Convention needed: Management of triple diagnosis

[rtmill]

Community request:

1. Management of triple date of diagnosis in Cancer OMOP

Problem The diagnostic journey to receive a cancer diagnosis is complex and studied in detail in many registries, especially epidemiological. As a general rule, most treated cancer patients have

• A date of imaging diagnosis
• A date of pathological diagnosis
• Generally, one of either a date of clinical diagnosis or biochemical diagnosis (e.g. via PSA)
• Potentially, date of referral for investigation from primary care or date patient told the diagnosis or date of diagnosis via secondary findings
• An algorithmically determined date of diagnosis according to European Cancer Registry Network standards (ECRN) that uses all the above in a prioritised hierarchy

Flatfile based registries collect these as separate fields and find it hard to map them to OMOP given there is only 1 date of diagnosis field that isn’t visit_occurence driven. Users: registries and hospitals wanting interoperability with registry norms Ask / proposed solution:

Short-term: clearer guidance as to how to handle date of diagnosis in a uniform way, e.g. “if you have a ECRN date, populate date of DX using that, if not..” Medium-term: clarity on how to get better interoperability on date of diagnosis between flat file derived data and richer data sets, such that “from the hospital” gives the same results as “from the registry”. This may require CDM development

[cgreich]

I think we already have this in place:

1. The date is condition_start_date. Not sure we need another date.
2. The source of the diagnosis is what the type concepts tell us:

• Imaging diagnosis: EHR radiology report, but we could change the name to something like "Imaging report"
• Pathological diagnosis: EHR Pathology report. We could omit the "EHR", to make it more generic.
• Clinical diagnosis: Any diagnosis not explicitly coming from another department is a clinical diagnosis. We should nail this in a Themis convention.
• Biochemical diagnosis: Lab, again, probably to be renamed "Lab report". You could say we shouldn't blame the lab for reporting a diagnosis, but if this is in the CONDITION_OCCURRENCE record the distinction may be philosophical, and we could nail it through Themis.
• Patient told: Patient self-report.
• Algorithmically determined: Standard algorithm. We could think of doing children of standard algorithms, but I would limit those only to generally accepted ones.

[golozara]

@odikia , @rtmill to add the types needed

[rtmill]

A relevant snippet from the docs regarding "initial diagnosis date":

https://ohdsi.github.io/OncologyWG/conventions.html#Date_of_Initial_Diagnosis

[odikia]

Thanks @golozara and @rtmill!

@rtmill, once we get close to closing this task, I’d like to update the convention documentation to reflect the use of “…_type _concept _id”. Adding as a blocked task? Do I understand that use of the project 2.0 nomenclature correctly?

[golozara]

Rimma will present her assessment on the discrepancies between CR and EHR.

[jam560]

I'm concerned about "hiding" the semantics of cancer initial diagnosis date in the type_concept_id, especially if we think CDMs might include very low quality sources like "patient reported". This data element is critical to many core oncology questions like incidence trends, time to treatment, time to progression, overall survival, etc. I've never seen an OMOP phenotype (in a demo or presentation) that pays attention to type_concept_id and I worry that relying on phenotypers to be aware that quality differences exist across type_concept_id is unwise. I admit that's an overall risk in OMOP, but this particular concept is both critical and complex. I suggest we should make the complexity apparent to the phenotyper so they have to think about it.

Having said that, this can be addressed with something like Kyle's OncVal proposal where the oncology data quality assessment tool would be aware of, and would surface, the varying source quality.

[jam560]

I've attached the SEER and ENCR Date of Diagnosis algorithms for reference. NAACCR uses the SEER algorithm. It would be interesting to find cancer registry standards for other parts of the world. I tried to find standards on the Australasian Association of Cancer Registries site but cannot.

ENCR Recommendation Dare of Incidence_Mar2022.pdf SEER Coding 2023 - Date of Diagnosis.pdf