Prediction of Other-Cause Mortality in Older Patients with Breast Cancer Using Comorbidity

Round 1

Reviewer 1 Report

This research is motivated by the need to inform rational selection of breast cancer patients for adjuvant treatments (more specifically, to inform the part of the selection process, which is related to the estimation of the risk of non-cancer death).

I have the following suggestions and comments regarding this manuscript:

1) In order to assess whether adjuvant treatments of breast cancer can or cannot provide appreciable benefits, considering the risk of death from other causes, it is meaningful to compare cumulative incidences for breath cancer-related deaths and cumulative incidences for other causes of death, of which only the latter is shown. 

2) Tumor stage, grade and estrogen receptor status variables are available, but they were not considered in the analysis of competing risks. For instance, using the approach reported by Wasif et al. (https://doi.org/10.1016/j.jamcollsurg.2019.03.013), women of black race diagnosed with stage I breast cancer at the age of 80 years would have an 8-year probability 24.6%  of death due to breast cancer and 30.0% of death due to other causes if ER-negative BRCA, but only 9.6% of death due to breast cancer and 28.5% of death of other causes if ER-positive BRCA. Psychiatric comorbidity in ER-negative BRCA would increase the risk of death due to cancer by 4% and the risk by other causes by 13.6%, which certainly indicates higher contribution of this comorbidity to non-cancer deaths relative to breast cancer death. However, considerable risk of death of breast cancer in patients with ER- BRCA with or without psychiatric comorbidities would be an important factor to consider for patient management, in addition to the high risk of deaths to other causes. This comment leads back again to the comment 1. 

3) Table 1 indicates TNM classification for staging, but other information implies prognostic staging (0-IV) rather than the TNM anatomical staging. The use of Arabic numerals in this context adds to the confusion. 

4) Criteria for inclusion of a comorbidity to the comorbidity count should be specified. The authors claim that in other study [19], comorbidity count was restricted to 77 comorbidities selected as relevant.  In this study, however, the inclusion criteria for comorbidities are not provided. What were the grounds for their inclusion with respect to the ICD-10 classification of diseases? How were, for example, multiple co-occurring mental and behavioral disorders (e.g. F60.31 and F90.0) counted? As one psychiatric comorbidity combined, or two comorbidities...?

5) Assumption of the proportionality of hazard ratio for using Fine and Gray's model needs to be validated where applicable (please, see PMC4386671 or similar text). 

6) In the absence of obvious difference between predictive value of the count of comorbidities versus Charlson Comorbidity Index (CCI), this reviewer cannot see an obvious benefit of using a simple comorbidity count, especially if it is derived from an unspecified list of possible comorbidities that can be included to the count, as opposed to the clearly defined CCI with existing algorithms for conversion of ICD10 entities to the CCI. The weights previously derived for calculation of the CCI are reasonably expected to have changed since this index was originally proposed, but the use of updated population-specific weights may produce CCI with more predictive value than a simple comorbidity count.  

7) Methods should be described in more details (which SW was used?) . Dutch Cancer Registry should be presented through citing appropriate reports published previously, and the URL to this Registry should be also provided for the benefit of readers of this manuscript. 

8) The introduction would benefit from more elaborated discussion of the previous research in this area, including inclusion of other relevant studies, such as Wasif et al. (please, see the point 2)

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

This manuscript aimed at comparing, in older patients with breast cancer, the predictive value on mortality of the Charlson index for other-cause mortality and the predictions with using a simple comorbidity count.

This paper is well written and clearly presented.

In addition to age, presence of comorbidities determines other-cause mortality. The expected result of the authors is to determine an optimal individualized comorbidity score that would help to choose the treatment strategy in the elderly. For this purpose, they compare in patients aged over 70 years with stage I-III breast cancer, the prediction in 5-year other-cause mortality of the usual Charlson index to that of the comorbidity count extracted from the Comprehensive Cancer Organization (IKNL).

Information on recurrence were retrospectively collected.

The comorbidity defined in the Charlson index are well-known and defined. Yet, how were defined (not selected) the comorbidities extracted from database to count individual comorbidities? For example, “obesity” (over than 1% of patients in most developed countries) is not listed.

