@pipeworx/meps

Connect: https://gateway.pipeworx.io/meps/mcp · Install: one-click buttons

Tools: 4

What Americans actually take, and what for — US prescription drug use by medical condition, from AHRQ’s Medical Expenditure Panel Survey.

MEPS surveys the whole US civilian non-institutionalized population: every payer, every age, commercially insured and uninsured alike. It is the only source in this catalog that links a prescription to the condition it was written for, which is what makes “what do people take for depression” answerable at all.

Tools

Conditions are taken in plain words — "depression", "high cholesterol", "type 2 diabetes" — or as a CCSR code such as MBD002. Drug names are MEPS’s cleaned generic names, so atorvastatin, not Lipitor.

These are survey estimates, and the sample is always in the response

Every respondent carries a weight projecting them onto roughly 18,000 Americans. A cell backed by three people therefore produces a confident-looking “55,000 people” that means almost nothing. So:

• every row carries sample_persons — the real respondents behind it;
• reliable: false marks anything under 60 respondents, which is AHRQ’s own publication floor;
• rows default to a minimum of 30 respondents, adjustable with min_sample.

Treat the numbers as rough magnitude and ranking, not as counts. Ranking the top few drugs for a common condition is what this data is good for. Precise totals, small subgroups and year-over-year differences of a few percent are not.

Two things that would otherwise mislead

Fewer than half of prescriptions carry a condition. 47.5% of fills in 2024 are linked to any diagnosis; the rest are not attributed to one. So a condition total is a subset of prescribing, not a census of it, and an empty result can mean “not captured” rather than “not prescribed”. Every response states the year’s actual coverage rather than implying completeness.

One fill can be linked to several conditions. That is correct for “is this drug used for this condition” and wrong for money: the condition tables carry no expenditure at all, because summing spend across conditions would invent national spending that was never spent. Money lives only in meps_drug_use and meps_top_drugs, where each fill is counted once.

Some “drugs” are therapeutic classes. MEPS substitutes a class name for the product where naming it would identify a respondent — INTERLEUKIN INHIBITORS, ANTINEOPLASTICS, ANTIDIABETIC AGENTS. In 2024 that is 49 of 554 names, but those 49 carry 31.9% of all recorded spending, so a ranking by cost is dominated by classes rather than products. meps_drug_use and meps_top_drugs say so in every response. A plural or category-sounding name is a class.

Data sources

• MEPS public use files, three per year, from https://meps.ahrq.gov/data_files/pufs/ — Prescribed Medicines, the Appendix to Event Files (the condition link), and Medical Conditions. Free, no credential. The file numbers per year come from AHRQ’s own index at https://meps.ahrq.gov/data_files/search_pufs.json; they are not derivable (2024 is HC-254A / HC-254I / HC-255), so they are looked up, never guessed.
• Condition category labels from the HCUP CCSR reference file, https://hcup-us.ahrq.gov/toolssoftware/ccsr/dxccsr.jsp.

Years 2019 onward. Earlier files code conditions with CCS rather than CCSR and are not comparable.

.github/workflows/meps-refresh.yml checks AHRQ’s index on the 3rd of each month and loads any year that has all three files and is not already held. AHRQ publishes a year roughly annually without pre-announcing it, so the check does nothing 11 months out of 12 — that is cheaper than noticing a release late.

MEPStrends is not the source. AHRQ’s own pre-computed estimates at meps.ahrq.gov/mepstrends/ sit behind HTTP Basic auth (WWW-Authenticate: Basic realm="Secured Data Tools"), so the estimates here are computed from the public files rather than read from theirs.

Tools

• meps_drugs_for_condition — Which prescription drugs Americans actually take for a given medical condition, with a national estimate of how many people take each one. Covers the whole US population — every payer, every age, comm
• meps_conditions_for_drug — What a drug is actually being prescribed for in the US population, ranked by estimated people. Answers off-label and multi-use questions that a label or an approval database cannot: what gabapentin, m
• meps_drug_use — National prescription volume and spending for one drug: how many Americans filled it, how many fills, total spend across all payers and how much patients paid out of pocket. Population-wide rather tha
• meps_top_drugs — The most-used prescription drugs in the United States, ranked by estimated number of people who filled them, with total and out-of-pocket spending. Answers what the most commonly taken medications in