PPE supplies and shortage per hospital in US

[Geoyi]

PPE shortage is coming with covid-19 patients' growth per hospital.

To answer the questions if we can model/simulate PPE (e.g. facial masks, surgical gloves, gowns, eye protections etc.) uses/shortages We need to think through:

• The medical procedure/treatment for a few group of covid-19 patients, e.g. critical care or not;
• staffs numbers per patient, including doctors and nurses;
• how many patients and how many doctors are working each day in each hospital, how frequently they need to use the PPE and other supplies. of course,
• the availability of the PPE and all supplies.

Note: The above table cited from a peer-reviewed article: Personal protective equipment in an influenza pandemic: a UK simulation exercise (2007).

NYT article Governments and Companies Race to Make Masks Vital to Virus Fight

President Trump resists using emergency powers to compel production, saying companies will voluntarily provide much-needed protective gear.

Doctors and hospital administrators have been warning for weeks that they face acute shortages of masks and lifesaving equipment such as ventilators. State and local officials in New York and elsewhere have been pleading for help from Washington and the industry.

[daveluo]

Thanks @Geoyi for starting this issue and working on this key part of the problem of protecting our healthcare workers!

Adding some thoughts below that may help with considering what needs to be modeled to estimate PPE supply needs.

This is mainly coming from talking through the problem w/ a clinician friend who sees inpatients, teaches residents, and needs to make decisions about conserving PPEs and minimizing exposure risk to them with every patient encounter (suspected or confirmed COVID cases):

• the number of COVID Patient encounters per Provider per Working Day or Shift may be a good field to base other estimates on. This tries to capture how many times a Provider needs to don and doff PPEs in a day's work, tied to the number of COVID pts they see (which may vary widely depending on the role of the provider, more on this below).
• This can then be used to represent the overall consumption rate of PPEs in a provider workforce. In ideal standard practice, this would be 1 set of PPEs per Pt encounter per Provider (since they're supposed to be single use) but in today's situation, providers are being asked to conserve and reuse. So this can be an adjustable estimate like 0.1 sets of PPEs per Pt encounter (representing 1 set of PPEs being used for 10 patient encounters on average before getting disposed and requiring a new set).
• The number of encounters a provider has with COVID pts also depends on the role. For example, A critcare nurse in a dedicated COVID ICU ward would be exposed to many more COVID pt encounters per day than a specialist consultant and need many more PPE sets per day.
• Staffing ratios for providers, nurses, techs, and anyone else needing to don PPEs with each patient encounter may become very dynamic and differ from usual practice. Health systems are putting in new policies to minimize provider exposure risk, usually by reducing the overall # of encounters between suspected/confirmed COVID pts and provider teams. For example in a teaching hospital, residents may no longer be seeing every patient in person ahead of the attending or there may be dedicated, smaller teams of 1 attending & 1 resident and the same for the nursing teams. This means the staffing ratios should be adjustable to account for this flux.

[Geoyi]

These are super helpful and thank you for talking through the problem with your clinician friend, @daveluo.

[daveluo]

It's also worth at least keeping in mind for ourselves or modeling supply needs distinguishing between the various types of PPEs: N95 respirators, surgical/procedure masks, eye protection, gown and gloves

and how they correspond to the different usual levels of isolation precautions: Contact, Droplet, Airborne

N95 masks and face shields seem to be the most common and needed requirement now but we also see use of surgical/procedure masks, gloves & gowns during screening and triaging patients. Or procedure masks in lieu of N95 when there aren't any.

[andrewwgithub]

I am reading reports that many facilities are minimizing their usage to 1 PPE (mask) per day.

[daveluo]

@andrewwgithub thanks for the input! To be more specific, do you mean facility-wide policy to be one mask (N95?) per provider per day?

[daveluo]

Also adding to thread as a reference: JAMA: Conserving Supply of Personal Protective Equipment—A Call for Ideas

The editors of JAMA recognize the challenges, concerns, and frustration about the shortage of personal protective equipment (PPE) that is affecting the care of patients and safety of health care workers in the US and around the world. We seek creative immediate solutions for how to maximize the use of PPE, to conserve the supply of PPE, and to identify new sources of PPE. We are interested in suggestions, recommendations, and potential actions from individuals who have relevant experience, especially from physicians, other health care professionals, and administrators in hospitals and other clinical settings. JAMA is inviting immediate suggestions, which can be added as online comments to this article.

[andrewwgithub]

@andrewwgithub thanks for the input! To be more specific, do you mean facility-wide policy to be one mask (N95?) per provider per day?

Yes one per provider per day until the supply situation improves.

