater than 8.5 Gy to the identified anatomical region were calculated. For N0 patients, a significant split remained (p Z 0.002, 546 patients), as well as for N2 patients (p Z 0.012, 256 patients). There was no significant difference in N1 patients (137 patients, with seven in the low-dose group) and N3 patients (67 patients with four in the low-dose group), possibly due to small patient numbers. Validation was performed with 89 patients treated with SABR, receiving 60 Gy in five fractions, median tumour volume 4 cm3 (0.1e38 cm3). The permutation testing showed the same pattern of response across the base of the heart, although was not significant in this small group of patients. The dose received by each patient to the identified anatomical region was collected. The BED that corresponds to a dose of 8.5 Gy for the cohort of patients treated with 20 fractions is 10.3 Gy. For the cohort of patients treated with SABR, this BED corresponds to a physical dose of 6.3 Gy. KaplaneMeier survival curves were plotted with patients grouped into those receiving greater than or less
110
A. McWilliam et al. / European Journal of Cancer 85 (2017) 106e113
Table 2 Univariate and multivariate survival analysis, hazard ratios with 95% confidence intervals are included. Univariate
Tumour volume (continuous) Dose to defined region (greater versus less median) Heart mean dose (continuous) Heart V5 (continuous) Heart V30 (continuous) Age (continuous) Mean lung dose (continuous) Gender (female versus male) Induction chemotherapy (yes versus no) T-stage (T1 reference) T2 T3 T4 N-stage (N0 reference) N1 N2 N3 Performance status (PS 0 as reference) 1 2 3
Multivariate
HR (95% CI)
P
HR (95% CI)
P
1.005 (1.004e1.006) 1.24 (1.01e1.53) 1.010 (0.99e1.03) 1.00 (0.99e1.01) 1.01 (0.99e1.02) 1.01 (1.00e1.02) 1.06 (1.04e1.08) 1.22 (1.07e1.39) 0.83 (0.68e1.03)
<0.001 0.04 0.32 0.61 0.35 0.89 <0.001 0.003 0.10 <0.001 0.01 <0.001 <0.001 0.01 0.02 0.01 0.01 0.03 0.09 0.06 0.01
1.01 1.25 e e e 1.01 1.01 1.22 0.92
<0.001 0.04 e e e 0.04 0.45 0.24 0.46 0.24 0.27 0.24 0.45 0.01 0.29 0.02 0.02 0.43 0.24 0.03 0.02
1.41 (1.13e1.76) 1.79 (1.41e2.27) 1.98 (1.53e2.55) 0.66 (0.41e1.06) 1.76 (1.08e2.85) 1.86 (0.85e4.07) 1.17 (0.98e1.40) 1.20 (0.99e1.45) 1.52 (1.11e2.07)
(1.01e1.01) (1.01e1.56)
(1.00e1.02) (0.98e1.04) (1.03e1.45) (0.74e1.15)
1.17 (0.89e1.54) 1.20 (0.88e1.64) 1.41 (1.00e1.96) 0.86 (0.66e1.13) 1.33 (1.06e2.24) 1.54 (1.07e2.24) 1.19 (0.89e1.60) 1.33 (0.98e1.82) 1.36 (0.88e2.11)
CI, confidence interval; HR, hazard ratio.
than this dose (Fig. 4). There is a statistically significant difference in survival between the two groups (log-rank p Z 0.016). In a multivariate analysis, the tumour volume, age and the dose to the heart split at 6.5 Gy was included. Tumour volume was not significant (continuous, HR 1.01; 95% CI 0.97e1.06, p Z 0.64), either was age (continuous, HR 1.01; 95% CI 0.97e1.05, p Z 0.74). The dose to the heart was significant with patients receiving less than 6.5 Gy, showing superior survival (HR 2.11; 95% CI 1.10e4.07, p Z 0.025).
4. Discussion This article has generated a strong hypothesis, which we believe can now aid in directing future studies. In an unselected group of 1101 patients, we have identified the base of the heart as a dose-sensitive region, where excess dose results in poorer patient survival (HR 1.25, 95% CI 1.01e1.56, p Z 0.04). For those patients where heart contours were available (386 patients), neither the mean dose, V5 or V30 showed significance in univariate or
Fig. 2. (A) Survival curves plotted split on quartiles of the dose delivered to the identified anatomical area of the heart (8.5 Gy, 16.3 Gy, 27.3 Gy). (B) KaplaneMeier curves with patients grouped on the first quartile dose of 8.5 Gy (275 patients in the low-dose group and 826 in the high-dose group), numbers at risk are provided for each group.
A. McWilliam et al. / European Journal of Cancer 85 (2017) 106e113
111
Fig. 3. Survival curves categorised on nodal stage for patients receiving less than 8.5 Gy (175 patients) (A) and greater than 8.5 Gy (926 patients) (B) to the identified anatomical region (N0 - 546 patients, N1 - 137 patients, N2 - 238 patients, N3 - 67 patients).
multivariate analysis. These measures have previously shown importance in analysis from the RTOG 0617 trial [3] and the pooled analysis of stage III NSCLC trials by Wang et al. [11]. Based on our analysis, the excess dose to the base of the heart seems to be a more important predictor of early death than dose-volume metrics currently identified (comparison within the sub-group of patients with available heart contours). For this subgroup (386 patients), we found a HR of 1.02 (95% CI 1.01e1.03, p Z 0.035) to the base of the heart, whereas
Fig. 4. Survival curves plotted for the validation cohort of patients treated with SABR (60 Gy in five fractions), patients grouped into those receiving greater than or less than 6.3 Gy, log-rank p Z 0.016 (68 patients in the low-dose group and 21 patients in the high-dose group), numbers at risk are provided. This is the dose with the same biological effective dose as used in the 20 fraction group (alpha/beta Z 2 Gy).
