# Deal with the questions from the ICB meeting on this ACS work
library(devtools)
library(icdb)
library(dplyr)
library(tidyr)
library(ggplot2)
library(grid)
library(rayshader)
library(viridis)
library(bigsnpr)
library(kableExtra)
rm(list=ls())
load_all()
# Connect to the server
svr <- icdb::server("XSW")
# Some cohort parameters
start <- "2018-07-01"
end <- "2021-12-31"
adm <- acsprojectns::get_codes(file_path = system.file("admission_method_codes/admission_method_codes.csv", package = "acsprojectns"))
ccg <- acsprojectns::get_codes(file_path = system.file("ccg_provider_codes/bnssg_ccg_provider_codes.csv", package = "acsprojectns"))
# The PCN for each practice
dat_pcn <- svr$MODELLING_SQL_AREA$swd_attribute %>%
distinct(practice_code, primary_care_network) %>%
mutate(practice_code = toupper(practice_code)) %>%
select(pcn=primary_care_network, prac_code=practice_code) %>%
run() %>%
mutate(pcn = as.factor(pcn))
# Get the first ACS episode within the time window
dat_sus <- svr$ABI$vw_APC_SEM_001 %>%
select(AIMTC_Pseudo_NHS,
AIMTC_Age,
Sex,
EthnicGroup,
AIMTC_PCTRegGP,
AIMTC_PracticeCodeOfRegisteredGP,
AdmissionMethod_HospitalProviderSpell,
AIMTC_ProviderSpell_Start_Date,
AIMTC_ProviderSpell_End_Date,
DiagnosisPrimary_ICD) %>%
filter(AIMTC_ProviderSpell_Start_Date > start,
AIMTC_ProviderSpell_End_Date < end,
AdmissionMethod_HospitalProviderSpell %in% !!adm$admission_method_code[grep("emergency", adm$admission_method_description)],
AIMTC_PCTRegGP %in% !!ccg$ccg_provider_code) %>%
codes_from("icd10/acs.yaml", DiagnosisPrimary_ICD) %>%
run() %>%
group_by(AIMTC_Pseudo_NHS) %>%
arrange(AIMTC_ProviderSpell_Start_Date, .by_group=T) %>%
fill(c(AIMTC_Age, Sex, EthnicGroup), .direction="downup") %>%
slice_min(AIMTC_ProviderSpell_Start_Date, with_ties = FALSE) %>%
ungroup() %>%
mutate(DiagnosisPrimary_ICD = drop_detail(.data$DiagnosisPrimary_ICD, level=2)) %>%
select(nhs_number = AIMTC_Pseudo_NHS,
age = AIMTC_Age,
sex = Sex,
ethnicity = EthnicGroup,
prac_code = AIMTC_PracticeCodeOfRegisteredGP,
epi_start = AIMTC_ProviderSpell_Start_Date,
epi_end = AIMTC_ProviderSpell_End_Date,
primary_diagnosis = DiagnosisPrimary_ICD)
# The SWD date range
dat_attr <- svr$MODELLING_SQL_AREA$primary_care_attributes %>%
select(nhs_number, attribute_period, sex, ethnicity, lsoa, prim_language) %>%
filter(nhs_number %in% !!dat_sus$nhs_number) %>%
run() %>%
group_by(nhs_number) %>%
mutate(swd_min = min(attribute_period, na.rm=T),
swd_max = max(attribute_period, na.rm=T)) %>%
arrange(attribute_period, .by_group=T) %>%
fill(c(sex, ethnicity, lsoa), .direction="downup") %>%
slice_max(attribute_period, with_ties = FALSE) %>%
ungroup() %>%
select(nhs_number, sex, ethnicity, lsoa, swd_min, swd_max, prim_language)
# LSOA : MSOA mapping
dat_msoa <- svr$MODELLING_SQL_AREA$swd_LSOA_descriptions %>%
select(lsoa = `LSOA code`,
lsoa_name = `LSOA name`) %>%
run() %>%
mutate(msoa = substr(.data$lsoa_name, 1, nchar(.data$lsoa_name)-1)) %>%
select(lsoa, msoa)
# Cambridge score
dat_camb <- svr$MODELLING_SQL_AREA$New_Cambridge_Score %>%
select(nhs_number, attribute_period, segment) %>%
filter(nhs_number %in% !!dat_sus$nhs_number) %>%
group_by(nhs_number) %>%
slice_max(attribute_period) %>%
dplyr::select(nhs_number, segment) %>%
icdb::run()
# IMD score
dat_imd <- svr$Analyst_SQL_Area$tbl_BNSSG_Datasets_LSOA_IMD_2019 %>%
select(imd = `Index of Multiple Deprivation (IMD) Decile`,
lsoa = `LSOA Code`) %>%
icdb::run()
# Create the cohort
ethnicity_codes <- get_codes(file_path = system.file("ethnicity_codes/ethnicity_codes.csv", package = "acsprojectns"))
