meta_analysis_aux.r

library(dplyr)
library(metafor)


get_effect_size <- function(dat, effect_size_mean, effect_size_sd) {
    #' Get the effect size and its standard error
    #'
    #' @param dat dataframe with the data
    #' @param effect_size string with the name of the effect size


    dat <- mutate(dat,
        SE = dat[[effect_size_sd]] / sqrt(dat$"# Patients")
    )
    dat <- mutate(dat, vi = dat$SE^2)
    dat <- mutate(dat, yi = dat[[effect_size_mean]])

    colnames <- colnames(dat)
    dat <- transform(dat, ci_low = dat$yi - 1.96 * sqrt(dat$vi), ci_upp = dat$yi + 1.96 * sqrt(dat$vi))
    colnames(dat) <- c(colnames, c("ci_low", "ci_upp"))

    return(dat)
}


prepare_subgroup_analysis <- function(dat, subgroup_strategy, subgroup_analysis = TRUE) {
    # Order data according to subgroup and yi

    dat <- mutate(dat, subgroup = dat[[subgroup_strategy]])
    dat <- dat[order(dat$"First author, year", decreasing = TRUE), ]

    subgroups <- sort(unique(dat$subgroup), decreasing = TRUE)
    subgroup_counts <- table(dat$subgroup)

    res_dict <- setNames(lapply(subgroups, function(x) list(count = subgroup_counts[[x]])), subgroups)

    # Entry for titles (last entry) + empty entry after each subgroup summary
    n_entries <- length((dat$yi)) + length(names(res_dict)) * 2 + 2

    return(list(dat, res_dict, n_entries))
}

test_heterogeneity <- function(data, overall_res, mods) {
    print(paste("overall I^2:", fmtx(overall_res$I2, digits = 2)))
    for (mod in mods) {
        overall_res_mods <- rma.uni(yi, vi, data = data, method = "REML", mods = ~ factor(data[[mod]]))
        print(paste(
            mod,
            "| I^2 with mods", fmtx(overall_res_mods$I2, digits = 2),
            "| relative decrease:", fmtx(((overall_res$I2 - overall_res_mods$I2) / overall_res$I2) * 100, digits = 2), "%",
            sep = " "
        ))
    }
    cat("\n")
}

# Independent Effects: Each moderator's effect on the outcome (effect size) is estimated while holding the other moderators constant.
test_heterogeneity_multivariate <- function(data, overall_res, mods, interaction) {
    print(paste("overall I^2:", fmtx(overall_res$I2, digits = 2)))

    if (interaction) {
        coll <- " * "
    } else {
        coll <- " + "
    }

    multivariate_string <- paste("~", paste(paste0("factor(data[[\"", mods, "\"]])"), collapse = coll))
    multivariate_mods <- as.formula(multivariate_string)

    overall_res_mods <- rma.uni(yi, vi, data = data, method = "REML", mods = multivariate_mods)
    print(paste(
        paste(paste0(mods), collapse = coll),
        "| I^2 with mods", fmtx(overall_res_mods$I2, digits = 2),
        "| relative decrease:", fmtx(((overall_res$I2 - overall_res_mods$I2) / overall_res$I2) * 100, digits = 2), "%",
        sep = " "
    ))
    cat("\n")
}