the mean heart dose provided a HR 1.01 (95% CI 0.99e1.03, p Z 0.32), heart V5 HR 1.00 (95% CI 1.00e1.01, p Z 0.61) and heart V30 HR 1.00 (95% CI 1.00e1.02, p Z 0.35). The strength of our methodology is that no assumptions, no outlines of organs at risk and no dosimetric parameters are required. Instead, a highresolution survival analysis was performed correlating planned dose distributions with patient survival for every voxel. Without the limitations of pre-defining structures, an anatomical region can be identified, generating a hypothesis for future bettertargeted investigations. This permutation testing methodology used in this paper is common in functional magnetic resonance imaging to solve the multiple testing issues [12,13]. It was translated and pioneered for RT in the prostate by Chen et al. [8] and Witte et al. [14], identifying anatomical regions correlated with biochemical failure and applied by van Luijk et al. [15] to locate regions associated with toxicity in head and neck cancer. We now apply this methodology in lung cancer patients and identified, for the first time in the context of high-dose thoracic RT for lung cancer, the base of the heart as a radiation-sensitive region. The identified anatomical region encompasses the aorta, the region of origin of the coronary arteries and the sinoatrial node. A review of the literature by Darby et al. [16] summarises the current state of knowledge for radiation-related heart disease. The studies described in this article are based on populations of patients with breast cancer or lymphoma. For these patients, cardiac toxicity manifests itself as an increase in coronary artery disease, myocardial fibrosis and myocardial infarctions. In patients who receive a mean dose greater than 30 Gy to the heart such events can become apparent within 12e24 months; however, when lower doses are delivered the latency period may exceed a decade. The difference in survival in
112
A. McWilliam et al. / European Journal of Cancer 85 (2017) 106e113
our study is seen much earlier, from 6 months after completion of RT and at very low doses to the base of the heart (8.5 Gy). This finding could be related to lung cancer patients typically being older and presenting with a higher burden of co-morbidities than breast cancer and lymphoma patients [17e19]. Specifically, many lung cancer patients will have pre-existing smoking-related cardiac co-morbidities including damage to the coronary arteries. The univariate and multivariate survival analyses show, expectedly, that larger tumour volume is significantly associated with poorer survival [20]. Larger tumours will also result in higher doses deposited across the mediastinum and the heart. However, our multivariate survival analysis shows that the dose to the identified region remains significant when controlling for tumour volume. Whereas the cardiac dose-volume metrics highlighted in recent studies showed no significance in the univariate or multivariate analyses. In addition, dose-volume approaches lose spatial information, which is not the case with this methodology, where we maintain spatial information allowing predictive anatomical regions to be identified. Interestingly, the mean lung dose was not significantly associated with poor survival in multivariate analyses, although this parameter does predict for lung toxicity [21]. Tumour volume is correlated with mean lung dose and, by including both variables in the multivariate analysis, the mean lung dose has lost its significance. Indeed, on removal of tumour volume from this multivariate analysis the mean lung dose becomes highly significant in predicting patient survival. Systemic anti-cancer therapy is known to be frequently associated with the development of cardiovascular events such as heart failure, myocardial infarction, hypertension, thromboembolism and cardiac rhythm disorders [22]. The risk of such cardiac complications varies according to the type, the combination and the dose intensity of the systemic anti-cancer therapy. Where these display as late effects, we may not see these in our methodology; and, certainly, chemotherapy was not significant in our multivariate analysis. A limitation of our analysis is the lack of data on patient co-morbidities and actual cause of death. The majority of lung cancer patients will have associated cardiac co-morbidities, including chronic cardiac and respiratory illnesses relating to smoking. Co-morbidities were not available for all patients, and, therefore, we could not control for these factors in the analysis. For a sub-set of 372 patients, we were able to access their ACE27 co-morbidity scores. However, we found no statistical significance in univariate analysis for patient survival. The ACE27 scale is based on all comorbidities, and more detailed information, specifically on cardiac co-morbidities, is required. In addition, the accurate evaluation of the cause of death is a major challenge, as death certificates often record the death as
‘lung cancer’ with no investigation of an accurate cause. These observations highlight the requirement for prospective studies with detailed patient data collection and a focus on cardiac history. Cardiac imaging studies, preand post-RT, should also investigate dose-related changes to anatomical structures such as the coronary arteries. In summary, the application of this methodology has removed the limitations of structure-based dosimetric analysis and has identified that the base of the heart is a dose-sensitive region, strongly correlated with patient survival. In our analysis, heart dose-volume-based metrics used in published studies did not show significance, where our approach of including the entire patient highlight a very significant region in the base of the heart. The impact on survival remains after controlling for nodal stage, performance status and tumour volume and is evident from 6 months after completion of RT. We have generated a strong hypothesis that can aid in informing the direction of future clinical studies. Ultimately, this work suggests that the sparing of specific sub-structures in the base of the heart could lead to significant improvements in survival in lung cancer patients. Conflict of interest statement The authors have no conflicts of interest to declare.
Acknowledgement This work was supported by Cancer Research UK via funding to the Cancer Research Manchester Centre [C147/A18083] and [C147/A25254].
