Analytic Challenges in Clinical Trials of Early Alzheimer’s Disease
, Suzanne B. Hendrix 1, Samuel P. Dickson 1 and Geert Molenberghs 3,4Round 1
Reviewer 1 Report
Comments and Suggestions for AuthorsThe topic of the manuscript is interesting. However, there is a major issue to be considered. It should be specified how "rapid progressors" have been defined in clinical trials. According to the literature, there is quite variability in defining rapid progressors among AD patients. With respect to the aim of the investigation, this point need to be addressed.
Comments on the Quality of English LanguageMinor editing is required.
Author Response
Comment:
The topic of the manuscript is interesting. However, there is a major issue to be considered. It should be specified how "rapid progressors" have been defined in clinical trials. According to the literature, there is quite variability in defining rapid progressors among AD patients. With respect to the aim of the investigation, this point need to be addressed.
Response:
Thank you very much for the great comment. Six references and additional text have been added as requested. As the reviewer notes, differences exist in the definition of rapid progressor. However, the key point for this paper is that Rapid Progressors are not a unique subgroup with distinguishing features. Instead, progression rate is a continuous outcome, with the continuous distribution of changes dichotomized into normal progression and rapid progression for conceptual understanding and convenience. Simply put, this is not a subgroup problem, it is a problem of non-normal data. Therefore, the 3rd and 4th paragraphs of the introduction have been modified as below, with new text in bold. The intent is to have readers see the problem of rapid progressors not as a subgroup problem, but as a problem of non-normal distribution of data.
The anticipated decline in an early symptomatic Alzheimer’s disease population on CDR-SB in a 78-week period is 1-2 points [Coric 2012; Egan 2019; Ostrowitzki 2017]. However, much larger declines for individual patients are common. [Abu-Rumeileh 2017; Wang 2022; Schmidt 2011]. Rapid progressing patients do not differ from other patients in demographic or baseline disease characteristics, comorbidities, concomitant medications, or the incidence of adverse events. There is no single clinical feature that differentiates rapidly progressing patients from other patients. Rapid progressing patients are not a unique subgroup with a clear definition and distinct features that separate them from the majority of patients. Instead, magnitude of progression follows a continuous distribution, with differences only in degree. Therefore, rapid progressing patients - and the resultant skewed data - are part of the reality of Alzheimer’s Disease and after the fact it is too late to address them in a completed randomized trial. However, current practice does not often include assessments of and sensitivity analyses for non-normal distribution of the data. Therefore, the analytic challenge is how to plan for the skewed data in the analyses of Alzheimer’s clinical trial data.
In addition, minor edits and clarifications to the manuscript have been provided.
Reviewer 2 Report
Comments and Suggestions for Authors-Could more sources be provided regarding the rapidly progressing patient population in AD? Such as more examples of studies that reflect this issue, supplementing support to the importance of the paper’s analysis.
-A brief introduction to the Hodges-Lehmann estimator could be provided.
-Shouldn't the robustness of results by HL and RR for the simulated set with missing data be interpreted in another way as MMRM handles missing data differently than MI. Or is there evidence showing that the imputation has not affected final results significantly.
Author Response
Comment 1: Could more sources be provided regarding the rapidly progressing patient population in AD? Such as more examples of studies that reflect this issue, supplementing support to the importance of the paper’s analysis.
Response 1: Thank you very much for the great comment. Six references and additional text have been added as requested. As the reviewer notes, differences exist in the definition of rapid progressor. However, the key point for this paper is that Rapid Progressors are not a unique subgroup with distinguishing features. Instead, progression rate is a continuous outcome, with the continuous distribution of changes dichotomized into normal progression and rapid progression for conceptual understanding and convenience. Simply put, this is not a subgroup problem, it is a problem of non-normal data. Therefore, the 3rd and 4th paragraphs of the introduction have been modified as below, with new text in bold. The intent is to have readers see the problem of rapid progressors not as a subgroup problem, but as a problem of non-normal distribution of data.
The anticipated decline in an early symptomatic Alzheimer’s disease population on CDR-SB in a 78-week period is 1-2 points [Coric 2012; Egan 2019; Ostrowitzki 2017]. However, much larger declines for individual patients are common. [Abu-Rumeileh 2017; Wang 2022; Schmidt 2011]. Rapid progressing patients do not differ from other patients in demographic or baseline disease characteristics, comorbidities, concomitant medications, or the incidence of adverse events. There is no single clinical feature that differentiates rapidly progressing patients from other patients. Rapid progressing patients are not a unique subgroup with a clear definition and distinct features that separate them from the majority of patients. Instead, magnitude of progression follows a continuous distribution, with differences only in degree. Therefore, rapid progressing patients - and the resultant skewed data - are part of the reality of Alzheimer’s Disease and after the fact it is too late to address them in a completed randomized trial. However, current practice does not often include assessments of and sensitivity analyses for non-normal distribution of the data. Therefore, the analytic challenge is how to plan for the skewed data in the analyses of Alzheimer’s clinical trial data.
Comment 2: A brief introduction to the Hodges-Lehmann estimator could be provided.
Response 2: The following text was added to the introduction. Also note that the HL estimator was further described in section 3.2.
One example of a non-parametric approach is the Hodges-Lehman estimator, which is essentially the median difference between groups.
Comment 3: Shouldn't the robustness of results by HL and RR for the simulated set with missing data be interpreted in another way as MMRM handles missing data differently than MI. Or is there evidence showing that the imputation has not affected final results significantly.
Response 3: The following text has been added to respond to the reviewer’s question.
MI, which was used for HL and RR, and MMRM make the same assumptions about missing data and lead to asymptotically similar results as size of data set and number of imputations increase. Therefore, the difference in results between the various methods is not due to different means of handling missing data.
In addition, minor edits and clarifications to the manuscript have been provided.