sex_codes <- get_codes(file_path = system.file("sex_codes/sex_codes.csv", package = "acsprojectns"))
cohort <- dat_sus %>%
left_join(dat_pcn, by="prac_code") %>%
left_join(dat_attr, by="nhs_number") %>%
left_join(dat_imd, by="lsoa") %>%
left_join(dat_msoa, by="lsoa") %>%
left_join(dat_camb, by="nhs_number") %>%
mutate(sex_code = do.call(dplyr::coalesce, across(contains("sex"))),
ethnicity_code = do.call(dplyr::coalesce, across(contains("ethnicity")))) %>%
left_join(sex_codes, by="sex_code") %>%
left_join(ethnicity_codes, by="ethnicity_code") %>%
select(nhs_number, age, sex, ethnicity, ethnicity_group, prim_language, lsoa, msoa, prac_code, pcn, imd, segment, epi_start, epi_end, primary_diagnosis, swd_min, swd_max)
# The prescriptions
med_filter_string <- "(?i)([A-z ]+)(\\d+\\.?\\d*)?([A-z]+)?[ ]*(?:[A-z]*)"
search_meds <- c("inclisiran", "ezetimibe", "atorvastatin", "rosuvastatin", "simvastatin", "fluvastatin", "pravastatin")
med_end <- as.Date(end) + lubridate::duration(1, units="years")
dat_meds <- svr$MODELLING_SQL_AREA$swd_activity %>%
select(nhs_number, arr_date, dep_date, pod_l1, spec_l1b) %>%
filter(nhs_number %in% !!dat_sus$nhs_number,
arr_date > start,
dep_date < med_end,
pod_l1 == "primary_care_prescription") %>%
run() %>%
filter(grepl(paste0(search_meds, collapse = "|"), spec_l1b, ignore.case=T)) %>%
dplyr::mutate(med_name = trimws(stringr::str_match(spec_l1b, pattern = med_filter_string)[,2]),
med_dose = stringr::str_match(spec_l1b, pattern = med_filter_string)[,3],
med_units = stringr::str_match(spec_l1b, pattern = med_filter_string)[,4]) %>%
mutate(lipid_strat = case_when(is.na(med_name) ~ "None",
med_name=="inclisiran" |
(med_name=="atorvastatin" & med_dose >=40) |
(med_name=="rosuvastatin" & med_dose >=20) |
(med_name=="simvastatin" & med_dose >=80) ~ "High intensity",
TRUE ~ "Low or Medium Intensity")) %>%
select(nhs_number, prescription_date = arr_date, med_name, med_dose, med_units, lipid_strat)
#############################################
## ANALYSIS
strategy_over_time <- purrr::map_df(.x = seq(3,12,3),
.f = function(win, win_unit="months", dat_sus, dat_attr, dat_meds){
cohort_with_swd_coverage <- cohort %>%
# filter out people that do not have swd data covering the period of interest
filter(if_any(c(swd_min, swd_max, epi_start, epi_end), ~!is.na(.x)),
lubridate::interval(epi_end, epi_end+lubridate::duration(win,win_unit)) %within% lubridate::interval(swd_min, swd_max))
avg_med_strat_in_window <- cohort %>%
# Join the medication data
left_join(dat_meds, by="nhs_number") %>%
# Filter out the prescriptions that are out of range
filter(prescription_date %within% lubridate::interval(epi_end, epi_end+lubridate::duration(win,win_unit))) %>%
# Group by each patient
group_by(nhs_number) %>%
# Get the lipid strategy for this time period for each patient
summarise(strat_mode = tail(names(sort(table(lipid_strat))), 1)) %>%
ungroup()
window_cohort_med_strat <- cohort_with_swd_coverage %>%
# Join in the medication strategy to the cohort
left_join(avg_med_strat_in_window, by="nhs_number") %>%
# Add the time period and replace the NAs with 'None'
mutate(period = as.factor(paste0(win, "_", win_unit)),
pcn = as.factor(pcn),
prac_code = as.factor(prac_code),
strat_mode = replace_na(strat_mode, "None"),
strat_mode = factor(strat_mode, levels=c("None","Low or Medium Intensity","High intensity")))
return(window_cohort_med_strat)
}, dat_sus=dat_sus, dat_attr=dat_attr, dat_meds=dat_meds)
#############################################
## ANALYSIS
strategy_over_time %>%
filter(period=="3_months") %>%
group_by(strat_mode) %>%
