Advertisement

Cardiac Function Modifies the Impact of Heart Base Dose on Survival: A Voxel-Wise Analysis of Patients With Lung Cancer From the PET-Plan Trial

Open AccessPublished:September 18, 2022DOI:https://doi.org/10.1016/j.jtho.2022.09.004

      Abstract

      Introduction

      Heart dose has emerged as an independent predictor of overall survival in patients with NSCLC treated with radiotherapy. Several studies have identified the base of the heart as a region of enhanced dose sensitivity and a potential target for cardiac sparing. We present a dosimetric analysis of overall survival in the multicenter, randomized PET-Plan trial (NCT00697333) and for the first time include left ventricular ejection fraction (EF) at baseline as a metric of cardiac function.

      Methods

      A total of 205 patients with inoperable stage II or III NSCLC treated with 60 to 72 Gy in 2 Gy fractions were included in this study. A voxel-wise image-based data mining methodology was used to identify anatomical regions where higher dose was significantly associated with worse overall survival. Univariable and multivariable Cox proportional hazards models tested the association of survival with dose to the identified region, established prognostic factors, and baseline cardiac function.

      Results

      A total of 172 patients remained after processing and censoring for follow-up. At 2-years posttreatment, a highly significant region was identified within the base of the heart (p < 0.005), centered on the origin of the left coronary artery and the region of the atrioventricular node. In multivariable analysis, the number of positron emission tomography–positive nodes (p = 0.02, hazard ratio = 1.13, 95% confidence interval: 1.02–1.25) and mean dose to the cardiac subregion (p = 0.02, hazard ratio = 1.11 Gy−1, 95% confidence interval: 1.02–1.21) were significantly associated with overall survival. There was a significant interaction between EF and region dose (p = 0.04) for survival, with contrast plots revealing a larger effect of region dose on survival in patients with lower EF values.

      Conclusions

      This work validates previous image-based data mining studies by revealing a strong association between dose to the base of the heart and overall survival. For the first time, an interaction between baseline cardiac health and heart base dose was identified, potentially suggesting preexisting cardiac dysfunction exacerbates the impact of heart dose on survival.

      Keywords

      Introduction

      Radiotherapy is an important modality for the treatment of lung cancer, being the most common form of treatment either as a standalone therapy or in combination with surgery, chemotherapy, or immunotherapy.
      National Cancer Registration Analysis Service
      Workbook 1: chemotherapy, radiotherapy and tumour resection by tumour & patient characteristics in England.
      On-set image guidance, intensity modulation, motion control strategies, and positron emission tomography (PET)–informed planning have substantially advanced the planning and delivery of lung radiotherapy in the past 20 years. Today’s treatments are more conformal and provide superior sparing of the healthy tissues, offering the potential for dose escalation and improved local control while maintaining acceptable levels of toxicity. Despite this, dose-escalation trials have failed and improvements in survival have been derived mainly from developments in systemic therapy, not radiotherapy.
      • Bradley J.D.
      • Paulus R.
      • Komaki R.
      • 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 p.
      ,
      • Antonia S.J.
      • Villegas A.
      • Daniel D.
      • et al.
      Durvalumab after chemoradiotherapy in stage III non–small-cell lung cancer.
      Lung cancer remains the global leading cause of cancer-related death, with a 5-year survival rate of just 10%.
      • Sung H.
      • Ferlay J.
      • Siegel R.L.
      • et al.
      Global cancer statistics 2020: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries.
      ,
      In 2015, the first major evidence of an association between cardiac dose and overall survival (OS) emerged after the publication of the RTOG 0617 trial.
      • Bradley J.D.
      • Paulus R.
      • Komaki R.
      • 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 p.
      This multicenter phase 3 study compared OS between high-dose (74 Gy) and standard-dose (60 Gy) prescriptions, both delivered in 2 Gy fractions and found OS to be worse in the high-dose arm with an 8-month reduction in median survival time (20.3 mo versus 28.7 mo). Multivariable analysis identified two cardiac dose-volume parameters (V5 and V30) to be significantly associated with OS (p = 0.004). This led to several retrospective studies analyzing the association between cardiac dose and OS, with highly inconsistent results.
      • Zhang T.W.
      • Snir J.
      • Boldt R.G.
      • et al.
      Is the importance of heart dose overstated in the treatment of non-small cell lung cancer? A systematic review of the literature.
      More recent studies have moved away from the use of whole heart dose parameters and instead analyze cardiac substructure doses or apply contour-independent voxel-wise methods. This allows cardiac subregions of enhanced dose sensitivity to be identified, which are lost in whole organ dosimetric analyses. Multiple such studies have identified dose to the base of the heart as an independent predictor of OS and a potential sparing region.
      • Stam B.
      • Peulen H.
      • Guckenberger M.
      • et al.
      Dose to heart substructures is associated with non-cancer death after SBRT in stage I–II NSCLC 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.
      • McWilliam A.
      • Khalifa J.
      • Vasquez Osorio E.
      • et al.
      Novel methodology to investigate the effect of radiation dose to heart substructures on overall survival.
      • Green A.
      • Vasquez Osorio E.
      • Aznar M.C.
      • McWilliam A.
      • van Herk M.
      Image based data mining using per-voxel Cox regression.
      Nevertheless, the literature is far from definitive in this area with significant associations between dose to almost all cardiac substructures and survival being reported.
      • Thor M.
      • Deasy J.O.
      • Hu C.
      • et al.
      Modeling the impact of cardiopulmonary irradiation on overall survival in NRG oncology trial RTOG 0617.
      • Xue J.
      • Han C.
      • Jackson A.
      • et al.
      Doses of radiation to the pericardium, instead of heart, are significant for survival in patients with non-small cell lung cancer.
      • Vivekanandan S.
      • Landau D.B.
      • Counsell N.
      • et al.
      The impact of cardiac radiation dosimetry on survival after radiation therapy for non-small cell lung cancer.
      • Wong O.Y.
      • Yau V.
      • Kang J.
      • et al.
      Survival impact of cardiac dose following lung stereotactic body radiotherapy.
      • Atkins K.M.
      • Chaunzwa T.L.
      • Lamba N.
      • et al.
      Association of left anterior descending coronary artery radiation dose with major adverse cardiac events and mortality in patients with non-small cell lung cancer.
      Furthermore, baseline cardiac disease is known to modify the heart’s response to radiation, but this interaction is poorly understood and rarely accounted for in dosimetric analyses.
      • Dess R.T.
      • Sun Y.
      • Matuszak M.M.
      • et al.
      Cardiac events after radiation therapy: combined analysis of prospective multicenter trials for locally advanced non-small-cell lung cancer.
      With 25% to 30% of patients with lung cancer exhibiting cardiac comorbidities, better understanding of their effect on dose-response is critical to realizing the full potential offered by sparing substructures of the heart.
      • Janssen-Heijnen M.L.G.
      • Schipper R.M.
      • Razenberg P.P.A.
      • Crommelin M.A.
      • Coebergh J.W.W.
      Prevalence of co-morbidity in lung cancer patients and its relationship with treatment: a population-based study.
      The PET-Plan phase 3 clinical trial (NCT00697333) investigated imaging-based target volume reduction in patients with locally advanced NSCLC.
      • Nestle U.
      • Schimek-Jasch T.
      • Kremp S.
      • et al.
      Imaging-based target volume reduction in chemoradiotherapy for locally advanced non-small-cell lung cancer (PET-Plan): a multicentre, open-label, randomised, controlled trial.
      Patients received either conventional target delineation on the basis of PET and computed tomography (CT) with elective nodal irradiation, or target delineation on the basis of PET alone, with isotoxically escalated prescription doses of 60 to 72 Gy. The risk of locoregional progression at 1 year in the PET group was found to be noninferior to the conventional group, and there was no significant change in toxicity. In this work, we present a voxel-wise dosimetric analysis of the PET-Plan trial data set which aims to validate the impact of heart base dose on OS. For the first time, a quantitative metric of baseline cardiac function will be included in the analysis to allow exploration of the impact of cardiac function on survival.

