R/adjust_uc_om.R

In multibias: Simultaneous Multi-Bias Adjustment

# Adjust for uncontrolled confounding and outcome misclassification

# the following functions feed into adjust_uc_om():
# adjust_uc_om_val() (data_validation input),
# adjust_uc_om_coef_single() (bias_params input),
# adjust_uc_om_coef_multinom() (bias_params input)

adjust_uc_om_val <- function(
    data_observed,
    data_validation) {
  if (!all(data_observed$confounders %in% data_validation$confounders)) {
    stop(
      "All confounders in observed data must be present in validation data.",
      call. = FALSE
    )
  }

  if (
    (length(data_validation$confounders) - length(data_observed$confounders) != 1) ||
      (is.null(data_validation$misclassified_outcome))
  ) {
    stop(
      paste0(
        "Attempting to adjust for unobserved confounding from one confounder and outcome misclassification.",
        "\n",
        "Validation data must have: 1) one more confounder than the observed data, 2) a true and misclassified outcome specified."
      ),
      call. = FALSE
    )
  }

  n <- nrow(data_observed$data)

  df <- data.frame(
    X = data_observed$data[, data_observed$exposure],
    Ystar = data_observed$data[, data_observed$outcome]
  )
  df <- bind_cols(
    df,
    data_observed$data %>%
      select(all_of(data_observed$confounders))
  )

  df_val <- data.frame(
    X = data_validation$data[, data_validation$true_exposure],
    Y = data_validation$data[, data_validation$true_outcome],
    Ystar = data_validation$data[, data_validation$misclassified_outcome]
  )

  uc <- setdiff(data_validation$confounders, data_observed$confounders)
  df_val$U <- data_validation$data[, uc]
  df_val <- bind_cols(
    df_val,
    data_validation$data %>%
      select(all_of(data_observed$confounders))
  )

  force_match(
    df$X,
    df_val$X,
    "Outcomes from both datasets must both be binary or both be continuous."
  )
  force_binary(
    df_val$U,
    "Uncontrolled confounder in validation data must be a binary integer."
  )
  force_binary(
    df$Ystar,
    "Outcome in observed data must be a binary integer."
  )
  force_binary(
    df_val$Ystar,
    "Misclassified outcome in validation data must be a binary integer."
  )
  force_binary(
    df_val$Y,
    "True outcome in validation data must be a binary integer."
  )

  y_mod <- glm(Y ~ X + Ystar + . - U,
    family = binomial(link = "logit"),
    data = df_val
  )

  y_mod_coefs <- coef(y_mod)
  y_pred <- y_mod_coefs[1]

  for (i in 2:length(y_mod_coefs)) {
    var_name <- names(y_mod_coefs)[i]
    y_pred <- y_pred + df[[var_name]] * y_mod_coefs[i]
  }

  df$Ypred <- rbinom(n, 1, plogis(y_pred))

  u_mod <- glm(U ~ X + Y,
    family = binomial(link = "logit"),
    data = df_val
  )

  u_mod_coefs <- coef(u_mod)
  u_pred <- u_mod_coefs[1]

  for (i in 2:length(u_mod_coefs)) {
    var_name <- names(u_mod_coefs)[i]
    var_name <- gsub("Y", "Ypred", var_name) # col Y is not in df
    u_pred <- u_pred + df[[var_name]] * u_mod_coefs[i]
  }

  df$Upred <- rbinom(n, 1, plogis(u_pred))

  final <- glm(
    Ypred ~ X + Upred + . - Ystar,
    family = binomial(link = "logit"),
    data = df
  )

  return(final)
}


adjust_uc_om_coef_single <- function(
    data_observed,
    u_model_coefs,
    y_model_coefs) {
  data <- data_observed$data
  n <- nrow(data)
  confounders <- data_observed$confounders
  len_c <- length(confounders)
  len_u_coefs <- length(u_model_coefs)
  len_y_coefs <- length(y_model_coefs)

  x <- data[, data_observed$exposure]
  ystar <- data[, data_observed$outcome]

  force_binary(ystar, "Outcome must be a binary integer.")
  force_len(
    len_u_coefs,
    3,
    paste0(
      "Incorrect length of U model coefficients. ",
      "Length should equal 3."
    )
  )
  force_len(
    len_y_coefs,
    3 + len_c,
    paste0(
      "Incorrect length of Y model coefficients. ",
      "Length should equal 3 + number of confounders."
    )
  )

