Stream: implementers
Topic: Locating clinical trials
David Hay (Jan 15 2017 at 22:49):
Question from a UK colleague:
The use case is for a clinical trials web service that will be able to return a list of trials that a patient may be suitable for based on certain clinical and demographic patient information.
For example:
Patient with a diagnosis of Stage 4, oestrogen receptor negative, adenocarcinoma of the left breast. Patient is 48yrs old and is also on Aspirin and has previous surgery - breast augmentation. She lives at W1T 4EU. What trials are open within a 25 mile radius?
Would this be a ServiceDefinition?
Geoff Low (Jan 15 2017 at 23:05):
I talked with Lloyd about this use case; as it's one of the most interesting cases for the Pharma industry.
I asked w.r.t. the ResearchStudy resource put together for the SDC project. Josh discussed trying to align the ResearchStudy resource with the representation used by ClinicalTrials.gov; ct.gov uses discrete terms for condition and intervention; the ResearchStudy bundles all of the above under the focus attribute (my level of understanding is still not strong enough to establish that it would be flexible enough).
Is your UK colleague in a position to discuss this with me? I'm interested to know who would be the broker for this information as it's multi-modal; sponsors maintain lists of studies they are running (often on their websites), CROs have a service they offer to sponsors to recruit subjects (through research networks, advertising, etc) and technology vendors like ourselves might consider it as a value add for our customers.
Lloyd McKenzie (Jan 16 2017 at 00:44):
Presuming that the codes used for "conditions" vs. "medications" vs. other things were discrete enough, that might be sufficient. But I think for this use-case, you wouldn't want ClinicalStudy.focus. All that would say is "this study focuses on breast cancer and drug XYZ". It might help to be able to find "breast cancer-related studies", but you'd then need to drill down to the Group referenced by the study that defines the eligibility criteria and evaluate the patient against those criteria.
David Hay (Jan 16 2017 at 02:21):
you mean ResearchStudy I presume? (I missed that :) )
So - if set up correctly, ResearchStudy.enrollment would refer to the Group that has the Group.characteristic's that qualify for inclusion in the group?
Lloyd McKenzie (Jan 16 2017 at 06:07):
Yes
Wai Keong Wong (Jan 16 2017 at 18:32):
This is very helpful. I was not aware the ResearchStudy resources was available. @David Hay I don't seem to be able to find ResearchStudy.enrollment
Looking at the Group Resource, I can see how it may support something like age perhaps not something more complex. For e.g. I struggle to understand how you specify something like 'Suspected Relapsed Breast Cancer'.
Lloyd McKenzie (Jan 16 2017 at 18:42):
Suspected Relapsed Breast Cancer *should* be expressible as a SNOMED code, so you'd have characteristic.code of "diagnosis" and characteristic.code.value of suspected relapsed breast cancer. The trickier part would be how many systems could match that against their patient's data. The fall-back would be human review against the criteria the computer doesn't understand after verifying membership against age, gender and other criteria that are easier to match on.
Wai Keong Wong (Jan 16 2017 at 18:52):
Yes, I am trying to avoid using post-coordinated SNOMED codes as I don't think many current EPRs use them. However the FHIR Condition profile does give one the option of specifying condition.clinicalStatus and condition.verificationStatus. However I cannot see how to use that within this Group resource
Lloyd McKenzie (Jan 17 2017 at 01:28):
You could define extensions if you wanted to, but I'm not sure that would help your interoperability more than post-coordinated SNOMED or text. You could propose allowing group criteria to nest to allow coordinating criteria, but it's not clear that level of sophistication would fall within the 80%. I think specifying high-level computable criteria that significantly narrow down the possible trials and then a few detailed non-computable criteria is probably more realistic for what we could get most systems to accomplish. That said, extensions are definitely an option within a controlled community.
Geoff Low (Jan 20 2017 at 13:12):
I'd agree for the moment to just focus on the condition for the MVP. It leads back to whom you'd expect to serve as the source for these clinical trial resources. Both @Pascal Pfiffner and @Josh Mandel had some great observations on the resources themselves and Josh's observation that there would be a benefit to alignment with the 'resource' representation in clinicaltrials.gov registry; we should also try to cover EUDRACT and WHO as well - luckily CDISC has completed some harmonisation work as part of https://www.cdisc.org/standards/foundational/ctr-xml . I'm an outsider, still learning about knowledge representation in healthcare (Biopharma is curiously prescriptive) so I'm not going to be a leader in this, just a very interested party! Just for those browsers, note that there is the #biopharma stream that could stand to be a lot more active!
Pascal Pfiffner (Jan 20 2017 at 13:32):
Interesting, the use-case you describe, @David Hay , is exactly the scenario we were addressing in a pilot 2-3 years back. One main issue is of course to get structured eligibility and location data for clinical trials; at least geodata is straightforward. The other main issue then is the logic when comparing eligibility criteria to data obtained about the patient, as was already brought up: SNOMED post-coordination vs. related resources (like a mutation Observation on a Condition). There's apparently a reason people invented things like Arden Syntax. ;)
Has somebody started to model a couple of actual studies as ResearchStudy and a handful of potential and not-so-potential participants as patients? We do have 100 synthetic but realistic breast cancer patients (they are in the SMART sandbox) and I have access to a dozen or so studies in the breast cancer realm that we could model and test.
Geoff Low (Jan 23 2017 at 17:57):
Seems to me that it would be an interesting project to model a transformer of the ct.gov xml to FHIR. This format is going to be more interesting as time goes on, given the final ruling (that results _must_ be reported to ct.gov) as we could use FHIR as an exchange format for results of clinical studies. This would be really interesting for enlightened doctors who would like to see the results of the clinical studies that contributed to the marketing approval for a interventional product, assuming that fhir is being used for Physician portals already in play. Again, it would be advantageous if there is reasonable concept alignment to the CT.gov model. Just my thoughts.
Last updated: Apr 12 2022 at 19:14 UTC
