Analyzing Clinical Significance: The Percentage-Change Approach"

In clinicalsignificance: A Toolbox for Clinical Significance Analyses in Intervention Studies

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>",
  fig.width = 7,
  fig.height = 5,
  fig.align = "center"
)

Introduction

The percentage-change approach defines clinical significance based on the relative change in a patient's score from baseline. This method is common in clinical trials where a specific reduction in symptoms, such as "a 50% improvement," is a primary endpoint for defining treatment response.

A change is considered clinically significant if it meets or exceeds a predefined Percentage-Change Cutoff (PCC). For example, if the PCC is 50% (or 0.5), any patient who shows at least a 50% reduction in symptoms would be classified as "Improved".

A key consideration for this method is that it is highly dependent on the patient's baseline score. A 10-point drop is a 50% improvement for a patient starting at 20 points, but only a 25% improvement for a patient starting at 40 points. This characteristic should be kept in mind when interpreting the results.

This vignette demonstrates how to use the cs_percentage() function to apply this approach.

library(clinicalsignificance)

Basic Analysis

Let's analyze the claus_2020 dataset. A common benchmark for depression treatment trials is a 50% reduction in symptom scores. We will set this as our pct_improvement cutoff.

# Analyze change using a 50% improvement cutoff
pct_results <- claus_2020 |>
  cs_percentage(
    id = id,
    time = time,
    outcome = bdi,
    pre = 1,
    post = 4,
    pct_improvement = 0.5
  )

summary(pct_results)

The summary shows that based on this 50% criterion, about 28% of patients are classified as having improved.

Visualizing the Results

The plot for the percentage-change approach looks similar to those for the anchor- and distribution-based methods. However, the shaded area is now determined by the percentage-change cutoff relative to each individual's starting score. This means the boundaries of the "unchanged" area are not parallel lines.

plot(pct_results)

Grouped Analysis

We can also explore if the proportion of "responders" differs between the treatment groups (TAU vs. PA).

# Grouped analysis with a 50% improvement cutoff
pct_grouped <- claus_2020 |>
  cs_percentage(
    id = id,
    time = time,
    outcome = bdi,
    pre = 1,
    post = 4,
    pct_improvement = 0.5,
    group = treatment
  )

summary(pct_grouped)

The results suggest that a much higher proportion of patients in the Placebo Amplification (PA) group (42.9%) achieved a 50% symptom reduction compared to the Treatment as Usual (TAU) group (10.5%).

The plot clearly visualizes this difference:

plot(pct_grouped)

Using Different Cutoffs for Improvement and Deterioration

A useful feature of cs_percentage() is the ability to set different cutoffs for improvement and deterioration via the pct_deterioration argument. For instance, in some contexts, a small worsening of symptoms (e.g., 20%) might already be considered a significant deterioration, while a larger change (e.g., 50%) is required for improvement.

```r pct_asymmetric <- claus_2020 |> cs_percentage( id = id, time = time, outcome = bdi, pre = 1, post = 4, pct_improvement = 0.5, pct_deterioration = 0.2 # A smaller threshold for worsening )

summary(pct_asymmetric)

clinicalsignificance documentation built on Nov. 27, 2025, 5:06 p.m.