      Materials and Methods

      The PET-Plan trial
      • Nestle U.
      • Schimek-Jasch T.
      • Kremp S.
      • et al.
      Imaging-based target volume reduction in chemoradiotherapy for locally advanced non-small-cell lung cancer (PET-Plan): a multicentre, open-label, randomised, controlled trial.
      was a multicenter, open-label, randomized controlled trial for patients with locally advanced NSCLC which aimed to evaluate the feasibility and effectiveness of image-based target volume reduction. The trial ran between 2009 and 2016, recruiting 205 patients from 24 centers across Germany, Austria, and Switzerland. All recruited patients had inoperable, locally advanced NSCLC, an Eastern Cooperative Oncology Group performance status of less than 3, and not received previous treatment. Patients were randomized 1:1 to target delineation on the basis of PET alone or “conventional” target delineation informed by PET and CT imaging, with elective nodal irradiation. Both groups received isotoxically escalated doses of 60 to 74 Gy in 2 Gy fractions with concurrent platinum-based chemotherapy. The primary end point of the trial was time to local progression, with the aim of testing the noninferiority of PET-based planning. Left ventricular ejection fraction (EF) measurements were performed according to local clinical protocols up to a maximum of 3 months before the start of treatment.
      A voxel-wise dosimetric analysis was performed to identify anatomical regions where a significant statistical association exists between dose and survival. Following the methodology described by Chen et al.
      • Chen C.
      • Witte M.
      • Heemsbergen W.
      • Herk M.V.
      Multiple comparisons permutation test for image based data mining in radiotherapy.
      and Palma et al.,
      • Palma G.
      • Monti S.
      • Cella L.
      Voxel-based analysis in radiation oncology: a methodological cookbook.
      the major steps of this analysis are as follows: spatial normalization of the dose distributions, calculation of the mean dose distributions for the alive and dead groups, calculation of the dose difference between the groups, and assessment of significance by permutation testing.
      All CT scans and dose distributions were deformed to a representative template anatomy, with small tumor burden distal from the mediastinum, using the NiftyReg package,

      SourceForge.net. Nifty Reg download. https://sourceforge.net/projects/niftyreg/. Accessed July 9, 2021.

