OHDSI Home | Forums | Wiki | Github

[Patient Level Prediction Workgroup] Meetings

Thanks Kristin for the nice insights in the work you are doing at Deloitte.

The recording and slides are on the wiki for those that have missed it.

Peter

The PLP TC tomorrow is cancelled since I will be travelling at this time.

Peter

Hi All,

In the Eastern Hemisphere meeting tomorrow Seng Chan You would like to share some work he has done in the area of Deep Learning and wants to discuss integration in the PLP Package.

Peter

Hi All,

Tomorrow we like to talk about an idea we have to use prior knowledge (for example from WikiPedia) when training prediction models. Martijn will share some very preliminary results on this.

I would also like to use some time of this TC to hear if you have ideas for future research directions related to PLP or topics to discuss.

Peter

Hi @Rijnbeek
As you know, Ajou university has two kind of dataset, one from hospital, one from national claim database. All medical services are based on national claim service in Korea. (Virtually all condition, drug and procedure codes in hospital data are included in claim data)
Claim database contains longitudinal medical histories of large population. But it doesnā€™t have enough measurement data.
Conversely, hospital data contains only hospital-based visit data of small population. But it does have high-resolution measurement data.

In terms of transfer learning, would it be possible to make predicting model pre-trained in claim database and then transfer this knowledge to the predicting model for hospital data?

Chan

Hi Chan,

I have not looked in to transfer learning a lot myself but I do know Nigham is using it for phenotyping:
https://openreview.net/pdf?id=B1_E8xrKe

I am not sure how this would work for your interesting problem.

Peter

Dear All,

Because of the daylight savings time change we will move the Western Hemisphere TC today to one hour earlier for Europe to keep the TC at noon in US.

Today Martijn will present preliminary results on model building that utilized prior knowledge (informed prior).
We would also like to have an interactive discussion about future research directions for PLP after that.

Hope to talk to you later today.

Peter and Jenna

Dear All,

Jenna and I like to start with a new series of PLP TCs in Western and Eastern Hemisphere.

We have changed the frequency to a monthly meeting starting this Wednesday with the Western Hemisphere meeting and next week with the Eastern Hemisphere meeting.

This Wednesday we will discuss some new developments in the framework, and some ongoing work you may find interesting.

I will send a new agenda item with the gotomeeting details to those that have been in our calls last year, if you did not receive it by the end of today please send me an email or message on the forum. All information is also on our wiki page.

Looking forward to a nice new series of TCs!

Peter and Jenna

The recording and the slides of the Western Hemisphere meeting are available on the wiki.

We discussed the following items:

  • Update on codebase and Atlas
  • Smooth Calibration Curves
  • HF Prediction in T2DM Network Study
  • Network study on Learning Curves
  • PLP POC Network study
  • Other ongoing work

Thanks for the nice discussion.

Peter and Jenna

Dear all,

Tomorrow Martijn will give a presentation about ā€˜An OHDSI approach to risk factor identificationā€™.

Peter

Dear All,

We will move the Western Hemisphere meeting of today to next week.

In that TC Jenna will present on a validation study she worked on.

Peter

Dear Team,

Tomorrow we will have two PLP TCs :smile: See the wiki for connection details and time.

Eastern Hemisphere: Ross Williams (Erasmus) will give a quick update on this study on HF prediction in T2DM patients.

Western Hemisphere: Jenna will present on a new validation network study.

Peter

Dear Team,

Today we will have the Western Hemisphere TC in which we will be running the cohorts for the stroke model validation network study as discussed in the last TC.

If you have access to data please join.

Furthermore, we will discuss the new network study skeleton package for prediction.

Peter and Jenna.

@Rijnbeek, Could you provide protocol of the study?

Iā€™m a little bit cautious to predict stroke because claim record for stroke is not validated in Korea. (The false-positive proportion of stroke seems to be high in the claim database). Though, Iā€™m willing to join!

@SCYou we can discuss this on our call on Wednesday.

@Rijnbeek Ok! please~

Dear All,

In the Western Hemisphere PLP TC tomorrow my colleague Jan Kors will present a literature review he has been performing in prediction of Dementia within the IMI project ROADMAP.

https://roadmap-alzheimer.org

We are considering to perform a validation study of the available models (will be hard) but also possibly adding a model development step. The validation will be performed on the source databases/disease cohorts but in parallel it will be good to see if we can help out within the OHDSI context.

Hope you will join this discussion.

Peter

1 Like

@Rijnbeek
This topic really important especially in Korea, since Korea is the fastest aging society in the world.

If we can use some cohort with genetic profile (eg. UK biobank), this will be much easier to validate existing model and develop powerful model. Because mutation status of APOE4 is critical in dementia.
Still, I believe that we can predict dementia with only clinical information, because baseline medical status and cardio-cerebrovascular disease are closely related with development of dementia.

Hi @Rijnbeek, I am interested in contributing to the dementia prediction validation project. My PhD work was on neuroimaging biomarkers in Alzheimerā€™s disease as well as on prediction of dementia in patients with mild cognitive impairment using the Alzheimerā€™s Disease Neuroimaging Initiative dataset. Thanks!

Dear all,

Unfortunately, we have to cancel the Western Hemisphere TC tomorrow.

Peter and Jenna

t