Helen edits

[HelenCEBM]

Main changes:

• Added a new study def that isn't run by week
  • Because of the possible 5-day delay between +ve test and treatment, running by week means that patients who become eligible in week n (e.g. on Saturday) but aren't treated til week n+1 (e.g. on Monday) will be on separate lines, and therefore excluded when we remove duplicates. (We'll also remove people who had two positive tests on separate weeks but who were otherwise eligible). It's also difficult to calculate the proportion of eligible patients who received treatment, as the two cohorts won't overlap neatly within each week.
  • The same R processing scripts can still be applied successfully
• Added the third currently available non-hospital drug throughout analyses (though low numbers so may want to remove from detailed outputs)
• For by-week analysis: changed date periods from between = ["index_date", "index_date + 7 days"] to +6 days to avoid double counting

Minor things:

• Removed filter by status so that all "Approved" or previously approved treatments are included.
• Changed test date to positive test date
• Fixed typos
• Adjusted expectations

Questions / to-do:

• output file table_elig_treat_redacted isn't being redacted (perhaps because the cols contain n (x%)? (Also need to ensure that additional values are redacted where just one in row/column is <=5 so that values can't be inferred).
• Flow chart figures also need redacting
• I think all results should be rounded e.g. to nearest 10 to minimise issues with small number diffs from week to week.
• Not sure we should use treated_within_10_days in eligibility/exclusion criteria? These patients were still initially eligible whether they received treatment or not... We could maybe count patients separately somewhere who appeared to have the treatment too late.
• On charts, sort legend labels by line order - some of the colours are quite hard to distinguish.
• Need to exclude people with more than 2 different drugs (though it's not necessarily bad to count them once for coverage purposes)

Less critical suggestions

• Can feather format be used instead? (Hopefully this would avoid having to define dtypes for each column, as well as saving on space/processing?)
• Can we shorten the clinical group names, i.e. remove "Patients with [a]"?
• Why is ronapreve in one of the R scripts?

[gitpod-io[bot]]

[acagreen17]

Main changes all look good. As discussed will alter extraction dates for variables to be min of eligible date (based on positive covid test) and treated date.

• Changed test date to positive test date

I'd extracted date of any test and later filter on positive only, just to save having to extract the variable twice as will use it to see how many of the treated patients without a positive test had been tested. But probably easier to have as separate variable.

• output file table_elig_treat_redacted isn't being redacted (perhaps because the cols contain n (x%)? (Also need to ensure that additional values are redacted where just one in row/column is <=5 so that values can't be inferred).
• Flow chart figures also need redacting
• I think all results should be rounded e.g. to nearest 10 to minimise issues with small number diffs from week to week.

Yes, need to need to add in redaction. Rounded to nearest 10 sounds sensible too.

• Not sure we should use treated_within_10_days in eligibility/exclusion criteria? These patients were still initially eligible whether they received treatment or not... We could maybe count patients separately somewhere who appeared to have the treatment too late.

Okay, will remove and describe.

• On charts, sort legend labels by line order - some of the colours are quite hard to distinguish.

Will do.

• Need to exclude people with more than 2 different drugs (though it's not necessarily bad to count them once for coverage purposes)

Done - patients now excluded based on receiving two different drugs within 2 weeks of each other.

• Can feather format be used instead? (Hopefully this would avoid having to define dtypes for each column, as well as saving on space/processing?)

I did consider using feather but sometimes R can be funny with reading in the fields with feather, which is why I prefer to be able to define dtypes for each column. Much more of a faff, but will revisit if it becomes a pain.

• Can we shorten the clinical group names, i.e. remove "Patients with [a]"?

Yes.

• Why is ronapreve in one of the R scripts?

Will be an old script from when we were doing some investigating. I will find and put in the graveyard folder.