OHDSI Home | Forums | Wiki | Github

How do people model 'relapsed cancers'?

Hi. I’m working on an analysis for acute leukemia in relapse and there isn’t a SNOMED concept for ‘acute lymphoid leukemia in relapse’, but even if there was, I don’t know if we can necessarily rely on specific ICD9/10 codes to identify it. Instead, a person who has ALL then has some gap period without treatment then had another diagnosis with treatment, that would seem to be a clear suggestion of relapse, even without the ‘relapse’ code. Does anyone have insight/experience into how best to capture relapse cancer?

3 Likes

Hi, @Jill_Hardin.

I had some experience in modeling cohorts for patients with lymphoma and lung cancer and faced with the same problem.
We also used approach based on some clear period to determine relapse/recurrence. Also notice, that in some cases change of chemotherapy regimen (for ex.adding new drug, starting some procedures etc. ) can be considered as demonstration of relapse despite the absence of gap. Therefore, to define precise gap and capture all signs of relapse you should research guidelines for treatment of acute leukemia.
P.S.: Let me predict the next stumbling-stone of your analysis - how to define death if such data is not present because of privacy reasons

Hi @Eldar. Thanks for the reply, we will investigate your suggestions/approach!

@Jill_Hardin, @Eldar: Please come to the Oncology Working Group and bring it on!! @rimma runs it.

What you describe is something the CONDITION_ERA table could possibly capture but currently does not.

You want to say that condition or condition_era C2 is related to prior condition C1. (fact relationship table comes to mind (as it does in many situations)

@Vojtech_Huser:

Come in as well.

@Jill_Hardin @Eldar @Christian_Reich @Vojtech_Huser One way to potentially surface cancer relapse/recurrence from EHR data is to look for condition_era of a cancer diagnosis followed by a condition era of a history of cancer diagnosis followed by a condition_era of a cancer diagnosis. Z85 and its descendants in ICD10 are supposed to be assigned to patients whose cancer is no longer being treated and no longer exists. See http://www.icd10data.com/ICD10CM/Codes/Z00-Z99/Z77-Z99/Z85-. I agree with @Vojtech_Huser that some kind of referencing between condition_era entries would make explicit relapse/recurrence. Though it might be hard to pull from any source system such explicit linkages. In practice, these linkages are not explicitly recorded in most EMRs. Tumor registries generally do a good job of recording primary site/histology and a not so great job at tracking recurrence. See https://www.ncbi.nlm.nih.gov/pubmed/24504926.

Found an article like this, actually like an explanation.
Minimalist Carcinogens: New Research Explains Why Virus-Associated Leukemia Thrives At Low Levels

t