You may have noticed that we provide all-payer extrapolations for facilities based on Medicare claims to estimate the total volume for various payer types for service lines at facilities. We derive this payer mix from an extrapolation model that we developed in-house, based on published best practices from the industry. Read on for the details.
As a starting point, we took the following sources of healthcare discharge volumes:
- Medicare Fee-for-service claims from the Inpatient Prospective Payment System (IPPS).
- Provider-level, state-provided datasets for 5 states that breaks down utilization by Provider (facility), DRG, and payer. These datasets are either all-payer claims datasets where available, or regulatory filings (like MonAHRQ) if claims data is not available.
- State-level HCUP-Netaggregate inpatient datasets for 35 states, that breaks down utilization at the state level by DRG and payer.
- National HCUP-Netaggregates where neither provider-level data nor state-level data were available.
By collecting data at different levels of granularity at the geography level, we’re then able to take our 100% Medicare numbers and use those as the baseline Medicare volumes, correlating them to each facility/state/national number for Medicare, where available.
By using known Medicare volumes and correlating them with data from HCUP-Net/States, we’re able to authoritatively build a payer mix for each DRG that’s comprised from known (Medicare) and discoverable (Medicaid, Private Payors) components.
Data Confidence classification
This approach gives us 3 levels of confidence for our extrapolations:
- P = facility-level, best-level confidence. This is either an exact count for the year, or an extrapolation based on an exact count for another year, if we don't have exact counts for that year.
- S = state level, second-level confidence
- N = National-level, lowest-level confidence within the extrapolation model
Fewer Physician Extrapolations
We only extrapolate procedure volumes for physicians when we have an all-claims database which can give us exact numbers (i.e., a state with P-level data), so we only perform these extrapolations for physician in a couple of states.
The reason we don't do state and national level extrapolations is because the source data is for hospitals, and we found that using those values didn’t seem reliable at the physician level.