Review Reports - [18F]FDG PET/CT-Derived Metabolic Characteristics and Tumour-Infiltrating Lymphocytes for Predicting Axillary Response to Neoadjuvant Chemo-(Immuno)Therapy in Clinically Node-Positive Breast Cancer

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

The predictive model requires a more rigorous assessment of its performance. The reported AUC of 0.80 is based on the same cohort used to develop the multivariable logistic regression model. Please clarify whether any form of internal validation (e.g., bootstrap resampling or cross-validation) was performed. If not, the authors should acknowledge that the reported AUC may represent apparent rather than validated model performance and should avoid implying that the model is ready for clinical application.
The variable-selection strategy should be described and justified in greater detail. Clinicopathological variables were selected for the multivariable model based on significant associations in univariable analyses, while SLR and TILs were included based on predefined clinical relevance. This data-driven selection approach may introduce model instability and should be discussed, particularly given the relatively modest number of axillary pCR events (46 patients).
Gharooni, M., Alikhani, A., Moghtaderi, H., Abiri, H., Mashaghi, A., Abbasvandi, F., Khayamian, M.A., Miripour, Z.S., Zandi, A., & Abdolahad, M. (2019). Bioelectronics of the cellular cytoskeleton: Monitoring cytoskeletal conductance variation for sensing drug resistance. ACS Sensors, 4(2), 353–362. https://doi.org/10.1021/acssensors.8b01142.

This study is relevant because it demonstrated that alterations in the cytoskeletal components of cancer cells, particularly actin and microtubules, can produce measurable changes in their electrical properties. The authors used electrical measurements to characterize cellular responses to cytoskeleton-targeting agents and demonstrated the potential of bioelectronic measurements for assessing cancer-cell drug responses and resistance. Therefore, this reference could provide useful context for the present findings and help place the reported observations within the broader field of cancer bioelectronics and electrical characterization of cellular phenotypes.

Author Response

Dear reviewer, thank you for your positive feedback.

Comment 1: The predictive model requires a more rigorous assessment of its performance. The reported AUC of 0.80 is based on the same cohort used to develop the multivariable logistic regression model. Please clarify whether any form of internal validation (e.g., bootstrap resampling or cross-validation) was performed. If not, the authors should acknowledge that the reported AUC may represent apparent rather than validated model performance and should avoid implying that the model is ready for clinical application.

Response 1: We thank the reviewer for this important comment. We agree that the reported AUC of 0.80 represents apparent performance, as it was estimated in the same cohort used to develop the model, and that no internal or external validation was performed. We have revised the manuscript accordingly:

The abstract, results and discussion now explicitly refer to "apparent" performance. In addition, we have added the following limitation to the discussion: ‘Fifth, as no internal or external validation was performed and model calibration was not assessed, the reported AUCs reflect apparent performance in the development cohort. As a result, model performance may be overestimated, and generalizability to independent populations remains to be established. Sixth, variable selection was partly based on univariable significance, which may result in unstable models and overestimated effect sizes, particularly given the limited number of axillary pCR events (n=46). Together with the absence of internal and external validation, this underlines that the model should be considered exploratory and hypothesis-generating.’ Furthermore, we have revised the conclusion to avoid implying that the model is ready for clinical application: ‘In conclusion, this study found that the model combining clinicopathological characteristics, splenic metabolic activity, and breast tumour TILs may be informative for the axillary response to NAC, but did not perform better than clinicopathological characteristics alone. Increased metabolic activity of ALNMs was associated with high breast tumour TILs, while an inverse relationship with splenic metabolic activity was observed. Validation in independent cohorts is required before any clinical application can be considered, and future studies should evaluate whether sufficiently accurate prediction models can support safe treatment de-escalation.’

Comment 2: The variable-selection strategy should be described and justified in greater detail. Clinicopathological variables were selected for the multivariable model based on significant associations in univariable analyses, while SLR and TILs were included based on predefined clinical relevance. This data-driven selection approach may introduce model instability and should be discussed, particularly given the relatively modest number of axillary pCR events (46 patients).

Response 2: We thank the reviewer for this important comment. We agree that selecting variables based on univariable significance is a data-driven approach that may lead to model instability and overfitting, particularly given the modest number of axillary pCR events (n = 46). We have expanded the description of the variable-selection strategy in the Methods section as follows: ‘The splenic metabolic activity and TILs were included in the multivariable logistic regression model given their predefined clinical relevance and the aim of the study. Additional clinicopathological variables showing a significant association in univariable regression (p<0.05) were selected for the multivariable logistic regression model, while taking into account recommendations regarding the number of events per variable (EPV) to limit model overfitting [32,33].’

