Associations between cardiac irradiation and survival in patients with non-small cell lung cancer: Validation and new discoveries in an independent dataset

Published:October 26, 2021DOI:https://doi.org/10.1016/j.radonc.2021.10.016

      Highlights

      • Associations between heart doses and survival following RT for LA-NSCLC were analyzed.
      • High left atrial wall volumes receiving 64–73 Gy were associated with poorer survival.
      • This result confirms earlier findings in an independent dataset.
      • Aortic valve volumes receiving 29–38 Gy were also negatively associated with survival.
      • Additionally, mean heart dose was negatively associated with survival.

      Abstract

      Introduction

      In ‘IDEAL-6′ patients (N = 78) treated for locally-advanced non-small-cell lung cancer using isotoxically dose-escalated radiotherapy, overall survival (OS) was associated more strongly with VLAwall-64-73-EQD2, the left atrial (LA) wall volume receiving 64–73 Gy equivalent dose in 2 Gy fractions (EQD2), than with whole-heart irradiation measures. Here we test this in an independent cohort ‘OX-RT’ (N = 64) treated routinely.

      Methods

      Using Cox regression analysis we assessed how strongly OS was associated with VLAwall-64-73-EQD2, with whole-heart volumes receiving 64–73 Gy EQD2 or doses above 10-to-70 Gy thresholds, and with principal components of whole-heart dose-distributions. Additionally, we tested associations between OS and volumes of cardiac substructures receiving dose-ranges described by whole-heart principal components significantly associated with OS.

      Results

      In univariable analyses of OX-RT, OS was associated more strongly with VLAwall-64-73-EQD2 than with whole-heart irradiation measures, but more strongly still with VAortV-29-38-EQD2, the volume of the aortic valve region receiving 29–38 Gy EQD2. The best multivariable OS model included LA wall and aortic valve region mean doses, and the aortic valve volume receiving ≥38 Gy EQD2, VAortV-38-EQD2. In a subsidiary analysis of IDEAL-6, the best multivariable model included VLAwall-64-73-EQD2, VAortV-29-38-EQD2, VAortV-38-EQD2 and mean aortic valve dose.

      Conclusion

      We propose reducing heart mean doses to the lowest levels possible while meeting protocol dose-limits for lung, oesophagus, proximal bronchial tree, cord and brachial plexus. This in turn achieves large reductions in VAortV-29-38-EQD2 and VLAwall-64-73-EQD2, and we plan to closely monitor patients with values of these measures still >0% (their median value in OX-RT) following reduction.

      Keywords

      Investigators have recently reported significant negative associations between heart irradiation and overall survival (OS) following radical radiotherapy (RT) for non-small cell lung cancer (NSCLC) [
      • Zhang T.W.
      • Snir J.
      • Boldt R.G.
      • Rodrigues G.B.
      • Louie A.V.
      • Gaede S.
      • et al.
      Is the importance of heart dose overstated in the treatment of non-small cell lung cancer? A systematic review of the literature.
      ,
      • McWilliam A.
      • Kennedy J.
      • Hodgson C.
      • Vasquez Osorio E.
      • Faivre-Finn C.
      • van Herk M.
      Radiation dose to heart base linked with poorer survival in lung cancer patients.
      ,
      • Thor M.
      • Deasy J.O.
      • Hu C.
      • Gore E.
      • Bar-Ad V.
      • Robinson C.
      • et al.
      Modeling the impact of cardio-pulmonary irradiation on overall survival in NRG Oncology trial RTOG 0617.
      ,
      • Guberina M.
      • Eberhardt W.
      • Stuschke M.
      • Gauler T.
      • Heinzelmann F.
      • Cheufou D.
      • et al.
      Heart dose exposure as prognostic marker after radiotherapy for resectable stage IIIA/B non-small-cell lung cancer: secondary analysis of a randomized trial.
      ,
      • Bradley J.D.
      • 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.
      ,
      • Tucker S.L.
      • Liu A.
      • Gomez D.
      • Tang L.L.
      • Allen P.
      • Yang J.
      • et al.
      Impact of heart and lung dose on early survival in patients with non-small cell lung cancer treated with chemoradiation.
      ,
      • McWilliam A.
      • Khalifa J.
      • Vasquez Osorio E.
      • Banfill K.
      • Abravan A.
      • Faivre-Finn C.
      • et al.
      Novel methodology to investigate the impact of radiation dose to heart sub-structures on overall survival.
      ,
      • Vivekanandan S.
      • Landau D.B.
      • Counsell N.
      • Warren D.R.
      • Khwanda A.
      • Rosen S.D.
      • et al.
      The impact of cardiac radiation dosimetry on survival after radiation therapy for non-small cell lung cancer.
      ]. We analysed OS in a cohort of patients, ‘IDEAL-6′, treated in the IDEAL-CRT phase 1/2 trial of isotoxically dose-escalated RT for locally-advanced NSCLC given in 30 fractions over 6 weeks concurrent with chemotherapy [
      • Vivekanandan S.
      • Landau D.B.
      • Counsell N.
      • Warren D.R.
      • Khwanda A.
      • Rosen S.D.
      • et al.
      The impact of cardiac radiation dosimetry on survival after radiation therapy for non-small cell lung cancer.
      ,
      • Landau D.B.
      • Hughes L.
      • Baker A.
      • Bates A.T.
      • Bayne M.C.
      • Counsell N.
      • et al.
      IDEAL-CRT: a phase 1/2 trial of isotoxic dose-escalated radiation therapy and concurrent chemotherapy in patients with stage II/III non-small cell lung cancer.
      ]. OS was significantly associated with one principal component (PC) of patients’ heart dose-distributions, which described fractional heart volumes receiving equivalent doses in 2 Gy fractions (EQD2) of 64–73 Gy (α/β = 3 Gy [
      • Schultz-Hector S.
      • Sund M.
      • Thames H.D.
      Fractionation response and repair kinetics of radiation-induced heart failure in the rat.
      ]), delivered largely to the left atrial (LA) wall. The best multivariable (MV) model of survival identified for IDEAL-6 included the fractional LA wall volume receiving 64–73 Gy EQD2 (VLAwall-64-73-EQD2) in preference to the corresponding whole-heart volume (VHeart-64-73-EQD2) [
      • Vivekanandan S.
      • Landau D.B.
      • Counsell N.
      • Warren D.R.
      • Khwanda A.
      • Rosen S.D.
      • et al.
      The impact of cardiac radiation dosimetry on survival after radiation therapy for non-small cell lung cancer.
      ].
      Here, we report a post-hoc analysis of survival in an independent cohort of locally-advanced NSCLC patients, ‘OX-RT’, treated with curative intent using RT alone or chemo-RT given in 2 Gy fractions. We assess the association between OS and VLAwall-64-73-EQD2, and compare it with associations between OS and other cardiac irradiation measures. MV models of OS in OX-RT are built from dose-volume measures and clinical factors, and judged according to the Akaike Information Criterion (AIC) and Harrell’s C-statistic.

