Revisiting the OGIPRO Trial: Dynamic Electronic Patient-Reported Outcomes Compared with EQ-5D-5L in HER2-Positive Breast Cancer
Round 1
Reviewer 1 Report
Comments and Suggestions for AuthorsThe prospective OGIPRO trial dataset is retrospectively analyzed in this manuscript by comparing weekly EQ-5D-5L outcomes (VAS and domain sums) in patients with HER2-positive breast cancer to (i) continuous, patient-initiated "dynamic" ePRO well-being gathered through the Medidux app and (ii) aggregated CTCAE-aligned symptom grades. The authors report a weaker, heterogeneous alignment between aggregated symptom grades and EQ-5D-5L domain sums and a strong alignment between ePRO well-being and EQ-5D-5L VAS using linear mixed-effects models. Although the topic is current and clinically relevant, interpretability, reproducibility, and the strength of the conclusions are currently limited by a number of methodological and reporting issues.
-Higher levels of the EQ-5D-5L domain items indicate worse health, while VAS indicates better health. To align domain sums with symptom burden and/or well-being, you must specify how you did so.
-The descriptive system (profile), index value (country tariff), and VAS are commonly used in EQ-5D-5L analysis. It is unusual to sum domain levels, and doing so may not be psychometrically or linearly valid. Explain how a sum may skew relationships and, at the very least, why a domain sum was chosen over an EQ-5D-5L index value (Swiss, German, or European value set, as appropriate).
-Clinical heterogenity and potential confounding are not taken into consideration. Early, locally advanced, and metastatic disease receiving systemic treatment are all included in the cohort. The frequency of symptom reporting and EQ-5D-5L outcomes can be influenced by treatment type/line, age, comorbidities, and performance status. Provide the treatment distribution and baseline characteristics at the very least, and if available, take into account adjusted models.
Author Response
Dear Reviewer,
Thanks for taking the time to review the manuscript. We believe we have improved the manuscript in accordance with your suggestions:
Comment 1: Higher levels of the EQ-5D-5L domain items indicate worse health, while VAS indicates better health. To align domain sums with symptom burden and/or well-being, you must specify how you did so.
Reply: It is true that EQ-5D-5L domain items and VAS are reversed. However, the same is true for the ePRO measures analyzed in the study: the ePRO symptom grade increases with worsening health, the ePRO well-being increases with better health. The constructs that we compare align. To avoid misunderstandings, we have clarified this in the manuscript.
Comment 2: The descriptive system (profile), index value (country tariff), and VAS are commonly used in EQ-5D-5L analysis. It is unusual to sum domain levels, and doing so may not be psychometrically or linearly valid. Explain how a sum may skew relationships and, at the very least, why a domain sum was chosen over an EQ-5D-5L index value (Swiss, German, or European value set, as appropriate).
Reply: Thank you for pointing this out. We had originally decided against using the index value, since we don't have a method for indexing the ePRO measures. However we agree that for comparability, it makes sense to analyze the index values: We have added a third model to the manuscript which compares dynamic ePRO symptom grades with EQ-5D-5L index value (for Germany, since there is no Swiss index value set). To keep to the 0-1 normalization of all values and the issue you mentioned in Comment 1, we used the disutility part of the index and scaled it (see manuscript) - these are linear transformations, so do not affect the statistics meaningfully, but make the results easier to read, in our opinion.
Comment 3: Clinical heterogenity and potential confounding are not taken into consideration. Early, locally advanced, and metastatic disease receiving systemic treatment are all included in the cohort. The frequency of symptom reporting and EQ-5D-5L outcomes can be influenced by treatment type/line, age, comorbidities, and performance status. Provide the treatment distribution and baseline characteristics at the very least, and if available, take into account adjusted models.
Reply: We have included a table 1 that highlights patient characteristics. We did not perform subgroup analysis, since we do not believe that the sample size is large enough to make this applicable.
Thank you again for your review, we believe that the manuscript has been substantially improved.
