Factors impacting survival after transarterial radioembolization in patients with hepatocellular carcinoma: Results from the prospective CIRT study

Background & Aims Transarterial radioembolization (TARE) with Yttrium-90 resin microspheres is an established treatment option for patients with hepatocellular carcinoma (HCC). However, optimising treatment application and patient selection remains challenging. We report here on the effectiveness, safety and prognostic factors, including dosing methods, associated with TARE for HCC in the prospective observational CIRT study. Methods We analysed 422 patients with HCC enrolled between Jan 2015 and Dec 2017, with follow-up visits every 3 months for up to 24 months after first TARE. Patient characteristics and treatment-related data were collected at baseline; adverse events and time-to-event data (overall survival [OS], progression-free survival [PFS] and hepatic PFS) were collected at every 3-month follow-up visit. We used the multivariable Cox proportional hazard model and propensity score matching to identify independent prognostic factors for effectiveness outcomes. Results The median OS was 16.5 months, the median PFS was 6.1 months, and the median hepatic PFS was 6.7 months. Partition model dosimetry resulted in improved OS compared to body surface area calculations on multivariable analysis (hazard ratio 0.65; 95% CI 0.46-0.92; p = 0.0144), which was confirmed in the exact matching propensity score analysis (hazard ratio 0.56; 95% CI 0.35-0.89; p = 0.0136). Other independent prognostic factors for OS were ECOG-performance status >0 (p = 0.0018), presence of ascites (p = 0.0152), right-sided tumours (p = 0.0002), the presence of portal vein thrombosis (p = 0.0378) and main portal vein thrombosis (p = 0.0028), ALBI grade 2 (p = 0.0043) and 3 (p = 0.0014). Adverse events were recorded in 36.7% of patients, with 9.7% of patients experiencing grade 3 or higher adverse events. Conclusions This large prospective observational dataset shows that TARE is an effective and safe treatment in patients with HCC. Using partition model dosimetry was associated with a significant improvement in survival outcomes. Impact and implications Transarterial radioembolization (TARE) is a form of localised radiation therapy and is a potential treatment option for primary liver cancer. We observed how TARE was used in real-life clinical practice in various European countries and if any factors predict how well the treatment performs. We found that when a more complex but personalised method to calculate the applied radiation activity was used, the patient responded better than when a more generic method was used. Furthermore, we identified that general patient health, ascites and liver function can predict outcomes after TARE. Clinical trial number NCT02305459.


Introduction
HCC is poor, with a life expectancy of 6-38 months, depending on the Barcelona Clinic Liver Cancer (BCLC) stage. 3 Only a minority of patients are eligible for curative-intent treatments, including surgical resection, liver transplantation, and ablative therapies. [4][5][6][7][8][9] In intermediate stages, transcatheter arterial chemoembolization (TACE) is standard of care; systemic treatments such as sorafenib, lenvatinib, and a combination of atezolizumab and bevacizumab have been approved for the first-line medical treatment of advanced and metastatic HCC based on convincing phase III trials. [10][11][12][13] Guidelines for the treatment of HCC also propose transarterial radioembolization (TARE, also known as selective internal radiation therapy [SIRT]) as an optional treatment modality for patients with liver dominant disease not eligible for surgical or ablative therapies, or who experienced no response, significant side effects or intolerance when treated with systemic therapies. 4,6-9 TARE is an interventional therapeutic procedure that involves the targeted delivery of high doses of radiation to liver tumours via the hepatic artery. Several studies have shown that TARE has a favourable safety profile and displays promising results in terms of local tumour control in patients with unresectable HCC limited to the liver in the intermediate and advanced stages. [14][15][16][17][18] Despite this, recent randomised controlled trials on TARE in HCC showed that compared to sorafenib alone, TARE or TARE plus sorafenib as a first-line treatment option for patients with unresectable HCC did not improve overall survival (OS) or progression-free survival (PFS) in these patient cohorts. 19,20 While questions have been raised regarding discrepancies in patient inclusion and site experience in administering TARE, which may have influenced outcomes, 21 recent research into dosimetry methods suggests that improving dose calculation and delivery could improve survival outcomes. 22 The prospective randomised DOSISPHERE-01 trial showed significantly better overall survival results with a personalised dosimetry model than the standard dose calculation model using glass microspheres in patients with unresectable, locally advanced HCC. 23,24 This suggests that further optimising selection of patients, treatment application and the dosimetry models may improve survival outcomes of patients with HCC.
The Cardiovascular and Interventional Radiological Society of Europe (CIRSE) initiated a European-wide observational study on the clinical application and outcomes of TARE with Y90 resin microspheres (SIR-Spheres® Y-90 resin microspheres, Sirtex Medical Pty Limited; St. Leonards, NSW, Australia). The study (NCT02305459) was open to all indications and recruited one of the largest cohorts on TARE in liver malignancies to date. 25 The objective of the current subgroup analysis was to investigate factors influencing survival in patients with HCC treated with TARE, including the effect that methods to calculate the prescribed dose have on effectiveness outcomes. The primary endpoint was OS, while secondary endpoints were PFS, PFS in the liver only (hepatic PFS [hPFS]), safety, and identification of potential prognostic survival factors, including an evaluation of the impact of methods to calculate the prescribed activity on survival outcomes.