Splenic metabolic activity (SLR) and TILs were included a priori based on their predefined clinical relevance and the study's aim, independent of their statistical significance. Additional clinicopathological variables were selected if they showed a significant association in univariable analysis, and the number of variables in the multivariable model was restricted in line with recommendations on events per variable (EPV) to limit overfitting. We acknowledge that this approach remains exploratory and that the resulting model may be unstable. We have therefore added this as a limitation to the Discussion: ‘Sixth, variable selection was partly based on univariable significance, which may result in unstable models and overestimated effect sizes, particularly given the limited number of axillary pCR events (n=46). Together with the absence of internal and external validation, this underlines that the model should be considered exploratory and hypothesis-generating.’

Comment 3: Gharooni, M., Alikhani, A., Moghtaderi, H., Abiri, H., Mashaghi, A., Abbasvandi, F., Khayamian, M.A., Miripour, Z.S., Zandi, A., & Abdolahad, M. (2019). Bioelectronics of the cellular cytoskeleton: Monitoring cytoskeletal conductance variation for sensing drug resistance. ACS Sensors, 4(2), 353–362. https://doi.org/10.1021/acssensors.8b01142.

This study is relevant because it demonstrated that alterations in the cytoskeletal components of cancer cells, particularly actin and microtubules, can produce measurable changes in their electrical properties. The authors used electrical measurements to characterize cellular responses to cytoskeleton-targeting agents and demonstrated the potential of bioelectronic measurements for assessing cancer-cell drug responses and resistance. Therefore, this reference could provide useful context for the present findings and help place the reported observations within the broader field of cancer bioelectronics and electrical characterization of cellular phenotypes.

Response 3: We thank the reviewer for suggesting this reference. We have carefully considered the study by Gharooni et al. However, this work focuses on bioelectronic measurements of cytoskeletal conductance in cancer cell lines as a tool to sense drug resistance in vitro. Our study is a clinical, retrospective cohort study investigating whether [¹⁸F]FDG PET/CT-derived metabolic characteristics, tumour-infiltrating lymphocytes, and clinicopathological variables can predict axillary pathologic response after neoadjuvant systemic therapy in patients with clinically node-positive breast cancer. As the topic, methodology, and outcome measures of the cited study differ substantially from ours, we feel that it would not provide relevant context for our findings and have therefore not included it in the revised manuscript. We would be happy to consider any other references the reviewer feels are more directly related to our study.

Please see the attachment for the revised manuscript.

Reviewer 2 Report

Comments and Suggestions for Authors

The topic of the manuscript is interesting and relevant, and it fits well within the aims and scope of the journal. The study addresses an important area; however, some aspects could be further strengthened to enhance its scientific clarity and methodological rigor.

The following suggestions are offered for the authors’ consideration:

Added value of PET/TIL variables
The authors should clarify the additional predictive value of the PET-derived variables and TILs. The model including tumour grade and breast cancer subtype already seems to have a very similar AUC to the full model. It would therefore be useful to show whether adding SLRmax and TILs results in a meaningful improvement in model performance.
Validation of the prediction model
The prediction model appears to have been developed and evaluated in the same dataset. Some form of internal validation, for example bootstrapping or cross-validation, would strengthen the analysis. Calibration should also be reported if the model is intended to have potential clinical use.
Missing data and patient selection
The authors should provide more detail on missing data and on the number of patients included in each multivariable analysis. In addition, only 135 of the 214 RISAS patients were included in the present study. The possible effect of this selection on the results should be discussed more clearly.
Interpretation of TILs and clinical implications
Some conclusions regarding TILs and the possible omission of axillary surgery seem somewhat strong. Breast tumour TILs were not independently associated with axillary pCR in the multivariable analysis, and the model has not yet been externally validated. I would therefore suggest a more cautious interpretation of these findings and of their potential use for treatment de-escalation.

Author Response

Dear reviewer, thank you for your positive feedback.

Comment 1: Added value of PET/TIL variables
The authors should clarify the additional predictive value of the PET-derived variables and TILs. The model including tumour grade and breast cancer subtype already seems to have a very similar AUC to the full model. It would therefore be useful to show whether adding SLRmax and TILs results in a meaningful improvement in model performance.

