OHDSI Home | Forums | Wiki | Github

**Searching for collaborators to validate APHRODITE phenotyping models**

Researchers
#1

Searching for collaborators to validate APHRODITE phenotyping models

Hi everyone!

I’m posting to gauge interest in collaborating on a phenotyping project that was presented at the OHDSI symposium last October. We have developed phenotyping classifiers for a range of conditions using the APHRODITE framework (T2DM, PVD, heart failure etc). To assess the portability of our model-building pipeline across sites, we are looking for collaborators who can run our models on their own OMOP databases.

If anyone has a manually labeled cohort of a particular condition (e.g. a population of known diabetics) – that would be ideal. However, even in the absence of manual labeling, it would be useful to see how the model performs against PheKB definitions which we can provide.

Please let me know if you’re interested in collaborating with us! The scripts are written, so it shouldn’t be a heavy lift for anyone who wants to get involved. We’d love to get your help in testing our phenotyping models – and if you’re not able to directly participate, we’d still appreciate your suggestions!

#2

On Fri, Feb 2, 2018 at 7:00 AM Mehr Kashyap noreply@ohdsi.org wrote:

#3

Hi - very interested but not sure if our dataset, which crosswalks
nursing home and hospital settings for a ~5K patients, will work. Could
you expand on the minimum requirements of an OMOP dataset (e.g., number
of patients, number/type of clinical notes, completeness of various
data, necessity of certain data types, and versions of vocab/CDM) ? ty. -R

#4

Hi @mehrkashyap, Columbia would love to help! Please let me know how best to engage. Looking forward to it!

#5

Hi, Seoul National University Bundang Hospital (SNUBH) would also love to help. We recently converted our EHR data to CDM.

#6

Thanks for your interest @rkboyce, @thomasfalconer, and @Sooyoung_Yoo. I’ll be reaching out to you shortly!

The APHRODITE classifiers require OMOP CDMv5 and the following tables: concept, concept_relationship, person, condition_occurrence, drug_exposure, procedure_occurrence, observation, visit_occurrence, and measurement. We’re not currently using any clinical notes in the classification task, and we should be able to work with a variety of dataset sizes/# of patients.

If anyone else would like to get involved, please don’t hesitate to reach out!

#7

@mehrkashyap
Nice job.
Sorry I forgot to reply to this thread.

Ajou university in Korea wants to join, too.
For DM patients, we can extract levels of HbA1c and to validate the definition.
For HF patients, we can also extract value of ejection fraction from echocardiography to validate definition (if the definition of heart failure is restricted to 'HF with reduced Ejection Fraction).

1 Like
t