OHDSI Home | Forums | Wiki | Github

Accessing OHDSI network hospitals for Machine learning grant

I am using the iNYP New York Presbyterian OHDSI network hospital data to run a machine learning analysis identifying precision treatments for individuals with schizophrenia. I need to validate what I am finding in NY with a larger more diverse sample, and was hoping someone from another network hospital in the US would be interested to collaborating? I am hoping to work with three other OHDSI network hospitals since we all share a common language, search engine, etc.
I am located at the Psychiatric Institute at the Columbia University Irving Medical Center and am writing a grant to support this research.
Thanks so much for your time, and I hope to hear from you soon!

Hi @nbareis!

I have a research interest in getting more psychiatry studies through the OHDSI network… so let me just say how much I appreciate seeing you reach out to the community for this initiative! :slight_smile:

You know, this is actually the beauty of the OMOP CDM. We clean up the language for you. If you’re not restricted to US sites, @mvanzandt is ring-fencing a global psychiatry effort. She may have a few suggestions.

Overall, what kind of timeline are you working around? Happy to help you find some support throughout the OHDSI network.

Tagging a few friends to see this chain and see if they have interest/bandwidth: @benomark @Andrew @pbiondich @nigam @abel @mgurley

Yup, I am using the iNYP Explore ATLAS open source application. I know it includes not only OMOP but other sources as well. I think at this point I would need to restrict my study to US sites. This would be for an NIMH grant with a submission date in October, so it would be much more feasible to only include 2 or 3 other hospital networks. I will be conducting the data analysis myself, and already have a dataset of individuals in the iNYP system.

An Atlas with non-OMOP data? That’s a new one for me. Sounds intriguing. :slight_smile:

This is like any other network study. You’ll write the study protocol, build a study package to disseminate and test it on your data before sending to the 2-3 sites you identify in the OHDSI network. Are you planning to use @jreps and @Rijnbeek PatientLevelPrediction framework? It’s a good resource. Peter’s team at Erasmus is also a great resource for ML innovations in the OHDSI community.

Aside from 2-3 centers raising their hands to help, what do you need for your grant?

Sorry I haven’t responded yet. This is great! I also heard there will be a meeting the coming Monday (9/16) about setting up a psychiatry network. Did I understand that correctly?

Also, since I’m totally new to all of this, can you direct me to the instructions on how to make this like the other network studies? Do you have a specific form the protocol should take?

I was hoping to possibly request data from another OHDSI center. I know this would involve an agreement and likely IRB for human subjects concerns etc, but it would be really helpful to have the information from a location other than New York. I also know this might be impossible, but thought I would ask.

Finally, is there a place I can go which has the list of all hospital systems in the OHDSI network? It would be good to find a few places in different geographical locations if they would be willing to sign on.

Anyway, hopefully you get this soon.

You’re correct! @mvanzandt and team are coordinating a Face-to-Face workgroup meeting after the main Symposium. Details here: Psychiatry Working Group Creation - F2F at the US Symposium. Mui can forward you the dial-in if you are not going to be in person.

Yes, there’s a few chapters in the Book of OHDSI (Study Steps and Network Research) that can help provide some frame of reference to the overall process. Study Steps is the general “how I run a study at my institution”. Network Research talks about all the considerations unique to running a network study.

Protocols are an important part of the process. Here’s an example of a few well written protocols:

  1. The 2019 US Symposium Women of OHDSI Network Study (led by @MauraBeaton)
  2. Concept Prevalence (led by @aostropolets)

You’ll notice there’s different levels of detail. Your study is more similar to #1 but I share Anna’s just to show that it can be an adapted format.

One big nuance here: OHDSI network studies do not share patient-level data. In other words, patient level data from different sites is never pooled in a central environment. As sites, we map our data to the OMOP Common Data Model. When we run network studies, we create study code that are “OMOP compliant” – meaning they can run on any data mapped to the OMOP CDM that upholds standard OMOP conventions. Study code packages create results files designed to be aggregate results (e.g. summary statistics, point-estimates, diagnostic plots, etc.) and do not share patient-level information.

The good news is you’ve got a whole network of friends who regularly do this research!

When you write a study in the OHDSI framework, you will find many institutions have expedited reviews for these projects. As @Christian_Reich jokes, “The damage is already done to the person.” These are low-risk retrospective studies. The CDM is designed to be de-identified and all data are maintained locally which helps with many of the privacy rules you have to uphold. It will be at the discretion of each participating institution what other agreements are needed, some may require data use agreements but not all.

Not impossible!! You can get information from locations other than New York.

Here is the 2019 OHDSI Data Network List, freshly updated for the 2019 US Symposium on Monday.

If you would like to set-up time to discuss your study, I would be happy to find time. I’m at the US Symposium this weekend. How’s late next week (Thurs or Fri)? Send me an email (kristin.kostka@iqvia.com).

This is all so helpful! Let me get back to you after I digest all of this!