loss_integrate: Calculate integrated metrics based on time-dependent metrics.

In survex: Explainable Machine Learning in Survival Analysis

Calculate integrated metrics based on time-dependent metrics.

Description

This function allows for creating a function for calculation of integrated metrics based on a time dependent metric. A possibility to cut off the data at certain quantiles is implemented, as well as weighting the integrated metric by max time and marginal survival function [1]

Usage

loss_integrate(loss_function, ..., normalization = NULL, max_quantile = 1)

Arguments

loss_function	A time dependent loss function taking arguments (y_true, risk, surv, times)
...	other parameters, currently ignored
normalization	either NULL, "t_max" or "survival". Decides what kind of weighting should be applied to the integrated metric. If "t_max", then the integral is calculated using dw(t) where w(t) = t/t_max. If "survival", then the integral is calculated using dw(t) where w(t) = (1 - S(t))/(1 - S(t_max)) and S(t) denotes the estimated marginal survival function. If NULL (default), the integral is calculated using dt.
max_quantile	a number from the interval (0,1]. The integral will be calculated only up to the time value of quantile(max_quantile) of the observed event/censoring times in y_true.

Value

a function that can be used to calculate metrics (with parameters y_true, risk, surv, and times)

References

• [1] Graf, Erika, et al. "Assessment and comparison of prognostic classification schemes for survival data." Statistics in Medicine 18.17‐18 (1999): 2529-2545.

survex documentation built on Oct. 25, 2023, 1:06 a.m.