Test case CohortMethod

Implementers
1

Hello,

Does anyone know a good set of target, comparator and outcome id to get the CohortMethod working on CMS SynPUF 1k?

To test the CohortMethod, I am using the SynPUF sample data. Just to try the functionality, I put in two frequent occurring drugs: Simvastatin as target drug and Lovastatin as comparator. As outcome condition, I made a cohort based on the conditions Diabetes and Essential Hypertension.
But when I run getDbCohortMethodData(), the following Warning is returned after Fetching the outcomes:

Warning messages:
1: In lowLevelQuerySql.ffdf(connection, sql) :
  Data has zero rows, returning an empty data frame
2: In lowLevelQuerySql.ffdf(connection, sql) :
  Data has zero rows, returning an empty data frame
3: In FeatureExtraction::getDbCovariateData(connection = connection,  :
  No data found

I call the Cohort Method with:

cmd <- getDbCohortMethodData(connectionDetails,
                             cdmDatabaseSchema = cdmDatabaseSchema,
                             targetId = 1539403, 
                             comparatorId = 1592085, 
                             outcomeIds = 3, # Created in cohort_test
                             outcomeDatabaseSchema = resultsDatabaseSchema,
                             outcomeTable = "cohort_test",
                             washoutPeriod = 0,
                             firstExposureOnly = FALSE,
                             removeDuplicateSubjects = FALSE,
                             excludeDrugsFromCovariates = FALSE,
                             covariateSettings = createCovariateSettings() )

I understand that this error is thrown because the combination of target, comparator and outcome does not give a suitable cohort. Does anyone have suggestions for a cohort or certain settings to adjust?

2

Hi, @MaximMoinat,

I looked at the Achilles reports for the synpuf database here:
http://www.ohdsi.org/web/achilles/#/CMS_SYNPUF_synthetic_data/conditions

And it looks like the two most prevalent things here are T2DM and EH.

I created a calypso feasibility to see if there were anyone who had their first diagnosis of T2DM before their first EH, and I found 157 people:among 672 people with T2DM.
http://www.ohdsi.org/web/calypso/#/2502
(Switch to the results tab and click the icon next to the Synpuf DB)

So if you want to run the cohort method looking for the outcome of EH among people with T2DM, make a cohort of people with their first diagnosis of T2DM (CoceptID: 201826 with descendants) and then make another cohort (for the outcome) of people with first diagnosis of EH (conceptID 320128) and cohort method should find people.

-Chris

3

Thanks for your answer. I will try this tomorrow.
Just to be sure, the target would be a cohort of T2DM and the outcome a cohort of EH? The comparator can be left empty?

In the end I would like to compare two drugs against a condition. So if anyone has a working parameter set for that, please share :).

4

Hi @MaximMoinat,
I’m not sure what targetID=1539403 represents (and the other IDs for comparatorID), but looking at the documentation here: https://raw.githubusercontent.com/OHDSI/CohortMethod/master/inst/doc/SingleStudies.pdf

Seems like you need to set up an exposureTable, an outcomeTable, and then pass in the IDs of the target, comparator and outcome.

So using the achilles report for identifying ingredient exposures in a population, we can take the top two drugs:
http://www.ohdsi.org/web/achilles/#/Demo_data_1K_synthetic_patients/drugeras

We can use T2DM as the outcome cohort, per my prior example.

Here’s a calypso feasiblity study that shows the population of T2DM that has 1) prior Simvastatinand no Lisinopril, and 2) prior Lisinopriland no Simvastatin. Based on the output of this calypso result, we see that people exist where the first diagnosis of T2DM preceeded by those with Simvastatin without Lisinopril, and those with Lisinopril without Simvastatin. These would make good target, comparator and outcome cohorts.

I’ve updated the calypso study:
http://www.ohdsi.org/web/calypso/#/2502

You’ll see here that among the 672 people with T2DM, we have 64 people treated with Lisinopril (and no Simvastatin) and 60 people treated with Simvastatin (and no Lisinopril ). It’s important that you don’t have people in both your target and comparator groups.

So, this should give you enough people to see something non-empty, but 64 and 60 people isn’t a lot of people (although we’re only talking a 1k person set anyways).

Have fun! You can look at the concept sets in calypso to see exactly what conceptIDs I used. And you can use CIRCE at http://www.ohdsi.org/web/circe/#/ to create the three distinct cohorts that you can use in CohortMethod. Let me know if you need help with that but you should try to solve this one for yourself, if only to get familiar with the tools. If you get stuck, I can lend a hand.

Hint: once you define a chort of 'first exposures to Lisinopril in CIRCE, you can click ‘show sql’ to get your database-specific dialect that will query the CDM for the people!

-Chris