  u1_0 <- u_model_coefs[1]
  u1_x <- u_model_coefs[2]
  u1_y <- u_model_coefs[3]

  y1_0 <- y_model_coefs[1]
  y1_x <- y_model_coefs[2]
  y1_ystar <- y_model_coefs[3]

  if (is.null(confounders)) {
    df <- data.frame(X = x, Ystar = ystar)
    df$Ypred <- rbinom(n, 1, plogis(y1_0 + y1_x * df$X + y1_ystar * df$Ystar))
    df$Upred <- rbinom(n, 1, plogis(u1_0 + u1_x * df$X + u1_y * df$Ypred))

    final <- glm(
      Ypred ~ X + Upred,
      family = binomial(link = "logit"),
      data = df
    )
  } else if (len_c == 1) {
    c1 <- data[, confounders]
    df <- data.frame(X = x, Ystar = ystar, C1 = c1)

    y1_c1 <- y_model_coefs[4]

    df$Ypred <- rbinom(
      n, 1, plogis(
        y1_0 + y1_x * df$X + y1_ystar * df$Ystar + y1_c1 * df$C1
      )
    )
    df$Upred <- rbinom(n, 1, plogis(u1_0 + u1_x * df$X + u1_y * df$Ypred))

    final <- glm(
      Ypred ~ X + C1 + Upred,
      family = binomial(link = "logit"),
      data = df
    )
  } else if (len_c == 2) {
    c1 <- data[, confounders[1]]
    c2 <- data[, confounders[2]]

    df <- data.frame(X = x, Ystar = ystar, C1 = c1, C2 = c2)

    y1_c1 <- y_model_coefs[4]
    y1_c2 <- y_model_coefs[5]

    df$Ypred <- rbinom(
      n, 1, plogis(
        y1_0 + y1_x * df$X + y1_ystar * df$Ystar +
          y1_c1 * df$C1 + y1_c2 * df$C2
      )
    )
    df$Upred <- rbinom(n, 1, plogis(u1_0 + u1_x * df$X + u1_y * df$Ypred))

    final <- glm(
      Ypred ~ X + C1 + C2 + Upred,
      family = binomial(link = "logit"),
      data = df
    )
  } else if (len_c == 3) {
    c1 <- data[, confounders[1]]
    c2 <- data[, confounders[2]]
    c3 <- data[, confounders[3]]

    df <- data.frame(X = x, Ystar = ystar, C1 = c1, C2 = c2, C3 = c3)

    y1_c1 <- y_model_coefs[4]
    y1_c2 <- y_model_coefs[5]
    y1_c3 <- y_model_coefs[6]

    df$Ypred <- rbinom(
      n, 1,
      plogis(
        y1_0 + y1_x * df$X + y1_ystar * df$Ystar +
          y1_c1 * df$C1 + y1_c2 * df$C2 + y1_c3 * df$C3
      )
    )
    df$Upred <- rbinom(n, 1, plogis(u1_0 + u1_x * df$X + u1_y * df$Ypred))

    final <- glm(
      Ypred ~ X + C1 + C2 + C3 + Upred,
      family = binomial(link = "logit"),
      data = df
    )
  } else if (len_c > 3) {
    stop(
      "This function is currently not compatible with >3 confounders.",
      call. = FALSE
    )
  }

  return(final)
}


adjust_uc_om_coef_multinom <- function(
    data_observed,
    u1y0_model_coefs,
    u0y1_model_coefs,
    u1y1_model_coefs) {
  data <- data_observed$data
  n <- nrow(data)
  confounders <- data_observed$confounders
  len_c <- length(confounders)
  len_u1y0_coefs <- length(u1y0_model_coefs)
  len_u0y1_coefs <- length(u0y1_model_coefs)
  len_u1y1_coefs <- length(u1y1_model_coefs)

  x <- data[, data_observed$exposure]
  ystar <- data[, data_observed$outcome]