      Materials and methods

       Patient data

      The independent cohort, OX-RT, was drawn from Oxford Cancer Centre. Following institutional approval, medical records were retrieved for 80 patients with locally-advanced NSCLC treated consecutively during 2010–2014. Of these, 64 had evaluable datasets with accessible electronic treatment plans including dose-distribution data and no re-planning during RT. Patient and treatment characteristics were collated for this cohort, along with time to last follow-up or death.
      RT was delivered using 3D conformal or volumetric modulated arc therapy (VMAT) techniques, as monotherapy or with sequential or concurrent chemotherapy comprising 3–4 or 2 cycles of platinum doublet respectively. For most OX-RT patients the prescribed dose was 66 Gy in 33 daily fractions over 6.5 weeks, but eight received doses ≤12% lower due to toxicity. Treatment characteristics are summarized in Table 1 for OX-RT and the IDEAL-6 cohort originally studied.
      Table 1Characteristics of OX-RT and IDEAL-6 patients and their treatments.
      CharacteristicOX-RT

      (No. = 64)
      IDEAL-6

      (No. = 78)
      p-value
      Age (years)
       median (range)71 (44–89)66 (43–84)0.002
      Gender (No.)0.10
       Female25 (39.1%)20 (25.6%)
       Male39 (60.9%)58 (74.4%)
      WHO PS (No.)0.03
       017 (26.6%)32 (41.0%)
       134 (53.1%)46 (59.0%)
       25 (7.8%)0
       31 (1.6%)0
       Missing7 (10.9%)
      Tumour stage (No.)
       T13 (4.7%)10 (12.8%)<0.001
       T217 (26.6%)20 (25.6%)
       T340 (62.5%)26 (33.3%)
       T40 (0%)22 (28.2%)
       Missing4 (6.3%)0
      Nodal status (No.)<0.001
       N0 or 129 (45.3%)13 (16.7%)
       N2 or 332 (50.0%)65 (83.3%)
       Missing3 (4.7%)
      Histology (No.)0.44
       Squamous36 (56.3%)42 (53.8%)
       Non-squamous24 (37.5%)36 (46.2%)
       Missing4 (6.3%)
      PTV (cm3)
       median (range)319 (82–1120)401 (139–1262)0.004
      4D-CT used for planning
       (No.)56 (87.5%)34 (43.6%)<0.001
      RT technique (No.)<0.001
       3D conformal48 (75.0%)75 (96.2%)
       VMAT16 (25.0%)3 (3.8%)
      OX-RT or IDEAL-6 prescribed dose (No.)
       66 Gy in 33# or 71.1–73 Gy in 30#56 (87.5%)20 (25.6%)
       64 Gy in 32# or 69.1–71 Gy in 30#3 (4.7%)11 (14.1%)
       62 Gy in 31# or 67.1–69 Gy in 30#2 (3.1%)10 (12.8%)
       60 Gy in 30# or 65.1–67 Gy in 30#2 (3.1%)15 (19.2%)
       58 Gy in 29# or 63–65 Gy in 30#1 (1.6%)22 (28.2%)
      Prescribed tumour EQD2* (Gy)
       median (range)66 (58–66)69.0 (63.5–75.6)<0.001
      Heart mean dose (Gy)
       median (range)7.6 (0.5–32.2)10.3 (1.1–32.2)0.19
      LA wall mean dose (Gy)
       median (range)12.5 (0.5–56.2)20.4 (1.3–63.0)0.004
      VLAwall-64-73-EQD2 (%)
       median (range)0.0 (0–25.5)1.9 (0–76.1)0.003
      Chemotherapy (No.)<0.001
       Concurrent17 (26.5%)78 (100%)
       Sequential16 (25.0%)0
       No chemotherapy29 (45.3%)0
       Missing2 (3.1%)0
      *EQD2s calculated using α/β = 10 Gy, no time-factor.
      Abbreviations: CT = computed tomography; EQD2 = equivalent dose in 2 Gy fractions; LA = left atrium; PTV = planning target volume; VLAwall-64-73-EQD2 = fraction of LA wall receiving 64–73 Gy EQD; VMAT = volumetric modulated arc therapy; WHO PS = World Health Organization performance status.

       Statistics

      Differences in patient and treatment factors between the cohorts were assessed using the Mann-Whitney test for continuous data, Fisher’s exact test for binary data, and the chi-square test for data with >2 categories. Reported confidence intervals (CIs) and significance-levels are 2-sided.
      OS was measured from treatment commencement, censored at last follow-up, and estimated using the Kaplan-Meier method. Significances of differences between survival curves were assessed using the log-rank test. MV models of OS were constructed from patient and treatment factors with p < 0.30 on univariable (UV) analysis, using bi-directional variable elimination to find the best models with the lowest AIC scores. MV model performance was measured using Harrell’s C-statistic [
      • Harrell F.E.
      • Lee K.L.
      • Mark D.B.
      Multivariable prognostic models: issues in developing models, evaluating assumptions and adeqaucy, an measuring and reducing errors.
      ,
      • Pencina M.J.
      • D'Agostino R.B.
      Overall C as a measure of discrimination in survival analysis: model specific population value and confidence interval estimation.
      ] which describes the fraction of all pairs of evaluable patients in which observed and modelled survivals are both shorter for the same patient. Where necessary, the false discovery rate after multiple hypothesis testing was limited to 10% via the Benjamini-Hochberg step-up procedure.
      For some OX-RT patients data was incomplete. These patients were omitted from UV analyses of the factors concerned, which were not carried forward to MV analysis since their associations with OS were insufficiently significant in the UV analyses.