References [1] Auperin A, Le Pechoux C, Rolland E, Curran WJ, Furuse K, Fournel P, et al. Meta-analysis of concomitant versus sequential radiochemotherapy in locally advanced non-small-cell lung cancer. J Clin Oncol 2010;28(13):2181e90. [2] Machtay M, Paulus R, Moughan J, Komaki R, Bradley JE, Choy H, et al. Defining local-regional control and its importance in locally advanced non-small cell lung carcinoma. J Thorac Oncol 2012;7(4):716e22. [3] Bradley JD, Paulus R, Komaki R, Masters G, Blumenschein G, Schild S, et al. Standard-dose versus high-dose conformal radiotherapy with concurrent and consolidation carboplatin plus paclitaxel with or without cetuximab for patients with stage IIIA or IIIB non-small-cell lung cancer (RTOG 0617): a randomised, two-by-two factorial phase 3 study. Lancet Oncol 2015;16(2): 187e99. [4] Dess RT, Sun Y, Matuszak MM, Sun G, Soni PD, Bazzi L, et al. Cardiac events after radiation therapy: combined analysis of prospective multicenter trials for locally advanced non-small-cell lung cancer. J Clin Oncol 2017;35(13):1395e402. [5] Emami B, Lyman J, Brown A, Coia L, Goitein M, Munzenrider JE, et al. Tolerance of normal tissue to therapeutic irradiation. Int J Radiat Oncol Biol Phys 1991;21(1):109e22.
A. McWilliam et al. / European Journal of Cancer 85 (2017) 106e113 [6] Gagliardi G, Constine LS, Moiseenko V, Correa C, Pierce LJ, Allen AM, et al. Radiation dose-volume effects in the heart. Int J Radiat Oncol Biol Phys 2011;76(Suppl. 3):S77e85. [7] http://sourceforge.net/projects/niftyreg/. [8] Chen C, Witte M, Heemsbergen W, van Herk M. Multiple comparisons permutation test for image based data mining in radiotherapy. Radiat Oncol 2013;8:293. [9] Lauk S, Ru¨th S, Trott KR. The effects of dose-fractionation on radiation-induced heart disease in rats. Radiother Oncol 1987;8: 363e7. [10] Schultz-Hector S, Sund M, Thames HD. Fractionation sensitivity and repair kinetics of radiation-induced heart failure in the rat. Radiother Oncol 1992;23:33e4. [11] Wang K, Eblan MJ, Deal AM, Lipner M, Zagar TM, Wang Y, et al. Cardiac toxicity after radiotherapy for stage III non-smallcell lung cancer: pooled analysis of dose-escalation trials delivering 70 to 90 Gy. J Clin Oncol 2017;35(12):1387e94. [12] Nichols TE, Holmes AP. Nonparametric permutation tests for functional neuroimaging: a primer with examples. Hum Brain Mapp 2002;15(1):1e25. [13] Winkler AM, Ridgway GR, Webster MA, Smith SM, Nichols TE. Permutation inference for the general linear model. Neuroimage 2014;92:381e97. [14] Witte MG, Heemsbergen WD, Bohoslavsky R, Pos FJ, AlMamgani A, Lebesque JV, et al. Relating dose outside the prostate with freedom from failure in the Dutch trial 68 Gy vs. 78 Gy. Int J Radiat Oncol Biol Phys 2010;77(1):131e8. [15] van Luijk P, Pringle S, Deasy JO, Moiseenko VV, Faber H, Hovan A, et al. Sparing the region of the salivary gland
[16]
[17]
[18]
[19]
[20]
[21]
[22]
113
containing stem cells preserves saliva production after radiotherapy for head and neck cancer. Sci Transl Med 2015;7(305). Darby SC, Cutter DJ, Boerma M, Constine LS, Fajardo LF, Kodama K, et al. Radiation-related heart disease: current knowledge and future prospects. Int J Radiat Oncol Biol Phys 2010;76(3):656e65. Kravchenko J, Berry M, Arbeev K, Kim Lyerly H, Yashin A, Akushevich I. Cardiovascular comorbidities and survival of lung cancer patients: Medicare data based analysis. Lung Cancer 2015; 88(1):85e93. Otake S, Ohtsuka T, Asakura K, Kamiyama I, Kohno M. Impact of comorbidity index on morbidity and survival in non-small cell lung cancer. Asian Cardiovasc Thorac Ann 2016;24(1):30e3. Lachina M, Green A, Jakobsen E. The direct and indirect impact of comorbidity on the survival of patients with non-small cell lung cancer: a combination of survival, staging and resection models with missing measurements in covariates. BMJ Open 2014;4(2). Remen B, van Loon J, van Baardwijk A, Wanders R, Borger J, Dingemans AM, et al. Total gross tumor volume is an independent prognostic factor in patients treated with selective nodal irradiation for stage I to III small cell lung cancer. Int J Radiat Oncol Biol Phys 2013;85(5):1319e24. Kwa S, Lebesque J, Theuws J, Marks LB, Munley MT, Bentel G, et al. Radiation pneumonitis as a function of mean lung dose: an analysis of pooled data of 540 patients. Int J Radiat Oncol Biol Phys 1998;42(1):1e9. Yeh ETH, Tong AT, Lenihan DJ, Yusuf SW, Swafford J, Champion C, et al. Cardiovascular complications of cancer therapy. Circulation 2004;109:3122e31.
