Access to Hematopoietic Stem Cell Transplant in Canada for Patients with Acute Myeloid Leukemia

Round 1

Reviewer 1 Report

Dear Authors 

The manuscript is overall well written, however  the following inputs would be necessary before acceptance .

(1) The access to HSCT analysed is specifically  is for AML patients . The title should be changed as it does not mention it to be specifically for AML patients. 

Is there a variation in protocols across centres in Canada as to when a patient would need a HSCT for AML. Is so please clarify. If there is a uniform protocol, then it should be described briefly in the methods

(2) Why is the Quebec region excluded from the analysis . please clarify

(3) IS there inter state referral of patients. Can a patient from BC recieve a HSCT in Toronto. If yes, has this been accounted for in the analysis

(4)Can a  age bias exist  in referral for HSCT e.g an old age patient may be less preferred than a younger one.  If so , do population demographics differ in between the regions .

Author Response

We would like to thank the reviewers for their thoughtful comments that we think improve the overall quality of the manuscript.

Reviewer 1:

• The reviewer suggested we adjust the title to reflect that the analysis was restricted to AML patients. This has been changed.  We suggest that access to allogeneic transplant for AML is a reasonable surrogate for access to transplant in general, as AML is by far the most common indication for transplant in adults.
• The reviewer asked if there were uniform protocols used to determine indications for transplant. There are no Canadian protocols, but there is fairly good agreement in the literature as to which patients need a transplant, and there are international guidelines (European LeukemiaNET, NCCN) that support this. We have added a section to the discussion section discussing these guidelines.
• The reviewer asked why Quebec was excluded from this analysis. Quebec is excluded from the analysis as it doesn’t participate in the CIHI Discharge Abstract Database.  This has already been described in the manuscript (see the first sentence of the Methods section).  For similar reasons, due to referral patterns, New Brunswick was excluded.
• The reviewer asked about interprovincial referrals. It is likely that there are interprovincial referrals, but this is a national database, and patients will be captured even if they have a transplant in a province other than the province where they were initially treated.
• The reviewer asked about a potential age bias in referrals for transplant. Yes, it is likely that age bias might exist.  This analysis was restricted to adults under age 65, in part to address this, and age was included as a variable to be controlled for in multivariable analysis.  Indeed, age was one of the strongest predictors of access to transplant in univariate analysis.

Reviewer 2 Report

The authors analyzed data from 619 non-Quebec Canadians under the age of 65 who were diagnosed with AML between 2004 and 2015 to determine facts that influenced Canadian alloSCT acquisition. The manuscript found that there is no statistically significant difference in transplant rates observed among the location of primary residence, low-income families, and reported race or ethnicity, but transplant rates varied by province of residence.

However, I still feel that the innovative aspects of the paper are not highlighted, which will make the paper difficult to attract potential readers.

Comments:

1. Detailed demographic data should be described in separate tables prior to the logistic regression model.

2. Donor availability is a key factor affecting HSCT rates. Please discuss this further.

3. Discuss whether the classification of AML and response to treatment might affect the incidence of HSCT.

Author Response

Thank you for your thoughtful comments that we think improve the overall quality of the manuscript.

Reviewer 2:

• The reviewer requested detailed demographic data be included as a new table. This is a good suggestion, and this has been added as Table 1.
• The reviewer suggested we comment on donor availability as a predictor of HSCT rates. We agree that this is important, and a new paragraph has been added to the discussion section.
• The reviewer asked about disease specific factors and their impact on access to HSCT. This has been addressed with an additional section to the discussion section.

Author Response File: Author Response.docx

Reviewer 3 Report

This is a big dataset study! The research team address some important risk factor that associate with allogeneic hematopoietic stem cell transplant. However, some major concern should be addressed:

1. The analyses were quite complicated, and I am unsure if these steps were applicable.
2. First, I don’t understand the meaning of the time period is?
3. The study design was a retrospective cohort? The result should be a relative risk instead of an odds ratio if this is a cohort.
4. Since the authors want to address the multicollinearity effect among the factors, they use Likelihood ratio testing <0.2 and Cramer’s V, Pearson, or Spearman correlation coefficients. I am not sure the procedures were correct. (I think I will use generalized linear mixed models with the Poisson distribution. To detect multicollinearity should use the regression model. The authors should first notice the multicollinearity and remove the factors before putting them in the model.) 
5. The authors should state what statistical software they use.
6. The authors state that “The multivariable model was constructed from the remaining variables using likelihood ratio testing with P-values ≤ 0.05 indicating significance. The relationship among continuous variables was modeled using restricted cubic splines. Akaike’s information criteria were used to select models with different numbers of knots”. I think the author can consider using generalized linear mixed models 

Author Response

Thank you for your thoughtful comments that we think improve the overall quality of the manuscript.

Reviewer 3:

• The reviewer commented that the statistical analysis was complicated. We agree, but feel our approach was required given the large number of patients in our dataset, and the interaction of multiple variables. 
• The reviewer asked us to clarify the use of “time period”. Time period refers to the year in which the transplant was performed, to account for changes in clinical practice over time, such as increased use of reduced intensity conditioning protocols in later years of the study period.  We think that “time period” is a reasonably appropriate descriptive term to describe this variable.  An extra sentence has been added to the methods to clarify this.
• The reviewer inquired as to whether relative risk should be used instead of an odds ratio, given the study design. However, ORs and RRs can be obtained from cohort studies. We used a logistic model for our analysis which calculates the log odds for each variable included in the model. The log odds is exponentiated to provide the ORs.
• The reviewer suggested that a linear mixed model with the Poisson distribution would have been a preferred statistical methodology, as opposed to the Likelihood ratio testing <0.2 and Cramer’s V, Pearson, or Spearman correlation coefficients. However, we did not use likelihood ratio testing to check for multicollinearity. Likelihood ratio testing was used to check univariable and multivariable significance of each variable in the models. Cramer’s V (checking categorical variables), Pearson or Spearman correlation coefficients (checking continuous variables), point-biserial correlations (checking continuous vs. binary variables), and eta-squared (checking continuous vs. variables with >2 categories) were used to test for multicollinearity. Each of these tests checks the association between the two variables. If it’s strong it could indicate that they are measuring the same concept (redundant/correlated).
• The reviewer requested we include the statistical software used. SAS version 9.4 and STATAMP version 14.2 were used and has been added to the methods section
• The reviewer requested we consider using generalized linear mixed methods, instead of likelihood ratio testing. However, generalized linear mixed models are used for continuous outcomes. Our outcome was binary. A logistic mixed model approach could have been used but estimating the associations for each province was of interest, so they were incorporated as indicator variables. This way we could see the ORs for each province, and account for the clustering.