Nothing expensive got cheaper
The CPI says prescription drugs fell 3.1%. In the federal NADAC file, 55 brand products sit above $2,000 a unit, and not one of them fell by half a percent.
Fringe Theory / Tool
NADAC brand file, July 2025 to July 2026
The prescription drug component of the CPI fell 3.1% in the twelve months to July 2026, its steepest decline since 1963 (BLS, July 2026 news release). Over roughly the same period, the Business Group on Health puts employer pharmacy trend at 11% to 12%. Both numbers are correct. This is the file that reconciles them: 1,539 brand drug products in the federal NADAC file, priced by what pharmacies actually paid, on two dates twelve months apart. Almost nothing here got cheaper because of competition.
CPI prescription drugs
12 months to July 2026. Deepest decline since 1963.
CPI hospital services
Same release, same month. Where hospital administered drugs are counted.
Employer pharmacy benefit
2026 trend, 12% before plan design changes and 11% after. Business Group on Health.
The last time this happened
Prescription drugs against all medical care, December 1959 to December 1968. It has happened three other times. See below.
Sources 1 and 6. The CPI prescription drug index samples retail, mail order and internet pharmacies only, and weights them by what consumers pay out of pocket. Drugs given in a hospital or a physician office are scored in a different index.
Has the drug price index fallen before?
Every stretch since 1948 in which the CPI prescription drug index posted sustained negative twelve month readings, against what the medical care index did over the same stretch. Baseline is the December before the first negative reading. Endpoints are December index levels, except 2026, which is July.
| Stretch | Prescription drugs | Medical care | What it was, and what is driving the current one |
|---|---|---|---|
| 1959 to 1968 | −15.1% | +40.8% | nine years of it; a single reading bottomed at −4.3% |
| 1970 to 1973 | +0.4% | +13.7% | three years, shallow; a single reading bottomed at −1.3% |
| 2017 to 2021 | +0.0% | +10.9% | four years, on and off; a single reading bottomed at −2.7% |
| 2025 to 2026 | −3.3% | +0.8% | Medicare negotiated prices and manufacturer list price cuts |
| Two more stretches contain a negative reading but no sustained decline | |||
| 1952 to 1953 | +0.0% | +3.5% | a single quarter below zero |
| 2012 to 2013 | +0.8% | +2.0% | two isolated months below zero |
There are six stretches in the table, including the two shallow ones, and in every one of them the cost of care went up. That is the argument in one table: a falling drug price index has never once coincided with a falling medical bill. The 1959 to 1968 stretch is the extreme case, nine years of falling drug prices against a 41% rise in the cost of care. What drove the rise is documented: coverage expanded, first through private hospital insurance and then through Medicare and Medicaid, which took effect on 1 July 1966 and enrolled 18.9 million people in Part A and 4 million in Medicaid in their first year, and CMS's own actuaries write that the faster growth that followed "reflected not only increased utilization, but also faster price growth for medical care (especially hospital services)." More people using more care. Why drug prices fell over the same years is a separate question, and not one this page has sourced.
Why does drug spend rise when prices fall?
Every figure on this page so far is a price. A plan does not buy prices, it buys prescriptions, and what it pays is the product. Here is the same twelve months on the other side of the multiplication sign, from the IQVIA Institute.
volume
not published on its own
spend
Volume is total prescription medicine use in 2025, 210 billion days of therapy. Spend is net US medicine spending, $548 billion to $606 billion, a $58 billion increase. GLP-1 and GIP agonists alone carried $14 billion of it, $9.6 billion in obesity products.
Source 7. Volume grew 1.5%. Spend grew 10.6%. A price index can hit a sixty year low inside that and change nothing a plan sponsor experiences, because the sponsor pays the product of the two terms and never either one alone.
What does NADAC measure, and what does it miss?
Every drug carries several prices at once. They measure different legs of the same transaction, and the gap between them is where most arguments about drug costs go wrong. This page uses NADAC throughout.
Schematic. Horizontal position shows which leg of the transaction each benchmark measures, and the rows run down the chain. AWP and WAC price the same leg, so AWP is drawn above WAC, which is also where it sits in price. Nothing else here is drawn to price level. That is the figure below.
Price level, for the two benchmarks that describe the same leg
Sources 10 and 11. Illustrative of the convention, not a measured average: the multiplier varies by product and publisher. The direction does not vary. AWP sits above the manufacturer's own list price, and the multiplier that puts it there was lowered by litigation, from 1.25 to 1.20, in September 2009. A benchmark a settlement can reset by five points was never a measurement. HHS's Inspector General puts it plainly: "neither AWP nor WAC is based necessarily on actual sales transactions." NADAC, further down the chain, is the first number on this page produced by asking anyone what they actually paid.
