Dynamic AUC and time dependent concordance

[tommyvcc]

Thank you so much for the hard work
I have one question regarding the AUC metric.
Is it a "cumulative dynamic AUC " or " incident dynamic AUC" ?
For example, if the time intervals are 0 to 1 , 1 to 2 and 2 to 3.
While the AUC is 0.6, 0.7, 0.8
Does this mean that the AUC for the period 0 till 3 is 0.8?
Or the AUC for the period 2 to 3 is 0.8?

[Frederik-Hartmann]

I am not completely following your questions, but I hope this helps anyway:

In the current version auton-survival uses scikit-survivals cumulative_dynamic_auc

auton-survival/auton_survival/metrics.py

Line 31 in 5dde465

from sksurv import metrics, util

auton-survival/auton_survival/metrics.py

Lines 258 to 265 in 5dde465

	def _cumulative_dynamic_auc(survival_train, survival_test, predictions, times, random_seed=None):
	idx = np.arange(len(predictions))
	if random_seed is not None:
	np.random.seed(random_seed)
	idx = np.random.choice(idx, len(predictions), replace=True)
	return metrics.cumulative_dynamic_auc(survival_train, survival_test[idx], 1-predictions[idx], times)[0]

In the docs of scikit-survival https://scikit-survival.readthedocs.io/en/stable/api/generated/sksurv.metrics.cumulative_dynamic_auc.html you can find more information.

Coming back to your question. If the time points are 0, 1 and 2. You will get the cumulative_dynamic_auc at these specific time points. e.g. the cumulative dynamic AUC at time point 1 would be 0.6 in your example. If I am not mistaken the cumulative dynamic AUC is not calculated over a time range, but rather compares the test samples to your time point.