      a B-spline based registration algorithm. The deformed scans were visually inspected to identify gross registration failures which were excluded from the analysis. The heart contours used for the treatment planning were defined according to the trial protocol (entire heart including the atria and the right ventricle) but were found to have highly inconsistent superior borders, in addition to multiple examples of major deviations from the protocol. Therefore, the heart contours were automatically generated using the deep learning auto segmentation component of ADMIRE V3 (Elekta AB, Stockholm, Sweden). The autosegmented heart contours were quality assured by visual inspection, with any required adjustments performed manually. The accuracy of the deformable registration process was then assessed by plotting the spatial distribution of the centroid coordinates of the autosegmented heart contours, deformed onto the reference anatomy. To account for this uncertainty in the results of the voxel-wise analysis, a three-dimensional Gaussian filter was applied to blur the dose distributions, with the filter’s width set to the SD of the center of mass coordinates of the deformed contours in each direction. Results obtained with and without blurring the dose distributions were compared.
      As described by Chen et al.,
      • Chen C.
      • Witte M.
      • Heemsbergen W.
      • Herk M.V.
      Multiple comparisons permutation test for image based data mining in radiotherapy.
      permutation testing was performed to evaluate the significance of any identified dose differences and to appropriately control for multiple comparisons. To enforce the null hypothesis of there being no association between dose and survival and therefore zero dose difference between the groups, the event labels were randomly permuted, with 1000 permutations performed. For each permutation, the summary test statistic was recalculated, with the maximum value across the whole image used to populate the null distribution of the test statistic for each permutation. This null distribution can then be used to determine the significance of any dose differences in the observed data. The summary test statistic used was the maximum t value, defined as the maximum value of the dose difference between groups divided by the SD computed over all voxels of the dose difference under permutation.
      The mean dose to the significant region, defined as 90% of the maximum t value, was extracted for all patients and univariable survival analyses were performed to determine whether dose to region and key clinical parameters (age, sex, performance status, tumor volume, T-stage, N-stage, number of PET-positive nodes, and baseline EF) were significantly associated with OS, with significance defined as a p value of less than 0.05. Kaplan-Meier survival curves were also plotted with the cohort split on median dose to the region.
      A region dose-EF interaction term was added to the analysis to probe for dose confounding of the impact of EF on survival, with EF values centered on the median value of the data set (60%). To verify whether region dose was acting a surrogate for mean heart dose, additional analyses were performed with region and mean heart dose exchanged (Supplementary Appendix Section 2). Model performance was assessed using the metrics of concordance and the Akaike information criterion (AIC). Concordance is a measure of the ability of a model to assign higher risk to individuals who experience an event (discrimination), whereas AIC is a relative measure of overall model performance which penalizes increasing model complexity. The logarithm of tumor volume was used to avoid a small number of unusually large tumors skewing the analysis. All parameters, regardless of their significance under univariable analysis, were then included in a multivariable Cox proportional hazards model to avoid the limitations associated with variable selection strategies.
      • Smith G.
      Step away from stepwise.
      Hazard ratios (HRs) and 95% confidence intervals (CIs) were computed for all variables.