summarise(n=n(),
sex_male = sum(sex=="male", na.rm=T)/n,
age = mean(age, na.rm=T),
eth_white = sum(ethnicity_group=="White", na.rm=T)/n,
eth_black = sum(ethnicity_group=="Black", na.rm=T)/n,
eth_asian = sum(ethnicity_group=="Asian", na.rm=T)/n,
pct_mixed = sum(ethnicity_group=="Mixed", na.rm=T)/n,
pct_other = sum(ethnicity_group=="Other", na.rm=T)/n,
pct_unkwn = sum(is.na(ethnicity_group))/n,
imd = mean(imd, na.rm=T),
segment = mean(segment, na.rm=T)) %>%
kbl() %>%
kable_minimal()
# Bar chart of average strategy use over time
strategy_over_time %>%
group_by(period) %>%
mutate(n_period=n()) %>%
group_by(period, strat_mode) %>%
summarise(pct = n()/n_period[[1]]) %>%
ggplot(aes(x=forcats::fct_rev(period), y=pct, fill=forcats::fct_rev(strat_mode))) +
geom_col() +
coord_flip()
# Bar chart of average strategy per PCN
strategy_over_time %>%
group_by(period, pcn) %>%
mutate(n_per_group=n()) %>%
group_by(period, pcn, strat_mode) %>%
summarise(pct = n()/n_per_group[[1]], n=n_per_group[[1]]) %>%
mutate(pcn = factor(pcn, levels=filter(.,period=="3_months",strat_mode=="None") %>% arrange(pct) %>% pull(pcn) %>% as.character(.))) %>%
filter(!is.na(pcn)) %>%
ggplot(aes(x=pcn, y=pct, fill=forcats::fct_rev(strat_mode))) +
geom_col() +
geom_text(data = . %>% group_by(pcn) %>% slice(n=1) %>% select(n, strat_mode),
aes(y=1, label=n), size=2, hjust = 1) +
labs(title = "Prescribed lipid lowering therapy post heart attack",
subtitle = "Any single prescription within timeframe; (n=number of PCN heart attack patients)",
fill="Lipid lowering therapy",
x = "Primary Care Network",
y = "Percentage") +
coord_flip() +
facet_wrap(~period, nrow=1)
# PCN data
df_pcn <- strategy_over_time %>%
group_by(period, pcn) %>%
mutate(n_per_group=n()) %>%
group_by(period, pcn, strat_mode) %>%
summarise(pct = n()/n_per_group[[1]], n=n_per_group[[1]]) %>%
mutate(pcn = factor(pcn, levels=filter(.,period=="3_months",strat_mode=="None") %>% arrange(pct) %>% pull(pcn) %>% as.character(.))) %>%
filter(!is.na(pcn)) %>%
ungroup()
# 5 PCN with most no therapy patients
pcn5 <- df_pcn %>%
filter(period=="3_months",
strat_mode=="None") %>%
mutate(num_none = pct*n) %>%
slice_max(num_none, n=5) %>%
pull(pcn)
# Demographics of the 5 worst PCNs
strategy_over_time %>%
filter(pcn %in% pcn5) %>%
group_by(pcn) %>%
summarise(n = n(),
min_age= min(age, na.rm=T),
av_age = mean(age, na.rm=T),
av_imd = mean(imd, na.rm=T),
av_camb= mean(segment, na.rm=T),
pct_stemi= sum(primary_diagnosis=="acs.stemi")/n(),
n_none = sum(strat_mode=="None"),
pct_none = sum(strat_mode=="None")/n(),
pct_low = sum(strat_mode=="Low or Medium Intensity")/n(),
pct_high = sum(strat_mode=="High intensity")/n()) %>%
arrange(desc(n_none)) %>%
kbl() %>%
kable_minimal()
# Bar chart of average strategy per practice
df_prac <- strategy_over_time %>%
group_by(period, prac_code) %>%
mutate(n_per_group=n()) %>%
group_by(period, prac_code, strat_mode) %>%
summarise(pct = n()/n_per_group[[1]], n=n_per_group[[1]]) %>%
mutate(prac_code = factor(prac_code, levels=filter(.,period=="3_months",strat_mode=="None") %>% arrange(pct) %>% pull(prac_code) %>% as.character(.))) %>%
filter(!is.na(prac_code)) %>%
ungroup()
# GP practice data
gp_prac_data = read.csv(system.file("gp_practice_details", "epraccur.csv", package="acsprojectns"),header = FALSE) %>%
select(prac_code = V1,
prac_name = V2,
prac_postcode = V10)
# 10 practices with most no therapy patients
gp10 <- df_prac %>%
filter(period=="3_months",
strat_mode=="None") %>%
mutate(num_none = pct*n) %>%
slice_max(num_none, n=10) %>%
left_join(gp_prac_data, by="prac_code") %>%
