fig3.Rmd

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

```{r}
# import libraries
library(tidyverse)
library(lubridate) # wrangling with dates
library(scales) # vectors and colour scaling
library(icd) # icd codes and comorbidities
library(haven)
library(import) # namespace handling
# library(xlsx)
library(patchwork)
library(ggplot2)
```

# import data

```{r}
### attendance records from stata file
stata = mclapply(list("AE_Attendance_transformed_20150101_20210917.csv"), import_csv, mc.cores=40)

stata = stata %>%
  mutate(year = lubridate::year(eventdate),
         eventdate = as.Date(eventdate)) %>% 
  filter(eventdate<max(eventdate)-28,
         year != 2015)
```

# Impute filtering condition info, group by age and gender

```{r}
# age and gender filtering conditions
demo_conditions = c("age>= 65 & Sex_num == 1","age >= 65 & Sex_num == 0")
# age and gender filtering conditions in readable form
demo_conditionsTextual = c("65+ & M","65+ & F")
# concat strings in dplyr-readable form
conditions = paste(paste0(demo_conditions,"~\"",demo_conditionsTextual,"\";"),collapse ='')

selected = stata %>%
  select(eventdate, year, age, Sex_num, death) %>%
  mutate(cond=case_when(
    !!! rlang::parse_exprs(conditions)
  ))
```

# 65+ daily attn and 28d death counts 

```{r}



wave.df <- import_waves(twosls = T)
date.df <- import_dates(wave.df=wave.df, 
                        start = min(selected$eventdate), 
                        end=max(selected$eventdate))


# impute after-wave periods
date.df = date.df %>%
  mutate(wave_num = ifelse(is.na(wave_num) & date>min(wave.df$start),"after-wave",wave_num))%>%
  mutate(wave_num = ifelse(date<min(wave.df$start),0,wave_num))

daily = selected %>% 
  group_by(cond) %>% 
  count(eventdate, death) %>%
  pivot_wider(names_from = "death", names_prefix = "n_d", values_from="n") %>%
  mutate(attn=n_d0+n_d1) %>%
  rename(death=n_d1) %>%
  select(-n_d0) %>%
  drop_na() %>%
  # mutate year and t, number of calendar days since 1st Jan
  mutate(Year = year(eventdate) - 2016,
         t=yday(eventdate)) %>% 
  merge(x=., y=date.df, by.x="eventdate", by.y="date")

model.stage1 = glm(attn~wave_num + Year + t + I(t^2) + I(t^3),daily,family="gaussian")
daily = daily %>%
  mutate(estimated.attn = predict(model.stage1, daily))
olddaily = daily
daily = bind_rows(olddaily %>%
            filter(year >= 2020) %>%
              mutate(year = as.factor(year)),
          olddaily %>%
            filter(year < 2020) %>%
            mutate(year="2016-2019 avg") %>%
            group_by(yday, cond) %>%
            mutate(estimated.attn = mean(estimated.attn)))
daily
  # group_by(wave_num, yday) %>%
  # summarise(estimated.attn = mean(estimated.attn))
# attn plot
fig3a = (
  daily  %>% filter(yday != 366) %>%
  ggplot() +
  geom_point(aes(x=yday(eventdate), y=attn, col = as.factor(year)), alpha=.1) +
  geom_line(aes(x=yday(eventdate), y=estimated.attn, col = as.factor(year))) +
  facet_wrap(~cond) +
  labs(x = "Calendar Day",
       y="Attendance Episode",
       col = "Year") +
  theme(legend.position = "bottom")
)

model.stage2 = glm(death~estimated.attn + Year + t + I(t^2) + I(t^3),olddaily,family="gaussian")
olddaily = olddaily %>%
  mutate(estimated.death = predict(model.stage2, daily))
daily = bind_rows(olddaily %>%
            filter(year >= 2020) %>%
              mutate(year = as.factor(year)),
          olddaily %>%
            filter(year < 2020) %>%
            mutate(year="2016-2019 avg") %>%
            group_by(yday, cond) %>%
            mutate(estimated.death = mean(estimated.death)))
daily
# death plot
fig3b = ggplot(daily) +
  geom_point(aes(x=yday(eventdate), y=death, col = as.factor(year)), alpha=.1) +
  geom_line(aes(x=yday(eventdate), y=estimated.death, col = as.factor(year))) +
  facet_wrap(~cond) +
  labs(y="28d Mortality", col = "Year") +
  theme(legend.position="bottom") +
  ylim(20,70)

fig3 = fig3a / fig3b & labs(x="Calendar Day") &
  theme(panel.grid.major = element_blank(), panel.grid.minor = element_blank(),
        panel.background = element_blank(), axis.line = element_line(colour = "black")) &
  plot_annotation(tag_levels = list(c("A", "B")))
fig3
ggsave(fig3, filename = file.path(out_dir, "fig3.svg"),height=7)
```