      Results

      After visual inspection, 13 patients were excluded from the voxel-wise analysis owing to deformable registration failure. Most of these failures were due to atelectasis or other atypical anatomy caused by extensive thoracic disease. Summary demographics for the 172 patients remaining in the analysis are given in Table 1. The SD of heart centroid coordinates for all 172 patients included in the analysis was 0.6 cm, 1.1 cm, and 0.6 cm in the lateral, craniocaudal, and anterior-posterior directions, respectively.
      Table 1Patient Characteristics
      n (%)/Median (IQR)
      Age65 (59–71)
      Gender
       Male125 (73)
       Female47 (27)
      Performance status
       089 (52)
       1, 283 (48)
      Primary GTV (mL)62 (28–125)
      T-stage
       112 (7.0)
       240 (23)
       351 (30)
       469 (40)
      N-stage
       016 (9.3)
       119 (11)
       297 (56)
       340 (23)
      No. PET +ve nodes4.00 (3.00–6.00)
      Ejection fraction
       <50%8 (5.1)
       50%–70%121 (77)
       ≥70%28 (18)
       Unknown15
      Note: For continuous variables, the median value is listed followed by the IQR. For categorical variables, the number and percentage in each category are listed.
      +ve, positive; GTV, gross tumor volume; IQR, interquartile range; PET, positron emission tomography.
      At 24 months posttreatment, a highly significant region was identified within the base of the heart (p < 0.005), centered on the origin of the left coronary artery and the region of the atrioventricular node (Fig. 1). Results obtained with the dose distributions blurred to account for deformable image registration uncertainty were similar, with the significant region slightly enlarged but still centered on the same cardiac anatomy. The region of significance was defined at 90% of the maximum t value, localized around the position of the highest t value in the data set, and mean doses were extracted from the unblurred distributions to be included in univariable and multivariable Cox proportional hazards analyses. The median dose to the region across all patients was 33.5 Gy, with a range of 1.8 Gy to 69.3 Gy.
      Figure thumbnail gr1
      Figure 1Axial, coronal, and sagittal views of the reference patient anatomy overlaid with isocontours of 80% (green), 90% (yellow), and 95% (red) of the maximum t value.
      In univariable analysis (Table 2), the number of PET-positive lymph nodes (HR = 1.12, 95% CI: 1.03–1.21, p = 0.01), mean dose to the region (HR = 1.02 Gy−1, 95% CI: 1.01–1.03, p < 0.01), and mean whole heart dose (HR = 1.03 Gy−1, 95% CI: 1.01–1.05, p < 0.01) were found to be significant predictors of survival. A separate survival analysis (Supplementary Table 1) was performed to test for an interaction between region dose and EF; a highly significant interaction term was found (p = 0.02).
      Table 2Univariable Survival Analysis Results
      VariableHR (95% CI)p Value
      Age1.02 (0.99–1.04)0.17
      Gender (ref: male)--
       Female0.85 (0.55–1.32)0.47
      Performance status (ref: 0)--
       1, 21.19 (0.81–1.74)0.37
      Gross tumor volume (log)1.43 (0.99–2.05)0.06
      T-stage (ref: T1)--
       T22.08 (0.80–5.40)0.13
       T31.53 (0.59–3.95)0.38
       T41.59 (0.63–4.01)0.33
      N-stage (ref: N0)--
       N10.52 (0.19–1.40)0.20
       N21.00 (0.50–2.01)1.00
       N31.14 (0.53–2.45)0.74
      No. of PET +ve nodes1.12 (1.03–1.21)0.01
      Ejection fraction1.00 (0.98–1.02)0.85
      Mean region dose1.02 (1.01–1.03)<0.01
      Mean whole heart dose1.03 (1.01–1.05)<0.01
      Note: Variables are continuous unless otherwise indicated. Bold red highlighting indicates signficant variables (p < 0.05).
      +ve, positive; CI, confidence interval; HR, hazard ratio; PET, positron emission tomography; ref, reference.
      Table 3Multivariable Survival Analysis Results
      VariableHR (95% CI)p Value
      Age1.019 (0.991–1.047)0.18
      Gender (ref: male)
       Female0.997 (0.607–1.637)0.99
      Performance status (ref: 0)
       1, 21.143 (0.741–1.762)0.55
      Gross tumor volume (log)1.278 (0.788–2.071)0.32
      T-stage (ref: T1)
       T20.778 (0.247–2.449)0.67
       T30.670 (0.198–2.263)0.52
       T40.575 (0.174–1.898)0.36
      N-stage (ref: N0)
       N10.514 (0.162–1.626)0.26
       N20.836 (0.368–1.896)0.67
       N30.847 (0.313–2.295)0.74
      No. of PET +ve nodes1.130 (1.023–1.248)0.02
      Mean region dose1.112 (1.019-1.213)0.02
      EF1.052 (0.995–1.113)0.08
      Interaction: region dose, EF0.999 (0.997–1.000)0.04
      Note: Variables are continuous unless otherwise indicated. Bold red highlighting indicates signficant variables (p < 0.05).
      +ve, positive; CI, confidence interval; EF, ejection fraction; HR, hazard ratio; PET, positron emission tomography; ref, reference.
      As both mean dose to the region and whole heart mean dose were significant predictors of OS, model performance was evaluated with the substitution of either parameter (Supplementary Table 4). The inclusion of either dose metric improved model performance relative to a model with no dose variable (Δ AIC = 4.1), with models including region dose performing best overall. No evidence of an interaction between whole heart mean dose and EF was found.
      In subsequent multivariable analysis, including the EF-region dose interaction term, the number of PET-positive nodal stations (p = 0.02, HR = 1.13, 95% CI: 1.02–1.25), mean dose to the cardiac subregion (p = 0.02, HR = 1.11, 95% CI: 1.01–1.21), and the interaction term (p = 0.04, HR = 0.99, 95% CI: 0.99–1.00) were found to be significantly associated with OS. To aid interpretation of the interaction term, contrast plots of log hazard ratio versus region dose were constructed at the 10th, 50th, and 90th percentiles of EF, equivalent to EF values of 50%, 60%, and 72%, respectively. Contrast plots reveal the effect of changing the region dose from a reference value (in this case, the mean) on the outcome, while holding the other interacting variable constant. As found in Figure 2, the plots indicate that increasing region dose above the mean value increases risk in both low and normal EF patients, with the effect being more pronounced for the low EF group. Region dose had no significant effect on log hazard ratio for high EF patients.
      Figure thumbnail gr2
      Figure 2Contrast plots of log (hazard ratio) versus region dose constructed at 10%, 50%, and 90% quantiles of the EF values. Contrast plots demonstrate the effect of changing the region dose from a reference value (in this case, the mean) on the outcome. The details of the model used to produce this plot are given in . EF, ejection fraction.
      Kaplan-Meier survival curves were constructed with the cohort split on median dose (33.5 Gy) to the identified cardiac region. As found in Figure 3, there is a clear split between the groups with the median survival of those receiving less than 33.5 Gy increased to 34 months, compared with 18 months for the above median dose group. The log-rank test of the significance of this difference gave a p value of less than 0.005.
      Figure thumbnail gr3
      Figure 3Kaplan-Meier survival curve split on median dose to the identified dose-sensitive region in the base of the heart.

