Team:
Congratulations to everyone for a very impressive representation of OHDSI at this week’s AMIA Joint Summits.
The buzz about OHDSI was palpable. Here’s only a few of the highlights that I saw or heard comments about…and I’m sure there was more and encourage you all to share your reflections and a-ha’s on this discussion thread.
On Tuesday, Daniella, Rimma, Michael Matheny, and Michael Kahn did a great job sharing their experiences of applying the OMOP CDM within their PCORNet efforts in pSCANNER, NYC, VA, and PedsNet, respectively. My main take-away: everyone shared that ETL to the OMOP CDM is challenging and time-consuming, but ultimately do-able at a high level of fidelity. Community support and open-source tools greatly facilitated the success of their work. Further efforts to make the ETL design and implementation process more efficient, transparent, and reliable are likely going to need to a priority if we hope to build the OHDSI community further. Kudos to Michael Kahn for being the only presenter at the conference to work a picture of a baby’s bottom and the word HELL into his scholarly presentation:)
On Wednesday, David Vawdrey, @hripcsa Jon and @nigam did a great job formally introducing OHDSI to the AMIA community, sharing our mission to generate reliable scientific evidence in the domains of clinical characterization, population-level estimation, and patient-level estimation. I was able to highlight the OHDSI methods library that @schuemie and @msuchard have been leading the development of. The room was packed, and there were a lot of great questions that pointed to various new directions that the community could expand into.
@Chris_Knoll unveiled CIRCE and CALYPSO, providing a live demo of how OHDSI’s open-source cohort definition tool can be successfully wrapped into an analytic application that facilitates the use case of clinical trial feasibility assessment. The feature set is truly impressive: the power of being able to create and edit a complex collection of inclusion criteria across all domains and attributes within the OMOP common data model, and then compile that criteria set into a platform-independent SQL script that will efficiently populate a Cohort table structure opens the door to many future analysis opportunities that were previously inconceivable.
@aguynamedryan gave a great talk showing how the tools OutcomesInsights have developed can be applied to an OMOP common data model. It’ll be great to see all of these tools running publicly live on the CMS SynPUF data that’ll be coming out of Ryan and Mark’s workgroup shortly.
Gal showed amazing progress on how inclusion criteria systematically extracted from studies in clinicaltrials.gov can begin to be automatically mapped into expressions that can become executable against the CDM.
Ending the day on a high note, Mary was awarded the TBI best student paper for her work, “Are All Vaccines Created Equal? Using Electronic Health Records to Discover Vaccines Associated With Clinician-coded Adverse Events”. Congratulations Mary!
Thursday was another impressive day within the OHDSI community. Jon blew my mind with his live demo of HERACLES, and Regenstrief’s seamless integration of NLP-based patient list extraction with the OHDSI cohort characterization tool. Who would have known patients with chili dogs mentioned in their records would also have abnormal heart failure measurements? HERACLES is absolutely going to be a game-changer for our community, it really takes the notion of ‘case discovery’ to a whole new level by complementing patient identification with a population summarization.
@rkboyce prepared and delivered a terrific talk on behalf of the LAERTES workgroup. @Vojtech_Huser had led the development of the submission on LAERTES, and I’m so glad he did, although we all missed seeing him at the meeting. The progress on LAERTES is truly remarkable. As we prepared the demo for the presentation, we applied the current evidencebase to explore potential adverse events associated with lisinopril: with a couple simple queries, we were able to pull a wealth of information about angioedema from clinical trials, case reports, and structured product labeling, and when we applied the tool to aplastic anemia, only to reveal that the tool automatically extracted ‘rare cases of bone marrow depression’ automatically from the SPLs, outsmarting me. We rattled off a dozen highly publishable works that should come out of LAERTES now that the infrastructure is in place…I encourage you all to think about how having systematic evidence of all drugs and all conditions could help you reframe and contextualize your research. LAERTES will soon be one of your best friends, right beside HERMES in making past knowledge a driver to your future insights.
