March 27, 2024
In this article, the authors described theirSchemper-Henderson measure to account for explained variation for survival outcomes, extended to allow for competing risks. They defied explained variation as therelative gain in predictive accuracy when prediction is based on prognostic
factors that replaces unconditional prediction. Essentially, the proportion of variation inthe survival outcomes ended up being attributed to single or multiple
prognostic factors. It also was alignedwith “Importance” of a prognostic factor. They called their measure, V. Their measure accounts for predictiveinaccuracy with covariates versus without accounting for covariates. They only allowed for one competing event and have not extended to multistate models. Their measure used the cumulative incidence function, CIF, to account for competing risks in the estimation of survival where it takes the ratio between observed
and expected cumulative incidence.
They then proposed direct and indirect estimates of theirmeasure, EVCR. The directestimates contrasted observed and expected outcomes and revealed the way
censored observations and CE are taken into account. The indirect estimate
replaced population values for CIFs by appropriate consistent point estimates
where the expectation was by averaging and the integration was by summing. Finally, they ran their methods throughsimulations and also through a real data example. They did admit that if the effect of theprognostic factor is high than it may explain little of the variability in the
outcome so once has to be careful in estimating hazard ratios. The authors had not shown alternativesurvival measures. The method was not perfect and then always depends upon
other factors in the estimation.
Written by,
Usha Govindarajulu,PhD
Keywords: explainedvariation, survival analysis, competing risks
References
Gleiss A, Gnant M, and Schemper M. (2024) “Explained variation anddegress of necessity and of sufficiency for competing risks survival data”.
Biometrical Journal. https://doi.org/10.1002/bimj.202300140
