In Public-Health-Scotland/hsmr: Reproducible Analytical Pipeline of Hospital Standardised Mortality Ratios Quarterly Publication

Setup

# Load environment file
source(here::here("setup_environment.R"))

# Tidylog messages make notebook output messy
detach("package:tidylog", unload = TRUE)

library(glue)      # for glue()
library(DT)        # for tables

source("checking_functions.R")

# What is the minimum % difference to show in tables?
table_pc_cutoff = 5

Get Data

get_t3_data = function(date_str, sheet_name = "Raw Data"){

  path = paste0(glue("{data_folder}{date_str}/output/{date_str}"),
                "-Table3-Crude-Mortality-population-based-and-30-day-from-discharge.xlsx")

  data = 
    read.xlsx(path, sheet = sheet_name) %>%
    select(location_name, quarter_short, agg_label, sub_grp, 
           deaths, pats, crd_rate)

  return(data)

}

t3_old_full = get_t3_data(previous_pub)
t3_new_full = get_t3_data(pub_day)

We can only compare the quarters that overlap, find those:

overlap_quarters = 
  semi_join(t3_new_full, t3_old_full, by = "quarter_short") %>%
  select(quarter_short) %>%
  distinct()

Now only keep the quarters that overlap:

t3_old = filter(t3_old_full, quarter_short %in% overlap_quarters$quarter_short)
t3_new = filter(t3_new_full, quarter_short %in% overlap_quarters$quarter_short)

# We also don't need all of the columns here
t3_old = select(t3_old, -agg_label)
t3_new = select(t3_new, -agg_label)

Compare to previous publication

Changes for the first overlapping quarter (1st quarter in this release, 2nd quarter in previous release), the last overlapping quarter (last quarter in previous release, 2nd last quarter in this release), and the remaining quarters are given in separate tables.

• First quarter shown separately to highlight any deaths or patients rolling over from the first quarter in the previous release.
• Last quarter shown separately as that is the most recent one in the previous publication and therefore the most likely to have seen changes to the SMR data.

Get a dataframe with the change from the last publication:

# Columns containing the labels
join_cols = c("location_name", "quarter_short", "sub_grp")

t3_diff = compare_df(t3_new, t3_old, join_by = join_cols, 
                     df1_name = "new", df2_name = "old")

Patients

t3_diff_px = 
  t3_diff %>%
  select(any_of(join_cols), starts_with("pats"))

writeLines(glue("Max % diff: {max(t3_diff_px[, 'pats.diff_pc'], na.rm = T)}
                 Min % diff: {min(t3_diff_px[, 'pats.diff_pc'], na.rm = T)}"))

Compare absolute % difference to previous value:

make_change_plot(t3_diff_px, "pats")

First Quarter

Now look more closely at the larger % differences.

Start with the first quarter of overlapping data only (i.e. first quarter in new publication):

make_change_table(t3_diff_px, measure = "pats",  
                  pc_cutoff = table_pc_cutoff, quarter_filter = "first only")

Last Quarter

Now last quarter of overlapping data (i.e. last quarter in previous publication):

make_change_table(t3_diff_px, measure = "pats", 
                  pc_cutoff = table_pc_cutoff, quarter_filter = "last only")

Middle Quarters

Now the rest of the quarters:

make_change_table(t3_diff_px, measure = "pats", 
                  pc_cutoff = table_pc_cutoff, quarter_filter = "middle only")

Deaths

t3_diff_deaths = 
  t3_diff %>%
  select(any_of(join_cols), starts_with("deaths"))

writeLines(glue("Max % diff: {max(t3_diff_deaths[, 'deaths.diff_pc'], na.rm = T)}
                 Min % diff: {min(t3_diff_deaths[, 'deaths.diff_pc'], na.rm = T)}"))

Compare absolute % difference to previous value:

make_change_plot(t3_diff_deaths, "deaths")

First Quarter

Now look more closely at the larger % differences.

Start with the first quarter of overlapping data only (i.e. first quarter in new publication):

make_change_table(t3_diff_deaths, measure = "deaths", 
                  pc_cutoff = table_pc_cutoff, quarter_filter = "first only")

Last Quarter

Now last quarter of overlapping data (i.e. last quarter in previous publication):

make_change_table(t3_diff_deaths, measure = "deaths", 
                  pc_cutoff = table_pc_cutoff, quarter_filter = "last only")

Middle Quarters

Now the rest of the quarters:

make_change_table(t3_diff_deaths, measure = "deaths",  
                  pc_cutoff = table_pc_cutoff, quarter_filter = "middle only")

Crude Mortality Rate

t3_diff_crd_rate = 
  t3_diff %>%
  select(any_of(join_cols), starts_with("crd_rate"))

writeLines(glue("Max % diff: {max(t3_diff_crd_rate[, 'crd_rate.diff_pc'], na.rm = T)}
                 Min % diff: {min(t3_diff_crd_rate[, 'crd_rate.diff_pc'], na.rm = T)}"))

Compare absolute % difference to previous value:

make_change_plot(t3_diff_crd_rate, "crd_rate")

First Quarter

Now look more closely at the larger % differences.

Start with the first quarter of overlapping data only (i.e. first quarter in new publication):

make_change_table(t3_diff_crd_rate, measure = "crd_rate", 
                  pc_cutoff = table_pc_cutoff, quarter_filter = "first only")

Last Quarter

Now last quarter of overlapping data (i.e. last quarter in previous publication):

make_change_table(t3_diff_crd_rate, measure = "crd_rate",
                  pc_cutoff = table_pc_cutoff, quarter_filter = "last only")

Middle Quarters

Now the rest of the quarters:

make_change_table(t3_diff_crd_rate, measure = "crd_rate",
                  pc_cutoff = table_pc_cutoff, quarter_filter = "middle only")

Check splits sum to totals

Get the difference between the sum of each split and the Scotland totals

t3_split_tot = split_tot_df(t3_new_full, hsmr_table = 3)

split_id_cols = c("agg_label", "quarter_short", "sub_grp")

t3_split_tot_deaths = 
  select(t3_split_tot, any_of(split_id_cols), starts_with("deaths"))

t3_split_tot_pats = 
  select(t3_split_tot, any_of(split_id_cols), starts_with("pats"))

Patients

make_split_table(t3_split_tot_pats)

Deaths

make_split_table(t3_split_tot_deaths)

Public-Health-Scotland/hsmr documentation built on June 24, 2024, 1:48 a.m.