76 datetime inputs

[loffleraSMH]

Closes #76

Main changes

• I added a new utility function in utils.R that's called convert_dt: It converts any date-time variables into the format expected by other Rgemini functions (typically: “ymd HM”, which is used as the default format by convert_dt) and returns warning messages informing users about invalid/missing date-times and (if relevant) about entries with missing timestamps/timestamps that are 00:00
• I replaced all ymd_hm calls with convert_dt (even in cases where it’s not technically needed, but just to standardize the warning messages about invalid/missing date-times)
• Removed any lubridate imports where they are not needed anymore
• Updated @params sections (where necessary) to indicate acceptable formats (character or POSIXct POSIXt)
• These changes affect the following functions: length_of_stay(), er_los(), day_time_of_admission(), hospital_fiscal_year(), stat_holidays_ON(), episodes_of_care(), readmission(), daily_census(), icu_entry(), icu_los(), mlaps()
• All functions should now run even when users provide non-character/pre-processed date-time inputs. I’ve tested all outputs of these functions to make sure the results are as expected/same as before.
• Note: For n_routine_bloodwork(), n_imaging(), and n_rbc_transfusions() I haven’t made any changes yet to avoid merge conflicts with Yishan’s PR for issue #125. I will open a separate issue for those functions because a lot of the date-times are currently handled within the SQL query (so we’d need to refactor those functions first before we can apply convet_dt()).

For a few functions, I made some additional changes:

• er_los(): As discussed with Yishan, I made a small change to this function to return genc_ids with an entry in er but missing/invalid ER date-time as er_los = NA (instead of er_los = 0). This only has a minimal impact on the function output (<1% of all entries in the er table are affected by this). I’ve also added a warning about this in the function and updated the documentation accordingly.
• icu_los(): Similar to er_los, any genc_ids with a missing/invalid date-time in ipscu are now returned as icu_los = NA.
• icu_entry(): I discussed this with Bijun, for any time-sensitive ICU flags (e.g., ICU within 24h), we now return genc_ids with 0 valid ICU admission date-times as NA. For any genc_ids with at least 1 valid date-time, any valid entries are considered in the calculation. I’ve updated the documentation accordingly.
• In day_time_of_admission(): I switched the code to standard lubridate functions (instead of str operations) to allow for more flexible date-time formats and to make sure timestamps with 00:00 are handled appropriately (the previous version returned any entries with 00:00 as NA for daytime). Otherwise, the output of the current versions is the same as before.

Here is an example for how to test the code for length_of_stay():

ipadmdad <- dbGetQuery(db, "select * from admdad LIMIT 100") %>% data.table() 

# pre-process date-times prior to running length_of_stay (would break old version but should work with new version)
ipadmdad[, discharge_date_time := ymd_hm(discharge_date_time)]
ipadmdad[, admission_date_time := ymd_hm(admission_date_time)]
# introduce some missing date-times (should be shown in warning message when calling new version of length_of_stay
ipadmdad[1:10, discharge_date_time := NA]

# source relevant functions
source('H:/GitHub/Rgemini/R/length_of_stay.R')
source('H:/GitHub/Rgemini/R/utils.R')

los <- length_of_stay(ipadmdad)

[loffleraSMH]

Thanks so much for the review and the helpful feedback @guoyi-smh! I fixed the bug and implemented your other suggestions. Let me know if there are any other changes you'd like me to make.

[guoyi-smh]

Hi @loffleraSMH , all the updates look good. I have resolved all the comments. Please feel free to go ahead and merge the branch.