In erincr/sweetspot: Find and Assess Treatment Effect Sweet Spots in Clinical Trial Data

library(sweetspot)

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>",
  fig.path = "man/figures/README-"
)

Sweet spot analysis

Find and assess treatment effect sweet spots in clinical trial data

Identifying heterogeneous treatment effects (HTEs) in randomized controlled trials is an important step toward understanding and acting on trial results. The method in this package exploits any existing relationship between illness severity and treatment effect, and identifies the "sweet spot": the contiguous range of illness severity where the estimated treatment benefit is maximized. We further compute a p-value to compare to the null hypothesis of no treatment effect heterogeneity, and we bias-correct our estimate of the conditional average treatment effect in the sweet spot. Finally, we provide a function for visualizing results. Because we identify a single sweet spot and p-value, we believe our method to be straightforward to interpret and actionable: results from our method can inform future clinical trials and help clinicians make personalized treatment recommendations.

Example 1: data with a sweet spot

We generate randomized trial data with a treatment effect sweet spot. We choose n=1000 trial participants with p=10 covariates that - together with treatment - determine probability of a positive outcome. Treatment assignment is randomized: every participant has a 50\% chance of receiving treatment.

A participant is in the "sweet spot" if their probability of a positive outcome (without treatment) is in the middle 20\% of probabilities. The average treatment effect for participants not in the sweet spot is 5\%: that is, treatment additively increases the probability of a positive outcome by 0.05. For participants in the sweet spot, this is instead 25\%.

set.seed(1234)
n <- 1000; p <- 10;
treated    <- sample(c(0,1), n, replace=TRUE)
covariates <- matrix(rnorm(n * p), nrow=n, ncol=p)
beta       <- rnorm(p)
outcome.prob <- 1/(1+exp(-(covariates %*% beta)))
in.sweet.spot <- !is.na(cut(outcome.prob, c(.4, .6)))
outcome.prob[treated==1 & !in.sweet.spot] <- outcome.prob[treated==1 & !in.sweet.spot] + .05
outcome.prob[treated==1 &  in.sweet.spot] <- outcome.prob[treated==1 & in.sweet.spot]  + .25
outcome.prob     <- pmin(outcome.prob, 1)
positive.outcome <- rbinom(n, 1, prob=outcome.prob)
negative.outcome <- 1-positive.outcome

We are now ready to run the sweet spot analysis.

result <- sweetspot(treated, covariates, negative.outcome, positive.outcome, "binomial")
plot_sweetspot(result, title="Sweet spot on simulated data")

We may also look at the predictions from the prevalidated predilection score model. For example, we can compute the AU-ROC of the predilection score model on untreated patients:

pos.scores <- 1/(1+exp(-result$predilection.scores[negative.outcome==1 & treated==0]))
neg.scores <- 1/(1+exp(-result$predilection.scores[negative.outcome==0 & treated==0]))
auc <- mean(sample(pos.scores,10000,replace=T) > sample(neg.scores,10000,replace=T))
auc

Example 2: data without a sweet spot

We now generate data without a sweet spot. We again choose n=1000 trial participants with p=10. The treatment effect for all patients is 5\%.

set.seed(1234)
n <- 1000; p <- 10;
treated    <- sample(c(0,1), n, replace=TRUE)
covariates <- matrix(rnorm(n * p), nrow=n, ncol=p)
beta       <- rnorm(p)
outcome.prob <- 1/(1+exp(-(covariates %*% beta)))
outcome.prob[treated==1] <- outcome.prob[treated==1] + .05
outcome.prob     <- pmin(outcome.prob, 1)
positive.outcome <- rbinom(n, 1, prob=outcome.prob)
negative.outcome <- 1-positive.outcome

result <- sweetspot(treated, covariates, negative.outcome, positive.outcome, "binomial")
plot_sweetspot(result, title="Sweet spot on simulated data")

References

For details of this method, please see the following preprint:

Erin Craig, Donald A. Redelmeier, and Robert J. Tibshirani. "Finding and assessing treatment effect sweet spots in clinical trial data." arXiv preprint arXiv:2011.10157 (2020). https://arxiv.org/abs/2011.10157

For more about the data, please see:

Dondorp, Arjen M., et al. "Artesunate versus quinine in the treatment of severe falciparum malaria in African children (AQUAMAT): an open-label, randomised trial." The Lancet 376.9753 (2010): 1647-1657. https://www.sciencedirect.com/science/article/pii/S0140673610619241

The data in this package was downloaded from Github: https://github.com/Stije/SevereMalariaAnalysis.

It was initially published to accompany the following research:

Watson, Leopold et al. Collider bias and the apparent protective effect of glucose-6-phosphate dehydrogenase deficiency on cerebral malaria eLife (2019). https://elifesciences.org/articles/43154

Leopold, Watson et al. Investigating causal pathways in severe falciparum malaria: a pooled retrospective analysis of clinical studies In Press, PLoS Medicine (2019). https://journals.plos.org/plosmedicine/article?id=10.1371/journal.pmed.1002858

Watson, J.A., Holmes, C.C. Graphing and reporting heterogeneous treatment effects through reference classes. Trials 21, 386 (2020). https://doi.org/10.1186/s13063-020-04306-1

erincr/sweetspot documentation built on April 14, 2022, 12:41 a.m.