Observational Health Data Sciences and Informatics

[4/19/17, 12:13:10 PM] Eric Prud'hommeaux: notes:

richard: presenting on Tue at the OHDSI call

[4/19/17, 12:13:20 PM] Eric Prud'hommeaux: … will get into F2F and projects

[4/19/17, 12:13:33 PM] Eric Prud'hommeaux: … right now we have two versions of OMOP-on-FHIR

[4/19/17, 12:13:50 PM] Eric Prud'hommeaux: … 1st was built ih-house – we can go back and forth to FHIR

[4/19/17, 12:14:08 PM] Eric Prud'hommeaux: … 2nd version is more widely applicable

[4/19/17, 12:14:40 PM] Eric Prud'hommeaux: … use cases: if you have an OMOP repo, want to drop in a FHIR server in front

[4/19/17, 12:14:47 PM] Eric Prud'hommeaux: … allows you to use FHIR apps

[4/19/17, 12:15:26 PM] Eric Prud'hommeaux: … 2nd: lots of interest in using FHIR to map from FHIR-enabled EHR to OMOP

[4/19/17, 12:15:45 PM] Eric Prud'hommeaux: … we built some static staging tables and procedures wrapped around them.

[4/19/17, 12:15:51 PM] Eric Prud'hommeaux: … we want it to work with CSV

[4/19/17, 12:16:14 PM] Eric Prud'hommeaux: … since we don't ahve a med school, we work with external data and have to be very flexible

[4/19/17, 12:16:33 PM] Eric Prud'hommeaux: … we don't have MBI functions so we're relying on source value for uniquenless

[4/19/17, 12:17:08 PM] Eric Prud'hommeaux: … of course for each incoming feed, you need a custom uniqueness system (if yo udon't have an MPI)

[4/19/17, 12:17:27 PM] Eric Prud'hommeaux: … the main prob we have is mapping the FHIR Observation to multiple tables

[4/19/17, 12:17:42 PM] Eric Prud'hommeaux: … also had to extend omop to include demographics data

[4/19/17, 12:18:12 PM] Eric Prud'hommeaux: … we have standardized rights and logic for pushing data into the tables

[4/19/17, 12:18:36 PM] Eric Prud'hommeaux: … we need to handle real-time load from individual fhir sources that happne in real life

[4/19/17, 12:18:59 PM] Eric Prud'hommeaux: … from my perspective, we see lots of interest in leveraging for FHIR mapping to OMOP

[4/19/17, 12:19:22 PM] Eric Prud'hommeaux: … i didn't persue any new FHIR apps (sitting on top of an existing CDM)

[4/19/17, 12:19:46 PM] Eric Prud'hommeaux: … @@1 probably has a different POV 'cause he's involved in FHIR, not OMOP

[4/19/17, 12:20:09 PM] Eric Prud'hommeaux: @@1: we developed a system with three componentns:

[4/19/17, 12:20:19 PM] Eric Prud'hommeaux: … .. web app for visualizing data

[4/19/17, 12:20:25 PM] Eric Prud'hommeaux: … .. emr for pulling data

[4/19/17, 12:20:39 PM] Eric Prud'hommeaux: … .. web api (developed by another group)

[4/19/17, 12:20:49 PM] Eric Prud'hommeaux: … we used those for the F2F meeting

[4/19/17, 12:21:31 PM] Eric Prud'hommeaux: … assuming your EHR is FHIR-enabled, the GT-FHIR server has OMOP V5 has a backend DB

[4/19/17, 12:21:48 PM] Eric Prud'hommeaux: … we have patients, medications, … to simulate the EHR

[4/19/17, 12:22:20 PM] Eric Prud'hommeaux: … i used a FHIR client to POST new data (medications) to the OMOP DB

[4/19/17, 12:22:36 PM] Eric Prud'hommeaux: … so it works as a FHIR-to-OMOP

[4/19/17, 12:23:04 PM] Eric Prud'hommeaux: … i can create lots of tables in OMOP (practitioner)

[4/19/17, 12:23:23 PM] Eric Prud'hommeaux: … i used all that to populate OMOP V5 and used the loaded server

[4/19/17, 12:23:56 PM] Eric Prud'hommeaux: richard: we were really only involved in the dissemination group

[4/19/17, 12:24:34 PM] Eric Prud'hommeaux: cali: Q re FHIR impl on OMOP: i understand that your FHIR server can read one patient data at a time?

[4/19/17, 12:24:44 PM] Eric Prud'hommeaux: … can it write to the OMOP db

[4/19/17, 12:25:14 PM] Eric Prud'hommeaux: @@1: you can read by patient or search [across patients]

[4/19/17, 12:25:22 PM] Eric Prud'hommeaux: … we write one resource at a time

[4/19/17, 12:25:29 PM] Eric Prud'hommeaux: … thinking about writing a bundle

[4/19/17, 12:26:10 PM] Eric Prud'hommeaux: richard: assuming existing patient

[4/19/17, 12:26:19 PM] Eric Prud'hommeaux: … otherwise we get NULLs in the CDM

[4/19/17, 12:27:04 PM] Eric Prud'hommeaux: ericP: limited to one by the FHIR API?

