| OHDSI Home | Forums | Wiki | Github |
Is there guidance from Themis about whether to map denied insurance claims or not? I looked over the conventions but didn’t see any information on this topic. Do some ETLs map denied insurance claims into OMOP, and if so how do they store the payment status in the OMOP CDM (using a type concept)?
Thanks!
I do not think you will find a documented convention for a denied claim. Even though the claim (payment) was denied, the medical event or medication probably happen so a medical event/drug record should be created. I assume by ‘payment status’ you mean cost. I do not know how detailed your source information is, but would it not be sufficient to say that paid_by_payer (the insurer) is 0?
Columns that end in ‘_type_concept_id’ are intended to identify the provenance (where the record came from in the source). It would be nice to change the column name because people assume that type should hold something like ‘denied claim’ rather then ‘Claim’ or ‘Payer system record (primary payer)’. But changing a CDM column name is not easy. If you really need to track denied claims, add a column to the appropriate table, but it is not something the CDM tracks.
I agree that denied medical claims probably happened unless they are for the wrong patient and reversed. However, denied pharmacy claims likely didn’t happen actually represents one of the greatest abilities of modern medicine to identify counterfactuals (in my opinion). Pharmacists don’t let you leave the store until a claim is adjudicated forcing a full system of prior authorization through a PBM.
Thanks @DTorok and @Kevin_Haynes!
By payment status I mean whether the information (e.g. drug_exposure record) came from a paid claim or a denied claim. This seems to fit the idea of provenance captured by the type concepts to me. Adding a column to track this information makes sense though.
Interesting but I’m not sure I follow completely. How would a denied claim let you identify a counterfactual?
How do you reference an existing comments like you did above?
And nice try bending the definition of provenance to distinguish “came from a paid claim or a denied claim”, but I am pretty sure they are both claim records. valid type concepts
The OMOP CDM is not a claims processing model. It aims at capturing what happened to the patient, not to the insurance company. It is the job of the ETLer to come up with a heuristic whether a denied claim means it didn’t happen or it was just not paid for whatever reason the stingy payer might have. Don’t throw this problem over the fence to the analyst, he/she has no more insight into how to interpret the situation. Most likely, she will just ignore your fine-tuned Type Concept. Sorry, man. 
Yep that makes sense @christian. So it sounds like there is no official Themis recommendation on whether denied claims contain information about what actually happened or not which makes sense since this is dataset specific. But in the US I guess people generally are mapping denied claims. Analysts take the records at face value as if they did indeed happen and rely on the ETLer’s decision.
Just to add my 2 cents here - it doesn’t seem obvious to me that a denied claim necessarily correctly represents what happened to a patient. Even if it does, you would need to somehow dedup a denied claim followed by a subsequent approved claim (which can be the same code+provider or a different code). And heuristic may have to be more complex if the claims have different codes or are far apart. I don’t have a readily available solution but wanted to suggest that it may not be that straightforward.
Stay very close to the data owners who have the data provenance knowledge and even then there is a chain from bedside/pharmacy/clinic to database that must be fully understood. I have filled claims as a pharmacist. A denied claim can occur for many unrecorded reasons. Take a medication needs a prior authorization and maybe on a phone call (not recorded in OMOP) it’s switched. If that’s not near real world randomization I don’t know what is.
I hope for the sake of US data that people are not generally mapping denied claims and losing the knowledge of denied. This is not straightforward. Denied claims are and should generally be excluded. However, the represent a unique opportunity to evaluate what could or might have happened only the real world knows and only a medical record validation study could confirm given a set of denied claims.
I see. Yes this is a rare opportunity to capture a real world counterfactual.
I agree. I was hoping there was a way to flag these claims if organizations decide to map them. I do agree that it is outside the scope of the OMOP CDM though. The events that get mapped are assumed to have happened.
This discussion is pretty old but I’m finding it still an open question of how to handle denied claims. I’ve found many examples in our raw data of routine procedures (eg EKG’s) for which the claim was denied and yet it seems quite possible (and even likely) that the event itself did occur.
So am I reading correctly @Adam_Black @Kevin_Haynes that net net you would not include denied claims in procedure / condition tables?
Claims being denied payment is an artifact of benefit design of the plan and provider contractual agreement. It can be independent of service delivery.
A service may be delivered to a patient for medical necessary reasons, this service may then be billed to third party payor by the provider of service. The third party payor may deny payment to the service provider based on benefit design and contractual agreement.
e.g. if there is a bundle payment agreement, then line item bill can be denied in favor of global payment
I will second that.
My opinion is that, with the exception of fraud (which are services that did not happen) - all time stamped events of service that happened for a person should be mapped to omop even if they are denied payments. Because it happened to the person.
Double counting should not be a problem, because even if we inflated the number of records with duplicates during etl - the number of era-fy records would stay the same. Cohorts are eras. Analysts use cohorts.
But do we know they actually happened? How often are claims denied due to billing errors (not necessarily fraud)?
If we’re confident that these events are accurately recorded and happened then of course we should map them. I think the question is about the veracity of the data on denied claims.
I think those come under claim reversals. Different idea to the current question.