[daveluo]

+data source:

https://gitter.im/covid19-healthsystemcapacity/community?at=5e77bc0e3a3acf64c1491943:

here's a continually updated #GetMePPEs spreadsheet for crowdsourced hospital facility needs and donation centers being set up: https://docs.google.com/spreadsheets/d/1txEanDkIrJ5GNfSk-zlXkTlB-bQPNRN_Y69qEwmdme8/htmlview?usp=sharing_eil&ts=5e77bb36&urp=gmail_link&sle=true

As a potential task after the 1st data release, someone could look at how we pull in this data automatically, join it to our facilities data, and visualize the crowdsourced needs and donation info per facility on our CCM map

[williamratcliff]

What is the status of data sources for the project?

[Geoyi]

Thanks for your inputs @daveluo and @andrewwgithub.

TODOs

I am planning to spend some time on the modeling this week and may need to reach out for guidance from you all. I have been talking to a few Andrew's colleagues at NM. The next key data points have all mentioned above are: PPE burn rates and usage as potential patients progress from testing to admission to discharge, a few admin and physitians from NM are willing to help to sort these key data points.

• get all the data for hospital capacity on the map;
• put convid-19 patients track data to the map;
• read the peer-reviewed articles what the team has put: 1 - 2015 study and 2 - critical care surge capacity planning. (other things I missed?)
• model and calculate hospital capacities and PPE needs around the Chicago area and:
  • reach out to NM to verify the outputs;
  • add missing data points, fix things;
• make a plan for the next steps (run and predict other states).

[lossyrob]

If you want to combine the those two papers to create an estimate of what's needed, I think one way to do that is:

• https://github.com/covidcaremap/covid19-healthsystemcapacity/blob/master/notebooks/processing/Generate_CCM_CareModel_Facility_Data.ipynb is a start to model Staff demands in the 3 scenarios mentioned in both of those papers: Conventional, Contingency, Crisis. It is based off of estimated ICU occupancy rates for each scenario, considering the number of ICU staffed beds in conventional and contingency scenarios, and All beds in the crisis scenario. It works backwards from there based on the table mentioned in the notebook to produce number of additional Physicians, Respiratory Therapists, and Critical Care Nurses required.
• We are using beds as a proxy for patients, assuming that all the beds of that scenario are taken up by a patient. If there is data on a patient's time through the hospital for positive COVID19, then you could estimate the number of patients based on bed counts
• With an estimated number of patients over time and estimated staff needed, if there's data around how many PPEs are need per staff per patient, that should let you estimate the burn rate.

These models are all based on percentages and I'm not sure how accurate they are, so there may be much better approaches to this problem. However the CareModel work might help serve as a starting point.

Let me know if you have any problems running the notebooks, using the data, or any workflow things. Ideally this repository makes getting running on an analytics challenge like the one you're tackling a bit easier to spin up on.

If you end up brining in data to use and make a PR, here's a quick rundown of how I've been approaching keeping track of everything:

• Download the data if it's not that big to be committed to the repo. If it is big, add downloading it to the processing/00_Download_Data.ipynb notebook. Add it to the data/external foldere.
• Add source information to data/README.md
• Data paths in notebooks should be built up with the data path helper methods; for example if you added external/data.csv, then you can get the full path (inside or outside docker) by:

  
  from covidcaremap.data import external_data_path

the_path = external_data_path('data.csv')

- The `covidcaremap` package has some constant values and some geospatial methods that are meant to be reused between notebooks. If you want to create reusable python functionality feel free to add it to that module.
- Any data output by the notebook should go into the `data/processed` directory. You can get a path there by using the `processed_data_path` method similar to external_data_path. I've been committing the `processed_data` files even though they can be regenerated by notebooks; I think this allows people to run notebooks selectively and also puts up data that people may find useful (but should use at their own risk).
- If a notebook genert data, it uses the `published_data_path`. This is data we'd advocate for people to use.
- If you need to install any python libs, please put them in `notebooks/requirements.txt`. That way the `docker/build` script will install them into the container and other users won't have to think to hard about the install environment.

If there's any suggestions on how to improve this workflow please feel free to comment, chat, or make changes in a PR! 

[lossyrob]

@williamratcliff

What is the status of data sources for the project?

I'm unsure what you are asking, can you clarify? Do you mean this project or the projects listed in the comments above yours? Thanks!

[Geoyi]

I need a few data layers today to put things together:

• 1) hospital capacity (staffed bed, ICU beds, # health workers)
• 2) covid patient prediction models (epi model --seems like Jaline Gerardin is working on this too, and will sync with here)
• 3) PPE supplies per day per hospital, the input is from the epi model from step 2)

I'd love to get some guidance on:

• which data layer for the hospital capacity I should use?
• who is working on covid patient prediction models (#9)? Has someone started?

