R Shiny开发：如何在模型函数中使用用户选择的多变量

Hey there! No need to apologize for your English—we’re all here to collaborate and help you get this working. Let’s walk through how to let users select multiple variables and plug them into your frailtypack model.

Key Steps to Implement #

The main challenge is converting the multiple selected variables (a character vector from selectInput) into a valid formula for your frailty model. Here’s a complete, working example that builds on your code:

library(shiny)
library(frailtypack)
data("readmission", package = "frailtypack")

ui <- fluidPage(
  # Multiple variable selection for model covariates
  selectInput("var", "Select covariates for the model", 
              choices = names(readmission), 
              multiple = TRUE,
              selected = c("sex", "age")), # Default selection for testing
  # Cluster variable selection
  selectInput("group", "Select cluster variable", 
              choices = names(readmission),
              selected = "id"), # Default cluster variable
  # Button to trigger model fitting (optional but improves user experience)
  actionButton("fit_model", "Fit Model"),
  # Output to display model results
  verbatimTextOutput("model_results")
)

server <- function(input, output) {
  # Reactive expression to build the model formula dynamically
  model_formula <- reactive({
    # Check if user selected any covariates
    if (length(input$var) == 0) {
      return(NULL)
    }
    # Collapse selected variables into a single string separated by "+"
    covars <- paste(input$var, collapse = " + ")
    # Build the full formula string (adjust Surv() arguments if your data uses different column names)
    formula_str <- paste0("Surv(time, event) ~ ", covars, " + cluster(", input$group, ")")
    # Convert the string to a valid formula object
    as.formula(formula_str)
  })
  
  # Fit the model only when the button is clicked
  fitted_model <- eventReactive(input$fit_model, {
    req(model_formula()) # Skip fitting if no covariates are selected
    # Use frailtyPenal as an example—swap with your specific frailtypack function if needed
    frailtyPenal(formula = model_formula(),
                 data = readmission,
                 n.knots = 8, # Adjust parameters to match your analysis needs
                 kappa = 1)
  })
  
  # Display model results or a friendly message if no covariates are selected
  output$model_results <- renderPrint({
    if (is.null(fitted_model())) {
      cat("Please select at least one covariate to fit the model.")
    } else {
      summary(fitted_model())
    }
  })
}

shinyApp(ui, server)

Explanations of Key Parts #

• Dynamic Formula Building: The model_formula() reactive takes the user’s selected variables, formats them into a string that fits the survival model syntax, then converts it to a formula object that frailtypack can use.
• Controlled Model Fitting: Using actionButton and eventReactive prevents the model from re-fitting every time the user changes a selection, which is more efficient and user-friendly.
• Error Handling: req(model_formula()) ensures the model doesn’t run if no covariates are chosen, and we add a clear message to guide the user in that case.
• Flexibility: Swap frailtyPenal with your preferred frailtypack function (like frailtyClust) and adjust parameters to match your specific analysis requirements.

Feel free to tweak UI elements (like adding help text) or model parameters to fit your exact needs!

内容的提问来源于stack exchange，提问作者Alexis