      Discussion

      This study has identified a dose-sensitive cardiac region within the base of the heart which is significantly associated with patient survival. Our findings are consistent with a growing body of evidence supporting the importance of heart base dose and the novel inclusion of a metric of baseline cardiac function in this study, which provides early evidence that poor cardiac function may exacerbate the effect of heart base dose on survival.
      The significant cardiac region reported here is partially consistent with the findings of a similar IBDM study conducted by McWilliam et al.,
      • 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.
      with overlap of the areas of highest significance, but without the major anterior and posterior extension. The difference in the identified regions may be partly explained by the different patient demographics of the two data sets, with the PET-Plan data set having a far higher proportion of performance status 0 patients, possibly indicating superior baseline cardiac function in this cohort. The median dose to the region is also considerably higher in this cohort at more than double that reported by McWilliam et al. (34 Gy versus 16 Gy). This is a result of both the escalated target doses delivered in the PET-Plan trial and the high proportion of patients with extensive nodal disease (N2, N3).
      Stam et al. also report a significant association between dose to the upper regions of the heart, specifically the left atrium and superior vena cava, and OS in a cohort of 803 patients with SI to II NSCLC SABR.
      • Stam B.
      • Peulen H.
      • Guckenberger M.
      • et al.
      Dose to heart substructures is associated with non-cancer death after SBRT in stage I–II NSCLC patients.
      Because of the proximity of the sinoatrial and atrioventricular nodes to the regions identified within the base of the heart, damage to the conduction system has been hypothesized as a potential physiological mechanism driving the association, but this has yet to be clinically validated. With the dose to the identified region much higher in this study than previously reported, it is possible that a threshold for activation of a different physiological mechanism of cardiac toxicity has been passed and is driving the dose-survival association in this region.
      Although similar anatomical regions were visible at earlier time points, only the IBDM results at 24 months posttreatment were statistically significant. This is most likely a consequence of the small data set (172 patients) causing the dose difference signal to be highly sensitive to the relative number of alive and dead patients, with an appropriate balance only being achieved at this time point. Regardless of this, death due to cardiac events is hypothesized to occur much earlier in lung cancer cohorts than the typical latency period of several years or even decades found in patients with breast and lymphoma, who receive much lower cardiac doses.
      • Dess R.T.
      • Sun Y.
      • Matuszak M.M.
      • et al.
      Cardiac events after radiation therapy: combined analysis of prospective multicenter trials for locally advanced non-small-cell lung cancer.
      Furthermore, the 24-month time point is consistent with multiple lung cancer studies which report a separation in survival curves, stratified by cardiac dose, which typically begins at approximately 6 months posttreatment and is sustained well beyond 2 years.
      • 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.
      ,
      • Stam B.
      • van der Bijl E.
      • van Diessen J.
      • et al.
      Heart dose associated with overall survival in locally advanced NSCLC patients treated with hypofractionated chemoradiotherapy.
      • Atkins K.M.
      • Bitterman D.S.
      • Chaunzwa T.L.
      • et al.
      Mean heart dose is an inadequate surrogate for left anterior descending coronary artery dose and the risk of major adverse cardiac events in lung cancer radiation therapy.
      • Johnson-Hart C.
      • Price G.
      • McWilliam A.
      • Green A.
      • Faivre-Finn C.
      • van Herk M.
      Impact of small residual setup errors after image guidance on heart dose and survival in non-small cell lung cancer treated with curative-intent radiotherapy.
      Appropriately handling the EF data for analysis is nontrivial owing to the nonlinear change in associated risk with increasing EF value; both high (>70%) and low EF values (<50%) are associated with poor prognosis, separated by a wide range of normal values (50%–70%), with a nadir of risk reported in a “supranormal” range of 60% to 65%.
      • Ponikowski P.
      • Voors A.A.
      • Anker S.D.
      • et al.
      2016 ESC Guidelines for the diagnosis and treatment of acute and chronic heart failure: The Task Force for the diagnosis and treatment of acute and chronic heart failure of the European Society of Cardiology (ESC) developed with the special contribution of the Heart Failure Association (HFA) of the ESC.
      • Wehner G.J.
      • Jing L.
      • Haggerty C.M.
      • et al.
      Routinely reported ejection fraction and mortality in clinical practice: where does the nadir of risk lie?.
      • Paonessa J.R.
      • Brennan T.
      • Pimentel M.
      • Steinhaus D.
      • Feng M.
      • Celi L.A.
      Hyperdynamic left ventricular ejection fraction in the intensive care unit.
      In accordance with the work of Wehner et al.,
      • Wehner G.J.
      • Jing L.
      • Haggerty C.M.
      • et al.
      Routinely reported ejection fraction and mortality in clinical practice: where does the nadir of risk lie?.
      we treated EF as a continuous variable, centered on the median value of the data set (60%), which coincided with the reported nadir in risk. For completeness, we also repeated the analysis with the EF data dichotomized into the three groups—reduced (<50%), preserved (50%–70%), and hyperdynamic (>70%) (Supplementary Appendix Table 5)—and found reduced EF to be a significant predictor of survival, with the major limitation of only eight patients falling into this group. A hyperdynamic EF was not found to be significantly associated with OS.
      The significance of the region dose-EF interaction term in the final analysis (and lack of effect of EF on survival in univariable analysis) may be cautiously interpreted as baseline EF only being of importance in patients receiving higher doses to the heart, in which group a higher EF is protective. As such sparing of the base of the heart could offer the potential to improve short-term survival, particularly in the group of patients with preexisting cardiac conditions who receive high heart base dose under current planning practices. Critically, identification of a partial heart volume for sparing, as opposed to whole heart sparing, offers a practical solution to realizing this potential. Implementation of heart base sparing may also allow realization of the potential offered by dose escalation which so far has failed to reveal improved OS in clinical trials.
      Despite being an established metric used to guide many clinical decisions, EF is not considered to be the most robust or independent measure of cardiac contractility. This is because EF values exhibit dependence on heart rate, preload and afterload,
      • Halliday B.P.
      • Senior R.
      • Pennell D.J.
      Assessing left ventricular systolic function: from ejection fraction to strain analysis.
      with reproducibility varying greatly between different measurement techniques.
      • Pellikka P.A.
      • She L.
      • Holly T.A.
      • et al.
      Variability in ejection fraction measured by echocardiography, gated single-photon emission computed tomography, and cardiac magnetic resonance in patients with coronary artery disease and left ventricular dysfunction.
      In addition, EF only measures one aspect of cardiac performance and there are multiple modes of chronic heart failure which preserve a normal EF. Saiki et al. have experimentally found that cardiac dose is causally linked to diastolic dysfunction and heart failure with preserved EF through damage to the coronary microvasculature, with this effect validated in a case-control study of patients with breast cancer.