Andrew W look me into a few discussions with NM and they produce an IDOH report.

[abarciauskas-bgse]

@Geoyi Adding what I posted in slack here in case it's helpful for you or others.

From https://usafacts.org/visualizations/coronavirus-covid-19-spread-map/ they have county level confirmed cases: https://static.usafacts.org/public/data/covid-19/covid_confirmed_usafacts.csv

Should we add a layer on counties' capacity (at least in terms of hospital beds) to handle the current and future estimated cases?

We will need to also merge in data on total population as the current layer is beds per 1000.

[lossyrob]

Looked over #68 - seems like a great start.

So if we take the county level confirmed cases mentioned above, could this be turned onto PPE estimates per county? That might be a first good step before the EPI model can project out caseload per county, if we get a model that can do that.

Where does the hospital/ICU capacity fit in? That could be supplied by the current dataset; updated ICU capacity numbers would be great but still looking for the best solution to collect that.

[Geoyi]

Thank you, @abarciauskas-bgse for sharing the data input.

Great points @lossyrob for using the confirmed cases at the county level to start. I talked to Jaline Gerardin at Northwestern U, she and her students have started EPI model for the Chicago area, and I think we may just wait a little to see what is coming out of it.

I'd say we wait until we have an updated and accurate hospital/ICU capacity, and make a plan from there. I am afraid everything is dynamically changing at the hospital side, for instance, they will make more beds/ICU beds if too many patients.

I am meeting with physicians again to debrief what data points we will need from them to track PPE needs at county level with the covid19 patient # changes.

Todo today : how can we estimate PPE needs when we have the confirmed cases at the county level

Qs to answer (☝️ look at the diagram):

• what is the ratio between the healthcare provider and patient at community testing to ED Presentation per day? what are the ratios look alike at In-Patient Care and Critical Care?

• what is the percentage of other non-covid19 patients who will need PPEs that can't be neglected, and what the PPE needs looks like? Are these needs can be estimated before convid19 hospitals' average ICU occupation?

• If we are looking at hourly/daily county-level covid19 confirmed cases, how should we envision the PPE needs through the whole patient flow? In the confirmed cases, we are only looking at after Home Care ☝️, and what will be the best way to compute PPE needs through In Patient Care to Critical Care because a covid19 patient will stay for a week or more?

• Last but not least, how do we define "PPE needs", large and well-funded hospitals have stocked up more PPEs but the physician in small hospitals are asked to reuse some of their PPE items?

I hope all the questions can be answered by physicians in the coming meeting.

[daveluo]

Got some feedback from clinician friends that I've consolidated into a range for staffing ratios they're seeing:

Service	Staff	Staff:Patient ratio per shift
Emergency Dept	Physicians, Nurse Practitioners (NP), Physician Assistants (PA)	?
	Nurses for general inpatient-level patients	1:6-9
	Nurses for ICU-level patients	1:4
General Inpatient	Physicians, Nurse Practitioners (NP), Physician Assistants (PA)	1:5-6
	Nurses and Patient Care Associates (PCA)	1:2-4
ICU	Physicians, NPs	1:3-4
	Respiratory Therapists	1:4-6
	Nurses and PCAs	1:1-2
Consultants	Specialist physicians like ID, pulmonary, nephrology	?
	Other expert consultants like physical therapists, nutrition, chaplains, palliative care	?

Also consider there's redundancy in teaching institutions: every COVID patient has at least 2 doctors – a resident and an attending physician. Re: PPE, unclear if both are always going in every day but likely they are.

[Geoyi]

PPE needs estimation has been done and updated here (cc: @andrewwgithub)

Note: current PPE needs were estimated from confirmed covid19 cases (dataset from USAFacts) and follow the above diagram ☝️ which can be underestimated for a few aspects;

• The estimation was updated until 25th, and most cases grew on 26th in the state, which was not captured yet but we will continue to work on improving;

• We're working with physicians and current PPEs needs through the patient flow (diagram above) still underestimated the needs, we are working on improving this and will reflect the improvement through the current codebase.

[andrewwgithub]

Still waiting on confirmation from some ED colleagues, here are some data from a 2014 survey for patients per hour:

https://www.acepnow.com/article/emergency-department-benchmarking-alliance-releases-2014-data-on-staffing-physician-productivity/2/

[andrewwgithub]

My ED colleagues have confirmed the patient per hour ratio looks good, some EDs have APPs practice independently, while at large academic centers, the APP must have an attending physician.

[Geoyi]

the online report for PPE need/demand is up here: https://rpubs.com/Geoyi/PPE_need_usa_convid19. The demand has been updated until April 3rd.

[andrewwgithub]

This is great thank you!