[4/19/17, 12:27:17 PM] Eric Prud'hommeaux: @@1: think we can write bundles

[4/19/17, 12:27:35 PM] Eric Prud'hommeaux: richard: we've talked about developing a cohort, which would be a huge packet

[4/19/17, 12:28:17 PM] Eric Prud'hommeaux: guoqian: when you load FHIR into OMOP, you created additional tables (beyond CDM)

[4/19/17, 12:28:40 PM] Eric Prud'hommeaux: richard: we created F-Person (demographcics, address, contact info)

[4/19/17, 12:28:52 PM] Eric Prud'hommeaux: @@1: OMOP has a location table as well

[4/19/17, 12:29:30 PM] Eric Prud'hommeaux: … so write Person/location info to the [CDM] IMOP location table

[4/19/17, 12:29:50 PM] Eric Prud'hommeaux: … if someone writes “street” and another writes “st”, that's a differetn addr

[4/19/17, 12:29:58 PM] Eric Prud'hommeaux: richard: that will come from the source value table

[4/19/17, 12:30:42 PM] Eric Prud'hommeaux: @@1: if the FHIR resource has a reference, we can use that ID in the source column

[4/19/17, 12:30:59 PM] Eric Prud'hommeaux: … but if the address is in the FHIR, we can't merge it against existing location records

[4/19/17, 12:31:13 PM] Eric Prud'hommeaux: richard: we can use USPS services for that

[4/19/17, 12:31:30 PM] Eric Prud'hommeaux: guoqian: so the purpose is to capture all FHIR information?

[4/19/17, 12:31:38 PM] Eric Prud'hommeaux: richard: i think it's useful info.

[4/19/17, 12:31:49 PM] Eric Prud'hommeaux: … the search might include names.

[4/19/17, 12:32:03 PM] Eric Prud'hommeaux: … (OMOP is considered de-identified)

[4/19/17, 12:32:17 PM] Eric Prud'hommeaux: … but we're trying to support other tools beyond OHDSI

[4/19/17, 12:32:37 PM] Eric Prud'hommeaux: @@1: we can use the Person ID has a name field

[4/19/17, 12:33:01 PM] Eric Prud'hommeaux: … those fields are not required in FHIR so nothing will break if they're not there.

[4/19/17, 12:33:12 PM] Eric Prud'hommeaux: … but there are tools that use them for [disambiugation?]

[4/19/17, 12:33:50 PM] Eric Prud'hommeaux: … if we can come up with a coding mapping, that would help people outside get interop

[4/19/17, 12:34:23 PM] Eric Prud'hommeaux: guoqian: from hwat i understand, OHDSI has its one concept ID. do you transformation or allignment?

[4/19/17, 12:34:37 PM] Eric Prud'hommeaux: richard: we use generic ETL procedures to do the translation

[4/19/17, 12:35:06 PM] Eric Prud'hommeaux: … we've talked about serveral optionns for what to do when we want to e.g. back-translate from CDM to ICD-9/10

[4/19/17, 12:35:29 PM] Eric Prud'hommeaux: @@1: our test env uses @@2

[4/19/17, 12:35:43 PM] Eric Prud'hommeaux: … the Vista EMR sends huge lists of different codings

[4/19/17, 12:35:58 PM] Eric Prud'hommeaux: … e.g. drugs: RxNorm/NDC

[4/19/17, 12:36:12 PM] Eric Prud'hommeaux: … so in our FHIR, we picked what we wanted

[4/19/17, 12:36:33 PM] Eric Prud'hommeaux: richard: we can still do those translations firom …

[4/19/17, 12:36:51 PM] Eric Prud'hommeaux: … the VA way is good. if you put them all in, the user can pick their fave

[4/19/17, 12:37:43 PM] Eric Prud'hommeaux: … the prob we've had (last week): what if you don't map to the source concept, e.g. an item ID from MIMIC? do yo give them free text?

[4/19/17, 12:38:01 PM] Eric Prud'hommeaux: guoqian:: need further discussion on two points:

[4/19/17, 12:38:20 PM] Eric Prud'hommeaux: … .. 1 collab on standard mappings between HFIR resources and OHDSI CDM

[4/19/17, 12:38:40 PM] Eric Prud'hommeaux: … .. 2 if we want to load FHIR into OHDSI CDM, what extesnions do we need?

