How Ballicules ranks the most-prescribed drugs

Every "top 200 drugs" list has to answer one question first: whose prescriptions are being counted. Most published lists answer it with Medicare Part D, because CMS publishes that file and it is free. Part D covers people 65 and older and people with disabilities. It is an excellent record of what that population takes and a poor proxy for what the country takes.

The distortion is large enough to change the answer. On the unmodified 2024 Part D ranking cetirizine sits 830th, cholecalciferol 1,295th and aspirin 555th — all three are among the most-used medicines in the United States, and all three are mostly taken by people younger than Medicare or bought without a Part D claim. Pulling the other way, diltiazem ranks 68th on Part D and donepezil 54th; across all payers they are 370th and 155th.

What we do instead

The ranking on this site estimates annual prescription fills across all payers for the US civilian non-institutionalized population. It is built from three federal sources:

MEPS, the Agency for Healthcare Research and Quality’s Medical Expenditure Panel Survey, is the anchor. MEPS asks households what was actually filled, regardless of who paid, which is the quantity we want. Where a drug family has at least 30 sampled people and a relative standard error of 30% or less, its MEPS weighted total is used directly.

CMS Medicare Part D and the Medicaid State Drug Utilization file cover everything the survey is too small to measure. For those drugs a model predicts the all-payer total from the two payer counts, fitted across 235 drug families with a separate term for each therapeutic group — which is what lets the model learn that a psychiatric drug’s Part D count understates it while a cardiovascular drug’s overstates it.

FDA product identity decides what counts as the same drug. NDC codes are matched to the current FDA directory by ingredient, route, dosage form and release characteristics, so an extended-release product is counted separately from the immediate-release one when the library treats them as different cards.

How well it works, stated honestly

The figures below come from the September 2026 modeling run, which reports its own validation. The ordering currently published on the site is an earlier run of the same estimate: the two agree closely at the top and less closely in the tail, and adopting the newer run is a curatorial decision rather than a mechanical one.

Held out from its own fitting, the model reproduced the survey’s ordering with a Spearman correlation of 0.85. Unadjusted Part D counts managed 0.74 on the same families. That is the case for doing this at all.

The median absolute percentage error was 29.4%. Drugs a few places apart are not meaningfully in order, and nobody should quote a single rank as a fact. The validation was against MEPS, which is itself a survey with sampling error, so it does not prove the estimate right in any absolute sense — only better than the alternative it replaces.

Two further limits. Ranks run among the medications this library carries, roughly 1,200 entities selected for teaching value, so a drug we do not cover cannot occupy a slot: "#200" means 200th of the ones we carry. And 128 selected entities are deliberately unranked — products first marketed after 2024, and facility-administered drugs whose pharmacy claims do not represent their real use.

The unmodified source

Because an estimate cannot be audited the way a published count can, the raw CMS ranking is published here too: the CMS Medicare Part D 2024 drug ranking, all 1,975 positions, exactly as CMS reports them. Where the two disagree, that page shows both numbers.

What rank is good for

Dispensing order is the fastest route to being useful: the first hundred drugs cover most of what walks through a clinic door. It is not a ranking of importance. Plenty of drugs far down the list decide an admission, and a few at the top are prescribed more often than they should be.