Run a LAGO optimization in a point and click app

Opens a Shiny app in your web browser that runs the whole LAGO optimization without writing R code. You load a dataset, choose the outcome and the goal, choose the intervention components with their bounds and costs, and say how the model accounts for centers and time. The app then finds the recommended intervention with lago_optimization() and shows the result, the confidence set and the R code that reproduces the run. From that result it can sweep one input with lago_sensitivity(), find the best outcome for a budget with lago_budget(), and save a report with lago_report().

Usage

lago_app(data = NULL, launch.browser = interactive(), port = NULL)

Arguments

data: A data.frame to start with, or NULL to start with the bundled BB_data example.
launch.browser: Whether to open the app in your web browser. By default it opens in an interactive session.
port: The port to listen on, or NULL for a free one.

Value

Invisibly, when you click Done or close the browser tab, a list with the last result of lago_optimization() (or NULL if nothing was run), the last sensitivity of lago_sensitivity() and budget of lago_budget() run from it (or NULL), the R code of all of them as a character vector, and the arguments of the optimization. The same list is stored in the lago_app_result option.

Details

The app runs on your own computer and is reachable only from it, at 127.0.0.1. Nothing is sent over the network.

The app has these steps.

Data: the data you pass, the bundled BB_data, R's mtcars, or a CSV file you upload. Each example opens on a setup that runs as is.
Outcome and goal: outcome_name, outcome_type, outcome_goal, outcome_goal_intention, and for a binary outcome the power goal: power_goal, num_centers_in_next_stage, patients_per_center_in_next_stage, power_goal_approach, icc and power_goal_cluster_id. The app builds the group column from the treatment arm column you choose.
Intervention and costs: intervention_components, their bounds, and the costs as unit_costs with default_cost_fxn_type and unit_cost_doses, or as cost_list_of_vectors. The "Shape the costs visually" button opens the cost designer of visualize_cost() in a dialog, which starts on the costs in use and puts the shaped costs in as cost coefficients. Interaction terms set include_interaction_terms and main_components, and the app builds each product column.
Centers, time and covariates: center_characteristics with their values, or fixed center effects with include_center_effects and center_effects_optimization_values, fixed time effects with include_time_effects and time_effect_optimization_value, and additional_covariates. The app builds the center and period columns from the columns you choose.

A step that needs attention says what to change, and the run button stays off until every step is ready. Warnings from the package are listed with the result.

After a run, these tabs start from its result.

Results: the recommendation, the confidence set with a CSV download, the outcome model, the intervention test, and the report download of lago_report(), which needs the rmarkdown, knitr and jsonlite packages and pandoc.
Sensitivity: lago_sensitivity() with parameter and values. It varies the outcome goal, a multiplier of all costs, and with a power goal the power goal, the centers and patients of the next stage, and the ICC. With time effects it varies the period. The values are even steps from a start to an end, or a typed list. A plot, a table and a CSV download show the result.
Budget: lago_budget() with budget and n_grid. The budget starts at the cost of the recommendation. The best outcome it buys can be put in as the outcome goal.
R code: the code of the run, followed by the code of the last sweep and budget search of that result.

Examples

if (interactive()) {
  out <- lago_app()
  out$result
  cat(out$code, sep = "\n")
}