       Validation in OX-RT of association between OS and VLAwall-64-73-EQD2

      Heart and left atrium were segmented on CT scans using a validated atlas [
      • Feng M.
      • Moran J.M.
      • Koelling T.
      • Chughtai A.
      • Chan J.L.
      • Freedman L.
      • et al.
      Development and validation of a heart atlas to study cardiac exposure to radiation following treatment for breast cancer.
      ]. LA wall was defined as the region lying ≤5 mm within the LA contour [
      • Vivekanandan S.
      • Landau D.B.
      • Counsell N.
      • Warren D.R.
      • Khwanda A.
      • Rosen S.D.
      • et al.
      The impact of cardiac radiation dosimetry on survival after radiation therapy for non-small cell lung cancer.
      ]. RT plans were imported into the Computational Environment for Radiotherapy Research (CERR) software, and dose-volume histograms (DVHs) were generated and exported to SPSS version 25 (IBM Corp, Armonk, NY) and R 4 (R Foundation, Vienna, Austria) for analysis.
      In the OX-RT validation cohort we determined how strongly OS was associated with VLAwall-64-73-EQD2 according to UV Cox proportional hazards regression. To further describe the association, OX-RT was dichotomized into groups with VLAwall-64-73-EQD2 values ≤ or > the median, plotting Kaplan-Meier OS curves for both groups.
      Additional UV analyses of OX-RT were carried out to determine strengths of associations of OS with VHeart-64-73-EQD2 and VHeart-10, 20, …, 70, the whole-heart fractional volumes receiving 64–73 Gy EQD2 or physical doses exceeding thresholds of 10 to 70 Gy rising in 10 Gy increments, and with heart and LA wall mean physical doses. We hypothesized that OS would not be associated as strongly with the whole-heart irradiation measures as with VLAwall-64-73-EQD2, following the pattern observed in IDEAL-6 [
      • Vivekanandan S.
      • Landau D.B.
      • Counsell N.
      • Warren D.R.
      • Khwanda A.
      • Rosen S.D.
      • et al.
      The impact of cardiac radiation dosimetry on survival after radiation therapy for non-small cell lung cancer.
      ].
      MV Cox regression analysis was performed to further characterize associations in OX-RT between OS and these dosimetric measures and patient and treatment factors potentially related to survival.

       Additional discovery work in OX-RT

       Whole-heart PC analysis and dose-localization

      PCs of whole-heart dose-volume histograms (DVHs) were obtained using varimax rotation to simplify their structure [
      • Vivekanandan S.
      • Landau D.B.
      • Counsell N.
      • Warren D.R.
      • Khwanda A.
      • Rosen S.D.
      • et al.
      The impact of cardiac radiation dosimetry on survival after radiation therapy for non-small cell lung cancer.
      ]. DVHs were approximated using linear combinations of ten PCs which accounted for >95% of the DVH variance. UV Cox regression was performed to determine associations between OS and patient-specific coefficients of PCs in the combinations approximating whole-heart DVHs.
      Dose-ranges described by PCs significantly associated with OS were identified from peaks in PC variable-loading plots, and heart substructures irradiated to these dose-levels were found using an approach described previously [
      • Vivekanandan S.
      • Landau D.B.
      • Counsell N.
      • Warren D.R.
      • Khwanda A.
      • Rosen S.D.
      • et al.
      The impact of cardiac radiation dosimetry on survival after radiation therapy for non-small cell lung cancer.
      ]. A single heart with typical volume and shape was selected as a reference geometry, and heart dose-distributions of all OX-RT patients were mapped to it via affine transformations derived from heart and left atrium outlines. Then 2D axial, coronal and sagittal projections through the heart were constructed, in which each pixel described the percentage of patients for whom the associated projection line ran through heart voxels irradiated to doses within the range identified. Having localized dose-ranges to specific heart regions, substructures within the regions were delineated on each patient’s CT scan using a validated atlas [
      • Feng M.
      • Moran J.M.
      • Koelling T.
      • Chughtai A.
      • Chan J.L.
      • Freedman L.
      • et al.
      Development and validation of a heart atlas to study cardiac exposure to radiation following treatment for breast cancer.
      ] and DVHs were calculated for them.

       Associations between OS and additional cardiac substructure dose-volume measures

      For substructures most commonly irradiated to the dose-ranges described by whole-heart PCs significantly associated with OS, we performed UV Cox regression of OS versus substructure volumes receiving these doses. Associations of OS with substructure mean doses and volumes receiving higher doses were also assessed. MV analysis was performed to determine the best model of OS in OX-RT, according to the AIC, that could be built from these measures and factors in the earlier MV model.

       Back-validating new discoveries in IDEAL-6

      For new substructures found to have dose-volume measures significantly associated with OS in MV analysis of OX-RT, we carried out additional subsidiary analyses of the original IDEAL-6 dataset to determine whether substructure irradiation was associated with OS in that cohort too.

      Results

      Characteristics of OX-RT and IDEAL-6 patients and treatments are compared in Table 1. At the cut-off point for our databases, median follow-up after commencing RT was 38 months for OX-RT patients and 25 months for IDEAL-6. Median OS was 28 months (95% CI, 21.9–34.1 months) for OX-RT and 39 months (95% CI not yet determinable) for IDEAL-6. Of 53 OX-RT patients for whom the relevant data was available, 17 (32%) had baseline cardiac comorbidity and 51 (96%) were ex/current smokers. For IDEAL-6 patients this data was not collected.
      In UV analyses of OX-RT, OS was significantly associated with VLAwall-64-73-EQD2 (HR, 1.08; 95% CI, 1.02–1.13; p = 0.006). Fig. 1(a) shows OS curves for OX-RT patients dichotomized by VLAwall-64-73-EQD2 equal to or >0%, the median value. The curves differed significantly with an HR of 2.46 (95% CI, 1.22–4.94; p = 0.009).
      Figure thumbnail gr1
      Fig. 1Kaplan-Meier survival curves for OX-RT patients dichotomized by: (a) fraction of left atrial wall receiving 64–73 Gy EQD2 (VLAwall-64-73-EQD2) equal to (light) or > (bold) the 0% median OX-RT value (log-rank p = 0.009); (b) mean dose to the wall of the left atrium (MD LA wall) ≤ (light) or > (bold) the 12.5 Gy median OX-RT value (log-rank p = 0.007).
      OS was associated more strongly with VLAwall-64-73-EQD2 than with VHeart-64-73-EQD2, VHeart-10, 20, …, 70, mean heart dose, patient characteristics or treatment factors (Table 2). However, OS was associated more strongly still with LA wall mean physical dose (HR, 1.05; 95% CI, 1.02–1.08, p = 2 × 10−4), and this association and that with VLAwall-64-73-EQD2 remained positive discoveries after allowing for multiple hypothesis testing. Fig. 1(b) shows OS curves for OX-RT patients dichotomized by LA wall mean physical dose ≤ or > the median of 12.5 Gy. Again, the two curves differed substantially (HR, 2.39; 95% CI, 1.25–4.58; p = 0.007).
      Table 2Associations between overall survival in the OX-RT cohort and LA wall and whole-heart dose-volume measures, and patient and treatment factors potentially related to survival. Unadjusted (univariable) results are shown together with the best multivariable model built from these factors.
      Unadjusted analysisBest multivariable model

