Getting Started
What Qope IsYour First 15 MinutesInterface TourPlans and Access
Drug Intelligence
Drug ProfilePatent CliffShortages TrackerInsightsTerritory Brief
Access & Organizations
MapSafety-Net IntelligenceEHR Vendor Intelligence
Workspace
Lead ListsRadarPins and Saved Views
Databases
Working with TablesRecord PagesExportsDataset Catalog
Account
Plans and BillingUsage and LimitsTeams and MembersProfile and Security
Data & Methodology
SourcesFreshnessData CaveatsGlossaryMethodology
Support
FAQTroubleshooting
Open the app
Documentation/Data & Methodology/Methodology

Methodology

How provider peer groups, percentiles, KOL tiers and similar providers are computed, and how Qope matches records across datasets.

This page explains the numbers Qope derives rather than copies: peer percentiles, KOL tiers and similar providers on provider pages, and the rules used to connect one dataset to another.

Every provider figure here comes from the CMS Medicare provider files for data year 2024 (see Sources and Freshness). Those files cover Medicare fee-for-service only. Commercial insurance, Medicaid, Medicare Advantage and cash business are not in them, so a provider with a small Medicare panel looks small here whatever the size of the practice.

Peer groups

A provider is compared with a peer group built from the Medicare provider summary file:

  1. Same specialty, same state -- every provider in the file with the same primary specialty, matched exactly, in the same state. The specialty is the one CMS reports in the Medicare file, which can differ from the NPPES taxonomy shown in the provider header.
  2. All providers in the state -- used instead when fewer than 6 providers in the state share the specialty, or when the specialty name contains a comma (such names cannot be filtered, see Data Caveats). If the whole state has 10,000 providers or fewer and at least 5 others share the exact specialty, the group narrows back to that specialty.

The provider itself is always left out of its own comparison.

Full group or sample

Group sizeWhat the provider is compared withHow the page labels it
10,000 providers or fewerEvery provider in the groupvs all 1,918 cardiology providers in CA
More than 10,000 providersA sample of about 1,000: 10 blocks of 100 providers, evenly spaced across the whole group ordered by NPIvs a 978-provider sample of 16,578 internal medicine providers in CA

For scale, California group sizes in 2024 data: cardiology 1,918, hematology-oncology 1,040, family practice 13,887, internal medicine 16,578, nurse practitioner 18,403, and 177,585 providers in the state overall. So a cardiologist or oncologist is ranked against every peer, while an internist or nurse practitioner in a large state is ranked against a sample.

The sampling method was checked on 2024 groups small enough to compare in full (Florida internal medicine, 9,603 providers, and California cardiology, 1,918). On the three KOL measures, 95% of ranks taken from the 10-block sample were within 2 percentile points of the full-group rank, and none was more than 3 points off. Near a tier boundary that is enough to move a provider by one KOL tier, which happened for 1.5% to 3.2% of providers in those checks. A sampled page says so in its label.

A peer group is loaded once and reused for up to a day. If it cannot be loaded, percentiles, the KOL tier and similar providers show as unavailable. They are never shown as zero, and a partial group is never presented as a complete one.

Percentiles

A provider's percentile on a measure is the share of peers with a lower value, rounded to a whole number from p0 to p100.

  • Only peers with a value above zero count, so the effective group for one measure can be smaller than the peer group.
  • No percentile is shown when the provider's own value is zero, or when fewer than 5 peers have a value.
  • "Typical" and "peer median" figures in the Practice Intelligence summary are medians of the same group, again over values above zero.
MeasureSource field in the Medicare file
Part B paymentsMedicare payment for Part B services
Part B servicesServices billed, including drug units
Part D drug costTotal Part D drug cost
Part D beneficiariesPart D beneficiary count
Brand share of Part D costBrand drug cost divided by total Part D drug cost
Opioid prescribing rateThe CMS opioid prescribing rate

Part B services count drug codes billed per unit, such as 1 mg of an injected drug, so providers who administer drugs rank high on services. A provider page skips the services percentile when listed drug units are 10% or more of the provider's services, or when the listed services cover less than half of the total.

The Providers directory marks a provider Brand-heavy when its brand share of Part D cost is at p90 or above in its peer group.

KOL tiers

The KOL tier uses exactly three percentiles from the peer group above: Part B payments, Part D drug cost and Part D beneficiaries.

TierRule
Tier 1 (Dream KOL)p90 or higher on at least two of the three measures
Tier 2 (Specialist)p90 or higher on one measure, or p75 or higher on two
Tier 3 (Watch list)p75 or higher on one measure

p90 or higher means roughly the top 10% of the group, and p75 or higher roughly the top 25%. At least two of the three measures must have a percentile, so a provider with Part B activity and no Part D prescribing gets no tier.

Service counts, publications, clinical trials, industry payments (Open Payments) and prescribing of any particular drug are not used. A tier ranks Medicare volume within a peer group; it is not a measure of scientific influence.

Similar providers

The Similar tab ranks candidates from the same peer group used for percentiles: every other same-specialty provider in the state, the sample when that group is larger than 10,000, or the whole state when the specialty group is too small.

Similarity runs from 0% to 100% and is not rescaled to the best match:

  • Volume closeness, 80%. For each of Part D drug cost, Part B services, Part D beneficiaries and Part B beneficiaries, the smaller of the two providers' values divided by the larger. Identical values score 1, and a measure where only one of the two has activity scores 0. Measures where neither has activity are skipped, and the rest are averaged.
  • Brand mix closeness, 20%. One minus the difference between the two providers' brand shares of Part D cost. Used only when both have Part D cost; otherwise similarity is the volume closeness alone.

The 20 highest scores are shown. When the peer group is the whole state, providers with the same specialty are listed first.

How datasets are matched

Most datasets share no common identifier, so Qope links them by name. Name-based joins can miss a match (a product renamed or abbreviated differently in two files) or over-match (a short name that is also a word in another product's name). Each join below states its rule.

LinkRule
Provider page: NPPES record, Medicare summary, Part B services, Part D drugsExact NPI. This is the one true key join. The NPPES and Medicare specialty vocabularies still differ
Drug Profile and Radar: FDA shortage, recall and Purple Book rows for a drugWhole-word text match on the drug's ingredient and brand names. A name must appear as a complete word or phrase, not inside a longer word, and a combination ingredient matches only when every component appears
NADAC pricesThe NDC description must start with the drug name, followed by a space, punctuation or the end of the text
Medicaid rebate section on Drug ProfileThe NDCs of the NADAC rows matched by name are looked up in the Medicaid Drug Rebate Program file by exact NDC, up to 20 NDCs
Manufacturers and applicants (Shortages)Names are uppercased, legal and generic suffixes such as Inc, LLC, Pharmaceuticals and Laboratories are removed, then a list of 10 known aliases merges variants, for example Teva Parenteral Medicines into Teva
Patent Cliff: Orange Book brands to Part D spendExact brand name, uppercased. Brands named differently in the two files are not matched and show --
Provider Access tab: shortage areasHPSA designations within 25 miles of the geocoded practice address, else the same ZIP code, else the same city name. The underserved-area match uses the county of the nearest HPSA, else a service area whose name contains the city

Qope does not join datasets on RxCUI or UNII codes, and apart from the Medicaid rebate step above it does not join on NDC. There is no link from a provider's NPI to a hospital's CMS Certification Number (CCN), so Qope cannot tell you which hospital a provider works at.

Scores on other pages

The contestability, supply-gap and priority scores on Patent Cliff and Shortages have their own inputs and weights, documented with each page:

  • Patent Cliff
  • Shortages