  force_binary(ystar, "Outcome must be a binary integer.")
  force_len(
    len_u1y0_coefs,
    3 + len_c,
    paste0(
      "Incorrect length of U1Y0 model coefficients. ",
      "Length should equal 3 + number of confounders."
    )
  )
  force_len(
    len_u0y1_coefs,
    3 + len_c,
    paste0(
      "Incorrect length of U0Y1 model coefficients. ",
      "Length should equal 3 + number of confounders."
    )
  )
  force_len(
    len_u1y1_coefs,
    3 + len_c,
    paste0(
      "Incorrect length of U1Y1 model coefficients. ",
      "Length should equal 3 + number of confounders."
    )
  )

  u1y0_0 <- u1y0_model_coefs[1]
  u1y0_x <- u1y0_model_coefs[2]
  u1y0_ystar <- u1y0_model_coefs[3]

  u0y1_0 <- u0y1_model_coefs[1]
  u0y1_x <- u0y1_model_coefs[2]
  u0y1_ystar <- u0y1_model_coefs[3]

  u1y1_0 <- u1y1_model_coefs[1]
  u1y1_x <- u1y1_model_coefs[2]
  u1y1_ystar <- u1y1_model_coefs[3]

  if (is.null(confounders)) {
    df <- data.frame(X = x, Ystar = ystar)

    p_u1y0 <- exp(u1y0_0 + u1y0_x * df$X + u1y0_ystar * df$Ystar)
    p_u0y1 <- exp(u0y1_0 + u0y1_x * df$X + u0y1_ystar * df$Ystar)
    p_u1y1 <- exp(u1y1_0 + u1y1_x * df$X + u1y1_ystar * df$Ystar)

    denom <- (1 + p_u1y0 + p_u0y1 + p_u1y1)

    u0y0_pred <- 1 / denom
    u1y0_pred <- p_u1y0 / denom
    u0y1_pred <- p_u0y1 / denom
    u1y1_pred <- p_u1y1 / denom

    df_uy_pred <- data.frame(
      U0Y0 = u0y0_pred,
      U1Y0 = u1y0_pred,
      U0Y1 = u0y1_pred,
      U1Y1 = u1y1_pred
    )
    df_uy_pred4 <- bind_rows(df_uy_pred, df_uy_pred, df_uy_pred, df_uy_pred)

    combined <- bind_rows(df, df, df, df) %>%
      bind_cols(df_uy_pred4) %>%
      mutate(
        Ubar = rep(c(1, 0, 1, 0), each = n),
        Ybar = rep(c(1, 1, 0, 0), each = n),
        pUY = case_when(
          Ubar == 0 & Ybar == 0 ~ U0Y0,
          Ubar == 1 & Ybar == 0 ~ U1Y0,
          Ubar == 0 & Ybar == 1 ~ U0Y1,
          Ubar == 1 & Ybar == 1 ~ U1Y1
        )
      )
    suppressWarnings({
      final <- glm(
        Ybar ~ X + Ubar,
        family = binomial(link = "logit"),
        weights = combined$pUY,
        data = combined
      )
    })
  } else if (len_c == 1) {
    c1 <- data[, confounders]

    df <- data.frame(X = x, Ystar = ystar, C1 = c1)

    u1y0_c1 <- u1y0_model_coefs[4]
    u0y1_c1 <- u0y1_model_coefs[4]
    u1y1_c1 <- u1y1_model_coefs[4]

    p_u1y0 <- exp(
      u1y0_0 + u1y0_x * df$X + u1y0_ystar * df$Ystar +
        u1y0_c1 * df$C1
    )
    p_u0y1 <- exp(
      u0y1_0 + u0y1_x * df$X + u0y1_ystar * df$Ystar +
        u0y1_c1 * df$C1
    )
    p_u1y1 <- exp(
      u1y1_0 + u1y1_x * df$X + u1y1_ystar * df$Ystar +
        u1y1_c1 * df$C1
    )

    denom <- (1 + p_u1y0 + p_u0y1 + p_u1y1)

    u0y0_pred <- 1 / denom
    u1y0_pred <- p_u1y0 / denom
    u0y1_pred <- p_u0y1 / denom
    u1y1_pred <- p_u1y1 / denom

    df_uy_pred <- data.frame(
      U0Y0 = u0y0_pred,
      U1Y0 = u1y0_pred,
      U0Y1 = u0y1_pred,
      U1Y1 = u1y1_pred
    )
    df_uy_pred4 <- bind_rows(df_uy_pred, df_uy_pred, df_uy_pred, df_uy_pred)

