Dear L3C Support, I wanted to reach out with a question about the paper, ?Identifying who has long COVID in the USA: a machine learning approach using N3C data? (Pfaff et al., 2022). I was wondering if you could provide any additional explanation of the rationale for the temporal windows for data inclusion. More specifically, we were curious about the motivation for including post-infection (beyond T+45; look-ahead period) information, given the exclusion of data from the acute infection period (T-45 to T+45). Thank you so much for your time and help. Best, Zach

Created by Zachary Butzin-Dozier zbutzin
Dear Emily, Thank you so much for your thorough response to my questions. This reply was incredibly helpful for my understanding of your rationale for temporal inclusion and exclusion. I really appreciate your taking the time to reply. Best, Zach
Hi Zach, Sure thing; I'll answer in two parts: **Why did we include post-infection data?** Our ML model is designed to be a computable phenotype, rather than a predictive model. This means that the model's primary function is to identify patients with long COVID, rather than determine who WILL get long COVID in the future based on past information. For this reason, we want to include all available data that could help us determine whether or not a patient has long COVID. The L3C Challenge is a different beast, in that it is asking specifically for a predictive model. This means that "extra" data will not be available to L3C models. **So why throw out any data at all, in that +/-45 day window?** The reason for the blackout is purely so the model does not get confused between long-lasting symptoms of acute COVID-19 and long COVID. The 45 days, admittedly, is somewhat arbitrary. At the time we built the model, the long COVID guidelines referred to symptoms lasting "four or more weeks" post acute COVID--however, many of our physician colleagues told us that a longer time period (e.g., 2-3 months) was a safer bet. We tried to split the difference. Hope this helps. Best, Emily

Temporal Inclusion in Pfaff et al. (2022) page is loading…