      • Saiki H.
      • Moulay G.
      • Redfield M.
      • et al.
      Experimental cardiac radiation exposure induces ventricular diastolic dysfunction with preserved ejection fraction.
      • Saiki H.
      • Petersen I.
      • Redfield M.
      • et al.
      Risk of heart failure with preserved ejection fraction in older women after contemporary radiotherapy for breast cancer.
      As such, it is clear that EF alone cannot provide enough information to fully characterize the complex interactions between radiotherapy dose and cardiac toxicity. A broader suite of cardiac function metrics acquired both pre- and post-treatment is required in future prospective studies, with global longitudinal strain proposed as an alternative and more sensitive index of cardiac contractility.
      • Halliday B.P.
      • Senior R.
      • Pennell D.J.
      Assessing left ventricular systolic function: from ejection fraction to strain analysis.
      A key strength of the voxel-wise image-based data mining methodology used in this work is that it is entirely contour independent. Reliance on predefined structures of interest can result in the failure to discover significant regions owing to them falling outside of, or across the boundaries of, the predefined structures. This problem is exacerbated in early discovery and hypothesis-generating studies when the regions of interest are unknown. Contour independence also circumvents the uncertainties introduced by inter- and intra-observer delineation, while also permitting analysis of data sets for which “appropriate” contours do not exist, both of which are significant issues in cardiac studies.
      • Gong Y.
      • Gore E.M.
      • Bar-Ad V.
      • et al.
      Variation of cardiac contours using different heart definitions for NSCLC patients enrolled on RTOG 0617.
      An inherent limitation of the application of image-based data mining to dose-outcome analysis is the intrinsic dose bias.
      • Palma G.
      • Monti S.
      • Pacelli R.
      • et al.
      Radiation pneumonitis in thoracic cancer patients: multi-center voxel-based analysis.
      The identification of significant anatomical regions with IBDM requires both dose levels high enough to initiate an effect and sufficient dose heterogeneity across the patient cohort to provide a signal in dose difference. As such, IBDM is insensitive to detecting significance in regions where dose is low or consistent. The spatial correlation of dose inherent to radiotherapy plans also introduces a dependence of IBDM results on current planning techniques and limits the spatial resolution of the analysis. The anterior regions of significance illustrated in Figure 1 are most likely an example of dose confounding resulting from common beam angles in patients with left-sided tumors. As such, the results of IBDM studies should not be interpreted as direct maps of radiosensitivity, but cautiously interpreted as providing information to guide further optimization of current treatment techniques. For more robust conclusions to be drawn, methods of characterizing the spatial correlations of dose distributions are required. Dosimetric connectograms and probabilistic independent component analysis have been proposed as potential solutions and require further evaluation.
      • Palma G.
      • Monti S.
      • Pacelli R.
      • et al.
      Radiation pneumonitis in thoracic cancer patients: multi-center voxel-based analysis.
      In common with most survival analysis studies, there is no differentiation by cause of death in this study. This lack of specificity presents a major limitation as any hypotheses generated from the cardiac dose survival association reported here are reliant on the unproven assumption that the deaths driving this association were due to cardiac causes. Studies have revealed that cardiovascular-related death is underreported in this cohort, and so even when the data are available, it may not be reliable.
      • Sun F.
      • Franks K.
      • Murray L.
      • et al.
      Cardiovascular mortality and morbidity following radical radiotherapy for lung cancer: is cardiovascular death under-reported?.
      There is no obvious solution to improve the quality of cause of death data, but instead the work presented here may be built on by prospective studies with a greater focus on collection of detailed comorbidity and cardiac function data. Pre- and post-treatment cardiac imaging studies such as ACCOLADE (NCT03645317) should facilitate elucidation of the physiological mechanism driving the association identified in this work.
      A limitation of this and almost all cardiac dose-outcome studies is the lack of consideration of both inter- and intra-fractional cardiac motion. The rapid expansion and contraction of the heart through the cardiac cycle changes the size and shape of the cardiac substructures, including the organ’s baseline position, creating a dose blurring effect which is not directly accounted for in this work. Vasquez Osorio et al. report that a margin of up to 7 mm is required to encompass the full extent of cardiac motion with heart contours delineated on 3D CT, and coincidentally, this is of a similar magnitude to the Gaussian blurring applied to the dose distributions to account for uncertainty in the deformable image registration process.
      • Vasquez Osorio E.
      • McCallum H.
      • Cobben D.
      • et al.
      Protecting the heart: A practical approach to account for the full extent of heart motion in radiation therapy planning.
      As such, some of the dose blurring due to cardiac motion is indirectly accounted for in this study, although the full complexity of cardiac and respiratory motion will not be captured by Gaussian blurring.
      The clinical implementation of heart base sparing requires establishment of a dose constraint which can be incorporated into routine treatment planning practice. Although there is no established dose limit for the base of the heart, previous work by McWilliam et al.
      • McWilliam A.
      • Khalifa J.
      • Vasquez Osorio E.
      • et al.
      Novel methodology to investigate the effect of radiation dose to heart substructures on overall survival.
      reported that a maximum EQD2 of 23 Gy was the optimal threshold in a cohort of 1000 patients, and splitting the PET-Plan cohort at this dose level does produce a significant difference in OS (p = 0.04). Nevertheless, any dose limit derived from retrospective studies suffers from the fundamental deficiency of not accounting for the increase in dose to surrounding structures caused by sparing the heart base. This has the potential to cause an unacceptable increase in toxicity, and therefore, the optimal dose limit can only be determined from prospective studies of heart base-sparing radiotherapy. The RAPID-RT study aims to optimize the heart base dose limit by applying the iterative approach of rapid learning to a cohort receiving heart base sparing as the standard of care.
      • Price G.
      • Devaney S.
      • French D.P.
      • et al.
      Can real-world data and rapid learning drive improvements in lung cancer survival? The RAPID-RT study.
      In conclusion, this work further validates the association between the heart base dose and OS in an independent cohort of patients with lung cancer. Importantly, for the first time, a metric of baseline cardiac function was included in the analysis. A significant interaction between pretreatment EF and dose to the base of the heart was identified, potentially suggesting preexisting cardiac dysfunction modifies the impact of heart dose on survival. Prospective studies are now required to understand the physiological mechanisms which drive this association and to evaluate the potential offered by heart substructure sparing.
      • Price G.
      • Devaney S.
      • French D.P.
      • et al.
      Can real-world data and rapid learning drive improvements in lung cancer survival? The RAPID-RT study.

