elderly_visit_b4_death.Rmd

---
title: "R Notebook"
output: html_notebook
---

```{r}
library(dplyr)
library(lubridate)
library(ggplot2)
library(tidyr)
library(tables)

library(doParallel)
all_cores <- parallel::detectCores(logical = FALSE)
registerDoParallel(cores = all_cores)
```

# triage

```{r}
# Jul2020-Jul2021 65+
ReferenceKeyWhitelist = stata %>% 
  filter(age >= 65, 
         eventdate >= as.Date("2020-07-01"), 
         eventdate < as.Date("2021-07-01")) %>% 
    select(ReferenceKey, DateofRegisteredDeath, eventdate, doadba, death, TriageCategory) %>%
  group_by(ReferenceKey) %>%
  arrange(eventdate) %>%
  summarise(DateofRegisteredDeath = last(DateofRegisteredDeath),
            TriageCategory = last(TriageCategory),
            death = last(death),
            doadba = last(doadba),
            .groups="drop") 

nrow(ReferenceKeyWhitelist)
# format is "yyy-mm-dd"
startdate = "2016-01-01"
elderly_death_visit = stata %>% 
  filter(year(eventdate) > 2015,
         eventdate < as.Date("2020-07-01"),
         eventdate >= as.Date(startdate)
         ) %>%
  filter(ReferenceKey %in% (ReferenceKeyWhitelist %>% pull(ReferenceKey))) 

df = elderly_death_visit %>% 
  select(ReferenceKey, eventdate) %>%
  mutate(period = ifelse(eventdate >= as.Date("2020-01-25"), "covid", "non-covid"))

df2 = df %>% 
  group_by(ReferenceKey, period) %>%
  summarise(visits = n()) %>%
  ungroup() %>% 
  # complete combinations of referencekey and period
  complete(ReferenceKey, period) %>%
  group_by(ReferenceKey) %>%
  summarise(covid = visits[period == "covid"],
            `non-covid` = visits[period == "non-covid"]) %>%
  replace_na(list(covid = 0, `non-covid` = 0)) %>%
  ungroup()


df3 = df2 %>%
  left_join(ReferenceKeyWhitelist, by=c("ReferenceKey")) %>%
  mutate(DateofRegisteredDeath = ifelse(is.na(DateofRegisteredDeath), "2021-07-01", DateofRegisteredDeath),
         covidIR = covid / as.numeric(difftime("2020-07-01","2020-01-25")),
         noncovidIR = `non-covid` / as.numeric(difftime("2020-01-24",startdate)),
         deltaIR = covidIR - noncovidIR)  %>% 
  mutate(change = ifelse(deltaIR < 0, "reduce", "increase"),
         change = ifelse(deltaIR == 0, "nochange", change),
         status = "survived",
         status = ifelse(death==1, "28d_death", status),
         status = ifelse(doadba==1, "doadba", status),
         ) %>%
  filter(`non-covid` > 8)

table(df3$change, df3$status, df3$TriageCategory)

ylims <- df3 %>%
  mutate(deltaIR =  deltaIR * 100) %>%
  group_by(status, TriageCategory) %>%
  summarise(Q1 = quantile(deltaIR, 1/4), Q3 = quantile(deltaIR, 3/4)) %>%
  ungroup() %>%
  #get lowest Q1 and highest Q3
  summarise(lowQ1 = min(Q1), highQ3 = max(Q3))

# cumulative plot
ggplot(data = df3%>% filter(!is.na(status), TriageCategory == 1), aes(x = deltaIR * 100, col = status))+
  geom_step(aes(y=..y..),stat="ecdf") +
  xlim(ylims$lowQ1, ylims$highQ3) +
  ylab("Cumulative portion") +
  xlab("Visit change per 100 days in covid period, against pre-covid period") 
  # facet_wrap(~old_home, labeller = labeller(old_home = labels)) 
  
```

# DOA/DBA

```{r}
# Jul2020-Jul2021 65+
ReferenceKeyWhitelist = stata %>% 
  filter(age >= 65, 
         eventdate >= as.Date("2020-07-01"), 
         eventdate < as.Date("2021-07-01"),
         doadba == 1) %>% 
    select(ReferenceKey, DateofRegisteredDeath, eventdate, TriageCategory) %>%
  group_by(ReferenceKey) %>%
  arrange(eventdate) %>%
  summarise(DateofRegisteredDeath = last(DateofRegisteredDeath),
            TriageCategory = last(TriageCategory),
            .groups="drop") 

nrow(ReferenceKeyWhitelist)
# format is "yyy-mm-dd"
startdate = "2016-01-01"
elderly_death_visit = stata %>% 