      C = 0.71, AIC = 265.2*
      Covariatep-valueHR (95% CI)p-valueHR (95% CI)
      VLAwall-64-73-EQD2 (%)0.0061.077 (1.022–1.134)
      Mean LA wall dose (Gy)2 × 10−41.049 (1.023–1.075)2 × 10−41.094 (1.043–1.148)
      VHeart-64-73-EQD2 (%)0.271.048 (0.965–1.139)
      VHeart-10 (%)0.041.012 (1.001–1.024)
      VHeart-20 (%)0.021.022 (1.003–1.040)0.060.958 (0.916, 1.002)
      VHeart-30 (%)0.021.028 (1.004–1.052)
      VHeart-40 (%)0.031.034 (1.003–1.066)
      VHeart-50 (%)0.191.027 (0.987–1.068)
      VHeart-60 (%)0.261.030 (0.978–1.085)
      VHeart-70 (%)0.108.326 (0.665–104.2)
      Mean heart dose (Gy)0.021.046 (1.007–1.087)
      Prescribed tumour EQD2 (Gy)0.180.905 (0.781–1.048)0.020.834 (0.718, 0.974)
      PTV (cm3)0.171.001 (1.000–1.002)0.031.002 (1.000, 1.003)
      Technique (3D conf vs VMAT)0.141.627 (0.818–4.081)
      RT alone vs chemo-RT0.760.905 (0.475–1.726)
      Age (years)0.030.965 (0.935–0.996)
      Gender (male vs female)0.261.457 (0.754–2.818)
      WHO PS 0 or 1 vs 2 or 30.950.971 (0.342–2.756)
      Nodal status (N0 or 1 vs 2 or 3)0.911.037 (0.543–1.983)
      Histology (non-squam vs squam)0.601.193 (0.616–2.309)
      Baseline cardiac comorbidity (1 present, 0 absent)0.570.805 (0.380–1.704)
      Smoker (1 at any time, 0 never)0.482.056 (0.279–15.17)
      Pack-year history0.841.001 (0.990, 1.013)
      Constructed from factors with p < 0.30 in univariable analyses, using bi-directional variable elimination to find the best multivariable model with the lowest AIC score.
      *AICs for univariable models based on mean LA wall dose, VHeart-20, mean heart dose or PTV alone were 267.4, 275.6, 278.8 and 278.8 respectively.
      For binary factors, an HR > 1 implies the risk of death is greater for the value listed first.
      Abbreviations: AIC = Akaike information criterion, C = Harrell’s C-statistic, HR = hazard ratio, LA = left atrium, PS = performance status, PTV = planning target volume, VStructure-X-Y-EQD2 = fraction of structure receiving X-Y Gy EQD2, VStructure-Z = fraction of structure receiving > Z Gy physical dose.
      The best MV survival model built for OX-RT patients from all these factors comprised LA wall mean physical dose, VHeart-20, prescribed dose and PTV size (Table 2). This model performed better than UV models based on LA wall mean physical dose or VLAwall-65-71-EQD2 alone, with a lower AIC score and a higher C-statistic.
      In OX-RT only whole-heart PC5 was significantly associated with OS (HR, 1.46; 95% CI, 1.11–1.92; p = 0.0074) (Supplementary Table 1), an association that remained a positive discovery after allowing for multiple hypothesis testing. PC5 had a prominent peak at 29–38 Gy EQD2 (Supplementary Fig. 1), most commonly delivered to a region around the aortic valve and left main coronary artery (Supplementary Fig. 2).
      Table 3 lists associations in OX-RT between OS and irradiation of the aortic valve and left main coronary artery volumes expanded by 5 mm to allow for cardiac motion [

      [14] Ganem GCA, Coelho RC and de Godoy CMG. (2020) Animation of atrial and ventricular external walls of a virtual 3D heart based on echocardiogram images. In: González Díaz C et al. (eds) VIII Latin American Conference on Biomedical Engineering and XLII National Conference on Biomedical Engineering. CLAIB 2019. IFMBE Proceedings, vol 75. Springer, Cham. https://doi-org.liverpool.idm.oclc.org/10.1007/978-3-030-30648-9_90.

      ], and with LA wall irradiation. In UV analyses OS was significantly associated with the aortic valve volume receiving 29–38 Gy EQD2 (VAortV-29-38-EQD2), the left main coronary artery volume receiving ≥38 Gy EQD2 (VLMCA-38-EQD2), and the mean doses in both regions. The association between OS and VAortV-29-38-EQD2 (HR, 1.07; 95% CI, 1.04–1.11; p = 7 × 10−5) was stronger than associations between OS and VLAwall-64-73-EQD2 or mean LA wall physical dose, and remained a positive discovery allowing for multiple hypothesis testing. For OX-RT patients dichotomized by whether VAortV-29-38-EQD2 was greater than the median value of 0%, OS curves differed significantly (HR, 2.39; 95% CI, 1.25–4.58; p = 0.006; Fig. 2).
      Table 3Associations between overall survival in the OX-RT cohort and cardiac substructure dose-volume measures. Unadjusted (univariable) results are shown together with the best multivariable model built from these factors and those included in the best MV model from Table 2. Best fits to bootstrap resampled data are also summarised.
      Unadjusted analysisBest multivariable model