pull(prac_code)
# Demographics of the 10 worst GP practices
strategy_over_time %>%
filter(prac_code %in% gp10) %>%
group_by(prac_code) %>%
summarise(n = n(),
min_age= min(age, na.rm=T),
av_age = mean(age, na.rm=T),
av_imd = mean(imd, na.rm=T),
av_camb= mean(segment, na.rm=T),
pct_stemi= sum(primary_diagnosis=="acs.stemi")/n(),
n_none = sum(strat_mode=="None"),
pct_none = sum(strat_mode=="None")/n(),
pct_low = sum(strat_mode=="Low or Medium Intensity")/n(),
pct_high = sum(strat_mode=="High intensity")/n()) %>%
arrange(desc(n_none)) %>%
left_join(gp_prac_data, by="prac_code")
# Bar chart of average strategy per practice
df_prac %>%
ggplot(aes(x=prac_code, y=pct, fill=forcats::fct_rev(strat_mode))) +
geom_col() +
geom_text(data = . %>% group_by(prac_code) %>% slice(n=1) %>% select(n, strat_mode),
aes(y=1, label=n), size=2, hjust = 1) +
coord_flip() +
labs(title = "Prescribed lipid lowering therapy post heart attack",
subtitle = "Cohort July 2018 - Dec 2021",
fill="Lipid lowering therapy",
xlab = "Percentage",
ylab= "GP practice") +
facet_wrap(~period, nrow=1)
# # Ethnicity across practices
# strategy_over_time %>%
# group_by(pcn) %>% # Number in each practice
# mutate(n_prac=n()) %>%
# group_by(period, pcn) %>% # Number in each practice in each period
# mutate(n_period_prac=n()) %>%
# group_by(period, pcn, strat_mode) %>% # % in each medication strategy, in each period, in each practice
# mutate(pct_strategy = n()/n_period_prac[[1]]) %>%
# mutate(pcn = factor(pcn, levels = filter(., period=="3_months",strat_mode=="None") %>% # Reorder practice code factor by the % of patients with "None" therapy at 3 months
# arrange(pct_strategy) %>%
# pull(pcn) %>%
# unique(.) %>%
# as.character(.))) %>%
# filter(!is.na(pcn)) %>% # Ditch NA practices
# group_by(pcn, ethnicity_group) %>% # % each ethnicity in each practice
# summarise(pct_ethnic_group = n()/n_prac[[1]],
# n_prac = n_prac[[1]]) %>%
# ungroup() %>%
# complete(pcn, ethnicity_group, fill=list(pct_ethnic_group=0)) %>%
# ggplot(aes(x=pcn, y=pct_ethnic_group, fill=forcats::fct_rev(ethnicity_group))) +
# geom_col() +
# coord_flip() +
# labs(title = "Prescribed lipid lowering therapy post heart attack",
# subtitle = "Cohort July 2018 - Dec 2021",
# fill="Lipid lowering therapy",
# xlab = "Percentage",
# ylab= "GP practice") +
# facet_wrap(~period, nrow=1)
# Maps - percentage not on lipid therapy at 3 months
ggplot() +
geom_polygon(data = gen_msoa_geo_data(dat=strategy_over_time %>%
filter(period=="3_months") %>%
group_by(msoa) %>%
mutate(n=n()) %>%
group_by(msoa, strat_mode) %>%
filter(strat_mode=="None") %>%
summarise(values=n()/n[[1]]) %>%
select(msoa, values)),
mapping = aes(x=long, y=lat, group=group, fill=values), alpha=0.9, colour="white", size=0.1) +
theme_void() +
coord_map() +
labs(title = "Percetage of heart attack patients not prescribed lipid lowering\ntherapy at 3 months post event",
subtitle = "BNSSG middle layer super output areas (July 2018 - Dec 2021)",
fill="Percentage") +
scale_color_discrete(guide = "none") +
scale_fill_viridis_c(option = "plasma", begin = 0.0, end=0.99) +
theme(plot.title = element_text(size= 16, hjust=0.01, color = "#4e4d47", margin = margin(b = -0.1, t = 0.40, l = 2, unit = "cm")),
plot.subtitle = element_text(size= 12, hjust=0.01, color = "#4e4d47", margin = margin(b = -0.1, t = 0.43, l = 2, unit = "cm")),
legend.position = c(0.20, 0.65)
)
# Maps - absolut numbers not on lipid therapy at 3 months
ggplot() +
geom_polygon(data = gen_msoa_geo_data(dat=strategy_over_time %>%
filter(period=="3_months") %>%
group_by(msoa) %>%
summarise(values=mean(n())) %>%