Available online at www.sciencedirect.com
ScienceDirect journal homepage: www.ejcancer.com
Original Research
Radiation dose to heart base linked with poorer survival in lung cancer patients Alan McWilliam a,b,*, Jason Kennedy b, Clare Hodgson c, Eliana Vasquez Osorio a, Corinne Faivre-Finn a,b,1, Marcel van Herk a,b,d,1 a Division of Molecular and Clinical Cancer Science, School of Medical Sciences, Faculty of Biology, Medicine and Health, University of Manchester, Manchester Academic Health Science Centre, UK b Department of Radiotherapy Related Research, The Christie NHS Foundation Trust, Manchester, UK c Clinical Outcomes Unit, The Christie NHS Foundation Trust, Manchester, UK d NIHR Manchester Biomedical Research Centre, Central Manchester University Hospitals NHS Foundation Trust, Manchester Academic Health Science Centre, UK
Received 28 April 2017; received in revised form 21 July 2017; accepted 31 July 2017
KEYWORDS Lung cancer; Radiotherapy; Cardiotoxicity
Abstract Background: Advances in radiotherapy (RT) have allowed an increased proportion of lung cancer patients to be treated curatively. High doses delivered to critical thoracic organs can result in excess mortality with tolerance doses poorly defined. This work presents a novel method of identifying anatomical dose-sensitive regions within the thorax. Methods: A high-resolution, normal-tissue dosimetric analysis was performed to identify regions in the heart that correlate with poorer survival. A total of 1101 patients treated with curative-intent RT were selected and all computed tomography imaging and dose distributions were deformed to a reference. Mean dose distributions were created for patients who survived versus those who did not at a set time point. Statistical significance of dose differences was investigated with permutation testing. The dose received by the most statistically significant region of the thorax was collected in all patients and included in a multivariate survival analysis. Results: The permutation testing showed a highly significant region across the base of the heart, where higher doses were associated with worse patient survival (p < 0.001). Cox-regression multivariate analysis showed region dose, tumour volume, performance status and nodal stage were significant factors associated with survival, whereas cardiac mean dose, V5 and V30 showed no significance. Survival curves, controlling for these factors, were plotted with patients receiving doses greater than 8.5 Gy to the identified region showing worse survival (log-rank p < 0.001, hazard ratio 1.2).
* Corresponding author: The Christie NHS Foundation Trust, Radiotherapy Related Research (Dept 58), Wilmslow Road, Manchester, M20 4BK, UK. E-mail address: [email protected] (A. McWilliam). 1 Joint last authors. http://dx.doi.org/10.1016/j.ejca.2017.07.053 0959-8049/ª 2017 Elsevier Ltd. All rights reserved.
A. McWilliam et al. / European Journal of Cancer 85 (2017) 106e113
107
Conclusion: The application of this novel methodology in lung cancer patients identifies the base of the heart as a dose-sensitive region for the first time. ª 2017 Elsevier Ltd. All rights reserved.
1. Introduction Radiotherapy (RT) plays a major role in curative-intent treatment of lung cancer. Advances in radiation technology have enabled curative-intent doses to be delivered to a larger proportion of locally advanced patients while keeping doses to normal tissues within accepted safety limits. This increase in the proportion of patients treated with curative intent is expected to improve local control, which is known to be associated with better survival [1,2]. However, over the last 2 years, evidence relating irradiation of the heart to excess mortality has emerged. Bradley et al. [3] reported the outcomes from RTOG 0617, a multicentre phase III study comparing 60 Gy versus 74 Gy delivered in 2 Gy fractions. They showed that a higher treatment dose was associated with increased mortality and multivariate analysis of survival identified heart doses, volume receiving 5 Gy (V5) and volume receiving 30 Gy (V30), as associated with patient survival. Subsequently, a study showed that higher mean heart dose was significantly associated with higher cardiac event rates [4]. There is now great interest in investigating how dose-to-anatomical sub-structures of the heart links to survival. Despite this emerging evidence, in current practice, heart dose constraints remain poorly defined, generic and date back to articles published over 25 years ago [5,6]. One of the major limitations of previous studies is the evaluation of dose to the whole heart as one structure or divided into a small number of pre-defined sub-structures. There is, therefore, a clear unmet need to define heart sub-structures at risk and heart dose constraints for this group of patients. The analysis of large populationbased cohorts of lung cancer patients treated with curative RT is an attractive strategy to identify dosesensitive heart structures. However, this approach is limited by the fact that heart sub-structures are not delineated routinely as part of the RT planning process. The methodology in this article, applied to lung cancer patients for the first time, does not require any delineation structures. Instead, performs a high-resolution, voxel-by-voxel dosimetric analysis, identifying regions correlated with patient survival. 2. Methods A total of 1163 lung cancer patients treated between 2010 and 2013 at a single academic cancer centre, with
routine curative-intent RT (55 Gy in 20 fractions), with or without induction chemotherapy, were randomly selected for analysis. Institutional approval had been gained to use these patients. Three-dimensional conformal and intensity-modulated radiation therapy (IMRT) plans were both used in this work. Patient images were deformable registered to a reference patient using the Nifty Registration package, (NiftyReg, UCL, UK [7]). To avoid potential effects from sliding tissue between the ribs and the lung, bone was excluded from the registration process by truncating Hounsfield Units at 100. The deformable registration is based on a Bspline parameterisation approach. The planned dose distribution was normalised to the reference by directly applying the derived deformable vector field. A visual validation of the registration was performed to ensure that all patients were successfully normalised into the same spatial reference. This approach allows large numbers of patients to be included in the analysis without the requirement for additional contouring. A sub-set of 386 patients, in whom the heart had been contoured by a clinical oncologist, were used to evaluate the accuracy of deformable registration. The uncertainty was estimated using the standard deviation of the centre of mass coordinates of deformed heart contours. To assess the influence of this uncertainty on the data mining results, dose distributions