- NADACused here
- National Average Drug Acquisition Cost. What retail pharmacies actually paid to buy the drug, from a voluntary monthly survey of invoices run for CMS. Published free and weekly. Three things about it decide how far the numbers on this page can be pushed, and all three are in the CMS methodology document. Only "chain and independent retail community pharmacies have been surveyed," with specialty and closed door pharmacies "excluded from the surveys at this time." Discounts or rebates "not reflected on the invoice at the drug line item level are not factored into the NADAC calculation." And brand rates are partly propagated rather than observed: "if the published price for a drug increases by 5%, then the NADAC rate for that drug is also increased by 5%," validated against later surveys. Where a survey cannot produce a rate, an existing brand rate stays on the file for up to twelve months.
- WAC
- Wholesale Acquisition Cost. The manufacturer's published list price to wholesalers. It is a list, not a transaction, and almost nobody nets out at it.
- AWPnot a real price
- Average Wholesale Price. Not a wholesale price and not an average. It sits above WAC, the manufacturer's own list price, and for most products the compendia set it by a fixed multiplier, conventionally AWP = WAC × 1.20. That multiplier is a publisher convention rather than a rule, it varies by product, and it used to be 1.25 until the major publishers cut it to 1.20 in September 2009 following litigation. Unlike WAC, which is defined in statute at section 1847A(c)(6)(B) as the manufacturer's list price to wholesalers, HHS's own Inspector General notes that AWP "is not defined in law or regulation," and says of both benchmarks that "neither AWP nor WAC is based necessarily on actual sales transactions." Most pharmacy contracts are still written as a discount off AWP.
- ASP
- Average Sales Price. The benchmark for the drugs Medicare pays for under Part B, not Part D. The split trips people up and it is worth stating flatly: Part D is the pharmacy benefit, the drugs a member picks up and takes home, and it has no single published price benchmark. Part B covers "drugs that are administered by infusion or injection in physician offices and hospital outpatient departments," and those are paid at ASP plus 6%. ASP is therefore the closest public analogue to NADAC for the half of the drug market NADAC cannot see.
- What the CPI measures
- The total amount the pharmacy is reimbursed for a prescription, member payment plus the payer's share. It is a further step down the chain than NADAC and it is not net of manufacturer rebates either. Its expenditure weight, separately, comes from consumer out of pocket spending only.
- Net cost to the plan
- What the plan pays after rebates flow back, months later. No public file carries it. Every figure on this page sits above it.
NADAC and the CPI are independent measures. BLS does not use NADAC. It prices prescriptions from its own outlet sample, and the secondary claims data it began using in October 2024 covers physicians' services and outpatient hospital services, not drugs. These are two different instruments pointed at two different legs of the same transaction. They can move in opposite directions with nobody being wrong. That is the point. A drop in one is not a drop in the other, and neither is a drop in what a plan spends.
Which brand drugs got cheaper, and which got dearer?
Each mark is one brand product. Horizontal position is its acquisition cost per unit in the NADAC weekly file dated 16 July 2025, on a log scale. Vertical position is the change to the file dated 15 July 2026. This is a twelve month change between two point in time snapshots, not a month over month move and not an average of the year between them.
For illustrative purposes only. This tool reports published acquisition costs from a federal file. It is not a quote, a benchmark, or an estimate of what any particular plan pays.
Source 5. NADAC weekly files dated 16 July 2025 and 15 July 2026, brand classified products matched on NDC and collapsed to distinct product descriptions. NADAC is what retail pharmacies paid to acquire the drug. It is not what a plan pays and it is not net of rebates. It is also not the whole brand market. NADAC surveys retail community pharmacies, so drugs dispensed through closed door specialty pharmacies or bought and billed by a clinic never appear. Keytruda, Ocrevus, Trikafta, Entyvio, Revlimid and Imbruvica all return zero rows in the file. The 55 products above two thousand dollars a unit are the most expensive brands that retail pharmacies buy, not the most expensive brands in the country.
How many brand drugs fell, by price band?
Read the last column. Below twenty dollars a unit, prices go both ways. Above two thousand, they only go one way.
| Acquisition cost per unit, Jul 2025 | Products | Rose | Flat | Fell | Median change | Share that rose | Has a generic |
|---|
Read the last two columns together. The share of products with a generic equivalent listed in the same file peaks at 56% in the $1 to $5 band, holds above half through $20, then falls away to 32%, 9%, 8% and finally zero. The share that rose runs the other direction, from 45% in the $5 to $20 band up to 80% at the top. That column, not the price itself, is what moves the number.
Which categories moved
Median twelve month change by therapeutic category. Select a bar to filter the plot above.
A gold dot marks a category whose median product costs more than a thousand dollars a unit. Every one of them rose. Only one category has a median that falls by more than a rounding margin, and it is not a competition story: 18 of the 33 oral diabetes products fell by more than 30%, and two manufacturers account for all 18. Insulin sits at −0.1% because 12 insulin products fell that far and 33 did not, which is what a list price cut on part of a portfolio looks like once a median is taken over the whole category. The same thing hides Eliquis, down 43%, inside an anticoagulant median of +2.0%. Medians are why this page also shows you every product.
