Explicit support and documentation for other time interval than days.

[jcblemai]

Is your feature request related to a problem? Please describe.
I am currently training some folks on the public-health side to do nowcasting, using this great package. There have been some rough edges (around the selection of data for training). I see some development to make better/simpler methods and I find that exciting -- so I will hold off on most of these comments to now. But one thing we have been struggling with epinowcast and baselinowcast is handling other reporting than daily. Most numbers have no units in the vignette and it is hard to know if it's counting occurrence (number stays the same between days and week) or days (multiply by 7 for week, 30 for month).

Describe the solution you'd like
If there is a philosophy on the date interval units shared by both package, a little box explaining it would be welcome. Otherwise, an example on NHSN data with week (e.g) would be interesting.

Describe alternatives you've considered
You can see how we handle that in the AMPH forecasting Suite (https://accidda.github.io/AMPH_Forecast_Suite/articles/02.1-nowcasting-demo.html). Hope there is now mistake.

Additional context
This issue is from my backlog (nov 2025) so there might have been progress I've missed since.

[jcblemai]

this goes in the right direction: #350, just noticed

[seabbs]

Thanks for this @jcblemai and thanks for the helpful example. For interest does the relatively new epinowcast table help at all for that package or more needed there as well (see: https://package.epinowcast.org/articles/features.html)

I think once we have #283 in this might all get a little easier but even with that we definitely need some docs on this.

My impression is we need to be clearer in the docs for baselinenowcast that it defines data using dates but the interval can set by the user (and show this when we do it more clearly)? Does that match up with what you think would be useful?

[jcblemai]

I missed this new epinowcast table, this would have been a huge timesaver. Incredible job. I remember fiddling with these while preparing the material. Amazing that this was added

Yes I am following #283, u I think this will help a lot. If I'm not mistaken this will make the 3 commands to get the reporting triangle easier ?

My impression is we need to be clearer in the docs for baselinenowcast that it defines data using dates but the interval can set by the user (and show this when we do it more clearly)? Does that match up with what you think would be useful?

yes, though now that I see the epinowcast table I think I would have been able to do with that. However, it is true that for the folk we are working with it would hope if that was included into baselinenowcast doc.

[kaitejohnson]

@jcblemai Thanks for posting this-- really useful.

I agree that an example on weekly NHSN data would likely be really helpful -- I think we could do that fairly easily as I know there is already one in place... I think it might be worthwhile to make a quick vignette in baselinenowcast that shows this workflow using epidatr (tagging you @SamuelBrand1 for awareness), and adding to the current getting started vignette the use of the delays_unit arg and what situations you would modify it (see below).

I know you said this was in your backlog, but did you try the workflow of going straight from the data.frame with incident case counts by reference date and report date to a reporting triangle e.g.

rep_tri <- baselinenowcast::as_reporting_triangle(
                  training_data,
                  delays_unit = "weeks") |> 
                 truncate_to_delay(max_delay = 4)

nowcast <- baselinenowcast(rep_tri)

Our intention was for the delays_unit argument in as_reporting_triangle to handle the issue specified in the title- explicit support for weekly-weekly or monthly-monthly data.

Let me know what you think about 1. modifying current vignette and 2. adding additional vignette, and I will make these into separate issues.

[jcblemai]

oh wow I did not see @SamuelBrand1 repository. This is extremely close to what we develop for the AMPH in purpose (and much better), too bad this did not exist when we wrote that vignette. We will monitor and improve our course as this get developed. These plots are awesome.

I hope NowcastedNHSN will offer NHSN nowcasted as a service, there is a real need for this in the community and I never got to something satisfactory. Would love to be able to just plug this n.

I did not know this workflow existed (perhaps it did not?), this is it is much clearer. I think most my questions have been answered at that point. I think an example with week and the "simpler path with all arguments precised" (like your snippet does) is useful if local public-health is an intended user. And more self contained in the sense that it explains each epinowcast function ("this remove the data older than X days in the data frame"). I would be happy to contribute instead of just opening an issue but it will only be able in a while.

[kaitejohnson]

I hope NowcastedNHSN will offer NHSN nowcasted as a service, there is a real need for this in the community and I never got to something satisfactory. Would love to be able to just plug this n.

Was the main issue with baselinenowcast the need for direct support for weekly data? I do not think we had the as_reporting_triangle(delays_unit) for all of November, so it's possible the older version of the package you were using just expected a matrix with rows as reference times and columns as delays, and for weekly data, you would have made a weekly row, weekly column matrix (I would have done this without epinowcast preprocessing, just directly from the data, as I think epinowcast might require dates and fill in missing ones... ).

I personally would be happy to make an issue to make a vignette (heavily based on @SamuelBrand1's) to demonstrate how to nowcast the NHSN signal using baselinenowcast and epidatr. @seabbs What do you think? @jcblemai Would this be useful and improve the workflow for your course and your operational needs?

[SamuelBrand1]

I hope NowcastedNHSN will offer NHSN nowcasted as a service, there is a real need for this in the community and I never got to something satisfactory. Would love to be able to just plug this n.

This is very much the goal! Currently, I'm fusing even more simple nowcasting (1 week lag only fitted to a LogNormal) with https://github.com/CDCgov/NowcastAutoGP to make a simple baseline forecaster but I wanted to use baselinenowcast and make nowcastNHSN as a script masquerading as a pkg.

Please NB: CDCgov/nowcastNHSN#12 where I add the actual nowcasting to the data munging, and I've added a couple of other uncertainty models.

[seabbs]

Yes I think an issue for that could be useful but we can also just point out to @SamuelBrand1 work rather than duplicating i.e in a resources or FAQ vignette. We could also use that to point at the epinowcast table for data munging vs reinventing the wheel here using dplyr chains.

[kaitejohnson]

Yes I think an issue for that could be useful but we can also just point out to @SamuelBrand1 work rather than duplicating i.e in a resources or FAQ vignette. We could also use that to point at the epinowcast table for data munging vs reinventing the wheel here using dplyr chains.

Yes I agree, the slight issue is that @SamuelBrand1 made a bunch of (very nice) custom functions in nowcastNHSN and so to run that these need to be installed.

I went ahead and made a very simple version in my scratch repo which just uses a few dplyr commands and a call to epidatr (still heavily based on @SamuelBrand1's so if we were to add this would add him as contributor of course).

Happy to also look into integration with epinowcast table and maybe we could add that as an alternative workflow?

[seabbs]

Hmm can't they just installl it from his instructions? Seems fine. Seems fine to add as a vignette as well but a bit of a duplication of effort. Its not an alternative workflow because this package doesn't have the tools for data munging.

[SamuelBrand1]

I personally would be happy to make an issue to make a vignette (heavily based on @SamuelBrand1's) to demonstrate how to nowcast the NHSN signal using baselinenowcast and epidatr. @seabbs What do you think? @jcblemai Would this be useful and improve the workflow for your course and your operational needs?

I think the "Getting started" vignette from CDCgov/nowcastNHSN#12 will do this, you can just link to it when it gets to main? E.g. have a bit in README for other repos that are using baselinenowcast.

Ultimately, the nowcastNHSN thing will be collection of S3 methods for different ways of grabbing NHSN reporting triangle and a few S3 methods for applying some kind of nowcasting on top of the reporting triangle (starting with baselinenowcast). And someway of presenting the nowcasts as a service to modellers (TBD).