However if we add the data of this patient (smoker) to the original model with as outcome 'not getting lungcancer' the model will be biased with the idea smoking --> not getting longcancer. Which is incorrect.

I would add a had_treatment boolean in your input variables. That way, the model should understand that "having had treatment" is a major predictor for not developing cancer.

Although the above suggestion is just a way to mitigate the problem that you mention, without suppressing it entirely. If you want to do things really properly, you should refrain from updating your model with new patients, whether you decided to give them treatment or not (if you update the model with patients whom you decided to to give treatment only, this will bias the data in favour of people who have an a-priori lower probability of getting cancer, and you probably don't want that).

So, in a nutshell, one of two ways:

• either include a boolean had_treatment (yes/no) in your input data
• or refrain from adding new patients (including the ones you decided not to give treatment)

However if we add the data of this patient (smoker) to the original model with as outcome 'not getting lung cancer' the model will be biased with the idea smoking --> not getting lung cancer. Which is incorrect.

I would add a had_treatment boolean in your input variables. That way, the model should understand that "having had treatment" is a major predictor for not developing cancer.

Although the above suggestion is just a way to mitigate the problem that you mention, without suppressing it entirely. If you want to do things really properly, you should refrain from updating your model with new patients, whether you decided to give them treatment or not (if you update the model with patients whom you decided to to give treatment only, this will bias the data in favour of people who have an a-priori lower probability of getting cancer, and you probably don't want that).

So, in a nutshell, one of two ways:

• either include a boolean had_treatment (yes/no) in your input data
• or refrain from adding new patients (including the ones you decided not to give treatment)

One last consideration: when adding new patients data, it's highly likely that this data is biased towards people being at risk. Why would a non-smoking 22 years old with no family antecedent come see you to have a check? Should you decide to include new patients data, you need to counter that bias one way or the other.