for each patient were blurred by a Gaussian filter with corresponding width. Results using original dose distributions and blurred distributions were compared. The difference in mean dose distributions for patients who survived versus those who did not survive at a given time point from end of treatment (3e36 months) was calculated, with patients censored for follow-up. To test if the dose difference between the two groups was statistically significant, permutation testing was used, with 1000 permutations performed. The maximum t-value was used over the average dose distribution to test for significance [8]. The test statistic, maximum t-value, is calculated from the difference in mean dose in a voxel between the two groups divided by the standard deviation of the voxels, with the maximum t-value selected. Permutations then generate random samples to determine the distribution of maximum t-values, this tests the null hypothesis that there is no difference between the two groups. This approach indicates areas of the anatomy, where the observed dose difference is related to a statistically significant difference in patient survival. To
108
A. McWilliam et al. / European Journal of Cancer 85 (2017) 106e113
ensure that the permutation testing was not influenced by variability in dose, profiles were plotted through the region of highest statistical significance for the dose difference, standard deviation of dose and the t-value. The area of highest statistical significance was used to define a small region of interest, defined at 90% of the maximum t-value. The mean of the individual doses received to this region was calculated for every patient. Using univariate analysis the parameters (patient and tumour characteristics) significantly associated with patient survival were determined (selected as those with p < 0.05). Factors that remained significant were included in a Cox-regression multivariate survival model. Tumour volume was collected for all patients, as volume will influence dose delivered to any anatomical region. Mean lung dose (total lung volume minus overlapping gross tumour volume) was also calculated, as this is a known predictor of lung toxicity. For comparison with published results the mean dose, the V5 and V30 were collected for the 386 patients with heart contours. Hazard ratios (HR) were calculated for all variables with 95% confidence intervals (CIs). KaplaneMeier curves were plotted as quartiles of dose received to the identified anatomical region, allowing an appropriate dose to be selected with which to group patients. Log-rank tests were performed to show any significance in overall survival. All statistical analysis was performed using the Statistical Package for Social Sciences, version 22 (SPSS, Chicago, IL). Validation was performed on a set of 89 sterotactic ablative radiotherapy (SABR) patients with stage I nonsmall cell lung cancer (NSCLC) treated with 60 Gy in five fractions (100% dose to cover 95% of the planning target volume). Patients were processed using the same methodology and reference patient. The identified region was transferred onto these patients, the dose to this area was collected and KaplaneMeier curves plotted. To calculate the appropriate dose to group the validation cohort, a biological effective dose (BED) calculation was performed. The dose selected to group the patients treated with 55 Gy in 20 fractions was used to calculate a BED (assuming an alpha/beta ratio of 2 Gy [9,10]). This BED was then used to calculate the BED with which to group the patients treated with SABR. 3. Results Visual inspection of the deformable image registration identified 62 registration failures, patients who displayed significant atelectasis or had previous surgery and were removed from subsequent analysis. Patient demographics for the 1101 patients who remained are included in Table 1. No discrimination of the patients was made on these characteristics. The standard deviation in centre of mass coordinates, for the 386 patient heart contours available, was
Table 1 Patient demographics from the 1101 patients selected for analysis. Not all patients had complete sets of records available; the ‘Total in group’ column shows the number of data points available for each variable. Variable
Sub-variable Sub-total
Total in group
Gender
Male Female
1101
Age (median) Tumour size (cm3) (median) Smoking history Current Ex-smoker Life-long non-smoker T-stage T1 T2 T3 T4 N-stage N0 N1 N2 N3 Induction chemotherapy Yes No
593 508 73 (38e95) 34.5 (0.5e528) 153 197 8 159 434 238 169 546 137 257 66 266 835
1101 1101 359
1000
1006
1101
found to be 3.9 mm in the lefteright, 4.9 mm anterioreposterior and 7.4 mm cranialecaudal direction after deformable registration. All time points investigated from 6 months post RT showed a statistically significant correlation between the mean dose distribution and survival (p < 0.001). Results from the original and blurred distributions were almost identical. The most significant dose difference was located across the base of the heart, where excess dose was associated with poorer patient survival. A representative result from the permutation testing using the blurred dose distribution is shown in Fig. 1, where patient survival at 12 months has been set. The region of highest significance (p-value uncorrected for multiple testing) overlaps with the aorta and includes the origin of the coronary arteries and sinoatrial node. The coronal and sagittal views show that the significant area is contained across the base of the heart (Fig. 1A). Profiles of dose difference (between patients who survived versus those who did not at 12 months), standard deviation of dose across all patients and the tstatistic were plotted (Fig. 1B). Profiles were taken through the origin of the left coronary artery. Note that the standard deviation is fairly homogenous over the relevant heart region, therefore the location of significance is defined by the dose difference between dead and living patients. The area of highest significance, defined at 90% of the maximum t-value, was used to define a region of interest. This was located in the base of the heart and limited at the boundary with the lung. The volume of the identified anatomical region was 12 cm3 with a median average dose, across all patients of 16.3 Gy. Dose to this
A. McWilliam et al. / European Journal of Cancer 85 (2017) 106e113
109
Fig. 1. In the top row (A) axial, sagittal and coronal views from the computed tomography scan of the reference patient. The statistical significance from the permutation testing is overlaid (p < 0.001). This shows that excess dose to the base of the heart was highly correlated with poorer patient survival. The lower row (B) shows profiles taken through the left coronary artery (as indicated on the patient anatomy). Dose difference, standard deviation of dose and the t-value from the permutation testing are plotted.