  filter(year(eventdate) > 2015,
         eventdate < as.Date("2020-07-01"),
         eventdate >= as.Date(startdate)
         ) %>%
  filter(ReferenceKey %in% (ReferenceKeyWhitelist %>% pull(ReferenceKey))) 

df = elderly_death_visit %>% 
  select(ReferenceKey, eventdate) %>%
  mutate(period = ifelse(eventdate >= as.Date("2020-01-25"), "covid", "non-covid"))

df2 = df %>% 
  group_by(ReferenceKey, period) %>%
  summarise(visits = n()) %>%
  ungroup() %>% 
  # complete combinations of referencekey and period
  complete(ReferenceKey, period) %>%
  group_by(ReferenceKey) %>%
  summarise(covid = visits[period == "covid"],
            `non-covid` = visits[period == "non-covid"]) %>%
  replace_na(list(covid = 0, `non-covid` = 0)) %>%
  ungroup()


df3 = df2 %>%
  left_join(ReferenceKeyWhitelist, by=c("ReferenceKey")) %>%
  mutate(DateofRegisteredDeath = ifelse(is.na(DateofRegisteredDeath), "2021-07-01", DateofRegisteredDeath),
         covidIR = covid / as.numeric(difftime("2020-07-01","2020-01-25")),
         noncovidIR = `non-covid` / as.numeric(difftime("2020-01-24",startdate)),
         deltaIR = covidIR - noncovidIR)  %>% 
  mutate(change = ifelse(deltaIR < 0, "reduce", "increase"),
         change = ifelse(deltaIR == 0, "nochange", change),
         )

table(df3$change,df3$TriageCategory)
chisq.test(df3$change, df3$old_home, correct = 1)
```

# DOA/DBA in covid and pre-covid

```{r}
ReferenceKeyWhitelist1 = stata %>% 
  filter(age >= 65, doadba == 1, eventdate >= "2018-07-01", eventdate < "2019-07-01") %>%
  select(ReferenceKey, DateofRegisteredDeath, eventdate, old_home) %>%
  group_by(ReferenceKey) %>%
  arrange(eventdate) %>%
  summarise(DateofRegisteredDeath = last(DateofRegisteredDeath),
            old_home = last(old_home),
            .groups="drop") 

ReferenceKeyWhitelist2 = stata %>% 
  filter(age >= 65, doadba == 1, eventdate >= "2020-07-01", eventdate < "2021-07-01") %>%
  select(ReferenceKey, DateofRegisteredDeath, eventdate, old_home) %>%
  group_by(ReferenceKey) %>%
  arrange(eventdate) %>%
  summarise(DateofRegisteredDeath = last(DateofRegisteredDeath),
            old_home = last(old_home),
            .groups="drop") 

trace_visits = function(ReferenceKeyWhitelist, 
                        startctrldate = "2019-01-01", 
                        endtrtdate = "2020-07-01", 
                        enddate = "2021-07-01",
                        minnoncovidvisits = 8, 
                        starttrtdate = "2020-01-25"){
  # format is "yyy-mm-dd"
  elderly_death_visit = stata %>% 
    filter(year(eventdate) > 2015,
           eventdate < as.Date(endtrtdate),
           eventdate >= as.Date(startctrldate)
           ) %>%
    filter(ReferenceKey %in% (ReferenceKeyWhitelist %>% pull(ReferenceKey))) 
  
  df = elderly_death_visit %>% 
    select(ReferenceKey, eventdate) %>%
    mutate(period = ifelse(eventdate >= as.Date(starttrtdate), "covid", "non-covid"))
  
  df2 = df %>% 
    group_by(ReferenceKey, period) %>%
    summarise(visits = n()) %>%
    ungroup() %>% 
    # complete combinations of referencekey and period
    complete(ReferenceKey, period) %>%
    group_by(ReferenceKey) %>%
    summarise(covid = visits[period == "covid"],
              `non-covid` = visits[period == "non-covid"]) %>%
    replace_na(list(covid = 0, `non-covid` = 0)) %>%
    # remove patients who did not visit AE once pre-covid
    filter(`non-covid` > minnoncovidvisits) %>%
    ungroup()
  
  df3 = df2 %>%
    left_join(ReferenceKeyWhitelist, by=c("ReferenceKey")) %>%
    mutate(
      # probly useless
      DateofRegisteredDeath = ifelse(is.na(DateofRegisteredDeath), enddate, DateofRegisteredDeath),