      CRediT Authorship Contribution Statement

      Matthew Craddock: Formal analysis, Methodology, Writing—original draft, Writing—review and editing.
      Ursula Nestle: Conceptualization, Writing—review and editing.
      Jochem Koenig, Tanja Schimek-Jasch, Stephanie Kremp, Stefan Lenz: Data curation.
      Kathryn Banfill, Ahmed Salem, Corinne Faivre-Finn: Writing—review and editing.
      Angela Davey: Methodology.
      Gareth Price: Supervision, Methodology.
      Marcel van Herk: Methodology, Software, Writing—review and editing.
      Alan McWilliam: Supervision, Conceptualization, Methodology, Writing—review and editing.

      Acknowledgments

      Mr. Craddock, Dr. McWilliam, Dr. Price, and Dr. Faivre-Finn are supported by Cancer Research UK by means of funding to the Cancer Research Manchester Centre [C147/A25254]. Drs. Faivre-Finn, McWilliam, and van Herk are also supported by the National Institute for Health and Care Research Manchester Biomedical Research Centre . Dr. Price is also supported by Cancer Research UK RadNet Manchester (C1994/A28701).

      References

        • National Cancer Registration Analysis Service
        Workbook 1: chemotherapy, radiotherapy and tumour resection by tumour & patient characteristics in England.
        • Bradley J.D.
        • Paulus R.
        • Komaki R.
        • 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 p.
        Lancet Oncol. 2015; 16: 187-199
        • Antonia S.J.
        • Villegas A.
        • Daniel D.
        • et al.
        Durvalumab after chemoradiotherapy in stage III non–small-cell lung cancer.
        N Engl J Med. 2017; 377: 1919-1929
        • Sung H.
        • Ferlay J.
        • Siegel R.L.
        • et al.
        Global cancer statistics 2020: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries.
        CA Cancer J Clin. 2021; 71: 209-249
        • Cancer Research UK
        Lung cancer survival statistics.
        • Zhang T.W.
        • Snir J.
        • Boldt R.G.
        • 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: 582-589
        • Stam B.
        • Peulen H.
        • Guckenberger M.
        • et al.
        Dose to heart substructures is associated with non-cancer death after SBRT in stage I–II NSCLC patients.
        Radiother Oncol. 2017; 123: 370-375
        • 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: 106-113
        • McWilliam A.
        • Khalifa J.
        • Vasquez Osorio E.
        • et al.
        Novel methodology to investigate the effect of radiation dose to heart substructures on overall survival.
        Int J Radiat Oncol Biol Phys. 2020; 108: 1073-1081
        • Green A.
        • Vasquez Osorio E.
        • Aznar M.C.
        • McWilliam A.
        • van Herk M.
        Image based data mining using per-voxel Cox regression.
        Front Oncol. 2020; 10: 1-11
        • Thor M.
        • Deasy J.O.
        • Hu C.
        • et al.
        Modeling the impact of cardiopulmonary irradiation on overall survival in NRG oncology trial RTOG 0617.
        Clin Cancer Res. 2020; 26: 4643-4650
        • Xue J.
        • Han C.
        • Jackson A.
        • et al.
        Doses of radiation to the pericardium, instead of heart, are significant for survival in patients with non-small cell lung cancer.
        Radiother Oncol. 2019; 133: 213-219
        • Vivekanandan S.
        • Landau D.B.
        • Counsell N.
        • 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: 51-60
        • Wong O.Y.
        • Yau V.
        • Kang J.
        • et al.
        Survival impact of cardiac dose following lung stereotactic body radiotherapy.
        Clin Lung Cancer. 2018; 19: e241-e246
        • Atkins K.M.
        • Chaunzwa T.L.
        • Lamba N.
        • et al.
        Association of left anterior descending coronary artery radiation dose with major adverse cardiac events and mortality in patients with non-small cell lung cancer.
        JAMA Oncol. 2020; 02115: 1-13
        • Dess R.T.
        • Sun Y.
        • Matuszak M.M.
        • 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: 1395-1402
        • Janssen-Heijnen M.L.G.
        • Schipper R.M.
        • Razenberg P.P.A.
        • Crommelin M.A.
        • Coebergh J.W.W.
        Prevalence of co-morbidity in lung cancer patients and its relationship with treatment: a population-based study.
        Lung Cancer. 1998; 21: 105-113
        • Nestle U.
        • Schimek-Jasch T.
        • Kremp S.
        • et al.
        Imaging-based target volume reduction in chemoradiotherapy for locally advanced non-small-cell lung cancer (PET-Plan): a multicentre, open-label, randomised, controlled trial.
        Lancet Oncol. 2020; 21: 581-592
        • Chen C.
        • Witte M.
        • Heemsbergen W.
        • Herk M.V.
        Multiple comparisons permutation test for image based data mining in radiotherapy.
        Radiat Oncol. 2013; 8: 1-10
        • Palma G.
        • Monti S.
        • Cella L.
        Voxel-based analysis in radiation oncology: a methodological cookbook.
        Phys Med. 2020; 69: 192-204
      1. SourceForge.net. Nifty Reg download. https://sourceforge.net/projects/niftyreg/. Accessed July 9, 2021.