The proportions of missing data on comorbidities, recurrences, and vital status are lacking.

The hypothesis is that mortality in women (with early stage cancer) dead without distant 5-year recurrence is the proxy of the other cause mortality. Yet, I’m troubled by incident (cardio, neurologic…) diseases related to cancer treatments (inherent to chemotherapy, immunotherapy…) in 5-year recurrence-free women. They should be considered as “cancer mortality” and not “other cause mortality”. Thus, in my opinion, the quality of death without recurrence as proxy of other cause mortality is questionable.

The conclusions of the analyses are that the predictive value of the Charlson index for 5-year other cause mortality did not differ from that of comorbidity count (line 214).

Thus, I don’t understand why the authors stipulate (line 221) “our findings are in line with a previous study that found that prediction of other cause mortality is better for comorbidity count than for Charlson index”. Their results do not support this affirmation. I’m not convince that the simplicity of the comorbidity count should argue to promote its extensive use in population-based studies.

Line 240 : 60% of the cohort had a Charlson score=0 of which 35% had 1 comorbidity and 16% had 2. Authors should provide which comorbidities were concerned. For example, in spite of its high frequency hypertension notification relies on “various” definitions in medical files so that it can reflect different levels of severity. Whatever the medical situation it counts as 1 in the comorbidity count, whereas the Charlson score consider the clinical burden of comorbidities in medical practices.

Women are less prone to die from other causes than men in accordance with sex differences that have been described (health-related behavior, such as smoking and physical activity, hormonal, psychosocial, environmental factors..). It would be interesting to include/compare the estimations in men and women before promoting the comorbidity count (other cancer location).

I agree with the authors that additional geriatric parameters would be really useful for decision making.

Conclusions of authors address the use of comorbidity count to clinicians for selecting older patients for adjuvant treatment in breast cancer (line 266). Decision making (individually or mostly during Multidisciplinary meeting) is already based on an overall clinical assessment of comorbidities and global health status. The topic “Methods for better comorbidities, social deprivation, health status… registration” is mostly dedicated to epidemiologists and researchers… that analyze large databases, in order to set up standardized international improved measures.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 3 Report

This retrospective registry review in over 7500 pts over 70 years shows that the Charleston Co-morbidity Index gives comparable result to simply adding the the comorbities. This paper is well done and makes an important point.

1. 85% of the cohort was ER and/or PR positive, yet only 56% of patients received endocrine therapy.  In 2003-2009 certainly there was an awareness of endocrine therapy. This could cause an increase breast cancer distant recurrences that affected results. More detail as what the tumor charactestics is necessary.
2. What was counted as a combordity in the "simple counting of comordities". It is not clear from the paper what constitutes a comorbidity. Specifically, in Table 2, under "Other frequently encountered comorbidities" is that the description of other comorbities that counted?
3. The counting of a breast cancer distant recurrence as surrogate for death is understandable, becausae of difficulities with cause specific mortality. However, 77 yo with a bone only recurrence and history of cardiovascuar disease, could well die of cardiovascular disease rather than breast cancer recurrence. Could you perform of "sensitviy analysis" of sorts where all patient that had distant recurrence were excluded and included to see if that substantially changed the result?

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

Authors have adequately addressed the reviewer's comments.

Author Response

There are no more responses requested by the reviewer. 

Reviewer 2 Report

thank you for the modifications the authors have introduced in the text.

I think that the percentage of excluded patients due to missing information/data (recurrence and vital status), should be mentioned in this nationwide population-based cohort study

This study reports results that I find challenging from a clinical point of view: The Charlson index has real limitations, but it is not, according to my experience, used in clinical setting.

My major criticism of this paper was related to the fact that such tools remain measurement tools (epidemiology, clinical research…) and not tools intended for geriatric clinic practice. Indeed, a quantitative and scored clinical approach questioned me.

But I hear that my point of view may have been radical to the extent that this study is well conducted and the discussion deserves to be raised.

I

Author Response

Please see the attachment.

Author Response File: Author Response.docx