      (C = 0.74, AIC = 262.5*)
      Bootstrap models
      Covariatep-valueHR (95% CI)p-valueHR (95% CI)Inclusion** (%)
      Substructure measures
       VAortV-29-38-EQD2 (%)7 × 10−51.069 (1.035–1.105)30.9
       VAortV-38-EQD2 (%)0.161.013 (0.995–1.030)0.020.956 (0.921–0.991)29.2
       Mean aortic valve dose (Gy)0.0071.036 (1.015–1.057)0.011.100 (1.019–1.186)33.8
       VLMCA-29-38-EQD2 (%)0.111.019 (0.996–1.042)28.6
       VLMCA-38-EQD2 (%)0.021.012 (1.002–1.024)34.4
       Mean LMCA dose (Gy)0.0041.027 (1.009–1.046)35.4
       VLAwall-64-73-EQD2 (%)0.0061.077 (1.022–1.134)29.9
       VLAwall-73-EQD2 (%)0.321.633 (0.628–4.248)32.0
       Mean LA wall dose (Gy)2 × 10−41.049 (1.023–1.075)0.041.059 (1.002–1.120)31.1
      Other factors
       VHeart-20 (%)0.021.022 (1.003–1.040)0.020.936 (0.887–0.989)29.8
       PTV (cm3)0.171.001 (1.000–1.002)0.0051.002 (1.001–1.004)25.8
       Prescribed tumour EQD2 (Gy)0.180.905 (0.781–1.048)0.0060.794 (0.674–0.935)23.0
      Constructed from factors with p < 0.30 in univariable analyses, using bi-directional variable elimination to find the best multivariable model with the lowest AIC score.
      *AICs for univariable models based on VAortV-29-38-EQD2, mean aortic valve dose, mean LA wall dose, VHeart-20, PTV or prescribed tumour dose alone were 268.4, 271.0, 267.4, 275.6, 278.8 and 278.8 respectively.
      **Percentage of the best models of survival in each of 1000 bootstraps of the OX-RT dataset in which a covariate is included.
      Abbreviations: AIC = Akaike information criterion, AortV = aortic valve region, C = Harrell’s C-statistic, HR = hazard ratio, LA = left atrium, LMCA = left main coronary artery, PTV = planning target volume, VStructure-X-Y-EQD2 = fraction of structure receiving X-Y Gy EQD2, VStructure-Y-EQD2 = fraction of structure receiving > Y Gy EQD2, VStructure-Z = fraction of structure receiving > Z Gy physical dose.
      Figure thumbnail gr2
      Fig. 2Kaplan-Meier survival curves for OX-RT patients dichotomized by: (a) fraction of the aortic valve region receiving 29–38 Gy EQD2 (VAortV-29-38-EQD2) equal to (light) or > (bold) the 0% median OX-RT value (log-rank p = 0.006); (b) mean heart dose (MHD) ≤ (light) or > (bold) the 7.6 Gy median OX-RT value (log-rank p = 0.045). Also shown are plots of correlations between: (c) LA wall mean dose and the aortic valve region mean dose; and (d) aortic volume receiving >38 Gy and aortic valve region mean dose.
      The best MV model built from all the substructure dosimetric indices together with factors included in the MV model of Table 2 comprised the volume of the aortic valve region receiving ≥38 Gy EQD2 (VAortV-38-EQD2), VHeart-20, mean physical doses in the aortic valve region and LA wall, PTV size and prescribed dose (Table 3). This model had a lower AIC score than the earlier MV model, and at 0.74 its Harrell’s C-statistic was good.
      Given these results we re-investigated survival in the original IDEAL-6 cohort, finding that VAortV-29-38-EQD2 was significantly associated with OS in UV analyses (Table 4). We also found the best MV model of OS in IDEAL-6 that could be built from several LA wall and aortic valve dose-volume measures and the factors considered in our original analysis of this cohort [
      • Vivekanandan S.
      • Landau D.B.
      • Counsell N.
      • Warren D.R.
      • Khwanda A.
      • Rosen S.D.
      • et al.
      The impact of cardiac radiation dosimetry on survival after radiation therapy for non-small cell lung cancer.
      ], excluding ‘any ECG change’ for which corresponding OX-RT data were unavailable. This model comprised VLAwall-64-73-EQD2, VAortV-29-38-EQD2, VAortV-38-EQD2, mean aortic valve dose and PTV size (Table 4), and had better AIC and C-index values than a model comprising only VLAwall-63-73-EQD2 and PTV size, the two factors included alongside ‘any ECG change’ in the best model in our original analysis of IDEAL-6 [
      • Vivekanandan S.
      • Landau D.B.
      • Counsell N.
      • Warren D.R.
      • Khwanda A.
      • Rosen S.D.
      • et al.
      The impact of cardiac radiation dosimetry on survival after radiation therapy for non-small cell lung cancer.
      ].
      Table 4Associations between OS in the IDEAL-6 cohort and cardiac substructure dose-volume measures. Unadjusted (univariable) results are shown together with the best multivariable model built from these factors and those considered in our original study of IDEAL-6 excluding ‘any ECG change’.
      Unadjusted analysisBest multivariable model

      C = 0.70, AIC = 193.0*
      Covariatep-valueHR (95% CI)p-valueHR (95% CI)
      Substructure measures
       VAortV-29-38-EQD2 (%)0.041.043 (1.002–1.086)0.0021.112 (1.040–1.188)
       VAortV-38-EQD2 (%)0.051.017 (1.000–1.034)0.041.045 (1.002–1.091)
       Mean aortic valve dose (Gy)0.271.017 (0.987–1.047)0.0090.895 (0.824–0.973)
       VLAwall-64-73-EQD2 (%)0.011.035 (1.008–1.063)0.051.040 (1.001–1.081)
       VLAwall-73-EQD2 (%)0.780.980 (0.852–1.127)
       Mean LA wall dose (Gy)0.261.016 (0.989–1.045)
      Other factors in best MV model
       PTV (cm3)0.041.002 (1.000–1.003)0.021.002 (1.001–1.004)
      Constructed from factors with p < 0.30 in univariable analyses, using bi-directional variable elimination to find the best multivariable model with the lowest AIC score.
      *AICs for univariable models based on VAortV-29-38-EQD2, VAortV-38-EQD2, mean aortic valve dose, VLAwall-64-73-EQD2 or PTV were 198.5, 199.0, 200.8, 197.4 and 198.2 respectively.
      Abbreviations: AIC = Akaike information criterion, AortV = aortic valve, C = Harrell’s C-statistic, HR = hazard ratio, LA = left atrium, PTV = planning target volume, VStructure-X-Y-EQD2 = fraction of structure receiving X-Y Gy EQD2, VStructure-Y-EQD2 = fraction of structure receiving > Y Gy.

