Kalderos collects chargeback data from multiple manufacturers and claims-level data from all fifty states. This data includes all Fee-for-Service (FFS) and Managed Medicaid (MCO) claims that have had or will have a Medicaid rebate paid on them. Once the data sets are ingested, our proprietary algorithm identifies the claims that need your review. Here’s how the process works:
Associate claims with covered entities. Through learned associations between 340B IDs and NPIs, Kalderos identifies which Medicaid claims are associated with which covered entity.
Match with chargeback data to determine which claims might be duplicates. Kalderos will identify if an associated claim is an eligible match based on whether a chargeback was invoiced by the covered entity within a specific amount of time from the dispense date.
Sort and prioritize potential claims. Of these candidate claims, Kalderos uses machine learning and historical responses to prioritize and select which claims to send. After your first couple inquiries, claim volume will be determined based on compliance history and you may receive few or no claims for that period.
We have continuously improved our machine learning model over time, improving precision so that covered entities only see the claims that are most relevant to them. Despite these improvements to the model, there are still some limitations due to a lack of currently available data.
Challenges around associating claims with covered entities
OPAIS information sometimes lists associations that are not “active,” so our database may make incorrect pharmacy associations. We provide a response option where covered entities can indicate this.
Where covered entities share a contract pharmacy, that pharmacy’s claims may be associated with more than one covered entity. State claims don’t always come with NPIs, meaning Kalderos needs to infer the association between the contract pharmacy and covered entity.
Challenges with determining whether Medicaid claims are associated with chargeback data
- Inventory practices at on-site and contract pharmacies differ, so there can be inconsistency in data.
- Chargebacks can be from a long time in the past or future relative to when a dispense occurred.
Challenges with prioritizing claims based on risk
- We require initial responses to learn about the covered entity and improve precision (typically this will be done over three response cycles). With machine learning technology, we are able to improve the specificity of our inquiries over time!