Tiger survival baseline

Hi, The Tiger survival baseline of 0.7 incorporates information from both clinical vriables and the TIL score. Challenge participants only have access to the TIL component so if the clinical data adds value, then submitted algorithms are unlikely to exceed this score. It would be very helpful to know what the baseline for just the TIL scoring component is as that gives a more realistic target to aim for. Thanks Anne

Dear Anne,

I am not entirely sure if I follow you. But the baseline model is developed with the same training data that we provided to all the participants.

The clinical variables are used in the R-script to compute the c-index. This incorporation works in the same way for all submitted algorithms.

You can find the pipeline for the baseline algorithm here

Best wishes, Mart

Dear Anne,

To add to my previous message.

Maybe you want to know the C-index when training a regression model based on the TILs scores from the TIGER baseline, without the other clinical variables.

We calculated this now for the TIGER baseline algorithm, and this is the result: 0.64 [0.45-0.81],

We computed this by replacing line 35 in the R-script:

fitcox=coxph(Surv(timerecurrence,recurrence)~ageclass+ morphologygroup + grade + moleculargroup + stage + surgerybreast + adjuvanttherapy + TILscore,data=train) # defines the fitting function

with this:

fitcox=coxph(Surv(timerecurrence,recurrence)~TILscore,data=train)

Furthermore, we think that participants can exceed the C-index 0.7 of the TIGER baseline result if their model predicts TILs scores with a higher prognostic value than the TIGER baseline algorithm. This morning a participant's submitted algorithm exceeded this score.

Best wishes, Mart

I think you did what Anne was asking. I agreed with Anne on her request, and we were discussing internally. I understand that there is a desire to include the clinical variables in a clinical application, but I think it is also good to test the TIL scoring algorithms without creating a model that includes the clinical variables. I would think that more care would be taken to include the clinical variables with the TIL score.

In your new results that you report below, why are you doing a regression?

There is no need to regress anything to calculate the C-index that compares the TIL score to the survival data.

Brandon Gallas

Hi Brandon,

Sorry for the late reply, we have had a few days of holidays in the Netherlands since you posted your question and this slipped off our radars a bit.

About considering clinical variables and TILs score together in the same model, this is something that we are doing to consider the prognostic value of a TILs score when put together other clinical variables, to see if it really adds something as an indipendent prognosticator or not. The reason why we do it by building a regression model together with the other variables is to have a practical way to evaluate this effect with a parameter (the c-index) that can be used to make a leaderboard. In post-challenge analysis we will also do additional statistical tests that we are discussing internally now, but for the challenge we found this to be a convenient way to assess the performance without ignoring the presence of other clinical variables and their potential interactions with the TILs.

About building a regression model to compute the c-index for the TILs alone, this is not something that we are using within TIGER but that will be part of the post-challenge analysis. In the result that Mart provided, for conveniency we just reused the R script that has been shared publicly, so that people can also reuse it locally and check their results, assuming it is a good approximation of what you would get by taking the raw TILs scores.

Regards, Francesco