Outbreak Work Package

[dirkschumacher]

Expected output:

• Descriptive epi package
• Design function interfaces (time, place, person) (see epiet case studies)
• 1 rmarkdown document (word document as target)
• (Bonus) case studies and walkthroughs

Descriptive Epi. package features:

In general epidemiological analyses are structured around Time, Place and Person. Within these categories certain routine analyses are performed and the aim of the package is to make these analyses as easy as possible.

Time

• Draw various kinds of epicurves
  • Stacked barcharts (by e.g. case definition)
  • Normal barcharts
  • Incidence vs counts
  • Low vs high case load. "For small datasets cases are shown as rectangles" Added a part in the vignette customize_plot to show individual cases reconhub/incidence#68

• Investigate what additional features are needed or if we can just use incidence

Place

• Various support for maps
• Choropleth maps, point maps
• Incidence vs counts
• Support getting base data from Open Streetmap
• Get "official" shapefiles
• Relevant packages: sf, tmap, osmdata, osmplotr

Person

• Make it easy to get and combine population data
• Attack-rate
• Incidence vs counts
• 2x2 tables
• Multivariate analysis using logistic regression (e.g. glm)
• CFR plots and estimation (consider risk adjustment)
• Epi. trees vis.
• R0 estimation and various other basic epi. indicators
• Age pyramid. Or at least some visualization by age group
• Make it easy to report on OR and RR including CIs in reports

[dirkschumacher]

I am thinking it would be good to maybe have one central package that offers a highlevel, unified api to a range of functionality (internally using other packages)...

They could handle standard base R input or specific linelist data.frames with attached meta data for magic analyses.

[aspina7]

I think probably most important function for basic field epi is one that outputs a nice summary table with counts, attack rates and OddsRratioss or RRs with 95%CIs and possibly a p-value.
A nice example of this is Dan Gardiner's sva function (in terms of output). Mostly msf epis wont even move on to do regressions....

The other nice thing that dan has done is adapt the recon incidence package, to make an epicurve function (I know, he should have pulled and then recontributed...) to plot single cases quite nicely, the current solution in recon doesnt come out quite as neatly (i feel...).

Otherwise just a nice wrapper function for tables with counts and proportions (as chris mentioned) so that you dont have to pull bits from different functions. Also to include the option of getting confidence intervals for proportions using prop.test (important for cross-sections and samples, which do a lot in small field studies)

[dirkschumacher]

@aspina7 thanks.

1. Will work on a function for counts, attack rates and ORs.
2. Reg. epicurves: I guess the best option would be to improve incidence? cc @zkamvar
3. Reg. the wrapper function: do you have some concrete examples? What would be a good interface?

Will create issues for all of those points to better keep track of the ideas

[zkamvar]

Reg. epicurves: I guess the best option would be to improve incidence? cc @zkamvar

I'm way ahead of you there: reconhub/incidence#72

[dirkschumacher]

Uhh nice 👍 👍 👍

[dirkschumacher]

Started on a simple markdown document to gather sample analyses.
https://github.com/R4EPI/outbreaks/blob/master/oubreak-report.md

Will do the same for the survey stuff probably. Idea is to evolve these into something useable. Extract potential functions into package and discuss.

[aspina7]

@dirkschumacher re 3. the wrapper function
Sorry, brainfart - for the most part you wouldnt need CIs in an outbreak study but see reply to CFR discussion