region for all patients was collected, and the Coxregression univariate analysis showed that survival was correlated with dose to this area (p Z 0.04), Table 2. However, the mean dose to the heart, V5 and V30 were not predictive of survival for the sub-group of patients (386 patients with heart contours analysed). These findings remained in the Cox-regression multivariate analysis, where the defined region in the base of the heart provided a HR of 1.25 (95% CI 1.01e1.56, p Z 0.04). This analysis indicates the importance of this region in patient survival. Nodal stage (categorical), performance status (categorical), age (continuous, HR 1.01; 95% CI 1.00e1.02, p Z 0.04) and tumour volume (continuous, HR 1.01; 95% CI 1.01e1.01, p < 0.001) were also strongly correlated in the multivariate analysis. T-stage, gender and mean lung dose showed significance in univariate analysis but did not maintain significance in multivariate analysis. Survival curves were plotted on the quartiles of dose to the identified heart area (1st quartile, 0e8.5 Gy, 2nd Quartile 8.5e16.3 Gy, 3rd quartile 16.3e27.3 Gy and 4th quartile 27.3e49), quartiles defined with equal numbers in each, Fig. 2A. Mean survival times of the quartiles were 28, 24, 21 and 17 months, respectively. The 1st quartile dose, 8.5 Gy, was selected as the dose to group the patients. The KaplaneMeier survival curves included with a clear split between the patient groups was seen, with those receiving a dose greater than
8.5 Gy, showing significantly worse survival, log-rank p < 0.001, Fig. 2B. The survival curves with nodal stage, set as a categorical variable, showed a similar response between those patients receiving a low dose to the identified anatomical area and those receiving a high dose, Fig. 3. Log-rank tests between the patients receiving less or greater than 8.5 Gy to the identified anatomical region were calculated. For N0 patients, a significant split remained (p Z 0.002, 546 patients), as well as for N2 patients (p Z 0.012, 256 patients). There was no significant difference in N1 patients (137 patients, with seven in the low-dose group) and N3 patients (67 patients with four in the low-dose group), possibly due to small patient numbers. Validation was performed with 89 patients treated with SABR, receiving 60 Gy in five fractions, median tumour volume 4 cm3 (0.1e38 cm3). The permutation testing showed the same pattern of response across the base of the heart, although was not significant in this small group of patients. The dose received by each patient to the identified anatomical region was collected. The BED that corresponds to a dose of 8.5 Gy for the cohort of patients treated with 20 fractions is 10.3 Gy. For the cohort of patients treated with SABR, this BED corresponds to a physical dose of 6.3 Gy. KaplaneMeier survival curves were plotted with patients grouped into those receiving greater than or less
110
A. McWilliam et al. / European Journal of Cancer 85 (2017) 106e113
Table 2 Univariate and multivariate survival analysis, hazard ratios with 95% confidence intervals are included. Univariate
Tumour volume (continuous) Dose to defined region (greater versus less median) Heart mean dose (continuous) Heart V5 (continuous) Heart V30 (continuous) Age (continuous) Mean lung dose (continuous) Gender (female versus male) Induction chemotherapy (yes versus no) T-stage (T1 reference) T2 T3 T4 N-stage (N0 reference) N1 N2 N3 Performance status (PS 0 as reference) 1 2 3
Multivariate
HR (95% CI)
P
HR (95% CI)
P
1.005 (1.004e1.006) 1.24 (1.01e1.53) 1.010 (0.99e1.03) 1.00 (0.99e1.01) 1.01 (0.99e1.02) 1.01 (1.00e1.02) 1.06 (1.04e1.08) 1.22 (1.07e1.39) 0.83 (0.68e1.03)
<0.001 0.04 0.32 0.61 0.35 0.89 <0.001 0.003 0.10 <0.001 0.01 <0.001 <0.001 0.01 0.02 0.01 0.01 0.03 0.09 0.06 0.01
1.01 1.25 e e e 1.01 1.01 1.22 0.92
<0.001 0.04 e e e 0.04 0.45 0.24 0.46 0.24 0.27 0.24 0.45 0.01 0.29 0.02 0.02 0.43 0.24 0.03 0.02
1.41 (1.13e1.76) 1.79 (1.41e2.27) 1.98 (1.53e2.55) 0.66 (0.41e1.06) 1.76 (1.08e2.85) 1.86 (0.85e4.07) 1.17 (0.98e1.40) 1.20 (0.99e1.45) 1.52 (1.11e2.07)
(1.01e1.01) (1.01e1.56)
(1.00e1.02) (0.98e1.04) (1.03e1.45) (0.74e1.15)
1.17 (0.89e1.54) 1.20 (0.88e1.64) 1.41 (1.00e1.96) 0.86 (0.66e1.13) 1.33 (1.06e2.24) 1.54 (1.07e2.24) 1.19 (0.89e1.60) 1.33 (0.98e1.82) 1.36 (0.88e2.11)
CI, confidence interval; HR, hazard ratio.
than this dose (Fig. 4). There is a statistically significant difference in survival between the two groups (log-rank p Z 0.016). In a multivariate analysis, the tumour volume, age and the dose to the heart split at 6.5 Gy was included. Tumour volume was not significant (continuous, HR 1.01; 95% CI 0.97e1.06, p Z 0.64), either was age (continuous, HR 1.01; 95% CI 0.97e1.05, p Z 0.74). The dose to the heart was significant with patients receiving less than 6.5 Gy, showing superior survival (HR 2.11; 95% CI 1.10e4.07, p Z 0.025).
4. Discussion This article has generated a strong hypothesis, which we believe can now aid in directing future studies. In an unselected group of 1101 patients, we have identified the base of the heart as a dose-sensitive region, where excess dose results in poorer patient survival (HR 1.25, 95% CI 1.01e1.56, p Z 0.04). For those patients where heart contours were available (386 patients), neither the mean dose, V5 or V30 showed significance in univariate or
Fig. 2. (A) Survival curves plotted split on quartiles of the dose delivered to the identified anatomical area of the heart (8.5 Gy, 16.3 Gy, 27.3 Gy). (B) KaplaneMeier curves with patients grouped on the first quartile dose of 8.5 Gy (275 patients in the low-dose group and 826 in the high-dose group), numbers at risk are provided for each group.