Zhe gave a really nice presentation on how the next frontier of clinical trial generalizability is here, showing how the GIST of clinical trial inclusion criteria can be assessed through standard population summaries from observational data. The method can be readily applied via CALYPSO to provide a formal quantitative summary of population comparisons. I see a lot of potential for this path of work, and some very obvious connections to various pieces of work across the OHDSI community.
Vibhu crystalized the importance of the phenotype work that Nigam and Juan shared with us at the OHDSI F2F earlier in the week, and the value that is going to come from developing out XPRESS/APHRODITE into a scalable community process. His mathematical proof about the information content in noisy classifiers is something that we all should be paying close attention to. He showed that, as long as you can come up with some proxy for a positive control that’s reasonably accurate, then an automated prediction approach can be shown to converge to equal (or sometimes better) operating characteristics from doing the old-school, manual, time-consuming, resource-intensive chart review process. This provides the first glimmer of hope that defining a universe of all health outcomes of interest is an achievable goal.
Rimma, Michael Kahn, and Chunhua presented in a terrific session on data quality. Under Michael’s leadership over the past couple years, we’ve seen tremendous progress on establishing frameworks for evaluating data quality and for starting the process of harmonizing data quality tools across data systems. Tiffany followed up this work with a nice poster showing the diversity of perspectives and tools being applied across various institutions…data quality remains an Achilles Heel for all of us, but at least we have Achilles Heel to help us sort it out:) This work is so fundamental to everything we do, but there’s still a lot we need to do to establish and apply best practices…perhaps a DataQualityPalooza is in order?
On Friday, Parsa and Christian represented the OMOP CDM in the ‘ontology tug-of-war’, and shared a hopeful vision for interoperability between the models that are actively in use across the various communities. ATHENA made her debut in this session, reinforcing the point that a common data model is as much about standardized content and semantics as it is about its structure. A key theme is that the analytic use case, not the underlying infrastructure, should be driving our focus, and much of the debate about the ‘best model’ will die away once people re-focus on using the CDM with the entire OHDSI toolkit to generate scientific evidence to meet their clinical needs.
A shout-out to @toanong , who has made an important step toward interoperability, by making an ETL from OMOP CDM to PCORNet publicly available on the OHDSI github: https://github.com/OHDSI/OMOPV4_PCORNetV1_ETL. Toan and Michael said they’d be updating this with their more recent work in the future. I hope other PCORI CRDNs can follow suit in sharing their code, and hopefully as a community, we can coalesce on one recommended solution so that other OHDSI members who may be considering joining PCORI could do so in an efficient and consistent manner. I know there’s been other threads about ETLs from i2b2->OMOP, OMOP->i2b2, PCORNet/MIniSentinel->OMOP,…for those who’ve gone through the pain of building these solutions, it’d be wonderful if you can share with the OHDSI community so nobody has to reinvent that wheel…
Another shoutout to Hua, who’s JAMIA paper on metformin repurposing was so highly regarded that it was the only publication called out in both the TBI and and CRI year-in-review. I’d love to see the OHDSI community, under Hua’s leadership, do a replication of these findings across our international network. After re-reading Hua’s paper, it seems a great candidate to also demonstrate the power of how CIRCE, combined with HERACLES and then CohortMethod, can provide an end-to-end analytical process that most of us have only dreamed of.
Last, but most importantly, I want to congratulate Chunhua for chairing an amazing Clinical Research Informatics conference. The meeting was absolutely terrific, I learned a lot from the sessions and posters, got tons of new ideas for collaborative research and development opportunities that I’ll be raising with the community.
For me, this meeting was exceptionally inspiring and energizing. The collaborative spirit and shared commitment to scientific evidence generation within OHDSI was front-and-center in all our work, and I heard a lot of excitement from others interesting in joining the journey with us.
Cheers, Big Ears!
Patrick