Study design
We analysed 422 patients with HCC collected in the CIRSE Registry for SIR-Spheres Therapy (CIRT) study. CIRT is a prospective, single device, multi-centre observational study of patients with primary and metastatic hepatic malignancies treated with TARE using Y90 resin microspheres as the standard of care. The CIRT methodology was published by Helmberger et al. 26 Sites were invited to participate if they had at least 40 TARE cases and 10 cases in 12 months prior to invitation. In total, 27 participating sites in eight countries were identified and enrolled from April 2014 until April 2017, of which 25 sites included patients with HCC. 25 Data was collected using a customised electronic data capturing system and electronic case report form that was developed by ConexSys Inc (Lincoln, RI, United States) and hosted on a local secure server in Vienna, Austria maintained by ITEA (Vienna, Austria). Statistical analyses were performed in SAS 9.4 (SAS Institute, Cary, NC, USA) and RStudio under R4.0.0 (R Foundation, Vienna, Austria, Supplementary CTAT Table).

Patient selection
Patients included in the analysis were adults diagnosed with HCC and scheduled to receive TARE with Y90 resin microspheres. There were no specific inclusion or exclusion criteria. The indication for TARE, the treatment design, the methods used for dose calculation and the follow-up regimen were based on the centres' internal standards. Participating sites contractually agreed to include all eligible patients consecutively. All included patients signed an informed consent form. This research project was performed in accordance with the ethical standards of the applicable institutional and/or national ethics committees and with the 1964 Helsinki declaration and its later amendments or comparable ethical standards.
Patient recruitment took place between 1 January 2015 and 31 December 2017. Follow-up data were collected until 31 December 2019. Sites were requested to follow-up with the patient every 3 months up to 24 months after the first TARE treatment. In addition, sites were encouraged to obtain followup information from referring physicians if follow-up evaluations were not performed at the site of the TARE treatment.

Assessments
At the time of first treatment, baseline data, demographics and treatment-related data were collected. Information concerning post-TARE treatments, safety data and time-to-event data were gathered at every follow-up visit. Time-to-event was defined from the date of the first TARE treatment until the date of the event. Liver function was described using the albumin-bilirubin (ALBI) formula developed by Johnson et al.: ALBI score = (log 10 bilirubin [lmol/L] × 0.66) + (albumin [g/L] × −0.0852). ALBI score < − −2.60 is grade 1, >−2.60 to < − −1.39 is grade 2, and >−1.39 is grade 3. 27 BCLC classifications were determined at the sites, but all classifications were evaluated according to the uniform BCLC staging standards set out by Reig et al. in the recent (2022) update. 28 Where necessary, patients were re-classified. Information on whether portal vein thrombosis (PVT) was malignant was not collected, but lobar and main PVT were considered malignant, while segmental PVT was considered malignant if the site classified the patient as BCLC C. Safety outcomes are described as severe day-of-treatment complications and occurrences of any adverse events after treatment, according to the Common Terminology Criteria for Adverse Events, version 4.03. Pre-defined serious adverse events (grade 3 and 4) were abdominal pain, fatigue, fever, nausea, vomiting, gastrointestinal ulceration, gastritis, radiation cholecystitis, radiation pancreatitis and radioembolization-induced liver disease (REILD). An open text field allowed us to collect details on other serious adverse events.