           covidIR = covid / as.numeric(difftime(endtrtdate, starttrtdate)),
           noncovidIR = `non-covid` / as.numeric(difftime(starttrtdate ,startctrldate)-1),
           deltaIR = covidIR - noncovidIR,
           IRR = covidIR / noncovidIR)  %>% 
    mutate(change = ifelse(IRR < 1, "reduce", "increase"),
           change = ifelse(IRR == 1, "no change", change),
           change = ifelse(IRR == 0, "no visit", change),
           )
  df3
}


df.covid = trace_visits(ReferenceKeyWhitelist2, minnoncovidvisits = 2)
df.precovid = trace_visits(ReferenceKeyWhitelist1, 
                           startctrldate = "2017-01-01", 
                           endtrtdate = "2018-07-01", 
                           starttrtdate = "2018-01-01", 
                           enddate = "2019-07-01",
                           minnoncovidvisits = 2)

table(df.covid$change, df.covid$old_home)
table(df.precovid$change,  df.precovid$old_home)
df = rbind(df.covid %>% mutate(period = "covid"),
                df.precovid %>% mutate(period = "precovid")) 
# chisq.test(df.covid$change, df.covid$old_home)
# chisq.test(df.precovid$change, df.precovid$old_home)

table(df[df$old_home=="N",]$period, 
      df[df$old_home == "N",]$change)
table(df[df$old_home=="Y",]$period,  
      df[df$old_home == "Y",]$change)
chisq.test(df[df$old_home=="N",]$period,
           df[df$old_home == "N",]$change)

chisq.test(df[df$old_home=="Y",]$period,
           df[df$old_home == "Y",]$change)

table(df$period,
           df$change)

chisq.test(df$period,
           df$change)
# # cumulative plot
# ggplot()+
#   geom_step(data = df.covid, aes(x = deltaIR * 100, y=..y..),stat="ecdf", col="red") +
#   geom_step(data = df.precovid, aes(x = deltaIR * 100, y=..y..),stat="ecdf", col="black") +
#   xlim(-3, 3) +
#   ylab("Cumulative portion") +
#   xlab("Attn rate (IR) change per 100 days in trt period, against ctrl period")
  # xlab("Attn rate (IR) change per 100 days in 25Jan2020-30Jun2020, against prior 1Jan2019-24Jan2020")
  # xlab("Attn rate (IR) change per 100 days in 1Jan2018-30Jun2018, against prior 1yr 1Jan2017-31Dec2017") 
  # facet_wrap(~old_home, labeller = labeller(old_home = labels)) 
library(ggpubr)
my_comparisons <- list( c("covid", "precovid"), c("covid", "precovid"))
ggboxplot(rbind(df.covid %>% mutate(period = "covid"),
                df.precovid %>% mutate(period = "precovid")), 
          x = "period", y = "IRR", 
          color = "old_home", palette = c("#00AFBB", "#E7B800"),
        ylab = "IRR", xlab = "Period")  + 
  scale_y_log10() + 
  stat_compare_means(aes(group = old_home))

ggboxplot(rbind(df.covid %>% mutate(period = "covid"),
                df.precovid %>% mutate(period = "precovid")), 
          x = "period", y = "IRR", 
          color = "period", palette = c("#00AFBB", "#E7B800"),
        ylab = "IRR", xlab = "Period")  + 
  scale_y_log10() + 
  stat_compare_means()

# density plot
ggplot()+
  geom_density(data = df.covid, aes(x = IRR ), col="red") +
  geom_density(data = df.precovid, aes(x = IRR ), col="black") +
  xlim(-1, 3) +
  facet_wrap(~old_home, nrow=2) 
  # ylab("Cumulative portion") +
  # xlab("Attn rate (IR) change per 100 days in trt period, against ctrl period")
signi_f = function(x){
  signif(x, 3) 
}
df.covid %>%
  group_by(old_home) %>%
  summarise(NumberOfPatients = n(),
            Q1 = quantile(IRR, 1/4), 
            median = median(IRR),
            Q3 = quantile(IRR, 3/4)) %>%
  mutate(across(where(is.numeric), signi_f)) %>%
  ungroup()

```

# Age group

```{r}
# Jul2020-Jul2021 65+
ReferenceKeyWhitelist = stata %>% 
  filter(age >= 65, eventdate >= as.Date("2020-07-01"), eventdate < as.Date("2021-07-01")) %>% 
