Thread by @ADAlthousePhD

@Jabaluck @_MiguelHernan @aecoppock I think (perhaps unsurprisingly) that this shows “different people from different fields see things differently because they work in different contexts” - the scenario you painted here is not really possible with how most *medical* RCTs enroll patients & collect baseline data

@Jabaluck @_MiguelHernan @aecoppock The workflow for most medical RCTs (excepting a few trial designs…which I’ll try to address at the end if I have time) is basically this:

@Jabaluck @_MiguelHernan @aecoppock 1. Clinics/practices/hospitals know that they are enrolling patients in such-and-such trial with such-and-such criteria.

@Jabaluck @_MiguelHernan @aecoppock 2. As potentially eligible patients present themselves, the designated staff may approach them and inquire whether they are interested in being part of the aforementioned trial.

@Jabaluck @_MiguelHernan @aecoppock 3. They will perform an eligibility assessment and confirm that the patient is eligible for the trial. Depending on how complex the eligibility criteria are, this may be done on the spot or may take days / weeks.

@Jabaluck @_MiguelHernan @aecoppock (Notice that at this point we haven’t even started talking about their treatment assignment or the randomization process yet)

@Jabaluck @_MiguelHernan @aecoppock 4. OK, so now the patient has been confirmed as “eligible” for the trial. In most modern RCTs this data is being recorded in an electronic data management system. Once the data are entered to confirm eligibility - again, all of this before anyone knows treatment assignment…

@Jabaluck @_MiguelHernan @aecoppock 4b. Those data are now locked and cannot be modified without filing a data change request form with the data coordinating center. There’s no monkeying with this data because they found out the patients treatment and decided to change something

@Jabaluck @_MiguelHernan @aecoppock 5. Now that the patient has been confirmed eligible, consented to be in the trial, any baseline data of interest should be collected. The patient *still* hasn’t actually been randomized yet. We want to collect as much of this data as possible before randomizing them…

@Jabaluck @_MiguelHernan @aecoppock 6. Only after the patient has been confirmed eligible and had most/all baseline data recorded and entered are they randomized in the trial. How does this part work?

@Jabaluck @_MiguelHernan @aecoppock 6b. The randomization is separately prepared, hidden from view of the staff (allocation concealment is a basic principle of trial implementation in medicine - the person enrolling patients in the trial should have no idea what the next assignment is, to avoid precisely this)

@Jabaluck @_MiguelHernan @aecoppock 6c. In this way we can assure that the eligibility & baseline data are unaffected by the knowledge of tx assignment - it’s all entered before the patient is assigned & by someone that has no ability to see what the next assignment is until the data are entered and locked

@Jabaluck @_MiguelHernan @aecoppock (Depending on the exact trial design - level of blinding possible - it’s possible they won’t even know *after* the patient is randomized what they are assigned to - but even if an open label trial, all of the baseline data has to be entered & confirmed before anyone knows the tx)

@Jabaluck @_MiguelHernan @aecoppock The point is, in most (modern, medical-environment) RCTs, there are no lists being merged or anything of the sort you’re worried about here. The data for each patient are being entered directly, in real time, before knowledge of tx assignment, and locked…

@Jabaluck @_MiguelHernan @aecoppock …and so can only be “changed” by filing a data change request (for even the smallest possible thing, like “I typed 90 and it should have been 80” - still requires a formal DCR to the data coordinating center for us to unlock and change that value)

@Jabaluck @_MiguelHernan @aecoppock All of this to say that so often these across-field “wait you do it that way? why?” discussions often eventually show that, surprise, there are other differences in the way we see and do things which influence them.

@Jabaluck @_MiguelHernan @aecoppock And I’ll admit that I’m quite skeptical running a few baseline t-tests will be a particularly sensitive method for detecting errors of the type: “the research staff screwed with the data and/or merged the data wrong”

@Jabaluck @_MiguelHernan @aecoppock Oh, also, there have been some interesting discussions over time about the merits of using “simple randomization” (every patient assigned with 50% probability, fully independent of past or future assignments) versus blocked randomization and whether the efficiency gains with…

@Jabaluck @_MiguelHernan @aecoppock …blocked randomization are worth the potential for research staff possibly inferring what the next treatment assignment will be based on the prior assignment(s). There are of course various safeguards you can put in place to protect against this as well…

@Jabaluck @_MiguelHernan @aecoppock …such as randomly permuted block sizes, or having multiple strata (which makes it harder for the staff at any one site to really follow the sequence and guess what’s next at their site) but some argue that you can avoid all that by using (what we call!) simple randomization