A. McWilliam et al. / European Journal of Cancer 85 (2017) 106e113
111
Fig. 3. Survival curves categorised on nodal stage for patients receiving less than 8.5 Gy (175 patients) (A) and greater than 8.5 Gy (926 patients) (B) to the identified anatomical region (N0 - 546 patients, N1 - 137 patients, N2 - 238 patients, N3 - 67 patients).
multivariate analysis. These measures have previously shown importance in analysis from the RTOG 0617 trial [3] and the pooled analysis of stage III NSCLC trials by Wang et al. [11]. Based on our analysis, the excess dose to the base of the heart seems to be a more important predictor of early death than dose-volume metrics currently identified (comparison within the sub-group of patients with available heart contours). For this subgroup (386 patients), we found a HR of 1.02 (95% CI 1.01e1.03, p Z 0.035) to the base of the heart, whereas
Fig. 4. Survival curves plotted for the validation cohort of patients treated with SABR (60 Gy in five fractions), patients grouped into those receiving greater than or less than 6.3 Gy, log-rank p Z 0.016 (68 patients in the low-dose group and 21 patients in the high-dose group), numbers at risk are provided. This is the dose with the same biological effective dose as used in the 20 fraction group (alpha/beta Z 2 Gy).
the mean heart dose provided a HR 1.01 (95% CI 0.99e1.03, p Z 0.32), heart V5 HR 1.00 (95% CI 1.00e1.01, p Z 0.61) and heart V30 HR 1.00 (95% CI 1.00e1.02, p Z 0.35). The strength of our methodology is that no assumptions, no outlines of organs at risk and no dosimetric parameters are required. Instead, a highresolution survival analysis was performed correlating planned dose distributions with patient survival for every voxel. Without the limitations of pre-defining structures, an anatomical region can be identified, generating a hypothesis for future bettertargeted investigations. This permutation testing methodology used in this paper is common in functional magnetic resonance imaging to solve the multiple testing issues [12,13]. It was translated and pioneered for RT in the prostate by Chen et al. [8] and Witte et al. [14], identifying anatomical regions correlated with biochemical failure and applied by van Luijk et al. [15] to locate regions associated with toxicity in head and neck cancer. We now apply this methodology in lung cancer patients and identified, for the first time in the context of high-dose thoracic RT for lung cancer, the base of the heart as a radiation-sensitive region. The identified anatomical region encompasses the aorta, the region of origin of the coronary arteries and the sinoatrial node. A review of the literature by Darby et al. [16] summarises the current state of knowledge for radiation-related heart disease. The studies described in this article are based on populations of patients with breast cancer or lymphoma. For these patients, cardiac toxicity manifests itself as an increase in coronary artery disease, myocardial fibrosis and myocardial infarctions. In patients who receive a mean dose greater than 30 Gy to the heart such events can become apparent within 12e24 months; however, when lower doses are delivered the latency period may exceed a decade. The difference in survival in
112
A. McWilliam et al. / European Journal of Cancer 85 (2017) 106e113
our study is seen much earlier, from 6 months after completion of RT and at very low doses to the base of the heart (8.5 Gy). This finding could be related to lung cancer patients typically being older and presenting with a higher burden of co-morbidities than breast cancer and lymphoma patients [17e19]. Specifically, many lung cancer patients will have pre-existing smoking-related cardiac co-morbidities including damage to the coronary arteries. The univariate and multivariate survival analyses show, expectedly, that larger tumour volume is significantly associated with poorer survival [20]. Larger tumours will also result in higher doses deposited across the mediastinum and the heart. However, our multivariate survival analysis shows that the dose to the identified region remains significant when controlling for tumour volume. Whereas the cardiac dose-volume metrics highlighted in recent studies showed no significance in the univariate or multivariate analyses. In addition, dose-volume approaches lose spatial information, which is not the case with this methodology, where we maintain spatial information allowing predictive anatomical regions to be identified. Interestingly, the mean lung dose was not significantly associated with poor survival in multivariate analyses, although this parameter does predict for lung toxicity [21]. Tumour volume is correlated with mean lung dose and, by including both variables in the multivariate analysis, the mean lung dose has lost its significance. Indeed, on removal of tumour volume from this multivariate analysis the mean lung dose becomes highly significant in predicting patient survival. Systemic anti-cancer therapy is known to be frequently associated with the development of cardiovascular events such as heart failure, myocardial infarction, hypertension, thromboembolism and cardiac rhythm disorders [22]. The risk of such cardiac complications varies according to the type, the combination and the dose intensity of the systemic anti-cancer therapy. Where these display as late effects, we may not see these in our methodology; and, certainly, chemotherapy was not significant in our multivariate analysis. A limitation of our analysis is the lack of data on patient co-morbidities and actual cause of death. The majority of lung cancer patients will have associated cardiac co-morbidities, including chronic cardiac and respiratory illnesses relating to smoking. Co-morbidities were not available for all patients, and, therefore, we could not control for these factors in the analysis. For a sub-set of 372 patients, we were able to access their ACE27 co-morbidity scores. However, we found no statistical significance in univariate analysis for patient survival. The ACE27 scale is based on all comorbidities, and more detailed information, specifically on cardiac co-morbidities, is required. In addition, the accurate evaluation of the cause of death is a major challenge, as death certificates often record the death as
‘lung cancer’ with no investigation of an accurate cause. These observations highlight the requirement for prospective studies with detailed patient data collection and a focus on cardiac history. Cardiac imaging studies, preand post-RT, should also investigate dose-related changes to anatomical structures such as the coronary arteries. In summary, the application of this methodology has removed the limitations of structure-based dosimetric analysis and has identified that the base of the heart is a dose-sensitive region, strongly correlated with patient survival. In our analysis, heart dose-volume-based metrics used in published studies did not show significance, where our approach of including the entire patient highlight a very significant region in the base of the heart. The impact on survival remains after controlling for nodal stage, performance status and tumour volume and is evident from 6 months after completion of RT. We have generated a strong hypothesis that can aid in informing the direction of future clinical studies. Ultimately, this work suggests that the sparing of specific sub-structures in the base of the heart could lead to significant improvements in survival in lung cancer patients. Conflict of interest statement The authors have no conflicts of interest to declare.