Statistical analysis
Data are presented as mean ± SD or median (IQR) for continuous variables and number (%) for categorical variables. Percentages are based on the whole cohort (N = 422) unless otherwise indicated. Patients who died during the study were categorized as having progression for the purpose of PFS and hPFS analysis. Patients alive and progression-free were censored on the day of last follow-up. The simultaneous occurrence of hepatic progression and extrahepatic progression was considered as hepatic progression.
Comparisons between groups were performed using the logrank test (Mantel-Haenszel version). The median OS, PFS and hPFS times were calculated with their associated 95% CIs. The group effect was calculated with a Cox proportional-hazards model with hazard ratio (HR) and 95% CIs. The development over time of the ALBI, bilirubin, albumin and international normalized ratio (INR) values were explored using a linear mixed model.
A multivariable analysis for OS, PFS and hPFS was performed using a Cox proportional-hazards model whereby the selection of variables was determined following a univariable analysis and a stepwise variable selection procedure, with a significance level of 0.2 used to determine whether to enter a predictor into the stepwise model. The model with the lowest Akaike information criterion value was considered the final model. All available data were used, and no imputations of missing data were made.
Additional analyses were performed to evaluate the impact on OS, PFS and hPFS of the two main methods to calculate the prescribed Y90 activity: partition model and (modified) body surface area ([m]BSA) methodology. For the comparison of partition model dosimetry (n = 177) with BSA/mBSA (n = 245), we considered a locally modified version of the partition model (n = 3) and voxel-based dosimetry (n = 1) as following the partition model. To compare the two groups, a propensity score analysis was performed. The propensity score is the probability of treatment assignment conditional on measured baseline covariates. Two approaches were used for the propensity score: 1. Matching: Greedy nearest neighbour matching within a calliper of 0.2 of the propensity score was used. Using this approach, a patient treated with partition model dosimetry is selected. This treated patient is then matched with a patient treated based on the BSA activity calculation, whose propensity score is closest to that of the treated patient, subject to the constraint that the differences between their propensity scores are less than a specified maximum (the calliper distance). To estimate the marginal treatment effect for OS, PFS and hPFS, a Cox model with a robust variance estimator that accounts for clustering within matched pairs was used. 2. Inverse probability of treatment weighting (IPTW): IPTW using the propensity score uses weights based on the propensity score to create a synthetic sample in which the distribution of measured baseline covariates is independent of treatment assignment.
To obtain appropriate estimates of variance, stabilised weights were used. For each patient, the stabilised weight is calculated by multiplying his or her original weight by the proportion of patients who received the treatment that he or she received. A Cox model, adjusted for stabilised weights, was used to estimate the relative treatment effect for OS, PFS, and hPFS.
A standardised difference between the two groups for each patient characteristic of interest was calculated to assess whether the covariates are well balanced between the partition model and BSA/mBSA models. The balance between groups was achieved if the magnitude of the standardised difference was less than 0.25.

Patient demographics
Four hundred and twenty-two patients with HCC from 25 centres in eight countries were included in this study (  Table S1. An increase in ALBI, bilirubin and INR, and a decrease in albumin values was observed 3 months after treatment (p <0.0001, Fig. S1).

Effectiveness
The median OS was 16.5 months (95% CI 14.2-19.3), median PFS was 6.1 months (95% CI 5.7-7.0), and median hPFS was 6.7 months for the entire population (95% CI 5.9-7.6). Survival was highest in patients with BCLC A (41.4 months; 95% CI 22.5-ND; p <0.0001) ( Table 3). The subgroup of patients with ALBI grade 1 lived longer (21.1 months; 95% CI 19.2-28.8; p <0.0001) than those with higher ALBI grades (grade 2: 14.0 months; 95% CI 11.5-16.5; p = 0.0005; grade 3: 7.8 months; 95% CI 2.7-12.9; p <0.0001). The number of tumour nodules, tumour location, presence of extrahepatic metastases, and PVT were also associated with survival (Table 3), as well as dose methodology (Fig. 1   Comparing outcomes between partition model dosimetry and BSA/mBSA To further evaluate the differences in survival outcomes between partition model dosimetry and BSA/mBSA in the univariable analysis and the multivariable analysis for OS, we used propensity score matching to evaluate differences between the patient groups for whom the determination of the prescribed Y90 activity was performed either by BSA/mBSA or the partition model. The covariates considered in the model were based on the outcomes of the multivariable analysis: cirrhosis, ascites, number of tumour nodules, bilobar, right or left-sided tumours, PVT, and ALBI grade (Table S6, see Table S7 for a comparison of all baseline values in the two groups  (Table S8).
Finally, comparing the prescribed activity between the partition model and BSA/mBSA revealed no significant differences when adjusted for tumour burden (Fig. S2) and the number of tumour nodules (Fig. S3).  (Table S9).

