Hello. Is there any script that calculate the C-index

Hello. I asked gpt how to calculate the C-INDEX. And it tells me that C-INDEX need event occurrences (1 if the event occurred, 0 otherwise) to calculate. Would you mind using the C-INDEX used in the competition to calculate the C-INDEX? Thanks.

Hi,

In order to better help you, could you please elaborate your problem and your question a little bit more?

Hello. I mean how to calculate the C-Index for the lung task. And do we need to predict the 'event' as well?Because I saw there are two keys in the ground_trush.json, {"survival_time_months": 43.7, "event": 0.0}

Hi,

We only ask that they predict survival time. You can find more info here: https://github.com/chaimeleon-eu/OpenChallenge/tree/master

hello,

Just to clarify the confusion. Think of the AUC in a binary classificaiton problem: it is a ranking metric equivalent to the following process:

1. Sample a pair of items in the test set, one of class 0 and one of class 1.
2. Compare their predicitions. The AUC rewards you if you assign higher probability to the class 1 item than to the class-0 item.
3. Repeat, until you have sampled all possible pairs of (class 0, class1) in your test set.

The c-index is a metric similar to the AUC, but it will take into account when an item is right-censored, i.e. when you predict a survival of T months, but after T the subject left the study and you don't know if he/she died or not, nor when. This is the "event=0" case in the data we have. The c-index will take this into account when forming valid pairs of items in the test set, as it would be unfair to use right-censored data in some pairs.

Summing up, you do not need to predict the event (it would be rather impossible to know if someone died or left the study early), it's just that the performance metric takes that into account. On the other hand, when training your models you probably want to take into account that it's not the same having (survival=T, event=0) than (survival=T, event=1).

Cheers,

Adrian

Thanks!