    combined <- bind_rows(df, df, df, df) %>%
      bind_cols(df_uy_pred4) %>%
      mutate(
        Ubar = rep(c(1, 0, 1, 0), each = n),
        Ybar = rep(c(1, 1, 0, 0), each = n),
        pUY = case_when(
          Ubar == 0 & Ybar == 0 ~ U0Y0,
          Ubar == 1 & Ybar == 0 ~ U1Y0,
          Ubar == 0 & Ybar == 1 ~ U0Y1,
          Ubar == 1 & Ybar == 1 ~ U1Y1
        )
      )

    suppressWarnings({
      final <- glm(
        Ybar ~ X + C1 + Ubar,
        family = binomial(link = "logit"),
        weights = combined$pUY,
        data = combined
      )
    })
  } else if (len_c == 2) {
    c1 <- data[, confounders[1]]
    c2 <- data[, confounders[2]]

    df <- data.frame(X = x, Ystar = ystar, C1 = c1, C2 = c2)

    u1y0_c1 <- u1y0_model_coefs[4]
    u1y0_c2 <- u1y0_model_coefs[5]

    u0y1_c1 <- u0y1_model_coefs[4]
    u0y1_c2 <- u0y1_model_coefs[5]

    u1y1_c1 <- u1y1_model_coefs[4]
    u1y1_c2 <- u1y1_model_coefs[5]

    p_u1y0 <- exp(
      u1y0_0 + u1y0_x * df$X + u1y0_ystar * df$Ystar +
        u1y0_c1 * df$C1 + u1y0_c2 * df$C2
    )
    p_u0y1 <- exp(
      u0y1_0 + u0y1_x * df$X + u0y1_ystar * df$Ystar +
        u0y1_c1 * df$C1 + u0y1_c2 * df$C2
    )
    p_u1y1 <- exp(
      u1y1_0 + u1y1_x * df$X + u1y1_ystar * df$Ystar +
        u1y1_c1 * df$C1 + u1y1_c2 * df$C2
    )

    denom <- (1 + p_u1y0 + p_u0y1 + p_u1y1)

    u0y0_pred <- 1 / denom
    u1y0_pred <- p_u1y0 / denom
    u0y1_pred <- p_u0y1 / denom
    u1y1_pred <- p_u1y1 / denom

    df_uy_pred <- data.frame(
      U0Y0 = u0y0_pred,
      U1Y0 = u1y0_pred,
      U0Y1 = u0y1_pred,
      U1Y1 = u1y1_pred
    )
    df_uy_pred4 <- bind_rows(df_uy_pred, df_uy_pred, df_uy_pred, df_uy_pred)

    combined <- bind_rows(df, df, df, df) %>%
      bind_cols(df_uy_pred4) %>%
      mutate(
        Ubar = rep(c(1, 0, 1, 0), each = n),
        Ybar = rep(c(1, 1, 0, 0), each = n),
        pUY = case_when(
          Ubar == 0 & Ybar == 0 ~ U0Y0,
          Ubar == 1 & Ybar == 0 ~ U1Y0,
          Ubar == 0 & Ybar == 1 ~ U0Y1,
          Ubar == 1 & Ybar == 1 ~ U1Y1
        )
      )

    suppressWarnings({
      final <- glm(
        Ybar ~ X + C1 + C2 + Ubar,
        family = binomial(link = "logit"),
        weights = combined$pUY,
        data = combined
      )
    })
  } else if (len_c == 3) {
    c1 <- data[, confounders[1]]
    c2 <- data[, confounders[2]]
    c3 <- data[, confounders[3]]

    df <- data.frame(X = x, Ystar = ystar, C1 = c1, C2 = c2, C3 = c3)

    u1y0_c1 <- u1y0_model_coefs[4]
    u1y0_c2 <- u1y0_model_coefs[5]
    u1y0_c3 <- u1y0_model_coefs[6]

    u0y1_c1 <- u0y1_model_coefs[4]
    u0y1_c2 <- u0y1_model_coefs[5]
    u0y1_c3 <- u0y1_model_coefs[6]