[4/19/17, 12:39:14 PM] Eric Prud'hommeaux: richard: we tried our best to do it in a new table so it wouldn't affect existing processes

[4/19/17, 12:39:36 PM] Eric Prud'hommeaux: guoqian: at Mayo, we used NoSQL tech to store extension data

[4/19/17, 12:39:47 PM] Eric Prud'hommeaux: richard: we're working with a Hadoop group

[4/19/17, 12:39:57 PM] Eric Prud'hommeaux: … but that's OMOP into Hadooop

[4/19/17, 12:40:55 PM] Eric Prud'hommeaux: ericP: should we start with yours?

[4/19/17, 12:41:17 PM] Eric Prud'hommeaux: richard: my work needs OMOP review to consider my assumptions

[4/19/17, 12:41:32 PM] Eric Prud'hommeaux: … we're using staging tables.

[4/19/17, 12:41:45 PM] Eric Prud'hommeaux: … i worked with different file structures from different providers

[4/19/17, 12:42:11 PM] Eric Prud'hommeaux: … so i tried to create standard staging tables for what i'd seen from differnent EHRs/claims systems

[4/19/17, 12:42:40 PM] Eric Prud'hommeaux: … so i think i've got the propor CDM process for them, but with the vocab changing and the mapping changing, i don't know if it's correct

[4/19/17, 12:43:17 PM] Eric Prud'hommeaux: ericP: tests cases?

[4/19/17, 12:43:28 PM] Eric Prud'hommeaux: richard: we use the MIMIC dataset for a lot of stuff

[4/19/17, 12:43:45 PM] Eric Prud'hommeaux: … that would allow us to see affects

[4/19/17, 12:43:59 PM] Eric Prud'hommeaux: … i'm not sure there's a consensus in the community about the proper mapping

[4/19/17, 12:44:17 PM] Eric Prud'hommeaux: … at the DC meeting, people were learning lots with each question

[4/19/17, 12:44:59 PM] Eric Prud'hommeaux: @@1: from my (FHIR) side, there's too much flexibility and extensibility. breaks interop

[4/19/17, 12:45:09 PM] Eric Prud'hommeaux: … we we still have to map FHIR to the staging table

[4/19/17, 12:45:41 PM] Eric Prud'hommeaux: … the OMOP side needs the right staging table and we need the right strategy table to map from FHIR to the staging table

[4/19/17, 12:45:55 PM] Eric Prud'hommeaux: … we're more confortable working with the staging table

[4/19/17, 12:46:48 PM] Eric Prud'hommeaux: … so folks in the field, testing, clinical data, etc, it would help to eval our strategy

[4/19/17, 12:54:25 PM] Eric Prud'hommeaux: @@3: Laurrie Phillips (dev) linked I2B2 to OMOP

[4/19/17, 12:54:36 PM] Eric Prud'hommeaux: … I2B2 is a star schema which can adapt to any data model

[4/19/17, 12:54:59 PM] Eric Prud'hommeaux: … deviated a bit from that to use the I2B2 cell to talk to an existing OMOP database

[4/19/17, 12:55:09 PM] Eric Prud'hommeaux: ,,, we just put out a releas to test it out

[4/19/17, 12:55:20 PM] Eric Prud'hommeaux: … i've been working on the FHIR API over I2B2

[4/19/17, 12:55:35 PM] Eric Prud'hommeaux: … we were tied to the default I2B2 model

[4/19/17, 12:55:49 PM] Eric Prud'hommeaux: … the focus was on running SmART apps on I2B2

[4/19/17, 12:56:05 PM] Eric Prud'hommeaux: … we're adapting that to the PicorNet data model

[4/19/17, 12:56:29 PM] Eric Prud'hommeaux: … limitation: reading, not writing

[4/19/17, 12:56:36 PM] Eric Prud'hommeaux: … … no population queries

[4/19/17, 12:57:04 PM] Eric Prud'hommeaux: guoqian: we can consider yours an indeirect map (I2B2-FHIR and I2B2-OMOP)

[4/19/17, 12:58:41 PM] Eric Prud'hommeaux: ericP: minimize imagination

[4/19/17, 12:58:55 PM] Eric Prud'hommeaux: guoqian: aggregate approaches and then review details

[4/19/17, 12:59:05 PM] Eric Prud'hommeaux: @@3: what does mapping mean?

[4/19/17, 12:59:18 PM] Eric Prud'hommeaux: … .. you can map to a Picornet, FHIR element

[4/19/17, 12:59:32 PM] Eric Prud'hommeaux: … .. but you can also translate the codes in the element

[4/19/17, 1:00:53 PM] Eric Prud'hommeaux: richard: we had translations to CDM

[4/19/17, 1:01:18 PM] Eric Prud'hommeaux: … i had problems finding terminology mappings

[4/19/17, 1:02:33 PM] Eric Prud'hommeaux: ericP: structural vs. open term mappings

[4/19/17, 1:02:53 PM] Eric Prud'hommeaux: @@1: right, FHIR focuses on clinical workflow

[4/19/17, 1:03:13 PM] Eric Prud'hommeaux: … my decision to ask if you'd use it in a workflow