Hi all. As I mentioned during our last meeting, we launched a survey in 2021 to collect data on exposure to repetitive head impacts and traumatic brain injury. The goal is to better understand these variables as potential risk factors for Parkinsonism/PD. We did an initial paper that is about to come out (and will share with the community). However, there’s lots of data here. If anyone has an interest, we should discuss about how to maximize use of this data. The leg work is in downloading/merging data sets.
3 Likes
@malosco it would be great to share a link to the paper in the thread Maya posted once released!
In the interim, would be curious to hear more about to data handling you’re referring to: could you discuss a little bit more about how that process went, who was involved, what you think was learned that might impact how you work with the data on a subsequent go-around?
@jgottesman I will definitely share when it is released in August. And, great questions. I do have the benefit of having the infrastructure of a data core and analytics team so we used this resource despite being an unfunded effort. After we registered and received access to the data, we selected the data sets of interest. It then required downloading each data set individually and then merging them for a full analytic set. There are logistics about ensuring the right study visit is selected and that all data is from that visit (depending on your study design). And, some times there are multiple visits that are equal distant from the visit of interest (we chose the most recent in this case). Myself and students guided study design but I credit our data programming team for the work of creating the data sets. Most of the time spent was matching visits between the data files.
We also encountered issues with missingness, based on the nature of the study (online, self-report, unsupervised, etc). In some cases it was challenging to determine the exact reason for missingness and decisions had to be made. Relatedly, we asked some overlapping questions in our survey that are asked as part of the Fox Insight study and there were discrepancies; however, we worked with the Fox Insight team and they helped to resolve the discrepancies here as it was related to a coding issue. The Fox team was super helpful and patient with us as we learned the data set. To this end, the Fox team also provided insight in appropriate use/recommendations of data including how to handle discrepant PD diagnoses (as discussed on another thread). I feel like a tip sheet or common FAQs of recommendations such as this would be a great resource (if it doesnt exist).
The Fox Insight and Den websites were quite informative. The data structure and output were user friendly. But, it does require a lot of hands and a team the first time around. Having direct input and access to the Fox Insight team was really helpful but I am not sure how feasible that is for every project. BUT, seems like this community is the place where that can happen!
3 Likes
Thanks! Would definitely love to see more about how/why you ended up making those design/analysis decisions (e.g., re: missingness). There is an FAQ of sorts at the beginning of the annotated data dictionary: link, but think this could definitely be expanded upon. @maya.sanghvi, could you add this to the running list if it’s not already on there? I think we have quite a few folks here who could contribute to questions/answers on this topic.
1 Like
Will do, Josh! I think this is a really core question/theme – I wonder if we’d want to split to a couple of Topics on design/analysis decisions for different datasets. Thanks for flagging!
@jgottesman @maya.sanghvi I wanted to elaborate more re: decision making for design and analyses.
-
Same measures administered twice. This has been somewhat discussed in previous posts. However, we did encounter the issue of the same questions being administered twice, mostly notably the question pertaining to having a current diagnosis of PD. I typically advocate for only administering questions once to decrease burden and to avoid discrepancies. Although, I appreciate that there are exceptions. In any case, we made the decision to use questions closest in time to the survey we launched. As previously discussed, participants discrepant responses on PD diagnosis at the time of our survey and their most recent FI visit were excluded; it is theoretically possible that these can conflict and be accurate but seems unlikely and more related to erroneous responses.
-
Our survey had unexpected missingness or responses that we had to make assumptions about. For instance, a person reported playing football but then failed to respond to what level they played at and/or years of play. It is unclear why the data was missing other than the participant just failed to complete the entire survey. Either way, for data looking at years of play, they were excluded. In general, people were included in analyses if they had the relevant data and excluded if they did not.
-
Much of our design considerations were on the front end when building and implementing the survey with the Fox Insight team. Josh, is the process we did typical? Here, we spent a lot of time discuss survey logic and testing the programming. The access to the program team at UCSF was incredibly helpful. If we think about working with Fox Insight in the future as part of a grant, I wonder if investigators should consider building funds into their budgets to provide support for the Fox Insight to build and program surveys (or other functions depending on the grant)? Are these services advertised by Fox Insight?
-
We did derive data points and I wonder how this plays into future use of the data by other investigators. As an example, our center calculates years of football play by summing years (or seasons) played at each level in the Fall (the primary season of football). We also have code to derive position at highest level played etc. While we describe this in the methods of the paper, I would think that if anyone wants to use this data in the future they should connect with our team (or it would be helpful). What does that process look like? If someone asks for the data, perhaps reference to the survey being developed at BU could be mentioned and that they might want to reach out for consultation with us? Just a thought.
Overall, I think all of the issues encountered are pretty standard for survey type research. However, it did spark thoughts about services Fox Insight may or may not be able to offer.
1 Like