Medicare Provider Outliers
Provider cost and volume comparisons using 2023 Medicare Part B/D data and the April 2026 NPPES registry. The mart contains 7,062,726 registry providers in admitted (NPPES taxonomy × state) groups, including providers without a Part B or Part D observation. Modeled in dbt on Snowflake; this site serves committed extracts — no live warehouse.
Registry providers in mart
MAD outliers
MAD outlier rate
z-score outliers
z-score rate
Part D spend ($B)
Coverage and denominator
The committed summary records 1,069,017 Part D prescribers and 1,076,323 Part B providers within the mart. These populations overlap; adding them does not give a distinct Medicare-provider count. Part D drug cost totals $205.7B and Part B payments $69.8B in this extract.
The headline flag rates divide by all 7,062,726 mart rows, including registry providers without observed Medicare activity. They are not rates among Medicare-active or metric-eligible providers, and they do not measure detection accuracy. A flag also requires at least 30 observed peers for its metric. State rates use the same all-mart-row basis within each state.
How to explore the snapshot
The ranked table contains the top 1,000 extracted rows per metric; the provider view contains six selected examples. These controls search the bundled extracts, not the full warehouse. The source documentation records the input vintages. Later registry taxonomy and location attributes classify 2023 utilization, so changes in those attributes can affect peer membership.
Comparing two scoring rules
At the configured thresholds, MAD flags 5.23% of mart rows and classical z-score flags 2.01%. MAD uses the median and median absolute deviation, which are less sensitive to extreme values than the mean and standard deviation. The thresholds also differ: 3.5 for modified z and 2.0 for classical z. No labeled review establishes which rule finds more actionable cases, or either rule's false-positive rate. A larger flagged set does not establish better detection.
Explore:
- Outlier Detection — top 1,000 extracted rows per metric
- Provider Drill-Down — six selected providers and their available peer observations
- Geography — outlier rate by state
- Methodology — peer groups, z vs MAD, and what a flag does not mean