Methodology

Four decisions shape every page on this site. Each exists because the obvious alternative would mislead.

1. The three kinds of money are never added

General payments, research funding and ownership interests are reported under different rules and mean different things. A meal and a clinical trial grant do not sum to anything a reader can use. You will never see a single combined total here, and any figure attributed to this site that combines them is a misreading.

2. A name never appears alone

More than a quarter of recipient profiles share a name with another profile, and a common name can match hundreds of people. Every name on this site is shown with a city, state or specialty beside it, in headings, search results and page titles alike, because a bare name identifies nobody.

3. Percentiles are within a specialty, or absent

Specialty medians run from about $75 to over $1,000 — a fourteenfold spread — so ranking a practitioner against everyone would say more about their specialty than about them. Where no honest rank exists, because there are no payments, no recorded specialty, or fewer than 100 paying profiles in that specialty, we show the total alone rather than a zero.

4. Truncated lists say so

Rankings are of one kind of money at a time — general payments unless a list says research — and never of the kinds added together. They show where they stop and how much they leave out. Shares in a leaderboard are measured against the whole program year, so a top-50 list sums to well under 100% and is never drawn as a chart that closes.

5. Charts never invite a sum

Where the three kinds of money are drawn side by side, they are drawn at the same size. Their amounts differ and are printed as numbers; giving them different weights would invite the eye to add them. Every chart value is also written out as text.

The state map is tinted against the largest state rather than against the national total. The largest state is only about a tenth of a year, so a linear scale against the total would paint every tile the same pale wash. The share is printed in each tile; the tint is a second cue.

An empty list or chart means nothing was reported for that slice. It is a finding, not a missing page.

Reading the figures on the home page

Three kinds of money. The diagram draws general payments, research payments and ownership interests as three separate lanes of equal height, with their amounts printed beside them. They are shown apart and never added together.

Where general payments landed. Each tile is a state’s share of the year’s reported general payments; a darker tile is a larger share, tinted on a square-root scale against the largest state. The share is printed in the tile. Territories and a collapsed “other” row are not drawn.

What general payments were for. General payments in the program year, by the nature of each payment as reported. A longer bar is a larger amount. Only the largest ten natures are shown, so the bars do not add up to the year.

Worked examples. The examples under “What you can find out” are the teaching hospital and company with the most reported general payments that year, and the two states with the most paid recipients. They are chosen by size so they are reproducible, not because anything about them is unusual.

CSV downloads

Every records list, leaderboard and the state table downloads as CSV, free and without signing in. A file holds at most 5,000 rows; longer lists download in consecutive files named by their range (keytruda-general-payments-rows-5001-10000.csv). Headers are plain English and keep a fixed order; the second row names the source, program years, filters, the rows in the file and the download date. Files are UTF-8 with a byte-order mark so spreadsheets read accented names correctly, and any cell that begins like a formula is prefixed with an apostrophe.

Display

CMS stores names and cities in block capitals. We case them for reading; the underlying data is never modified. Where a page shows an identifier — NPI, CCN, CMS profile id — it is reproduced exactly as reported.