Multiple outcome timepoints#125

Hiya! I feel like the measurement time point box is not very helpful/confusing. Different variables are measured at different time points and only having one box doesn’t account for this.

Different variables are measured at different time points

How common is this? I don’t read as much lit as you, but it’s infrequent I see this.

I do often see multiple outcome time points reported (in which most data elements are uniformly reported). We created the two time points paradigm (baseline + follow up) simplification to enable combinability of the data and simplify the extraction interface. At some point, we’ll need to open things up to multiple outcome timepoints.

And thanks for reporting the confusion / difficulty, btw :) Do you have any proposals for how it could be improved?

I’ll use stroke papers as an example since we work a lot in that space. In a given stroke paper there will be several different measurement timepoints for outcomes. NIHSS at 24 hours. mRS at 90 days. 30/90 day mortality. Also, things like final TICI that are measured at the end of the procedure. That’s at least 3 time points there and those are the most common ones.

To complicate things further, in other papers, there are mean follow up time periods. Even if a study is going for 3 years, the mean follow up time is usually much less and varies.

In most of the projects I’ve worked on, I’ve basically just put the max time followed up, but it doesn’t really help us in any way since different variables have different measurement time points and it is hard/impossible to choose one that reflects them all.

Hopefully this makes sense? I’m happy to clarify.

Also, I’d say this problem is extremelyyyyyyy common.

I think we should add a column for time period. Like right after sample size.

Got it! I think my perception is biased because most interventions I look at aren’t acute.

We’ll have to take this under consideration. When we look into supporting multiple time points, I think your idea is exactly how we’ll go about this.

As long as this wasn’t fixed and you could change it, I’d be on board!

Yep, definitely not fixed. Based on your feedback, I’ll prioritize above dual RoB/extraction we had planned (this shouldn’t push our timeline back more than a week). I’ll bring a spec to our design meeting next week and get your +1 on what we come up with. May have questions for you before then.

Thanks for pushing on this again, I like knowing what sucks about NK :)

Of course! Thanks for being willing to fix this. I wish I had brought this up sooner. It’s been on my mind for too long.

To recap some discussion we had tonight:

• In AutoLit, gatherers will be able to specify multiple, arbitrary outcome timepoints within a {study, data element}
  • One of these timepoints will be designated as “the” outcome timepoint.
• In Synthesis, and our other export utilities, only “the” outcome timepoint will be shown.

To facilitate posthoc modification of the “the” outcome timepoint, we will provide a bulk action specifying, e.g. for data element {x} set "the" outcome timepoint to any metric at a time between and {a} and {b} days. Corner cases will be handled:

• when a {study, data element} has no outcome timepoint between a and b, “the” outcome will be made empty.
• when a multiple timepoints in a {study, data element} fall between a and b, we’ll pick one of them (e.g. the earliest or closest to the center).

There’s a couple interesting/complicated elements here:

• The choice to force selection of the “the” outcome timepoint in AutoLit is an instance of deciding who should have more narrative control of the data - authors or consumers
• This feature has potential to make dual extraction adjudication a mega headache :p
  • My early instinct says we should reuse baseline and “the” outcome as the datasets to be compared.

Thank you, all, for your patience & help specifying this feature.

You’ll find the (backwards compatible) updates to extraction with release 1.35.0. We’ve added the ability to create > 2 timepoints, and more naturally specify measurement timepoints at the data element level.

These are big changes, so please reach out with any questions!