Thread (for prospective PI's and lead authors) on collaborating with a statistician in clinical research:
(Some of this may end up sounding cranky but it's really not meant to be)
This is also just my personal preference for collaborative workflow; much of it feels somewhat like common sense to me, though others may well disagree
Informed by some early career experiences where I did a *lot* of work for people that never followed through and completed their papers...I now try as much as possible to follow these steps:
(obviously this should be calibrated according to the study design, funding mechanism, time of writing, etc - this mostly refers to analysis phase, after data have been collected)
1. Investigator lets statistician know of desire to write a paper using data from (study) on (question).
2. Statistician and investigator have a meeting to discuss the study question, outline specific analytic objectives and (if appropriate) corresponding hypotheses, and agree on statistical analysis plan for that specific paper.
3. Statistician goes into their hole and performs first round of analysis.
4. Statistician emails collaborator with first round results; team holds meeting to discuss the output(s), make sure everything is clearly presented, ask questions or request things that are missing.
(4b. "What could be missing?" - just my experience, but sometimes little things that weren't clearly spelled out at first - for example, I used to do lots of survival analyses and would often forget to report the specific number of patients that had events in each group)
5. Statistician returns to their hole and performs "second round" of analyses, addressing any questions that came up at first round meeting (possibly making adjustments to Figures, labeling axes as preferred by PI, etc)
(5b. In my experience, final preparation of Figures takes a little more time than most people think. I've tried to get better about asking early if the PI has specific preferences for axis labels, legend, things like that because otherwise that can suck more time than expected)
It feels a bit ridiculous to send 5 emails back and forth with different versions of a Figure when each one has 1 change - e.g. "Can you change that bar from green to red?" - "OK, now can you change the title to read..." - "OK, what if we move the legend from top right to..."
6. Statistician sends collaborator the "second round" of the analysis packet (with any changes or additions that were made at the previous meeting).
AT THIS TIME: the PI really *ought* to have enough to write a full *first draft* of the paper.
AT THIS TIME: the PI really *ought* to have enough to write a full *first draft* of the paper.
If they don't, IMO, that *may* fall on them - either they are veering off course from their original plan (p-hacking, HARKING, or just "oh gee we didn't think of that, let's try this instead"-ing) or there was poor communication about the analysis plan.
7. Of course, the statistician can & should support as needed to complete the paper - help write statistical methods & results, any last "fill in the blank" requests or adjustments to Tables and Figures, etc - but IMO at this point it is fair for the statistician to ask to see...
...at least a *draft* of the paper before proceeding with further analysis requests. This helps the statistician protect their effort, and make sure it is spent on projects that are moving forward.
(I say this having experienced a few dozen projects where dozens of pages of statistical output never translated to a finished paper, because PI/lead author was too busy, got distracted by other projects, etc)
It's also being respectful of your own clinical colleagues, who may also be vying for that statistician's time / effort, and wondering why it's taking so long for the statistician to get to *their* paper, unaware that they've had to revisit *your* analysis 3 times this week.
If your statistician has already provided 62 pages of analysis with 17 Tables & 24 Figures, and you tell them something like "I just need (2 more things) and then I'll be able to start writing" - that's likely to draw an incredulous reaction (at least, from me...)
RELATED: you don't need the final presentation of the Figures to write a first draft. If you tell the statistician you're waiting to write the paper until you get Figures with the correct axis label, come on. There's no reason you have to wait for that to write a first draft.
8. Statistician should ideally see the "final" version of the paper (after all other co-authors have had their say) one last time before submission to make sure that any revisions suggested by co-authors are still consistent with the analyses that were done and reporting.
9. After submission - during peer review process & revisions - please engage the statistician fully! Don't send them vague requests like "We got comments back from reviewers. Can you redo the regression analyses and make the lines thicker on Figure 2?"
9b. Better to share the submitted paper with the reviewer comments, so they can see the context of each comment and help you address appropriately. We want to help, but
as a full participant in the process - it helps us provide a more useful response if we see the whole thing
as a full participant in the process - it helps us provide a more useful response if we see the whole thing
Anyways, just some suggestions (hopefully useful for statisticians and PI's in managing their collaborative relationships). I've semi-formally discussed this with most PI's that I work with, and in some cases even written down a form of this as a 'policy' for their studies.
Oh, and also: don't do this.
