OHDSI Home | Forums | Wiki | Github

LAERTES Inclusion of animal studies in drug-event pubmed extractions


(Nicholas Tatonetti) #1

Hi all,

During our latest LAERTES meeting Rich brought up that the automatic extraction of drug/event pairs as described by Avillach et al. pulls out a lot of animal studies results in addition to human results.

This thread is to discuss how to handle these data.

I would like to include these data, as they can be useful for investigation, however, I believe that they should be clearly delineated. Then if users do not want to use animal studies results they can easily exclude them. In fact, if this is the normal use case then we can set up LAERTES so that you must explicitly request to include animal results. This way it does not mistakenly get used by unwitting users.


(Rkboyce) #2

The RDF graphs used for the drill down use case do keep track of MeSH tags and it appears that some MEDLINE entries tag that the adverse effect is in an animal. For example, PMID 24155430 indicates Dog Diseases (D004283)). Further study of this topic will be necessary before v1.0 is released. Also, are there any HOIs that are clearly non-human and so would not be present in SNOMED? .


(Rkboyce) #3

This post is to revive an important topic related to LAERTES. The question of how to treat animal and human evidence in the evidence base. Currently, no distinction is made between scientific literature evidence involving humans and that involving animals or in silico systems. I think it is reasonable that the system make it easy for users to filter on those types if needed. One approach would be to create new evidence types that would distinguish those broad categories. I prefer to keep the evidence types small and rather just tag the specific evidence items with a separate flag indicating human or non-human so that folks can use that as needed. Please let me know if you have any thoughts one way or another, or other comments/questions on this topic.

thanks,
-Rich


(Christian Reich) #4

@rkboyce:

Isn’t it more than that? We would then have to also grab all the products only approved for animals, and their ingredients and all that, to give the problem a fair treatment. What do you think?


(Lili Peng) #5

HI Richard, Can you present a use case of how animal vs human data in drug-HOI relationships are used differently in interpreting medical outcomes? You argue that these two should be delineated separately, but why? Is there a strong need for labelling evidence types as human vs. non-human that I’m unaware of?


(Rkboyce) #6

I will make some comments but, since I am not clinically trained, I am hoping that @Christian_Reich and @hergchan will also chime in. In working with other clinical scientists, I have often heard that there a number of potential issues with animal data that make pharmacologic observations in such systems less trustworthy for supporting claims about what can occur in humans. These include differences in biochemical pathways, gene and protein function and structure, and the difficulty in identifying phenotypes that map to human conditions (e.g., depression). For this reason, it makes sense to me to provide some tag that will help client systems show those users who care when an evidence item is based on animal research. This costs almost nothing to do and the potential benefit for certain investigations would be large. I think of Laertes as being oriented toward helping research drug-HOI associations that occur in humans. I think @Christian_Reich’s question about bringing in all products approved only for animals might make sense if someone wanted to work deeply with animal data. Right now, my proposal is to address the fact that more than a quarter and maybe even more than half of the evidence from certain literature sources (SemMed and now CTD) will be probably be from animal studies.


LAERTES Team Meeting Agenda
(Lili Peng) #7

I agree with this. Animal data is usually acquired during preclinical testing of the drug development pipeline, whereas human data is acquired during clinical stage. I’m not formally trained as a drug safety practitioner, but if I were and wanted to find data on drug-HOI relationships for a particular drug or outcome, and I were to find both animal and human data on LAERTES (presuming the data would be tagged accordingly), I would lend more credence to the human data than animal data, since human data is more clinically relevant.

Not distinguishing between the two types would be a disadvantage to LAERTES.

By the way, Richard, I saw your email to the LAERTES working group regarding your focusing next week on upgrading/updating LAERTES. Do these upgrades have anything to do with these discussions on ‘modality’ and tagging animal vs. human data?


(Rkboyce) #8

The upgrade is primarily focused on the issues tagged here: https://github.com/OHDSI/KnowledgeBase/milestones/January%202016%20Data%20Load

I will also to follow up on the discussion of modality and animal data by making appropriate changes to the process so we can discuss the results afterwards. I take an iterative approach to this but believe strongly in having something built to look at and discuss.


(Waydes) #9

I am the current president of VMDB - Veterinary Medical Databases. A group of veterinary schools and organizations is looking at using the Common Data Model for animal records. I know this is an old thread but it was the only one that discussed animals. I have some thoughts on how to do this and would like to bounce ideas of of others. It would be really good to use the various tools, especially White Rabbit / Rabbit in a Hat, Atlas and others. Any comment and suggestions appreciated. Are the are specific groups that discuss software. Thanks.


t