Response 1: We thank the reviewer for this important comment and agree that the difference in predictive value of SLRmax and TILs beyond breast cancer subtype and tumour grade should be addressed more clearly. The apparent performance of the model including only tumour grade and breast cancer subtype was AUC 0.79 [95% CI 0.71-0.87]). Adding SLRmax and breast tumour TILs resulted in an apparent AUC of 0.80 [95% CI 0.73–0.88]. This difference is minimal, and therefore we have revised the fourth paragraph of the Results as follows: ‘Results of univariable and multivariable logistic regression analyses of factors associated with axillary pCR are presented in Table 4. Multivariable logistic regression combining tumour grade (grade 3 vs. 1-2; OR 2.19 [95% CI 0.87-5.55], p=0.097), breast cancer subtype (Ref HR+/HER2-; HR+/HER2+ OR 10.38 [95% CI 2.95-36.46], p<0.001; HR-/HER2+ OR 13.66 [95% CI 2.84-65.75], p=0.001; TN OR 5.13 [95% CI 1.44-18.29], p=0.012), SLRmax (OR 0.01 [95% CI 0.00-2.74], p=0.111), and breast tumour TILs (OR 1.25 [95% CI 0.45-3.48], p=0.666), resulted in an apparent AUC of 0.80 [95% CI 0.73-0.88]; Figure 3). In comparison, the multivariable model including only SLRmax and breast tumour TILs resulted in an apparent AUC of 0.63 [95% CI 0.53-0.73], while the multivariable model including only breast cancer subtype and tumour grade resulted in an apparent AUC of 0.79 [95% CI 0.71-0.87].’

In addition, we have revised the first paragraph of the Discussion as follows: ‘Patients with cN+ breast cancer who achieve axillary pCR after NAC may not benefit from any axillary surgery. Therefore, accurate non-invasive identification of patients achieving axillary pCR may allow omission of axillary surgery in the future. In this study, we explored the value of baseline [18F]FDG PET/CT-derived metabolic characteristics and TILs to predict axillary pCR in cN+ breast cancer patients treated with NAC by creating a multivariable model. The model combining [18F]FDG PET/CT-derived SLRmax and breast tumour TILs demonstrated moderate performance (apparent AUC 0.63 [95% CI 0.53-0.73]). A model including only the clinicopathological characteristics breast cancer subtype and tumour grade performed considerably better (apparent AUC 0.79 [95% CI 0.71-0.87], underscoring the importance of including tumour biology in response prediction. Adding SLRmax and breast tumour TILs to this clinicopathological model yielded a similar performance (apparent AUC 0.80 [95% CI 0.73-0.88]), so an incremental predictive value of these biomarkers could not be demonstrated in our cohort. Second, the association between [18F]FDG PET/CT-derived metabolic characteristics and TILs revealed higher ALNM metabolic activity and lower splenic metabolic activity in patients with high breast tumour TILs.’

The conclusion was revised as follows: ‘In conclusion, this study found that the model combining clinicopathological characteristics, splenic metabolic activity, and breast tumour TILs may be informative for the axillary response to NAC, but did not perform better than clinicopathological characteristics alone. Increased metabolic activity of ALNMs was associated with high breast tumour TILs, while an inverse relationship with splenic metabolic activity was observed. Validation in independent cohorts is required before any clinical application can be considered, and future studies should evaluate whether sufficiently accurate prediction models can support safe treatment de-escalation.’

In addition, the abstract conclusion was adapted accordingly: ‘A model combining clinicopathological, metabolic, and immune-related characteristics may be informative for predicting axillary response to NAC, but did not perform better than clinicopathological characteristics alone. Future studies should evaluate whether accurate prediction models can support safe axillary treatment de-escalation.’

Comment 2: Validation of the prediction model
The prediction model appears to have been developed and evaluated in the same dataset. Some form of internal validation, for example bootstrapping or cross-validation, would strengthen the analysis. Calibration should also be reported if the model is intended to have potential clinical use.