    u1y1_c1 <- u1y1_model_coefs[4]
    u1y1_c2 <- u1y1_model_coefs[5]
    u1y1_c3 <- u1y1_model_coefs[6]

    p_u1y0 <- exp(
      u1y0_0 + u1y0_x * df$X + u1y0_ystar * df$Ystar +
        u1y0_c1 * df$C1 + u1y0_c2 * df$C2 + u1y0_c3 * df$C3
    )
    p_u0y1 <- exp(
      u0y1_0 + u0y1_x * df$X + u0y1_ystar * df$Ystar +
        u0y1_c1 * df$C1 + u0y1_c2 * df$C2 + u0y1_c3 * df$C3
    )
    p_u1y1 <- exp(
      u1y1_0 + u1y1_x * df$X + u1y1_ystar * df$Ystar +
        u1y1_c1 * df$C1 + u1y1_c2 * df$C2 + u1y1_c3 * df$C3
    )

    denom <- (1 + p_u1y0 + p_u0y1 + p_u1y1)

    u0y0_pred <- 1 / denom
    u1y0_pred <- p_u1y0 / denom
    u0y1_pred <- p_u0y1 / denom
    u1y1_pred <- p_u1y1 / denom

    df_uy_pred <- data.frame(
      U0Y0 = u0y0_pred,
      U1Y0 = u1y0_pred,
      U0Y1 = u0y1_pred,
      U1Y1 = u1y1_pred
    )
    df_uy_pred4 <- bind_rows(df_uy_pred, df_uy_pred, df_uy_pred, df_uy_pred)

    combined <- bind_rows(df, df, df, df) %>%
      bind_cols(df_uy_pred4) %>%
      mutate(
        Ubar = rep(c(1, 0, 1, 0), each = n),
        Ybar = rep(c(1, 1, 0, 0), each = n),
        pUY = case_when(
          Ubar == 0 & Ybar == 0 ~ U0Y0,
          Ubar == 1 & Ybar == 0 ~ U1Y0,
          Ubar == 0 & Ybar == 1 ~ U0Y1,
          Ubar == 1 & Ybar == 1 ~ U1Y1
        )
      )

    suppressWarnings({
      final <- glm(
        Ybar ~ X + C1 + C2 + C3 + Ubar,
        family = binomial(link = "logit"),
        weights = combined$pUY,
        data = combined
      )
    })
  } else if (len_c > 3) {
    stop(
      "This function is currently not compatible with >3 confounders.",
      call. = FALSE
    )
  }

  return(final)
}


adjust_uc_om <- function(
    data_observed,
    data_validation = NULL,
    bias_params = NULL,
    level = 0.95) {
  if (
    (!is.null(data_validation) && !is.null(bias_params)) ||
      (is.null(data_validation) && is.null(bias_params))
  ) {
    stop(
      "One of data_validation or bias_params must be non-null.",
      call. = FALSE
    )
  }

  if (!is.null(data_validation)) {
    final <- adjust_uc_om_val(
      data_observed,
      data_validation
    )
  } else if (!is.null(bias_params)) {
    if (all(c("u", "y") %in% names(bias_params$coef_list))) {
      final <- adjust_uc_om_coef_single(
        data_observed,
        u_model_coefs = bias_params$coef_list$u,
        y_model_coefs = bias_params$coef_list$y
      )
    } else if (
      all(
        c("u1y0", "u0y1", "u1y1") %in%
          names(bias_params$coef_list)
      )
    ) {
      final <- adjust_uc_om_coef_multinom(
        data_observed,
        u1y0_model_coefs = bias_params$coef_list$u1y0,
        u0y1_model_coefs = bias_params$coef_list$u0y1,
        u1y1_model_coefs = bias_params$coef_list$u1y1
      )
    } else {
      (
        stop(
          paste0(
            "bias_params must specify parameters for ",
            "uncontrolled confounding and outcome misclassification"
          ),
          call. = FALSE
        )
      )
    }
  }

  est <- summary(final)$coef[2, 1]
  se <- summary(final)$coef[2, 2]
  alpha <- 1 - level

  estimate <- exp(est)
  ci <- c(
    exp(est + se * qnorm(alpha / 2)),
    exp(est + se * qnorm(1 - alpha / 2))
  )

  return(list(estimate = estimate, ci = ci))
}