    select(ReferenceKey, DateofRegisteredDeath, eventdate, doadba, death, old_home, age) %>%
  group_by(ReferenceKey) %>%
  arrange(eventdate) %>%
  summarise(DateofRegisteredDeath = last(DateofRegisteredDeath),
            doadba = last(doadba),
            death = last(death), 
            old_home = last(old_home),
            age = last(age),
            .groups="drop") 

nrow(ReferenceKeyWhitelist)
# format is "yyy-mm-dd"
startdate = "2016-01-01"
elderly_death_visit = stata %>% 
  filter(year(eventdate) > 2015,
         eventdate < as.Date("2020-07-01"),
         eventdate >= as.Date(startdate)
         ) %>%
  filter(ReferenceKey %in% (ReferenceKeyWhitelist %>% pull(ReferenceKey))) 

df = elderly_death_visit %>% 
  select(ReferenceKey, eventdate) %>%
  mutate(period = ifelse(eventdate >= as.Date("2020-01-25"), "covid", "non-covid"))

df2 = df %>% 
  group_by(ReferenceKey, period) %>%
  summarise(visits = n()) %>%
  ungroup() %>% 
  # complete combinations of referencekey and period
  complete(ReferenceKey, period) %>%
  group_by(ReferenceKey) %>%
  summarise(covid = visits[period == "covid"],
            `non-covid` = visits[period == "non-covid"]) %>%
  replace_na(list(covid = 0, `non-covid` = 0)) %>%
  # remove patients who did not visit AE once pre-covid
  filter(`non-covid` > 8) %>%
  ungroup()


df3 = df2 %>%
  left_join(ReferenceKeyWhitelist, by=c("ReferenceKey")) %>%
  mutate(DateofRegisteredDeath = ifelse(is.na(DateofRegisteredDeath), "2021-07-01", DateofRegisteredDeath),
         covidIR = covid / as.numeric(difftime("2020-07-01","2020-01-25")),
         noncovidIR = `non-covid` / as.numeric(difftime("2020-01-24",startdate)),
         deltaIR = covidIR - noncovidIR)  %>% 
  mutate(change = ifelse(deltaIR < 0, "reduce", "increase"),
         change = ifelse(deltaIR == 0, "nochange", change),
         status = "survived",
         status = ifelse(death==1, "28d_death", status),
         status = ifelse(doadba==1, "doadba", status),
         age_gp = case_when(
           age < 75 ~ "65-74",
           age >=75 & age < 85 ~ "75-84",
           age >=85 & age < 95 ~ "85-94",
           TRUE ~ "95+"
         ))

table(df3$change, df3$status, df3$age_gp)
# table(df3$change, df3$old_home)
chisq.test(df3$change, df3$old_home, correct = 1)
```

```{r}
ylims <- df3 %>%
  mutate(deltaIR =  deltaIR * 100) %>%
  group_by(status, old_home) %>%
  summarise(Q1 = quantile(deltaIR, 1/4), Q3 = quantile(deltaIR, 3/4)) %>%
  ungroup() %>%
  #get lowest Q1 and highest Q3
  summarise(lowQ1 = min(Q1), highQ3 = max(Q3))

# labels = c("Non-RCHE", "RCHE")
# names(labels) = c("N", "Y")


# cumulative plot
ggplot(data = df3%>% filter(!is.na(status)), aes(x = deltaIR * 100, col = status))+
  geom_step(aes(y=..y..),stat="ecdf") +
  xlim(ylims$lowQ1, ylims$highQ3) +
  ylab("Cumulative portion") +
  xlab("Visit change per 100 days in covid period, against pre-covid period") +
  # facet_wrap(~old_home, labeller = labeller(old_home = labels)) 
  facet_wrap(~age_gp)

signi_f = function(x){
  signif(x, 3) * 100
}
df3 %>%
  group_by(status, old_home) %>%
  summarise(NumberOfPatients = n(),
            Q1 = quantile(deltaIR, 1/4), 
            median = median(deltaIR),
            Q3 = quantile(deltaIR, 3/4)) %>%
  mutate(across(where(is.numeric), signi_f)) %>%
  ungroup()
```

```{r}

ggplot(data = df3%>% filter(!is.na(status)), aes( x = deltaIR * 100, fill = status)) +
    geom_density(alpha=.1) +
    xlim(ylims$lowQ1, ylims$highQ3) +
    # xlab("Frequency") +
    xlab("Visit change per 100 days in pandemic") +
  facet_wrap(~old_home)
```

```{r}
source("./scripts/icd.R")
comorbidities_quan_deyo <- stata %>% 
          select(AENumber, ReferenceKey, death, paste0("DR", 1:10)) %>%