Response 2: We thank the reviewer for this important comment and agree that the model was developed and evaluated in the same dataset, and that the reported AUCs therefore represent apparent rather than validated performance. Given the limited number of axillary pCR events (n = 46), we considered that internal validation would be unstable, and, as no independent cohort was available, external validation was not feasible. Accordingly, we have revised the manuscript to consistently refer to "apparent" performance. In addition, we have explicitly acknowledged the absence of internal and external validation and calibration assessment as a limitation, as follows: ‘Fifth, as no internal or external validation was performed and model calibration was not assessed, the reported AUCs reflect apparent performance in the development cohort. As a result, model performance may be overestimated, and generalizability to independent populations remains to be established. Sixth, variable selection was partly based on univariable significance, which may result in unstable models and overestimated effect sizes, particularly given the limited number of axillary pCR events (n=46). Together with the absence of internal and external validation, this underlines that the model should be considered exploratory and hypothesis-generating.’ Furthermore, we have revised the Conclusion to state that validation in independent cohorts is required before any clinical application can be considered: ‘In conclusion, this study found that the model combining clinicopathological characteristics, splenic metabolic activity, and breast tumour TILs may be informative of for the axillary response to NAC, but did not perform better than clinicopathological characteristics alone. Increased metabolic activity of ALNMs was associated with high breast tumour TILs, while an inverse relationship with splenic metabolic activity was observed. Validation in independent cohorts is required before any clinical application can be considered, and future studies should evaluate whether sufficiently accurate prediction models can support safe treatment de-escalation.’

Comment 3: Missing data and patient selection
The authors should provide more detail on missing data and on the number of patients included in each multivariable analysis. In addition, only 135 of the 214 RISAS patients were included in the present study. The possible effect of this selection on the results should be discussed more clearly.

Response 3: We thank the reviewer for this comment. Of the 214 RISAS patients, 135 were included in the present study. The selection process, including reasons for exclusion, is shown in Figure 2: 60 patients were excluded because their [¹⁸F]FDG PET/CT was acquired in four centres without EARL-reconstruction, 16 because no baseline [¹⁸F]FDG PET/CT was available, and 3 because neither a breast nor an ALNM core needle biopsy was available. These exclusions were necessary to ensure that quantitative PET measurements were comparable across centres and that all patients had the data required for the analyses. We acknowledge that the resulting sample size, and in particular the limited number of axillary pCR events (n = 46), reduces the statistical power and the stability of the multivariable model. We considered this a justified trade-off, as including non-harmonised scans or patients without baseline imaging would have compromised the validity of the comparisons.

We acknowledge that the exclusion of non-EARL centres may have introduced selection bias and may limit the generalisability of our findings to centres with different acquisition protocols. This was added to the limitations of our study as follows: ‘This study has several limitations. First, [18F]FDG PET/CT exams and CNBs were collected retrospectively. Although this enabled a centralised reassessment of both [18F]FDG PET- and TILs measurements by a single radiologist and pathologist, interobserver variability was not assessed. Second, the absence of post-NAC [18F]FDG PET/CT imaging precluded the assessment of changes in quantitative PET parameters (e.g. SUV) in both the breast tumour and ALNMs as well as in non-target organs, such as the liver and spleen, which may provide additional information. Third, [18F]FDG PET/CT exams were collected from 13 different centres in the Netherlands. Because acquisition and reconstruction protocols differ between centres, we included only scans from EARL-accredited centres (nine centres) for this retrospective analysis. This selection may have introduced selection bias and limits the generalisability of our findings. Although EARL accreditation harmonises quantitative measurements across scanners, minor residual differences in scanner type, reconstruction protocol and acquisition parameters cannot be fully excluded. . Fourth, detailed information on patient’s medical history was lacking. Therefore, patients with conditions that may have influenced immune function such as chronic inflammation, autoimmune disorders, or used immunomodulatory medication were not excluded from analysis while these are known to potentially influence both splenic metabolic activity and levels of TILs [42,43]. Fifth, as no internal or external validation was performed and model calibration was not assessed, the reported AUCs reflect apparent performance in the development cohort. As a result, model performance may be overestimated, and generalizability to independent populations remains to be established. Sixth, variable selection was partly based on univariable significance, which may result in unstable models and overestimated effect sizes, particularly given the limited number of axillary pCR events (n=46). Together with the absence of internal and external validation, this underlines that the model should be considered exploratory and hypothesis-generating.’

Comment 4: Interpretation of TILs and clinical implications
Some conclusions regarding TILs and the possible omission of axillary surgery seem somewhat strong. Breast tumour TILs were not independently associated with axillary pCR in the multivariable analysis, and the model has not yet been externally validated. I would therefore suggest a more cautious interpretation of these findings and of their potential use for treatment de-escalation.

