Are the multiple drugs in the column `drug` in data files the group of drugs being compared in the trial at the same time? about clinical-trial-outcome-prediction

[futianfan]

clinical-trial-outcome-prediction/benchmark/data_split.py

Line 398 in f8f3159

remove placebo

this is the code we remove placebo, it did promote the accuracy.

[futianfan]

yes, each line represents a trial, which may contain multiple drugs in the trial.

[jasperhyp]

Thanks! But how do I know which drug in a list is the drug that the label of success/fail is assigned to?

[futianfan]

good question, trials may test one drug or drug combination, so we do not need to discriminate which one is the tested drug.

[jasperhyp]

Hmmm interesting. I am thinking about two scenarios. First, in some cases a certain drug that does not work (usually placebos, sometimes vitamins perhaps, lists of drugs that contain them are present in the dataset) is used for the control group, and I guess they should be tested. Second, in some cases a known effective drug might be the control group (aka Active Comparator), especially in some cancer treatment studies. This might not lead to information leakage if we assume the choice of experiment for a certain drug to be unbiased, but probably would make it ambiguous as to which drug/drug combinations are being tested. Anyways, thanks for collecting this nice dataset!

[futianfan]

thanks for pointing out it. That is really a problem. But for the first scenario, we did remove the placebo. For second scenario, we have not found good solution to do it. manually labelling it is too costly. Thanks a lot again!

[jasperhyp]

clinical-trial-outcome-prediction/benchmark/data_split.py

Line 398 in f8f3159

remove placebo

this is the code we remove placebo, it did promote the accuracy.

Wonderful! Thanks for referring me to this. I assume the second issue might be ameliorated if the notes on clinicaltrials.gov for the specific trial can be analyzed, maybe through some NLP information extraction methods. It might contain a general description of which drug/drug combination is being tested. But your paper is already wonderful! Thanks again for sharing the dataset.

[futianfan]

thanks for your attention and suggestions! we will consider your suggestions!