Skip to content

Identifying parameters of a distribution given first moment and c-statistic

License

Unknown, MIT licenses found

Licenses found

Unknown
LICENSE
MIT
LICENSE.md
Notifications You must be signed in to change notification settings

resplab/mcmapper

Repository files navigation

mcmapper

The goal of mcmapper is to enable estimating the parameters of two-parameter distributions for risks (e.g., beta, logit-normal) based on the expected value and c-statistic of the distribution. The theory behind this approach is explained in Sadatsafavi et al.

Some of the recent developments in predictive analytics, such as new methods for sample size calculations for risk prediction models, require knowing the distribution of predicted risks. However, original reports seldom provide such information. On the other hand, prediction model development and validation studies most often report the expected values of predicted or observed risks, and the c-statistic of the model for predicting the outcome. For a given family of distributions, one can use this package to recover the parameters of such a distribution.

The uniqueness of the characterization is established in a forthcoming paper.

Installation

You can install the development version of mcmapper from GitHub with:

# install.packages("devtools")
devtools::install_github("resplab/mcmapper")

Example

Gupta et al validated their risk prediction model for the risk of deterioration in hospitalized patients with COVID-19 infection. In the external validation sample (from London, UK), the prevalence of deterioration was 43%, the model had a c-statistics of 0.77. Assuming that calibrated risks had a logit-normal distribution, we can recover the parameters:

library(mcmapper)

res <- mcmap(c(m=0.43, c=0.77), type="logitnorm")

print(res)
#> $type
#> [1] "logitnorm"
#> 
#> $value
#>         mu      sigma 
#> -0.3605462  1.1836304 
#> 
#> attr(,"class")
#> [1] "mcmapper_output"

About

Identifying parameters of a distribution given first moment and c-statistic

Resources

License

Unknown, MIT licenses found

Licenses found

Unknown
LICENSE
MIT
LICENSE.md

Stars

Watchers

Forks

Packages

No packages published

Languages