# Feedback for XiaoDan's paper

I think, overall the paper is written well. I think the improvements are mainly about coherence, creating the flow and most importantly placing concepts in context in every part of the paper.

## May

1. Who is paper targeted to?
2. The section about non-imaging datatypes is to me, the weakest section of this paper. The section about metrics is the strongest.
3. Do we really need the non-imaging data section section to be as elaborate as it is? Some parts of the section elaborates heavily on foundation concepts, but also makes sweeping generalisations. For example the definition of non-imaging data can be complex with theoretical definitions. Is DateTime a data type separate from continuous and categorical values? Suggestion: Find references with definitions of these data types.
6. Overall, storyline is not yet coherent; reader needs reminder about context of concepts throughout paper. Suggestion: Put introduction paragraphs for each section and subsection, to describe what the sections are about and why they are relevant and its context within the paper. Enumerate subsections so they don't come as a surprise. Put summary paragraphs after each section.

## Iain

8. The title makes it sound as though there will be a focus on how synthetic data can be used to train **trustworthy** AI but there is not a strong case that synthetic data makes AI more trustworthy. We should either
1. define trustworthiness and make the argument that synthetic data is important for it, or
2. not mention it or include it as one of several motivations for synthetic data.
9. Other than privacy, is there anything about this review which is specific to medical data? Could it be easily adapted to be a general review of tabular synthetic data generation? May: This is a good idea
10. It might be nice to have a summary of each of the section 5 statistical models for comparison (either in a table at the end of the section or a sentence at the beginning of each sub-section). It could include; what kind of technique it is, synonyms, whether it is applicable to continuous or discrete variables, whether it is applicable to time-series data, privacy concerns, notable implementations.