---
title: "Anchored Binary Analysis"
date: "`r Sys.Date()`"
output: rmarkdown::html_vignette
bibliography: references.bib
csl: biomedicine.csl
vignette: >
  %\VignetteIndexEntry{Anchored Binary Analysis}
  %\VignetteEngine{knitr::rmarkdown}
  %\VignetteEncoding{UTF-8}
---

```{r, include = FALSE}
knitr::opts_chunk$set(warning = FALSE, message = FALSE)
```

# Loading R packages

```{r}
# install.packages("maicplus")
library(maicplus)
```

Additional R packages for this vignette:

```{r}
library(dplyr)
```

# Illustration using example data

This example reads in `centered_ipd_twt` data that was created in `calculating_weights` vignette and uses `adrs_twt` dataset to run binary outcome analysis using the `maic_anchored` function by specifying `endpoint_type = "binary"`.

```{r}
data(centered_ipd_twt)
data(adrs_twt)

centered_colnames <- c("AGE", "AGE_SQUARED", "SEX_MALE", "ECOG0", "SMOKE", "N_PR_THER_MEDIAN")
centered_colnames <- paste0(centered_colnames, "_CENTERED")

weighted_data <- estimate_weights(
  data = centered_ipd_twt,
  centered_colnames = centered_colnames
)

# get dummy binary IPD
pseudo_adrs <- get_pseudo_ipd_binary(
  binary_agd = data.frame(
    ARM = c("B", "C", "B", "C"),
    RESPONSE = c("YES", "YES", "NO", "NO"),
    COUNT = c(280, 120, 200, 200)
  ),
  format = "stacked"
)

result <- maic_anchored(
  weights_object = weighted_data,
  ipd = adrs_twt,
  pseudo_ipd = pseudo_adrs,
  trt_ipd = "A",
  trt_agd = "B",
  trt_common = "C",
  normalize_weight = FALSE,
  endpoint_type = "binary",
  endpoint_name = "Binary Endpoint",
  eff_measure = "OR",
  # binary specific args
  binary_robust_cov_type = "HC3"
)
```

There are two summaries available in the result: descriptive and inferential. In the descriptive section, we have summaries of events.

```{r}
result$descriptive
```

In the inferential section, we have the fitted models stored (i.e. logistic regression) and the results from the `glm` models (i.e. odds ratios and CI).

```{r}
result$inferential$summary
```

Here are model and results before adjustment.

```{r}
result$inferential$fit$model_before
result$inferential$fit$res_AC_unadj
result$inferential$fit$res_AB_unadj
```

Here are model and results after adjustment.

```{r}
result$inferential$fit$model_after
result$inferential$fit$res_AC
result$inferential$fit$res_AB
```

```{r, echo = FALSE, eval = FALSE}
# heuristic check
# merge in adrs with ipd_matched

ipd <- adrs_twt
ipd$weights <- weighted_data$data$weights[match(ipd$USUBJID, weighted_data$data$USUBJID)]

pseudo_ipd <- pseudo_adrs
pseudo_ipd$weights <- 1

# Change the reference treatment to C
ipd$ARM <- stats::relevel(as.factor(ipd$ARM), ref = "C")
pseudo_ipd$ARM <- stats::relevel(as.factor(pseudo_ipd$ARM), ref = "C")

binobj_dat <- glm(RESPONSE ~ ARM, ipd, family = binomial(link = "logit"))
binobj_dat_adj <- suppressWarnings(glm(RESPONSE ~ ARM, ipd, weights = weights, family = binomial(link = "logit")))
binobj_agd <- glm(RESPONSE ~ ARM, pseudo_ipd, family = binomial(link = "logit"))

bin_robust_cov <- sandwich::vcovHC(binobj_dat_adj, type = "HC3")
bin_robust_coef <- lmtest::coeftest(binobj_dat_adj, vcov. = bin_robust_cov)
bin_robust_ci <- lmtest::coefci(binobj_dat_adj, vcov. = bin_robust_cov)

exp(summary(binobj_dat)$coef[2, "Estimate"])
exp(summary(binobj_dat_adj)$coef[2, "Estimate"])

bin_robust_ci
exp(bin_robust_ci)

res_AC <- res_BC <- list()
res_AC$est <- bin_robust_coef[2, "Estimate"]
res_AC$se <- bin_robust_coef[2, "Std. Error"]

res_BC$est <- summary(binobj_agd)$coefficients[2, "Estimate"]
res_BC$se <- summary(binobj_agd)$coefficients[2, "Std. Error"]

res_AB <- bucher(res_AC, res_BC, conf_lv = 0.95)
print(res_AB, exponentiate = TRUE)
```

# Using bootstrap to calculate standard errors

If bootstrap standard errors are preferred, we need to specify the number of bootstrap iteration (`n_boot_iteration`) in `estimate_weights` function and proceed fitting `maic_anchored` function. Then, the outputs include bootstrapped CI. Different types of bootstrap CI can be found by using parameter `boot_ci_type`.

```{r}
weighted_data2 <- estimate_weights(
  data = centered_ipd_twt,
  centered_colnames = centered_colnames,
  n_boot_iteration = 100,
  set_seed_boot = 1234
)

result_boot <- maic_anchored(
  weights_object = weighted_data2,
  ipd = adrs_twt,
  pseudo_ipd = pseudo_adrs,
  trt_ipd = "A",
  trt_agd = "B",
  trt_common = "C",
  normalize_weight = FALSE,
  endpoint_type = "binary",
  endpoint_name = "Binary Endpoint",
  eff_measure = "OR",
  boot_ci_type = "perc",
  # binary specific args
  binary_robust_cov_type = "HC3"
)

result_boot$inferential$fit$boot_res_AB
```