Discussion
The HCC cohort collected in the CIRT study is one of the largest prospectively collected cohorts on the use of TARE in Europe. Despite the heterogeneous patient population, the multivariable analysis found that, compared with BSA, the partition model was a predictor of improved OS, but not PFS and hPFS. Additional propensity score matching, using the exact matching model and the IPTW model, found that patients whose prescribed activity was calculated with the partition model had better OS and PFS when compared with patients with similar baseline characteristics, but whose activity was prescribed based on BSA. Additionally, the IPTW model found an improved hPFS following the partition model.    ECOG >0, right-sided tumours, presence of ascites and PVT, and ALBI >1. The median overall survival of our entire cohort of 422 patients was 16.5 months. This is in the range of 12.8-20.5 months reported by prior studies in similar real-life settings. 15,17,18,29 On the other hand, in the TARE arms of two prospective studies with randomisation against Sorafenib (SIRveNIB and SARAH), the median OS was only 8.0 and 8.8 months, respectively. 19,20 An important reason for the differences in survival is the selection of patients, e.g. while only one-third of our patients were classified as BCLC C, the randomised studies recruited a much higher percentage of patients in this advanced group (68% in the SARAH study and 48.4% in the SIRveNIB study in the intention-to-treat population). Finally, our study used Y90 resin microspheres instead of glass microspheres, but a comparison of both spheres in consecutive patients in a single-centre study (resin, n = 41; glass, n = 36) revealed no difference in survival between groups. 22 In line with previous study outcomes, our multivariable analyses showed that ECOG >0, presence of ascites and extrahepatic disease were negative prognostic factors for OS. [30][31][32] Our data also confirms previous findings that the extent of PVT influences OS, 31,[33][34][35] although in our analysis, lobar PVT was found to only influence hPFS, while segmental PVT and main PVT were found to impact OS. Our study further establishes that the ALBI grade was a strong independent predictor of OS, mirroring previous findings 27, 36 and strengthening the justification for its inclusion in the recent BCLC strategy. 28 Independent prognostic factors associated with OS identified in previous studies were albumin, alpha-fetoprotein, alkaline phosphatase and tumour size <5 cm. 22,29,31,32,37 The BCLC stage has also been identified as a prognostic factor for survival outcomes in several prior studies. [37][38][39] In our cohort, BCLC staging was a significant predictor of OS, PFS, and hPFS in the univariable analysis, but it was only a predictor for PFS in the multivariable analysis. A possible explanation is that BCLC is a composite variable consisting of variables that were found to be independent predictors, such as ECOG, PVT, and extrahepatic disease, and may thus not be independent from these variables. Additionally, a recent comparison of prognostic scoring systems in a cohort of patients receiving TARE ranked BCLC lower than other prognostic factors for this treatment. 28 The apparent difference in survival based on tumour location in the right vs. the left liver lobe found in our cohort may reflect the complexity and variation of blood supply to the liver, as recently described by Choi et al. 40 The variations in the blood supply of the left liver lobe may require a more meticulous positioning of suited microcatheters to ensure a consistent and robust dose distribution compared to the right liver lobe, explaining differences in outcome if not considered thoroughly. [41][42][43] In terms of safety and toxicity, our cohort confirms previous reports on the favourable safety outcomes of TARE. 15,29,33,37,44 We observed a worsening of the liver function after TARE in terms of INR, bilirubin and albumin values (and therefore ALBI score), which mirrors the results of the SORAMIC randomised controlled trial, where in the TARE + sorafenib group, poorer ALBI scores after 4 and 6 months were observed compared to the sorafenib alone group. 45 Our study reported that 1.4% of patients experienced REILD, which was grade 3 or higher in half of affected patients. This occurrence of REILD is on the lower end of the studies used in the systematic review by Braat et al., who identified that the incidence of symptomatic REILD varied between 0 and 31%, although, in most reports, the incidence was 0-8%. 46 In our cohort, the multivariable analysis and the propensity score analyses showed that the partition model activity calculation led to better OS outcomes compared to BSA and mBSA, and improved PFS and hPFS in the propensity score analysis. The BSA and mBSA methods rely on an assumed correlation between BSA and the tumour burden to estimate Y90 activity. Ignoring the variability of the tumour-to-normal-liver ratio in individual patients, it sacrifices accuracy for simplicity, 47 and may result in wide variations of radiation dose absorbed by both the tumour and the surrounding non-tumoural liver parenchyma. 48,49 On the other hand, personalised dosimetry such as partition model relies on differential tumour-to-non-tumour perfusion evaluated on pretreatment Technetium-99m-macroaggregated albumin singlephoton emission computer tomography combined with computer tomography to predict dose distributions between the "partitions" tumoural liver, non-tumoural liver, and lung. It has been demonstrated that personalised dosimetry models can increase the tumour-absorbed dose while keeping the dose in the non-tumoural liver and the lung low. 22 However, compared to the BSA model, these models are very resource-intensive and require a good collaboration between nuclear medicine physicians and interventional radiologists. Nevertheless, the data presented in this study strongly suggests that partition model dosimetry improves OS, PFS and hPFS outcomes in patients with unresectable HCC. These outcomes reflect recent studies examining the relationship between tumour-absorbed dose and survival outcomes in patients with HCC.
