---
title: "Model for Survival"
engine: knitr
---

```{r}
#| label: setup
library(here)
library(survival)

source(here("code/helpers.R"))
options(arrow.skip_nul = TRUE)

a <-  load_accounts() %>%
  filter(!has_moved) %>%
  filter(locked == FALSE) %>%
  anti_join(., arrow::read_feather(here("data/scratch/individual_moved_accounts.feather")), by=c("username"="moved_acct", "server"="moved_server")) %>%
  inner_join(arrow::read_feather(here("data/scratch/metadata.feather")), by="server") %>%
  filter(created_at >= "2023-11-01") %>%
  filter(created_at <= "2023-11-30") %>%
  #inner_join(activity, by="server") %>%
  filter(created_at < last_status_at) %>%
  mutate(active_time = as.integer(active_time)) %>%
  mutate(status = ifelse(active, 0, 1)) %>%
  mutate(jm = server %in% arrow::read_feather(here("data/scratch/joinmastodon.feather"))$domain) %>% 
  mutate(follows_someone = following_count > 0) %>%
  mutate(has_a_follower = followers_count > 0)
```

```{r}

server_summary <- a %>%
  group_by(server) %>%
  summarize(cohort_size = n(), .groups = 'drop')

sel_a <- a %>%
  mutate(is_ms = server == "mastodon.social") %>%
  ungroup() %>%
  inner_join(server_summary, by = "server")

cx <- sel_a %>%
  mutate(c_size = as.factor(as.integer(log10(cohort_size)))) %>%
  mutate(follows_someone = following_count >= 14) %>%
  filter(followers_count > 0) %>%
  filter(following_count > 0) %>%
  coxph(Surv(active_time_weeks, status) ~ cluster(server) + jm + is_ms, data = ., x=TRUE, robust = T) # cluster(server)

cz <- cox.zph(cx)
#plot(cz)
cx
cz
```


```{r}
#| label: fig-survival-plot
#| fig-cap: Survival plot for accounts created in November 2023. Accounts created on mastodon.social had a higher hazard than accounts created on other servers.
#| fig-width: 6.75
#| fig-height: 3
library(ggsurvfit)
sel_a %>%
  mutate(c_size = as.factor(as.integer(log10(cohort_size)))) %>%
  mutate(no_followers = !has_a_follower) %>%
  filter(followers_count > 0) %>%
  filter(following_count > 0) %>%
  filter(statuses_count > 0) %>%
  mutate(small = cohort_size < 10) %>%
  mutate(large = cohort_size > 100) %>%
survfit2(Surv(active_time_weeks, status) ~ jm + strata(is_ms), data = .) %>% # cluster(server) 
  ggsurvfit() +
  add_confidence_interval() +
  #scale_y_continuous(limits = c(0, 1)) +
  labs(
    y = "Overall survival probability",
    x = "Time (days)",
  ) +
  theme_bw_small_labels() +
  theme(axis.text.x = element_text(angle = 45, hjust = 1))
```