Date
November 20th, 2020
Time
1:00pm - 1:45pm
Attendees
MITRE → Caroline PotteigerSteve Bratt Salim Semy Zach Lister Lauren Levine
BreastCancerTrials.org → Elly Cohen
UTSW → Melanie Hullings Shaalan Beg Brandi Cantarel
ACS CAN → Brittany A. McKelvey Devon Adams Mark Fleury
Action items
- Caroline Potteiger → cancel the team sync next week.
- Caroline Potteiger → set up small working group on Phase 1 analysis after Thanksgiving.
- Mark Fleury → look into IBM contacts and make an introduction.
Discussion notes in blue. Decisions in green. Action items in red.
Planned Agenda Topics
- <Hold for hot topics from project team>
- Welcome Brittany!
- Operating Committee meeting next week
- Past Action Items 11/13
- Thanksgiving next week. Reschedule? Cancel?
- Cancel next week.
- Engagement Update
- TrialJectory - initial kickoff on 11/12. Scheduling another for next week or the week after.
- Ciitizen - having internal discussions.
- TriNetX - having internal discussions
- No updates
- Inspirata - continuing discussions.
- Could work with UCSF dataset. Inspirata has an NLP engine that could work with unstructured data.
- Informa - having internal discussions.
- Phase 2 Site Engagement - met up this week to discuss a plan. Several touch points scheduled for the future.
- RXL CRO - no updates
- IBM Watson Health - CodeX had a few initial discussions, but it dropped off. Mark to look into IBM contacts and make an introduction.
- Phase 1 Update
- Working on IRB approvals for UTSW (Phase 1B) and MITRE/Cancer Insights (Phase 1A)
- Phase 1A -
- MITRE IRB has been re-submitted and DUA has been approved.
- Phase 1B -
- MITRE is working on the DUA.
- 3 main options - DUA with UTSW and CodeX, DUA with MITRE (and MITRE is able to re-distribute data), DUA with MITRE and each matching company.
- Will develop a model that is helpful for other CodeX use cases.
- MITRE legal and UTSW legal to discuss.
- Phase 1A -
- Massive Bio - in progress.
- Getting initial results, but they need some work. Troubleshooting now.
- TrialScope issue - labeling problem. Is getting fixed.
- BreastCancerTrials.org - still working on finding the problem.
- UCSF - consult held on 11/16.
- Didn't have many answers for legal/DUA/privacy questions, but Elly can access the data and share with us what's included.
- If the data is what we need, the next step is a security consultation, which will answer some of those legal/DUA/privacy questions.
- Elly's affiliate status allows her to access the tool.
- UCSF has 2 datasets (structured and unstructured). This tool allows us to look at the structured dataset. We would need a UCSF team to look at the unstructured data.
- Working on IRB approvals for UTSW (Phase 1B) and MITRE/Cancer Insights (Phase 1A)
- Phase 1 Analysis
- Plan for analysis?
- Potential elements to add - measurable disease (bone metastasis is not eligible), line of therapy
- Potential elements to remove - elements that may not be readily available (histology, biomarkers)
- If something is not present, how do we know it's because it's absent or it just wasn't recorded?
- Caroline meeting with spec team next week.
- Our analysis on how many "true matches" there are is going to depend on how many total matches come back
- 1 match in 5 = good filtering, 20 matches in 100 = bad filtering
- Patient data elements we add must be populated by both clinical trials and what's available from the patient record.
- False negatives are ones we really want to eliminate and make sure they don't happen.
- Mark put together a document for Phase 1 Goals - Trial Matching - Documents
How discriminatory are our 7 variables in matching patients when compared to a match on the full record?
- Can we drop some of 7 and have just as good of a match?
- May have to drop because of feasibility, but should understand hierarchy.
- Test all seven variables,
- Test each permutation of 6,
- Test permutations of 5 based on either variable-extraction feasibility or based on results from 6-variable permutations showing least important variables
- Should we add/replace variables to get a sufficient match?
- How does the match quality with these variables vary by cancer type?
- How does geographic restrictions play into overall number of trials returned (We may be able to filter on fewer clinical variables if geographic restrictions reduce trial count significantly)
- May have to drop because of feasibility, but should understand hierarchy.
- Can we drop some of 7 and have just as good of a match?
- Right now all trials are breast cancer, but we need to look at other cancers.
- Blood cancers as another cancer? Breast cancer falls under the solid cancer category.
- Mark does not know of any blood cancer matching services
- mCODE is currently tailored towards solid cancers, but CodeX's Registry Reporting use case is focused on blood cancer.
- https://www.lls.org/research/bringing-precision-medicine-to-aml-patients?ds_rl=1279598
- Complexity of treatment information may be different between cancer type.
- Other cancer type data could be taken from the UCSF data
- Blood cancers as another cancer? Breast cancer falls under the solid cancer category.
- We need to publish the data from this to be transparent.
- Wouldn't expect the same results from each matching service.
- In the future, we may want to consider some sort of threshold for matching services to become "mCODE-enabled".
- Semi-commercial matching services may compete against each other, but other matching services that are disease specific aren't as competitive.
- Put together a smaller group to work on this analysis plan.
- Brittany, Caroline, Salim
- Come up with a reasonable set of goals and matching services
- Goal is not to find all the knowledge, but make sure the next Phase will have good enough results
- Plan for analysis?