      Discussion

      In the OX-RT validation cohort, OS was associated more strongly with VLAwall-64-73-EQD2 than with any whole-heart irradiation measure investigated including VHeart-64-73-EQD2, as we had previously found in the original IDEAL-6 patient group [
      • Vivekanandan S.
      • Landau D.B.
      • Counsell N.
      • Warren D.R.
      • Khwanda A.
      • Rosen S.D.
      • et al.
      The impact of cardiac radiation dosimetry on survival after radiation therapy for non-small cell lung cancer.
      ]. The validation is encouraging, since patient- and treatment-related factors differed significantly between the routinely-treated single-centre OX-RT and dose-escalated multi-centre IDEAL-6 cohorts (Table 2).
      In IDEAL-6, OS was associated with one PC of patients’ whole-heart dose-distributions, which described 64–73 Gy EQD2s most often delivered to the LA wall [
      • Vivekanandan S.
      • Landau D.B.
      • Counsell N.
      • Warren D.R.
      • Khwanda A.
      • Rosen S.D.
      • et al.
      The impact of cardiac radiation dosimetry on survival after radiation therapy for non-small cell lung cancer.
      ]. In OX-RT, OS was also significantly associated with one whole-heart PC, which had a small peak at 67–72 Gy EQD2 (Supplementary Fig. 1) but a larger peak describing 29–38 Gy EQD2s typically delivered to the region around the aortic valve and left main coronary artery. Correspondingly, OS was associated with VAortV-29-38-EQD2 more significantly than with VLAwall-64-73-EQD2 in UV analyses of OX-RT, but less significantly in UV analyses of IDEAL-6. This may reflect dosimetric differences between the two cohorts, since in OX-RT VAortV-29-38-EQD2 took a wider range of values than VLAwall-64-73-EQD2, 0–76% versus 0–45%, but in IDEAL-6 it took a narrower range, 0–26% versus 0–46%.
      The best MV model of OS in OX-RT included VAortV-38-EQD2, VHeart-20 and mean doses delivered to the aortic valve region and LA wall. These measures were inter-correlated (Pearson r2 values of 0.250–0.781, Fig. 2) but their variance-inflation factors were <10 (3.21–9.41) and each retained significance in the best model [
      • O’brien R.M.
      A caution regarding rules of thumb for variance inflation factors.
      ] which was selected using the AIC to avoid over-fitting. HRs were >1 for the mean dose factors but < 1 for VAortV-38-EQD2 and VHeart-20, suggesting the mean dose terms alone may over-penalize volumes receiving high doses. Notably, the model included terms describing irradiation of the LA wall and aortic valve, as did the best model of OS in IDEAL-6. In 1000 bootstrap resamples [
      • Abram S.V.
      • Helwig N.E.
      • Moodie C.A.
      • DeYoung C.G.
      • MacDonald A.W.
      • Waller H.G.
      Bootstrap enhanced penalized regression for variable selection with neuroimaging data.
      ] of OX-RT, the best MV survival models included measures of irradiation of the LA wall, aortic valve and left main coronary artery with roughly equal frequencies and with an average of 2.9 measures per model, similar to the 3 in the fit to the original data (Table 3).
      Our results parallel studies that found OS was associated with irradiation of the base-of-heart [
      • McWilliam A.
      • Kennedy J.
      • Hodgson C.
      • Vasquez Osorio E.
      • Faivre-Finn C.
      • van Herk M.
      Radiation dose to heart base linked with poorer survival in lung cancer patients.
      ] and left atrium and superior vena cava [
      • Stam Barbara
      • Peulen Heike
      • Guckenberger Matthias
      • Mantel Frederick
      • Hope Andrew
      • Werner-Wasik Maria
      • et al.
      Dose to heart substructures is associated with non-cancer death after SBRT in stage I-II NSCLC patients.
      ], although in those studies the measures identified as most strongly associated with OS were physical dose >8.5 Gy to the base-of-heart in regularly fractionated patients [
      • McWilliam A.
      • Kennedy J.
      • Hodgson C.
      • Vasquez Osorio E.
      • Faivre-Finn C.
      • van Herk M.
      Radiation dose to heart base linked with poorer survival in lung cancer patients.
      ], and the maximum left atrium physical dose in SABR patients (6.5 and 77.6 Gy median and maximum values in SABR patients) [
      • Stam Barbara
      • Peulen Heike
      • Guckenberger Matthias
      • Mantel Frederick
      • Hope Andrew
      • Werner-Wasik Maria
      • et al.
      Dose to heart substructures is associated with non-cancer death after SBRT in stage I-II NSCLC patients.
      ]. In other studies, however, MV models have been built from combinations of atrial, pericardial, right-sided cardiac substructure, ventricular and lung irradiation measures [