          get_dx_comorbidities(icd9map = icd::icd9_map_quan_deyo)


comorbidities_quan_deyo = comorbidities_quan_deyo %>%
  left_join(stata %>% select(AENumber, age, eventdate), by = "AENumber") 
comorbidities_quan_deyo = comorbidities_quan_deyo %>%
  pivot_longer(cols = MI:HIV,names_to = "comorb", values_to = "score") %>%
  # get only the positives
  filter(score)
comorbidities_quan_deyo_shortlisted = comorbidities_quan_deyo %>%
  filter(eventdate < as.Date("2020-07-01"),
         eventdate >= as.Date(startdate),
         ReferenceKey %in% (ReferenceKeyWhitelist %>% pull(ReferenceKey))) 

comorbidities_quan_deyo_shortlisted = 
  aggregate(score~AENumber + age + ReferenceKey + eventdate + death, data=comorbidities_quan_deyo_shortlisted,sum) %>%
  group_by(ReferenceKey) %>%
  arrange(eventdate) %>%
  summarise(eventdate = last(eventdate),
            score = last(score))

elderly_death_visit = 
  comorbidities_quan_deyo_shortlisted %>% 
  filter(year(eventdate) > 2015,
         eventdate < as.Date("2020-07-01"),
         eventdate >= as.Date(startdate)
         ) 

df = elderly_death_visit %>% 
  select(ReferenceKey, eventdate, score) %>%
  mutate(period = ifelse(eventdate >= as.Date("2020-01-25"), "covid", "non-covid"))

df2 = df %>% 
  group_by(ReferenceKey, period, score) %>%
  summarise(visits = n(),
            score = score) %>%
  ungroup() %>% 
  # complete combinations of referencekey and period
  complete(ReferenceKey, period) %>%
  replace_na(list(score = 0, visits = 0)) %>%
  group_by(ReferenceKey) %>%
  summarise(covid = visits[period == "covid"],
            `non-covid` = visits[period == "non-covid"],
            score = score) %>%
  replace_na(list(covid = 0, `non-covid` = 0)) %>%
  ungroup()


df3 = df2 %>%
  left_join(ReferenceKeyWhitelist, by=c("ReferenceKey")) %>%
  mutate(DateofRegisteredDeath = ifelse(is.na(DateofRegisteredDeath), "2021-07-01", DateofRegisteredDeath),
         covidIR = covid / as.numeric(difftime("2020-07-01","2020-01-25")),
         noncovidIR = `non-covid` / as.numeric(difftime("2020-01-24",startdate)),
         deltaIR = covidIR - noncovidIR)  %>% 
  mutate(change = ifelse(deltaIR < 0, "reduce", "increase"),
         change = ifelse(deltaIR == 0, "nochange", change),
         status = "survived",
         status = ifelse(death==1, "28d_death", status),
         status = ifelse(doadba==1, "doadba", status),
         ) 

ylims <- df3 %>%
  mutate(deltaIR =  deltaIR * 100) %>%
  group_by(status, score) %>%
  summarise(Q1 = quantile(deltaIR, 1/4), Q3 = quantile(deltaIR, 3/4)) %>%
  ungroup() %>%
  #get lowest Q1 and highest Q3
  summarise(lowQ1 = min(Q1), highQ3 = max(Q3))


# cumulative plot
ggplot(data = df3%>% filter(!is.na(status)), aes(x = deltaIR * 100, col = status))+
  geom_step(aes(y=..y..),stat="ecdf") +
  # xlim(ylims$lowQ1, ylims$highQ3) +
  ylab("Cumulative portion") +
  xlab("Visit change per 100 days in covid period, against pre-covid period") +
  # facet_wrap(~old_home, labeller = labeller(old_home = labels)) 
  facet_wrap(~score)

df3 %>%
  group_by(status, score) %>%
  summarise(NumberOfPatients = n(),
            Q1 = quantile(deltaIR, 1/4), 
            median = median(deltaIR),
            Q3 = quantile(deltaIR, 3/4)) %>%
  mutate(across(c("Q1", "median", "Q3"), signi_f)) %>%
  ungroup()
```


```{r}
summary = stata %>% 
  filter(age >= 65, doadba==1) %>%
  mutate(age_gp = case_when(
           age < 75 ~ "65-74",
           age >=75 & age < 85 ~ "75-84",
           age >=85 & age < 95 ~ "85-94",
           TRUE ~ "95+"
         )) %>%
  count(year = year(eventdate), age_gp) 
summary %>%
  pivot_wider(names_from = age_gp, values_from = n)
```