The randomised phase II DOSISPHERE-01 trial comparing patients with unresectable locally advanced HCC receiving personalised dosimetry with standard dosimetry showed that objective response was achieved in 20/28 patients (71%; 95% CI 51-87) in the personalised dosimetry group vs. 10/28 (36%; 95% CI 19-56) in the standard dosimetry group (p = 0.0074). This translated into a median OS of 26.6 months (95% CI 11.7-NR) in the personalised dosimetry group compared to 10.7 months (95% CI 6.0-16.8) in the standard dosimetry group. Furthermore, patients who received a tumour dose of 205 Gy or higher had an OS of 26.6 months (95% CI 13.5-NR) compared to 7.1 months (95% CI 4.6-14.8) in those that received a tumour dose of less than 205 Gy (HR 0.33; 95% CI 0.15-0.71; p = 0.0029). 23 Of note is that to achieve a high tumourabsorbed dose without increasing the dose absorbed by the nontumoural liver, the DOSISPHERE-01 trial included only patients with tumours showing arterial phase hyperenhancement. Additionally, a secondary analysis of 120 patients from the SARAH study showed that participants who received at least 100 Gy (n = 67) had longer OS than those who received less than 100 Gy (median, 14.1 months [95% CI 9.6-18.6] vs. 6.1 months [95% CI 4.9-6.8], respectively; p = 0.001). In the patient group that was available for response analysis (n = 109), tumour radiation-absorbed dose was higher in patients with disease control vs. those with progressive disease (median, 121 Gy [IQR 86-190 Gy] vs. 85 Gy [IQR 58-164 Gy]; p = 0.02). 50 Unfortunately, the present study did not collect any data on the tumour target dose or radiationabsorbed dose and thus cannot provide any suggestions on optimal dosage. Analysis of differences in prescribed activity between BSA/mBSA and the partition model found no significant differences, when adjusted for tumour volume, percentage of tumour activity and number of tumour nodules. Nevertheless, the multivariable analysis and the propensity score matching suggest that patients whose dosages were calculated with the partition model generally performed better than patients whose dosage was calculated with BSA or mBSA, which is in line with the aforementioned studies. Our study adds to the findings from DOSISPHERE-01 and SARAH, by showing that personalised dosimetry methods also improve effectiveness outcomes in a reallife clinical context with Y90 resin microspheres, compared to activity calculation methods based on BSA, irrespective of a site's experience.
A limitation of the study is the observational study design, whereby important confounding factors may not have been accounted for. The heterogeneity of the patient population reflects the real-life clinical practice in participating centres and thus its diversity in patient selection and clinical outcomes. We used the propensity score method and multivariable analysis to alleviate, to some degree, the effect of this heterogeneity and multiple methods of analysis were used to show the similarity of outcomes despite the differences in analysis. Of course, certain confounding factors which were not considered in these methods could have contributed to the outcomes and should be considered when interpreting the results.
The study was designed to explore the clinical outcomes of TARE and therefore focused less on dosimetry-specific data. This means that retrospectively important data points such as precise administered activity and tumour-absorbed dose were not included in the evaluation at the time of study design. Furthermore, as an observational study, the design was non-prescriptive for tumour response assessment, which was performed with various criteria (e.g., RECIST, modified RECIST or PET Response Criteria in Solid Tumours) according to local practice and expertise of centres. This prevented us from including tumour response in the analysis.
We attempted to collect quality-of-life data from patients on a voluntary basis at the time of treatment and at every follow-up visit until study exit. The relevance of the collected dataset is currently being evaluated. The relatively high number of patients lost to follow-up can introduce bias regarding the interpretation of OS, and imprecise follow-up imaging intervals should be considered when interpreting PFS and hPFS. A potential explanation might be that TARE requires a comprehensive infrastructure with patients being referred to specialised centres for the treatment while being followed up by their local physician. In those cases, sites were encouraged to obtain follow-up information by contacting the referring physician. If this was not possible, the patient was considered as lost to follow-up. Selection bias can be expected in an observational study and regular remote monitoring was performed to verify that all eligible patients were included. Remote monitoring was done to improve data quality; however, no source data verification was performed.
This large prospective observational data set suggests that TARE with resin Y90 microspheres has a favourable toxicity profile and that patients with good liver function and no extrahepatic disease are ideal candidates for this therapy. Furthermore, our data revealed that optimising the application of the therapy by using the partition model instead of BSA models, can significantly improve survival outcomes. It is thus recommended that activity calculations with the partition model are considered when designing future randomised controlled trials on TARE.