Acknowledgement This work was supported by Cancer Research UK via funding to the Cancer Research Manchester Centre [C147/A18083] and [C147/A25254].
References [1] Auperin A, Le Pechoux C, Rolland E, Curran WJ, Furuse K, Fournel P, et al. Meta-analysis of concomitant versus sequential radiochemotherapy in locally advanced non-small-cell lung cancer. J Clin Oncol 2010;28(13):2181e90. [2] Machtay M, Paulus R, Moughan J, Komaki R, Bradley JE, Choy H, et al. Defining local-regional control and its importance in locally advanced non-small cell lung carcinoma. J Thorac Oncol 2012;7(4):716e22. [3] Bradley JD, Paulus R, Komaki R, Masters G, Blumenschein G, Schild S, et al. Standard-dose versus high-dose conformal radiotherapy with concurrent and consolidation carboplatin plus paclitaxel with or without cetuximab for patients with stage IIIA or IIIB non-small-cell lung cancer (RTOG 0617): a randomised, two-by-two factorial phase 3 study. Lancet Oncol 2015;16(2): 187e99. [4] Dess RT, Sun Y, Matuszak MM, Sun G, Soni PD, Bazzi L, et al. Cardiac events after radiation therapy: combined analysis of prospective multicenter trials for locally advanced non-small-cell lung cancer. J Clin Oncol 2017;35(13):1395e402. [5] Emami B, Lyman J, Brown A, Coia L, Goitein M, Munzenrider JE, et al. Tolerance of normal tissue to therapeutic irradiation. Int J Radiat Oncol Biol Phys 1991;21(1):109e22.
A. McWilliam et al. / European Journal of Cancer 85 (2017) 106e113 [6] Gagliardi G, Constine LS, Moiseenko V, Correa C, Pierce LJ, Allen AM, et al. Radiation dose-volume effects in the heart. Int J Radiat Oncol Biol Phys 2011;76(Suppl. 3):S77e85. [7] http://sourceforge.net/projects/niftyreg/. [8] Chen C, Witte M, Heemsbergen W, van Herk M. Multiple comparisons permutation test for image based data mining in radiotherapy. Radiat Oncol 2013;8:293. [9] Lauk S, Ru¨th S, Trott KR. The effects of dose-fractionation on radiation-induced heart disease in rats. Radiother Oncol 1987;8: 363e7. [10] Schultz-Hector S, Sund M, Thames HD. Fractionation sensitivity and repair kinetics of radiation-induced heart failure in the rat. Radiother Oncol 1992;23:33e4. [11] Wang K, Eblan MJ, Deal AM, Lipner M, Zagar TM, Wang Y, et al. Cardiac toxicity after radiotherapy for stage III non-smallcell lung cancer: pooled analysis of dose-escalation trials delivering 70 to 90 Gy. J Clin Oncol 2017;35(12):1387e94. [12] Nichols TE, Holmes AP. Nonparametric permutation tests for functional neuroimaging: a primer with examples. Hum Brain Mapp 2002;15(1):1e25. [13] Winkler AM, Ridgway GR, Webster MA, Smith SM, Nichols TE. Permutation inference for the general linear model. Neuroimage 2014;92:381e97. [14] Witte MG, Heemsbergen WD, Bohoslavsky R, Pos FJ, AlMamgani A, Lebesque JV, et al. Relating dose outside the prostate with freedom from failure in the Dutch trial 68 Gy vs. 78 Gy. Int J Radiat Oncol Biol Phys 2010;77(1):131e8. [15] van Luijk P, Pringle S, Deasy JO, Moiseenko VV, Faber H, Hovan A, et al. Sparing the region of the salivary gland
[16]
[17]
[18]
[19]
[20]
[21]
[22]
113
containing stem cells preserves saliva production after radiotherapy for head and neck cancer. Sci Transl Med 2015;7(305). Darby SC, Cutter DJ, Boerma M, Constine LS, Fajardo LF, Kodama K, et al. Radiation-related heart disease: current knowledge and future prospects. Int J Radiat Oncol Biol Phys 2010;76(3):656e65. Kravchenko J, Berry M, Arbeev K, Kim Lyerly H, Yashin A, Akushevich I. Cardiovascular comorbidities and survival of lung cancer patients: Medicare data based analysis. Lung Cancer 2015; 88(1):85e93. Otake S, Ohtsuka T, Asakura K, Kamiyama I, Kohno M. Impact of comorbidity index on morbidity and survival in non-small cell lung cancer. Asian Cardiovasc Thorac Ann 2016;24(1):30e3. Lachina M, Green A, Jakobsen E. The direct and indirect impact of comorbidity on the survival of patients with non-small cell lung cancer: a combination of survival, staging and resection models with missing measurements in covariates. BMJ Open 2014;4(2). Remen B, van Loon J, van Baardwijk A, Wanders R, Borger J, Dingemans AM, et al. Total gross tumor volume is an independent prognostic factor in patients treated with selective nodal irradiation for stage I to III small cell lung cancer. Int J Radiat Oncol Biol Phys 2013;85(5):1319e24. Kwa S, Lebesque J, Theuws J, Marks LB, Munley MT, Bentel G, et al. Radiation pneumonitis as a function of mean lung dose: an analysis of pooled data of 540 patients. Int J Radiat Oncol Biol Phys 1998;42(1):1e9. Yeh ETH, Tong AT, Lenihan DJ, Yusuf SW, Swafford J, Champion C, et al. Cardiovascular complications of cancer therapy. Circulation 2004;109:3122e31.