select(msoa, values)),
mapping = aes(x=long, y=lat, group=group, fill=values), alpha=0.9, colour="white", size=0.1) +
theme_void() +
coord_map() +
scale_fill_continuous(name = 'Num. ACS patients', low="lightgray", high="black") +
geom_point(data = cbind(gen_msoa_geo_data(dat=strategy_over_time %>%
filter(period=="3_months") %>%
group_by(msoa) %>%
mutate(n=n()) %>%
filter(strat_mode=="None") %>%
summarise(values=n()/n[[1]]) %>%
select(msoa, values)),
gen_msoa_geo_data(dat=strategy_over_time %>%
filter(period=="3_months" & (strat_mode=="None")) %>%
group_by(msoa) %>%
summarise(values=n()) %>%
select(msoa, values)) %>%
select(n=values)) %>%
group_by(msoa) %>%
summarise(lat=mean(lat), long=mean(long), value=mean(values), n=mean(n)),
mapping = aes(x=long, y=lat, color=value, size=n)) +
scale_colour_gradient(name = '% no therapy', low="green", high="red") +
scale_size_continuous(name = 'Num. no therapy', range = c(1, 6), breaks = seq(0, 20, 5)) +
# scale_size_continuous(range = c(1, 5)) +
labs(title = "Number of heart attack patients not prescribed lipid lowering\ntherapy at 3 months post event",
subtitle = "BNSSG middle layer super output areas (July 2018 - Dec 2021)",
fill="Heart attack patients") +
theme(plot.title = element_text(size= 16, hjust=0.01, color = "#4e4d47", margin = margin(b = -0.1, t = 0.40, l = 2, unit = "cm")),
plot.subtitle = element_text(size= 12, hjust=0.01, color = "#4e4d47", margin = margin(b = -0.1, t = 0.43, l = 2, unit = "cm")),
legend.position = c(0.20, 0.65)
)
# Target graphs
# Number (and %) of ACS patients per PCN
# Number (and %) of ACS patients on statins in year after ACS
#
#
# # What proportion of heart attack patients in this cohort are current smokers?
# smokers <- svr$MODELLING_SQL_AREA$primary_care_attributes() %>%
# select(nhs_number, attribute_period, smoking) %>%
# filter(nhs_number %in% !!cohort$AIMTC_Pseudo_NHS) %>%
# show_query() %>%
# run()
#
# # Does deprivation impact outcomes and costs?
# imd_centile
#
# Are outcomes different in areas with higher adjusted event rates (e.g. Weston)?
# Is the pattern of spend that is observed in the follow up costs (e.g. 4/12% patients = 25/50% costs) also observed at the index event?
# Do people with high index costs also have high follow up costs?
# What are the most frequent diagnoses for the subsequent hospital admissions following a heart attack admission?
# Regarding secondary prevention uptake and outcomes:
# How many people attend a cardiac rehabilitation programme?
# Who attends cardiac rehabilitation programmes and who does not?
# How do outcomes differ between those patients that attended a cardiac rehabilitation programme and those that did not?
# How do patients that do and do not have follow up checks (BP, HbA1c, cholesterol) differ in terms of the following [measures].
# How do patients that do and do not receive high-intensity lipid lowering therapy differ in terms of the following [measures].
# Measures:
# • MSOA age-adjusted ACS rate
# • Age
# • Sex
# • Ethnicity
# • English as first spoken language
# • Deprivation
# • Cambridge score
# Look at who has SWD data by practice
ggplot(data = cohort %>% group_by(pcn) %>% summarise(n=n(), pct_swd=sum(in_swd)/n()) %>% ungroup(),
mapping = aes(x=forcats::fct_reorder(pcn, pct_swd, .desc=T), y=pct_swd)) +
geom_bar(stat="identity", fill="lightblue") +
geom_text(aes(label=n), vjust=0.4, hjust=2) +
coord_flip() +
ylab('Percent of ACS patients with follow up SWD data') +
scale_y_continuous(labels = scales::percent)
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.