      • Thor M.
      • Deasy J.O.
      • Hu C.
      • Gore E.
      • Bar-Ad V.
      • Robinson C.
      • et al.
      Modeling the impact of cardio-pulmonary irradiation on overall survival in NRG Oncology trial RTOG 0617.
      ,
      • McWilliam A.
      • Khalifa J.
      • Vasquez Osorio E.
      • Banfill K.
      • Abravan A.
      • Faivre-Finn C.
      • et al.
      Novel methodology to investigate the impact of radiation dose to heart sub-structures on overall survival.
      ].
      The prominence of base-of-heart structures in several analyses concurs with a pre-clinical study in which heart failure followed more rapidly after whole-heart than heart-minus-atria irradiation [
      • Kitahara T.
      • Liu K.
      • Solanki K.
      • Trott K.R.
      Functional and morphological damage after local heart irradiation and/or adriamycin in Wistar rats.
      ], suggesting that survival might be reduced more by base-of-heart toxicity. Clinically, in LA-NSCLC patients base-of-heart doses might also be more damaging because they are relatively high [
      • Turtle Louise
      • Bhalla Neeraj
      • Willett Andrew
      • Biggar Robert
      • Leadbetter Jonathan
      • Georgiou Georgios
      • et al.
      Cardiac-sparing radiotherapy for locally advanced non-small cell lung cancer.
      ]: in IDEAL-6, mean EQD2s in the left and right atrial walls were 19.7 and 9.0 Gy compared to 5.3 and 3.6 Gy for the ventricles. Fibrosis can be visualized via late gadolinium enhancement (LGE) on MRI [
      • Fukumoto K.
      • Habibi M.
      • Ipek E.G.
      • Zahid S.
      • Khurram I.M.
      • Zimmerman S.L.
      • et al.
      Association of left atrial local conduction velocity with late gadolinium enhancement on cardiac magnetic resonance in patients with atrial fibrillation.
      ] and is found in LA walls of patients with atrial fibrillation [
      • Burstein B.
      • Nattel S.
      “Atrial fibrosis: Mechanisms and clinical relevance in atrial fibrillation.
      ]. In oesophageal cancer patients LGE was more evident in areas of the heart receiving higher radiation doses [
      • Umezawa R.
      • Ota H.
      • Takanami K.
      • Ichinose A.
      • Matsushita H.
      • Saito H.
      • et al.
      MRI findings of radiation-induced myocardial damage in patients with oesophageal cancer.
      ].
      Given the variety of cardiac irradiation measures reported to be associated with OS, and the possibility that measures most closely associated with survival might be specific to particular RT techniques or cohorts, we have investigated the usefulness of reducing a broad measure, mean heart dose. In OX-RT this measure was significantly associated with OS (HR, 1.046; 95% CI, 1.007–1.087; p = 0.02), survival at 2 years post-treatment being 29% higher in patients with mean heart doses <7.6 Gy, the median value (Fig. 2(b)). Similarly, in a large retrospective study the all-cause death-rate at 2 years was respectively 40% and 52% in patents with mean heart doses less or greater than 10 Gy [
      • Atkins K.M.
      • Rawal B.
      • Chaunzwa T.L.
      • Lamba N.
      • Bitterman D.S.
      • Williams C.L.
      • et al.
      Cardiac radiation dose, cardiac disease, and mortality in patients with lung cancer.
      ].
      In a planning study carried out for LA-NSCLC patients treated using VMAT we found that by introducing a mean heart dose penalty into plan optimization, in addition to the penalty used routinely to control cardiac hot-spots, mean heart doses could be reduced by an average 4.8 Gy while respecting tumour coverage protocol limits and without markedly increasing irradiation of the lungs, oesophagus, proximal bronchial tree, cord or brachial plexus [
      • Turtle Louise
      • Bhalla Neeraj
      • Willett Andrew
      • Biggar Robert
      • Leadbetter Jonathan
      • Georgiou Georgios
      • et al.
      Cardiac-sparing radiotherapy for locally advanced non-small cell lung cancer.
      ]. This reduction amounted to 36% of the average baseline mean heart dose, and on the basis of the OX-RT data corresponds to a predicted hazard ratio for death of 0.81. Furthermore, mean heart dose reductions led to knock-on reductions in many cardiac substructure dose-volume measures. In particular, VLAwall-64-73-EQD2, VAortV-29-38-EQD2 and mean left and right ventricle doses fell by 68%, 100%, 41% and 51% relative to baseline values, insuring against survival being related more specifically to these measures than to whole-heart irradiation. Presently we are preparing a cardiac-sparing RT trial in which mean heart doses are reduced in this way.