Reviewer 2 Report
Comments and Suggestions for AuthorsAndreas Trojan and Colleagues conducted an interesting study comparing dynamic electronic patient-reported outcomes (ePRO) with the established EQ-5D-5L questionnaire in patients with HER2-positive breast cancer. The manuscript is well written, methodologically sound, and addresses an important gap in the literature regarding the validation of continuous digital PRO tools. I request the authors to address some of my concerns before final acceptance.
1.In aggregation of symptom data, combining 91 symptom types into a single score may obscure meaningful clinical patterns; justification is needed.
2.In Statistical Analysis, more details should be provided regarding model specifications, including fixed and random effects, and whether assumptions (e.g., normality, homoscedasticity) were checked. Additionally, provide how missing data were handled, especially given the dynamic nature of ePRO reporting.
3.For Clinical Relevance of Findings, the weaker agreement for symptom domains requires deeper discussion. Are certain symptom clusters (e.g., pain, fatigue) more aligned with EQ-5D-5L domains than others?
4.Regarding, Sample Size and Generalizability, the analysis is based on 53 patients, While the within-patient repeated measures increase statistical power, the small cohort may limit generalizability. Discuss the representativeness of the sample and whether results might differ in other cancer types or demographic groups.
5.Figure 5 in particular seems to contain important data on symptom entry distribution but it is incorrectly placed in the main text (after figure 5, figure 3 and 4 appears).
6.The term “Medidux” is sometimes spelled “Meddux” (e.g., page 7). Please ensure consistency throughout the manuscript.
7.The discussion highlights the potential for AI and machine learning in analyzing ePRO data. Provide specific research directions, such as the development of symptom-specific digital biomarkers or the integration of ePRO data with electronic health records for predictive modeling.
- Add “HER2” to the abbreviation list.
Author Response
Dear Reviewer,
Thank you for taking the time to review our manuscript and for your valuable feedback. Find our reply below.
Comment 1: In aggregation of symptom data, combining 91 symptom types into a single score may obscure meaningful clinical patterns; justification is needed.
Reply: We absolutely agree with your assessment. We believe that combining the symptoms into one score is necessary given the relatively small sample size. However we plan to explore more advanced methods of analysis in future work that will hopefully make use of the rich data in its fullest. We have expanded on the limitation in the manuscript.
Comment 2: In Statistical Analysis, more details should be provided regarding model specifications, including fixed and random effects, and whether assumptions (e.g., normality, homoscedasticity) were checked. Additionally, provide how missing data were handled, especially given the dynamic nature of ePRO reporting.
Reply: We have provided more information on the model specifications and parameters, and have specified how we dealt with missing data.
Comment 3: For Clinical Relevance of Findings, the weaker agreement for symptom domains requires deeper discussion. Are certain symptom clusters (e.g., pain, fatigue) more aligned with EQ-5D-5L domains than others?
Reply: We believe they are, which ties back to your first comment. Exploratory analysis using basic statistics methods did not yield significant results, probably due to the relatively small sample size, but we plan to use more advanced methods in future work.
Comment 4: Regarding, Sample Size and Generalizability, the analysis is based on 53 patients, While the within-patient repeated measures increase statistical power, the small cohort may limit generalizability. Discuss the representativeness of the sample and whether results might differ in other cancer types or demographic groups.
Reply: We have added a comment on generalizability to the discussion.
Comment 5: Figure 5 in particular seems to contain important data on symptom entry distribution but it is incorrectly placed in the main text (after figure 5, figure 3 and 4 appears).
Reply: This was an error, and has been fixed.
Comment 6: The term “Medidux” is sometimes spelled “Meddux” (e.g., page 7). Please ensure consistency throughout the manuscript.
Reply: It should be correctly spelled everywhere now.
Comment 7: The discussion highlights the potential for AI and machine learning in analyzing ePRO data. Provide specific research directions, such as the development of symptom-specific digital biomarkers or the integration of ePRO data with electronic health records for predictive modeling.
Reply: We have expanded somewhat on future research directions in the discussion and conclusion.
Comment 8: Add “HER2” to the abbreviation list.
Reply: Done.
Thank you again for your review, we believe that the manuscript has been substantially improved.