Response 4: We thank the reviewer for this valuable comment and agree that our findings should be interpreted more cautiously. We have therefore revised the manuscript as follows:

First, the Discussion and Conclusion now state explicitly that an incremental predictive value of TILs and PET-derived variables could not be demonstrated. The first paragraph of the Discussion was revised as follows: ‘Patients with cN+ breast cancer who achieve axillary pCR after NAC may not benefit from any axillary surgery. Therefore, accurate non-invasive identification of patients achieving axillary pCR may allow omission of axillary surgery in the future. In this study, we explored the value of baseline [18F]FDG PET/CT-derived metabolic characteristics and TILs to predict axillary pCR in cN+ breast cancer patients treated with NAC by creating a multivariable model. The model combining [18F]FDG PET/CT-derived SLRmax and breast tumour TILs demonstrated moderate performance (apparent AUC 0.63 [95% CI 0.53-0.73]). A model including only the clinicopathological characteristics breast cancer subtype and tumour grade performed considerably better (apparent AUC 0.79 [95% CI 0.71-0.87], underscoring the importance of including tumour biology in response prediction. Adding SLRmax and breast tumour TILs to this clinicopathological model yielded a similar performance (apparent AUC 0.80 [95% CI 0.73-0.88]), so an incremental predictive value of these biomarkers could not be demonstrated in our cohort. Second, the association between [18F]FDG PET/CT-derived metabolic characteristics and TILs revealed higher ALNM metabolic activity and lower splenic metabolic activity in patients with high breast tumour TILs.’ The conclusion was revised as follows: ‘In conclusion, this study found that the model combining clinicopathological characteristics, splenic metabolic activity, and breast tumour TILs may be informative for the axillary response to NAC, but did not perform better than clinicopathological characteristics alone. Increased metabolic activity of ALNMs was associated with high breast tumour TILs, while an inverse relationship with splenic metabolic activity was observed. Validation in independent cohorts is required before any clinical application can be considered, and future studies should evaluate whether sufficiently accurate prediction models can support safe treatment de-escalation.’

Second, we have added to the conclusion that validation in independent cohorts is required before any clinical application can be considered: ‘In conclusion, this study found that the model combining clinicopathological characteristics, splenic metabolic activity, and breast tumour TILs may be informative for the axillary response to NAC, but did not perform better than clinicopathological characteristics alone. Increased metabolic activity of ALNMs was associated with high breast tumour TILs, while an inverse relationship with splenic metabolic activity was observed. Validation in independent cohorts is required before any clinical application can be considered, and future studies should evaluate whether sufficiently accurate prediction models can support safe treatment de-escalation.‘

Third, these points have been added to the Limitation section of our manuscript, as follows: ‘This study has several limitations. First, [18F]FDG PET/CT exams and CNBs were collected retrospectively. Although this enabled a centralised reassessment of both [18F]FDG PET- and TILs measurements by a single radiologist and pathologist, interobserver variability was not assessed. Second, the absence of post-NAC [18F]FDG PET/CT imaging precluded the assessment of changes in quantitative PET parameters (e.g. SUV) in both the breast tumour and ALNMs as well as in non-target organs, such as the liver and spleen, which may provide additional information. Third, [18F]FDG PET/CT exams were collected from 13 different centres in the Netherlands. Because acquisition and reconstruction protocols differ between centres, we included only scans from EARL-accredited centres (nine centres) for this retrospective analysis. Although EARL accreditation harmonises quantitative measurements across scanners, minor residual differences in scanner type, reconstruction protocol and acquisition parameters cannot be fully excluded. each centre uses different PET/CT acquisition and reconstruction protocols, we included only those that adhered to EARL guidelines for this retrospective analysis. Fourth, detailed information on patient’s medical history was lacking. Therefore, patients with conditions that may have influenced immune function such as chronic inflammation, autoimmune disorders, or used immunomodulatory medication were not excluded from analysis while these are known to potentially influence both splenic metabolic activity and levels of TILs [42,43]. Fifth, as no internal or external validation was performed and model calibration was not assessed, the reported AUCs reflect apparent performance in the development cohort. As a result, model performance may be overestimated, and generalizability to independent populations remains to be established. Sixth, variable selection was partly based on univariable significance, which may result in unstable models and overestimated effect sizes, particularly given the limited number of axillary pCR events (n=46). Together with the absence of internal and external validation, this underlines that the model should be considered exploratory and hypothesis-generating.’