Financial support
The CIRT study was funded by an independent investigator-initiated research grant from SIRTEX Medical Europe GmbH (Bonn, Germany). CIRSE, the Cardiovascular and Interventional Radiological Society of Europe, is responsible for the independent execution of the CIRT study and has sole ownership of the data.

Data availability statement
Data access is limited by ethical and regulatory considerations. chairperson of the CIRT Steering Committee between 2015 and 2018 and is now enjoying his well-deserved retirement. CIRSE, the Cardiovascular and Interventional Radiological Society of Europe, is responsible for the independent execution of the CIRT study and has sole ownership of the data. The electronic data capturing system was developed and supported by ConexSys Inc (Lincoln, RI, United States). ITEA GmbH (Vienna, Austria) developed and maintained the necessary infrastructure and a customised data management system was designed by Joaquin Padilla Montani, Vienna, Austria. Finally, the authors thank the CIRSE Central Office and the CIRSE Clinical Research department staff for their support during the design and setup of the study and drafting of the manuscript.

Supplemental information
Factors impacting survival after transarterial radioembolization in patients with hepatocellular carcinoma: Results from the prospective CIRT study  Levels of significance: p <0.05 (Wilcoxon-Mann-Whitney-Test). We compared the prescribed activity (GBq) between patients whose prescribed activity was calculated with BSA or mBSA and partition model. Patients treated with the partition model generally had smaller tumours (A, in cc) and livers (B) but the percentage of the liver affected by tumour (C) was similar as patients treated with BSA/mBSA. The prescribed activity using BSA/mBSA was significantly higher than using partition model (D, p=0.003), but adjusted for tumour volume (E) or percentage of affected liver (F), no significant difference in prescribed dose was found. Levels of significance: p <0.05 (Wilcoxon-Mann-Whitney-Test). We compared the prescribed activity (GBq) between patients whose prescribed activity was calculated with BSA or mBSA and partition model. Adjusting for number of tumour nodules, we found no significant differences between the prescribed activity using BSA/mBSA or the partition model for         Table S8. The effect of centre expertise in survival outcomes between the BSA/mBSA and the partition model cohorts.
The hypothesis is that centres that perform partition model dosimetry are generally better-performing hospitals, because partition model dosimetry requires a certain level of expertise and infrastructure dedicated to TARE. If this hypothesis would be true, we would expect better OS/PFS/hPFS outcomes of the BSA/mBSA patients that are treated in those hospitals.
In the below table we compared the survival outcomes of the following three groups: 1) partition model dosimetry patients; 2) dose calculation with BSA/mBSA for patients that were treated in sites that also performed partition model dosimetry; and 3) dose calculation with BSA/mBSA for patients that were treated in sites that did not perform partition model dosimetry.