Prognostic-weighted survey GLM Prognostic-weighted survey GLM

Description

Fits a survey-weighted logistic regression model using stabilized prognostic score weights derived from a model predicting the outcome conditional on baseline covariates and excluding the exposure effect. The function supports design-based inference under complex survey sampling while adjusting for confounding through prognostic weighting

Usage

final_prog_svyglm(
  data,
  dep_var,
  exposure,
  covariates,
  id_var,
  strata_var,
  weight_var,
  outcome_covariates = NULL,
  level = 0.95,
  ...
)

Arguments

Value

A list with:

• prog_model: Prognostic svyglm.

• final_model: Weighted outcome svyglm.

• OR_table: Odds ratios with CI.

• AUC: Weighted AUC.

• data: Data with prognostic weights.

Examples

set.seed(123)
n <- 1000
dat <- data.frame(
  psu = sample(1:10, n, replace = TRUE),
  strata = sample(1:5, n, replace = TRUE),
  weight = runif(n, 0.5, 2),
  age = rnorm(n, 50, 10),
  sex = factor(sample(c("Male", "Female"), n, replace = TRUE)),
  exposure = rbinom(n, 1, 0.5)
)
dat$outcome <- rbinom(n, 1, plogis(-2 + 0.03*dat$age + 0.5*dat$exposure))
fit<-final_prog_svyglm(data = dat,
 dep_var = "outcome",
 exposure="exposure",
 covariates = c("age", "sex"),
 id_var = "psu",
 strata_var = "strata",
 weight_var = "weight",
 level = 0.95
)
names(fit)
fit$OR_table

Propensity-weighted survey GLM

Description

Calculates IPTW weights and fits survey-weighted GLM. Supports binary, multinomial, or continuous exposures.

Usage

final_prop_svyglm(
  data,
  dep_var,
  covariates,
  exposure,
  id_var,
  strata_var,
  weight_var,
  exposure_type = "binary",
  outcome_covariates = NULL,
  level = 0.95,
  ...
)

Arguments

Value

A list with:

• ps_model: Propensity score svyglm model.

• final_model: Weighted outcome svyglm model.

• OR_table: Odds ratios with CI and p-values.

• AUC: Weighted AUC.

• data: Data with IPTW and predictions.

Examples

set.seed(123)
n <- 1500
dat <- data.frame(
  psu = sample(1:10, n, replace = TRUE),
  strata = sample(1:5, n, replace = TRUE),
  weight = runif(n, 0.5, 2),
  age = rnorm(n, 50, 10),
  sex = factor(sample(c("Male", "Female"), n, replace = TRUE)),
  exposure_bin = rbinom(n, 1, 0.5)
)
dat$outcome <- rbinom(n, 1, plogis(-2 + 0.03*dat$age + 0.5*dat$exposure_bin))
## ---- Example 1: Binary exposure ----
fit_bin_exp<-final_prop_svyglm(dat, dep_var="outcome",
                  covariates=c("age","sex"),
                  exposure="exposure_bin",
                  id_var="psu", strata_var="strata",
                  weight_var="weight", outcome_covariates = NULL)
fit_bin_exp$OR_table
## ---- Example 2: Continuous exposure ----
fit_cont_exp <- final_prop_svyglm(
  dat,
  dep_var     = "outcome",
  covariates  = c("sex"),
  exposure    = "age",
  id_var      = "psu",
  strata_var  = "strata",
  weight_var  = "weight",
 exposure_type = "continuous",
 outcome_covariates = NULL)
fit_cont_exp$OR_table
#### ---- Example 1: Multinomial exposure ----
dat$exposure_3cat <- cut(dat$age,
breaks = quantile(dat$age, probs = c(0, 1/3, 2/3, 1)),  # tertiles
 labels = c("Low", "Medium", "High"),
include.lowest = TRUE)
# Numeric coding for exposure effect
exp_eff <- ifelse(dat$exposure_3cat == "Low", 0,
                ifelse(dat$exposure_3cat == "Medium", 0.6, 1.2))
dat$outcome <- rbinom(n,1,plogis(-3 +0.02 * dat$age +0.4  * (dat$sex == "Male") +exp_eff))
fit_multi_cat <- final_prop_svyglm(dat, dep_var     = "outcome",
covariates  = c("age", "sex"), exposure    = "exposure_3cat",
id_var      = "psu", strata_var  = "strata", weight_var  = "weight",
exposure_type = "multinomial",
outcome_covariates = NULL)
fit_multi_cat$OR_table

Final Survey-Weighted GLM

Description

Fits a survey-weighted logistic regression model (quasibinomial) using raw survey variables. Returns ORs, confidence intervals, p-values, and model discrimination statistics.

Usage

final_svyglm(
  data,
  dep_var,
  covariates,
  id_var,
  strata_var,
  weight_var,
  family = "binomial",
  level = 0.95,
  interaction_terms = NULL
)

Arguments

Value

A list containing:

• model: Survey-weighted logistic regression model.

• results_table: Odds ratios with confidence intervals and p-values.

• AUC: Survey-weighted AUC (Somers' C).

• data: Input data with predicted probabilities.

• design: Survey design object.

Examples

set.seed(123)
n <- 100
dat <- data.frame(
  psu = sample(1:10, n, replace = TRUE),
  strata = sample(1:5, n, replace = TRUE),
  weight = runif(n, 0.5, 2),
  age = rnorm(n, 50, 10),
  sex = factor(sample(c("Male", "Female"), n, replace = TRUE)),
  exposure = rbinom(n, 1, 0.5)
)
dat$outcome <- rbinom(n, 1, plogis(-2 + 0.03*dat$age + 0.5*dat$exposure))
fit_simple<-final_svyglm(dat, dep_var="outcome", covariates=c("age","sex"),
             id_var="psu", strata_var="strata", weight_var="weight")
fit_simple$OR_table

Print method for svyCausal objects

Description

Print method for svyCausal objects

Usage

## S3 method for class 'svyCausal'
print(x, ...)

Arguments

Value

The object x, invisibly.

Weighted ROC Curve for Survey-Weighted Models

Description

Produces a weighted ROC curve and reports weighted AUC for survey-based models.

Usage

viz_auc_svyglm(
  fit_object,
  title = "Weighted ROC Curve",
  line_color = "#0072B2"
)

Arguments

Details

AUC is computed using, consistent with complex survey weighting.

Value

A ggplot object.

Examples

set.seed(123)
n <- 100
dat <- data.frame(
  psu = sample(1:10, n, replace = TRUE),
  strata = sample(1:5, n, replace = TRUE),
  weight = runif(n, 0.5, 2),
  age = rnorm(n, 50, 10),
  sex = factor(sample(c("Male", "Female"), n, replace = TRUE)),
  exposure = rbinom(n, 1, 0.5)
)
dat$outcome <- rbinom(n, 1, plogis(-2 + 0.03*dat$age + 0.5*dat$exposure))
fit_example<-final_svyglm(dat, dep_var="outcome", covariates=c("age","sex"),
             id_var="psu", strata_var="strata", weight_var="weight")
viz_auc_svyglm(fit_object=fit_example)