Please see the attachment for the revised manuscript

Reviewer 3 Report

Comments and Suggestions for Authors

The paper " The Value of [18F]FDG PET/CT-Derived Metabolic Characteristics and Tumour-Infiltrating Lymphocytes to Predict Axillary Pathologic Response After Neoadjuvant Chemo-(immuno)Therapy in Clinically Node-Positive Breast Cancer" addresses a relevant research topic. However, the manuscript requires major revision for the following reasons:

1.The title of the manuscript requires reformatting.

2.Was standardization of measurements carried out during data collection (e.g., scanning parameters in the position on back or stomach )?

3.Why did the authors choose PET/CT rather than PET/MRI?

4.Was attenuation correction performed during PET/CT acquisition? Was the same attenuation correction method applied consistently across all centers?

Author Response

Dear reviewer, thank you for your positive feedback.

Comment 1: The title of the manuscript requires reformatting.

Response 1: We thank the reviewer for this comment. The title has been revised in accordance with the journal's guidelines to ‘[¹⁸F]FDG PET/CT-derived metabolic characteristics and tumour-infiltrating lymphocytes for predicting axillary response to neoadjuvant chemo-(immuno)therapy in clinically node-positive breast cancer’

Comment 2: Was standardization of measurements carried out during data collection (e.g., scanning parameters in the position on back or stomach)?

Response 2: We thank the reviewer for this question. Patients from 13 Dutch hospitals were initially considered for inclusion. To ensure harmonisation of quantitative measurements, only [¹⁸F]FDG PET/CT scans from EARL-accredited centres were included. Scans from four centres without EARL accreditation were excluded, leaving scans from nine centres for analysis. This harmonises quantitative measurements across scanners and allows pooling of the quantitative data. We verified the acquisition protocol for all participating hospitals and confirmed that all scans were acquired in supine position. We have added this information to the Methods section as follows: ‘All included patients underwent whole-body [¹⁸F]FDG PET/CT imaging in supine position before the start of NAC, using a standard acquisition protocol.’

In addition, all patients fasted for at least 4 hours, and image acquisition started approximately 45-60 minutes after tracer injection. Metabolic parameters were all measured centrally using the same software and segmentation method. As the scans were acquired in routine clinical care, some residual variability in acquisition parameters cannot be excluded, and we have acknowledged this in the limitations as follows: ‘Third, [¹⁸F]FDG PET/CT exams were collected from 13 different centres in the Netherlands. Because acquisition and reconstruction protocols differ between centres, we included only scans from EARL-accredited centres (nine centres) for this retrospective analysis. This selection may have introduced selection bias and limits the generalisability of our findings. Although EARL accreditation harmonises quantitative measurements across scanners, minor residual differences in scanner type, reconstruction protocol and acquisition parameters cannot be fully excluded.’

Comment 3: Why did the authors choose PET/CT rather than PET/MRI?

Response 3: We thank the reviewer for this question. We chose [¹⁸F]FDG PET/CT because it is the standard imaging modality in clinical practice for staging patients with breast cancer, and it is widely available. PET/MRI is currently available in only two hospitals in the Netherlands, which would have limited the size and representativeness of the study cohort. Moreover, PET/CT and PET/MRI could not be combined in a single analysis, as quantitative measurements differ between the two modalities (e.g., due to differences in attenuation correction and reconstruction) and are therefore not directly comparable. To ensure a homogeneous study population, we therefore included only patients who underwent PET/CT.

Comment 4: Was attenuation correction performed during PET/CT acquisition? Was the same attenuation correction method applied consistently across all centers?

Response 4: We thank the reviewer for this question. Only scans from EARL-accredited centres were included, and EARL accreditation requires standardised reconstruction settings, including attenuation correction for each scanner. Nevertheless, minor differences in scanner type and reconstruction algorithm between centres cannot be fully excluded. We have acknowledged this in the limitation section of our manuscript: ‘Third, [¹⁸F]FDG PET/CT exams were collected from 13 different centres in the Netherlands. Because acquisition and reconstruction protocols differ between centres, we included only scans from EARL-accredited centres (nine centres) for this retrospective analysis. This selection may have introduced selection bias and limits the generalisability of our findings. Although EARL accreditation harmonises quantitative measurements across scanners, minor residual differences in scanner type, reconstruction protocol and acquisition parameters cannot be fully excluded.’