On 2017 Mar 19, NLM Trainees’ Data Science Journal Club commented:

The NLM Trainees Data Science Journal Club for this discussion consisted of several NLM staff members, including one of the paper’s authors. In brief, the authors employed a visual and a computational method to explore the contribution of individual drugs to a drug class-level signal in the context of adverse events. The authors used both techniques to identify class effects caused by all members of a drug class individually, as well as class effects based on a subset of drug class members.

As background information regarding adverse events, one group member pointed out that there has to be enough evidence between a drug or class and an adverse event for it to be formally labeled an adverse event. As such, the type of comparison performed in this study has the potential to introduce false positives, where there is a co-occurrence between a drug and an event that is not to the level of a “label-able” event.

In terms of methodology, the group discussed that, while this study is biased towards using case reports, the strength of evidence would likely be higher if based on randomized controlled trials; however, RCT’s are generally more focused on drug efficacy than safety.

The question was raised as to whether study results might differ if the drugs or adverse events were mapped using different terminologies than ATC and MeSH, respectively. In response, the author pointed out that the FDA uses MEDRA to report events, but for the purposes of this study, it was more complicated to use than MeSH, as it would have required hand curation (beyond the automated curation the UMLS offers) to be useable. That said, the general methodology used by the authors is largely terminology agnostic and can be used with any terminology that is reasonably hierarchical.

One group member was curious whether this process could be used for clinical decision support. Discussion revealed that this methodology is not suited for CDS, as the evidence would need to be clearer. For CDS use, the process would need to be based on drug labels or drug information systems, which are usually expressed at the individual drug level. It might be interesting to know if a class was affecting adverse events at the class level or based on a subset of the class, but this would not be the primary concern in a primary care setting. However, the results of the study could be valuable to drug safety professionals and those building adverse event repositories to support their review or decisions in regard to adverse event relationships.

On 2017 Mar 19, NLM Trainees’ Data Science Journal Club commented:

The NLM Trainees Data Science Journal Club for this discussion consisted of several NLM staff members, including one of the paper’s authors. In brief, the authors employed a visual and a computational method to explore the contribution of individual drugs to a drug class-level signal in the context of adverse events. The authors used both techniques to identify class effects caused by all members of a drug class individually, as well as class effects based on a subset of drug class members.

As background information regarding adverse events, one group member pointed out that there has to be enough evidence between a drug or class and an adverse event for it to be formally labeled an adverse event. As such, the type of comparison performed in this study has the potential to introduce false positives, where there is a co-occurrence between a drug and an event that is not to the level of a “label-able” event.

In terms of methodology, the group discussed that, while this study is biased towards using case reports, the strength of evidence would likely be higher if based on randomized controlled trials; however, RCT’s are generally more focused on drug efficacy than safety.

The question was raised as to whether study results might differ if the drugs or adverse events were mapped using different terminologies than ATC and MeSH, respectively. In response, the author pointed out that the FDA uses MEDRA to report events, but for the purposes of this study, it was more complicated to use than MeSH, as it would have required hand curation (beyond the automated curation the UMLS offers) to be useable. That said, the general methodology used by the authors is largely terminology agnostic and can be used with any terminology that is reasonably hierarchical.

One group member was curious whether this process could be used for clinical decision support. Discussion revealed that this methodology is not suited for CDS, as the evidence would need to be clearer. For CDS use, the process would need to be based on drug labels or drug information systems, which are usually expressed at the individual drug level. It might be interesting to know if a class was affecting adverse events at the class level or based on a subset of the class, but this would not be the primary concern in a primary care setting. However, the results of the study could be valuable to drug safety professionals and those building adverse event repositories to support their review or decisions in regard to adverse event relationships.