      Conclusion

      In UV analyses of two independent cohorts of LA-NSCLC patients, OX-RT and IDEAL-6, OS was associated more strongly with VLAwall-64-73-EQD2 than with any measure of whole-heart irradiation investigated. OS was also significantly associated with VAortV-29-38-EQD2 in both cohorts, and in OX-RT was additionally significantly associated with mean heart dose. In a separate planning study we found that by penalizing cardiac irradiation and re-optimizing plans, mean heart doses could be reduced by an average 36% relative to baseline while respecting protocol limits on irradiation of other normal tissues, in the process achieving large reductions in VLAwall-64-73-EQD2 and VAortV-29-38-EQD2. We therefore intend to trial a treatment in which mean heart doses are reduced like this, closely monitoring patients in whom VLAwall-64-73-EQD2 and VAortV-29-38-EQD2 are not consequentially reduced to 0%, the median value of both measures in the OX-RT cohort.

      Declaration of Competing Interest

      The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

      Appendix A. Supplementary data

      The following are the Supplementary data to this article:

      References

        • Zhang T.W.
        • Snir J.
        • Boldt R.G.
        • Rodrigues G.B.
        • Louie A.V.
        • Gaede S.
        • et al.
        Is the importance of heart dose overstated in the treatment of non-small cell lung cancer? A systematic review of the literature.
        Int J Radiat Oncol Biol Phys. 2019; 104: 582-589https://doi.org/10.1016/j.ijrobp.2018.12.044
        • McWilliam A.
        • Kennedy J.
        • Hodgson C.
        • Vasquez Osorio E.
        • Faivre-Finn C.
        • van Herk M.
        Radiation dose to heart base linked with poorer survival in lung cancer patients.
        Eur J Cancer. 2017; 85: 106-113https://doi.org/10.1016/j.ejca.2017.07.053
        • Thor M.
        • Deasy J.O.
        • Hu C.
        • Gore E.
        • Bar-Ad V.
        • Robinson C.
        • et al.
        Modeling the impact of cardio-pulmonary irradiation on overall survival in NRG Oncology trial RTOG 0617.
        Clin Cancer Res. 2020; 26: 4643-4650https://doi.org/10.1158/1078-0432.CCR-19-2627
        • Guberina M.
        • Eberhardt W.
        • Stuschke M.
        • Gauler T.
        • Heinzelmann F.
        • Cheufou D.
        • et al.
        Heart dose exposure as prognostic marker after radiotherapy for resectable stage IIIA/B non-small-cell lung cancer: secondary analysis of a randomized trial.
        Ann Oncol. 2017; 28: 1084-1089https://doi.org/10.1093/annonc/mdx069
        • Bradley J.D.
        • 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: 187-199https://doi.org/10.1016/S1470-2045(14)71207-0
        • Tucker S.L.
        • Liu A.
        • Gomez D.
        • Tang L.L.
        • Allen P.
        • Yang J.
        • et al.
        Impact of heart and lung dose on early survival in patients with non-small cell lung cancer treated with chemoradiation.
        Radiother Oncol. 2016; 119: 495-500https://doi.org/10.1016/j.radonc.2016.04.025
        • McWilliam A.
        • Khalifa J.
        • Vasquez Osorio E.
        • Banfill K.
        • Abravan A.
        • Faivre-Finn C.
        • et al.
        Novel methodology to investigate the impact of radiation dose to heart sub-structures on overall survival.
        Int J Radiat Oncol Biol Phys. 2020; 108: 1073-1081https://doi.org/10.1016/j.ijrobp.2020.06.031
        • Vivekanandan S.
        • Landau D.B.
        • Counsell N.
        • Warren D.R.
        • Khwanda A.
        • Rosen S.D.
        • et al.
        The impact of cardiac radiation dosimetry on survival after radiation therapy for non-small cell lung cancer.
        Int J Radiat Oncol Biol Phys. 2017; 99: 51-60https://doi.org/10.1016/j.ijrobp.2017.04.026
        • Landau D.B.
        • Hughes L.
        • Baker A.
        • Bates A.T.
        • Bayne M.C.
        • Counsell N.
        • et al.
        IDEAL-CRT: a phase 1/2 trial of isotoxic dose-escalated radiation therapy and concurrent chemotherapy in patients with stage II/III non-small cell lung cancer.
        Int J Radiat Oncol Biol Phys. 2016; 95: 1367-1377https://doi.org/10.1016/j.ijrobp.2016.03.031
        • Schultz-Hector S.
        • Sund M.
        • Thames H.D.
        Fractionation response and repair kinetics of radiation-induced heart failure in the rat.
        Radiother Oncol. 1992; 23: 33-40https://doi.org/10.1016/0167-8140(92)90303-C
        • Harrell F.E.
        • Lee K.L.
        • Mark D.B.
        Multivariable prognostic models: issues in developing models, evaluating assumptions and adeqaucy, an measuring and reducing errors.
        Stat Med. 1996; 15: 361-387https://doi.org/10.1002/(SICI)1097-0258(19960229)15:4<361::AID-SIM168>3.0.CO;2-4
        • Pencina M.J.
        • D'Agostino R.B.
        Overall C as a measure of discrimination in survival analysis: model specific population value and confidence interval estimation.
        Stat Med. 2004; 23: 2109-2123https://doi.org/10.1002/sim.1802
        • Feng M.
        • Moran J.M.
        • Koelling T.
        • Chughtai A.
        • Chan J.L.
        • Freedman L.
        • et al.
        Development and validation of a heart atlas to study cardiac exposure to radiation following treatment for breast cancer.
        Int J Radiat Oncol Biol Phys. 2011; 79: 10-18https://doi.org/10.1016/j.ijrobp.2009.10.058
      1. [14] Ganem GCA, Coelho RC and de Godoy CMG. (2020) Animation of atrial and ventricular external walls of a virtual 3D heart based on echocardiogram images. In: González Díaz C et al. (eds) VIII Latin American Conference on Biomedical Engineering and XLII National Conference on Biomedical Engineering. CLAIB 2019. IFMBE Proceedings, vol 75. Springer, Cham. https://doi-org.liverpool.idm.oclc.org/10.1007/978-3-030-30648-9_90.

        • O’brien R.M.
        A caution regarding rules of thumb for variance inflation factors.
        Qual Quant. 2007; 41: 673-690https://doi.org/10.1007/s11135-006-9018-6
        • Abram S.V.
        • Helwig N.E.
        • Moodie C.A.
        • DeYoung C.G.
        • MacDonald A.W.
        • Waller H.G.
        Bootstrap enhanced penalized regression for variable selection with neuroimaging data.
        Front Neurosci. 2016; 10: 344https://doi.org/10.3389/fnins.2016.00344
        • Stam Barbara
        • Peulen Heike
        • Guckenberger Matthias
        • Mantel Frederick
        • Hope Andrew
        • Werner-Wasik Maria
        • et al.
        Dose to heart substructures is associated with non-cancer death after SBRT in stage I-II NSCLC patients.
        Radiother Oncol. 2017; 123: 370-375https://doi.org/10.1016/j.radonc.2017.04.017
        • Kitahara T.
        • Liu K.
        • Solanki K.
        • Trott K.R.
        Functional and morphological damage after local heart irradiation and/or adriamycin in Wistar rats.
        Radiat Oncol Investig. 1993; 1: 198-205
        • Turtle Louise
        • Bhalla Neeraj
        • Willett Andrew
        • Biggar Robert
        • Leadbetter Jonathan
        • Georgiou Georgios
        • et al.
        Cardiac-sparing radiotherapy for locally advanced non-small cell lung cancer.
        Radiat Oncol. 2021; 16https://doi.org/10.1186/s13014-021-01824-3
        • Fukumoto K.
        • Habibi M.
        • Ipek E.G.
        • Zahid S.
        • Khurram I.M.
        • Zimmerman S.L.
        • et al.
        Association of left atrial local conduction velocity with late gadolinium enhancement on cardiac magnetic resonance in patients with atrial fibrillation.
        Circ Arrhythm Electrophysiol. 2016; 9e002897https://doi.org/10.1161/CIRCEP.115. 002897
        • Burstein B.
        • Nattel S.
        “Atrial fibrosis: Mechanisms and clinical relevance in atrial fibrillation.
        J Am Coll Cardiol. 2008; 51: 802-809https://doi.org/10.1016/j.jacc. 2007.09.064
        • Umezawa R.
        • Ota H.
        • Takanami K.
        • Ichinose A.
        • Matsushita H.
        • Saito H.
        • et al.
        MRI findings of radiation-induced myocardial damage in patients with oesophageal cancer.
        Clin Radiol. 2014; 69: 1273-1279https://doi.org/10.1016/j.crad. 2014.08.010
        • Atkins K.M.
        • Rawal B.
        • Chaunzwa T.L.
        • Lamba N.
        • Bitterman D.S.
        • Williams C.L.
        • et al.
        Cardiac radiation dose, cardiac disease, and mortality in patients with lung cancer.
        J Am Coll Cardiol. 2019; 73: 2976-2987https://doi.org/10.1016/j.jacc.2019.03. 500