My suggestions are as follow:
1. As international readers may not be familiar with the home health care of your country, please explain how the patients are recruited into this service and how they are terminated from the service. It is stated, ‘Approximately 24% of patients experienced prior hospitalization in the 6 months prior to the start of the HHC episode’ which for me, may mean most patients are not recruited into this service due to hospitalization.
2. It is stated, ‘Awareness of a patient’s risk status allows healthcare providers to better monitor patients for worsening symptoms and provide early interventions when needed.’ The authors may need to explain how the ‘home visits’ or tele-contact are scheduled, and if acute visits can be requested etc. Also, how ‘better monitor patients for worsening symptoms and provide early interventions when needed.’ can be achieved in the current system.
3. Although ‘unstructured clinical notes analyzed by natural language processing’ may have great scientific interests, you may need to discuss its clinical usefulness. If and only if the clinical notes were processed immediately after the clinical notes submitted and important information were abstracted, analysized and summerized in a specific area, otherwise, it may not have much usage in clinical settings. One of the reasons that structured data forms are developed and used in clinical settings is that it can enhance the identification of specific pieces of information. The authors may need to discuss how the ‘unstructured clinical notes analyzed by natural language processing’ can complement, supplement or replace structured data forms.
4 It is stated, ‘This study population included patients who received HHC services during 1/1/2015 and 12/31/2017’ is a bit old and ‘We used both OASIS-C1 released in
2015 and OASIS-C2 released in 2017, as well as a crossover between the two.’ I would suggest the author to use the newer data which all used OASIS-C2 (i.e. say 2018 to 2020) and rerun the analysis. It is expected that the re-running of analysis will not be difficult (as all syntaxes are ready) and the results should not be very different from this dataset.
5. It is stated, ‘Cluster 2 is labeled “high comorbidity burden” defined as multiple comorbidities’. However, I cannot see the author has added the number of conditions up to form a variable called “multiple comorbidities” nor explain why the cluster should be called “high” comorbidity burden. As it is stated, ‘For example in Cluster 3, older age was clustered together with the risk of cognitive impairments (e.g., dementia) and the risk of having sensory impairments (e.g., difficulty seeing, hearing, and speaking)...In addition, elderly patients with cognitive decline are more prone to having pressure ulcers and requiring greater assistance with ADLs/IADLs due to their vulnerability to poor self-care and decreased mobility’, hence cluster 3 may also be considered as “multiple comorbidities”. (I have no concerns about Cluster 2 itself, my doubt is whether Cluster 2 should be labeled as “high comorbidity burden”.)
6. As “NLP to the clinical notes“ is not available to the international nursing community and some nurses from other countries may be interested in considering to introduce such system, I would suggest the author also to conduct the A K-means cluster analysis with the “36 variables derived from the OASIS assessment or the EHR data” and the “9 variables derived from the application of NLP” separately and see if both resulted in 3 clusters and if the three clusters are of similar clinical features. Then, the author may check which datasets (i.e. the current one, OASIS assessment or NLP) have a better prediction power for hospitalizations or emergency department visits. Anyway, develop and implement a “NLP to the clinical notes“ is expensive, so, some data for consideration of cost-effectiveness should be provided.
7. Some items in the reference list do not follow the APA-7 style in using small and capital letters.

I have no further comments.