3a_6_Outcome_model_fit_alternative_splines.Rmd

---
author: "Leon Di Stefano"
date: "`r Sys.Date()`"
output: 
  html_document:
    keep_md: false
params:
  fit_name: "main_fit_alternative_splines"
  outcome_min: 28
  outcome_max: 35
title: "`r paste0('Outcome model fit', params$outcome_min, params$outcome_max, params$fit_name, sep = '-')`"
---

```{r}
knitr::opts_chunk$set(echo = TRUE)
require(here)
require(brms)
require(tidybayes)
require(mice)
require(cmdstanr)

here::i_am(file.path("hcq_pooling_analysis", "3a_6_Outcome_model_fit_alternative_splines.Rmd"))

source(here("hcq_pooling_analysis", "common.R"))

set_cmdstan_path(here::here("hcq_pooling_analysis", ".cmdstan", "cmdstan-2.27.0"))

out_stub         <- paste(params$outcome_min, params$outcome_max, sep = '-')
output_dir       <- here("hcq_pooling_analysis", "output", out_stub)
output_model_dir <- file.path(output_dir, params$fit_name)
if(!dir.exists(output_model_dir)) {
  dir.create(output_model_dir, recursive = TRUE)
}
```

```{r}
params
```

Read multiply-imputed data:

```{r}
mice_df_list <-
  read_rds(here(output_dir, "mice_complete_df_list.rds"))

patients <-
  read_rds(here(output_dir, "patients.rds"))
```

Specify the primary model formula:

```{r}
sap_primary_model <-
  brms::bf(
    niaid_outcome ~
        treat*(
          sex_model +
          rms::rcs(age_model, 4) +
          rms::rcs(bmi_model, 4) +
          rms::rcs(comorbidity_count, 4) +
          niaid_baseline_numeric_model
        ) +
        (1 + treat || siteid) +
        (1 + treat || niaid_baseline_fct)
  )
```

Have a look at the default priors:

```{r}
get_prior(sap_primary_model, data = mice_df_list[[1]])
```

Fit the primary outcome model:

```{r message=FALSE}
ncores <- parallel::detectCores() - 1
brm_fit <-
  brm_multiple(
      formula = sap_primary_model,
      family = cumulative,
      prior = prior(student_t(3, 0, 10), class = sd), # Defaults may have changed—SAP has broader priors
      data = mice_df_list,
      cores = ncores,
      control = list(adapt_delta = 0.999), # helps avoid divergent transitions 
      iter = 3000,
      thin = 3,
      backend = 'cmdstanr',
      seed = 20200524
      # file_refit = "on_change"
      )
write_rds(brm_fit, file.path(output_model_dir, paste0(params$fit_name, ".rds")))
```


```{r}
sessionInfo()
```



```{r}
Sys.time()
```