        • Smith G.
        Step away from stepwise.
        J Big Data. 2018; 5
        • Stam B.
        • van der Bijl E.
        • van Diessen J.
        • et al.
        Heart dose associated with overall survival in locally advanced NSCLC patients treated with hypofractionated chemoradiotherapy.
        Radiother Oncol. 2017; 125: 62-65
        • Atkins K.M.
        • Bitterman D.S.
        • Chaunzwa T.L.
        • et al.
        Mean heart dose is an inadequate surrogate for left anterior descending coronary artery dose and the risk of major adverse cardiac events in lung cancer radiation therapy.
        Int J Radiat Oncol Biol Phys. 2021; 110: 1473-1479
        • Johnson-Hart C.
        • Price G.
        • McWilliam A.
        • Green A.
        • Faivre-Finn C.
        • van Herk M.
        Impact of small residual setup errors after image guidance on heart dose and survival in non-small cell lung cancer treated with curative-intent radiotherapy.
        Radiother Oncol. 2020; 152: 177-182
        • Ponikowski P.
        • Voors A.A.
        • Anker S.D.
        • et al.
        2016 ESC Guidelines for the diagnosis and treatment of acute and chronic heart failure: The Task Force for the diagnosis and treatment of acute and chronic heart failure of the European Society of Cardiology (ESC) developed with the special contribution of the Heart Failure Association (HFA) of the ESC.
        Eur Heart J. 2016; 37: 2129-2200
        • Wehner G.J.
        • Jing L.
        • Haggerty C.M.
        • et al.
        Routinely reported ejection fraction and mortality in clinical practice: where does the nadir of risk lie?.
        Eur Heart J. 2020; 41: 1249-1257
        • Paonessa J.R.
        • Brennan T.
        • Pimentel M.
        • Steinhaus D.
        • Feng M.
        • Celi L.A.
        Hyperdynamic left ventricular ejection fraction in the intensive care unit.
        Crit Care. 2015; 19: 288
        • Halliday B.P.
        • Senior R.
        • Pennell D.J.
        Assessing left ventricular systolic function: from ejection fraction to strain analysis.
        Eur Heart J. 2021; 42: 789-797
        • Pellikka P.A.
        • She L.
        • Holly T.A.
        • et al.
        Variability in ejection fraction measured by echocardiography, gated single-photon emission computed tomography, and cardiac magnetic resonance in patients with coronary artery disease and left ventricular dysfunction.
        JAMA Netw Open. 2018; 1e181456
        • Saiki H.
        • Moulay G.
        • Redfield M.
        • et al.
        Experimental cardiac radiation exposure induces ventricular diastolic dysfunction with preserved ejection fraction.
        Am J Physiol - Hear Circ Physiol. 2017; 313: H392-H407
        • Saiki H.
        • Petersen I.
        • Redfield M.
        • et al.
        Risk of heart failure with preserved ejection fraction in older women after contemporary radiotherapy for breast cancer.
        Circulation. 2017; 135: 1388-1396
        • Gong Y.
        • Gore E.M.
        • Bar-Ad V.
        • et al.
        Variation of cardiac contours using different heart definitions for NSCLC patients enrolled on RTOG 0617.
        Int J Radiat Oncol Biol Phys. 2014; 90: S739
        • Palma G.
        • Monti S.
        • Pacelli R.
        • et al.
        Radiation pneumonitis in thoracic cancer patients: multi-center voxel-based analysis.
        Cancers (Basel). 2021; 13: 1-17
        • Sun F.
        • Franks K.
        • Murray L.
        • et al.
        Cardiovascular mortality and morbidity following radical radiotherapy for lung cancer: is cardiovascular death under-reported?.
        Lung Cancer. 2020; 146: 1-5
        • Vasquez Osorio E.
        • McCallum H.
        • Cobben D.
        • et al.
        Protecting the heart: A practical approach to account for the full extent of heart motion in radiation therapy planning.
        Int J Radiat Oncol Biol Phys. 2020; 108: 1082-1090
        • Price G.
        • Devaney S.
        • French D.P.
        • et al.
        Can real-world data and rapid learning drive improvements in lung cancer survival? The RAPID-RT study.
        Clin Oncol (R Coll Radiol). 2022; 34: 407-410