Towards a Machine Learning Model for Detection of Dementia Using Lifestyle Parameters
Round 1
Reviewer 1 Report
In this manuscript, the authors proposed a model for early dementia detection using lifestyle data from the National Health and Ageing Trends Study. The proposed model has good results. The experimental results demonstrate the effectiveness of the model. This research has certain clinical application value. There are some issues that need to be addressed before it is ready to be published. The followings are my comments and suggestions:
1. The authors should briefly introduce the principle of random forest algorithm.
2. In Section 2, which artificial neural network model to use should be clarified.
3. The content of Figure 2 is better represented by a formula.
4. More experiments can be added in Section 3 to verify the innovativeness of the paper.
5. There are some typos in the paper.
Author Response
Please see the attachment
Author Response File: 
Author Response.pdf
Reviewer 2 Report
The title of the paper implies that using ML model detection of dementia can be accomplished using lifestyle parameters. I was expecting to see methods that would enable the early detection of dementia and some separation of symptoms of normal aging and the start of dementia and Alzheimer's The paper leaves me in the dark on those two points. The abstract stated that this method would aid in flagging high-risk individuals, but to me, your method seemed to be best at identifying individuals with advanced disease. The lifestyle parameters, such as the inability to draw a clock would seem to be an advanced stage of Alzheimer's. They used the term data markers for diagnosis. I am not clear if these are phenotypes of digital biomarkers.
In general, the paper contains much redundancy. For example, line 61 repeats line 42. Sentences are repeated almost verbatim after a few intervening lines. I also find the material not well-organized.
What I would like to see discussed is what are the clues that would lead one to take the survey. What are some of the first clues that suggest the onset of dementia? Table 1 identifies forgetfulness and losing track of time as early-stage indicators, but we all do that. So how do we distinguish normal aging and dementia?
I like the section on the standard protocol for dementia, but this gets lost in future sections of the paper. The reference to Vijay S. Nori does predict the onset of dementia. How does that approach differ from what you are doing?
Your study uses NHATS data. The excellent description of that resource was useful. The list of your contributions did not seem original. the 4th bullet needs discussion. Are you saying that you identified out of the set of data contained in NHATS data that goes beyond what is conventionally used to detect dementia? The temporal aspect of your work is enabled by the fact that NHATS was a temporal database. I am looking for something original.
In Table 2 I think you mean housing, not hosing. How did you select these specific general classifications of parameters? Is this part of NHATS? I like the introduction to the naming convention. Is this your work, or is it part of the naming convention of NHATS?
The detain on the value set for the parameters did not add much to understanding what you were working with. It is obvious that the value set would be present, not present, or not answered. I agree that the list of the data elements you selected is informative. I would like some discussion about why you selected these items and not others. I would like to know, for example, if the person watches the news on TV. I can think of other questions that I would think would be relevant. Figure 3 is a small font and difficult to see. I think it has little value unless you discuss how it is used and by whom.
I would like to see some discussion of Figures 5 and 6, or at least refer to them in the text. I am confused by Table 14, the final testing questions. Are you suggesting that these questions are the ones you have selected that are adequate to detect dementia or Alzheimer's? I would like to see a discussion of the time value of each question. Which questions help with early detection? Which questions can differentiate between normal and disease states? Finally, do you distinguish between dementia and Alzheimer's?
The paper is poorly organized and has a lot of redundancy.
Author Response
Please see the attachment
Author Response File: Author Response.pdf
Round 2
Reviewer 2 Report
The authors did an excellent job of addressing my previous comments. The paper is still a bit long but acceptable.