What this does and does not show
Above two thousand dollars a unit there are fifty five products in this file and none of them declined by as much as half a percent. The largest move in the wrong direction is a Humira presentation at −0.26%, which is rounding. Say the sentence carefully, because it is doing a lot of work: fifty five products in this file, and the file only sees what retail pharmacies buy. With fifty five observations and no declines, the honest upper bound on the true rate is still around 5%, not zero.
The temptation is to conclude that expensive drugs rise because they are expensive. The file says something narrower and more useful. Not one brand product above two thousand dollars a unit has a generic equivalent listed. Price level is standing in for the absence of a competitor. Where a competitor exists, the median brand moved 0.1%. Where none exists, it moved 2.4%.
Here is the part that took a second look to see. Forty products in this file fell by more than 30%. Thirty eight of them have no generic equivalent listed at all. Generic entry did not do this. A list price decision did, taken by a handful of companies on their own products.
Keep readingHow four companies did it, why a negotiated Medicare price is not what moved these prices, and what any of it does to a plan's net cost Tap to see more
Whether competitive pressure is what prompted those decisions is a fair question and not one this file can answer. What the file can say is that the price move arrived as a published corporate decision rather than as a competitor's product showing up on the shelf.
Humira is the exception that explains the rule. It is the only molecule above two thousand dollars a unit facing real biosimilar competition, and it is the only one whose flat result has a competitive explanation. Its biosimilar is in the file too, though outside this panel, because NADAC classifies Hadlima as a biosimilar rather than plain brand: Hadlima runs about $1,258 per milliliter against a July 2026 median of $3,367 across the four Humira 40 mg presentations, roughly 37 cents on the dollar, and both were flat over the year. Competition arrived, the price gap opened, and then both stopped moving. Four of the eleven flat products above two thousand dollars a unit are not Humira at all, three Perseris presentations and the Annovera ring, and neither of those faces a biosimilar.
Look at how they fall and the mechanism is obvious. Every empagliflozin product moved by the same −44.4%, across sixteen presentations spanning four brands. Both Farxiga strengths moved −36.9%, four hundredths of a point apart. All three Eliquis presentations moved −43.0%. Generic entry does not produce identical percentages across a manufacturer's entire portfolio. A manufacturer changing its own list price does. Boehringer Ingelheim published the notice: it would be "reducing the WAC prices for the Glyxambi, Jardiance, Synjardy, Synjardy XR, and Trijardy XR products listed below, effective January 1, 2026."
The timing is documented and it is not gradual. In December 2025 the National Community Pharmacists Association told its members that "the biggest WAC reductions in the history of the pharmaceutical industry" were about to land, and published the schedule: Imbruvica on 22 December 2025, then Eliquis, Farxiga, Jardiance and the rest of the Boehringer Ingelheim family on 1 January 2026, then Eli Lilly's insulins on 1 February 2026. The alert was about a practical problem. A pharmacy holding stock bought at the old price gets reimbursed at the new one.
1 January 2026 is also the date the first Medicare negotiated prices took effect, and that overlap is worth slowing down on, because a maximum fair price is not automatically what a pharmacy pays. CMS effectuates it as a refund. The pharmacy buys the drug at the manufacturer's price and the manufacturer refunds the difference afterward, through a clearing house CMS calls the Medicare Transaction Facilitator, with "up to 14 days to process the data" before the refund is even instructed. Nothing in that route lowers acquisition cost, which is what this file measures. A manufacturer that lowers its own list price toward the negotiated price does lower it, and narrows the refund it owes at the same time. Some did. Most did not.
| Selected drug | Change | What is behind it |
|---|---|---|
| NovoLog / Fiasp | split | One selected drug, two brands, opposite outcomes. Fiasp fell 75.1%. NovoLog did not move, because its own list was reset two years earlier |
| Jardiance | −44.4% | Boehringer applied the identical 44.4% cut to Synjardy, Glyxambi and Trijardy XR, none of which is a selected drug |
| Eliquis | −43.0% | All three presentations, same percentage |
| Farxiga | −36.9% | Both strengths, same percentage. No generic listed in the file on either date |
| Januvia | +0.0% | Flat |
| Xarelto | +2.3% | Rose |
| Entresto | +2.9% | Rose |
| Enbrel | +5.0% | Rose |
| Stelara | +5.0% | Rose |
| Imbruvica | not here | Zero rows in the file. Its own WAC reduction took effect 22 December 2025, and a retail acquisition cost survey never saw it |
Three of the ten moved down across every presentation, a fourth moved on one of its two brands and not the other, four rose, one was flat, and one is invisible to this file. A negotiated Medicare price did not, by itself, move what a pharmacy pays. The second list points the same way: of the fifteen drugs with a negotiated price scheduled for 2027, nine are priced here and only Linzess cut, by 50%, while semaglutide sold as Ozempic, Rybelsus and Wegovy rose about 3%, Trelegy Ellipta 3.0%, Otezla 4.8%, Vraylar 4.9% and Austedo 5.0%. And the move is wider than the program. Of the 40 products in this file that fell by more than 30%, 12 have no negotiated price at any point, including Tresiba, Lyumjev and the concentrated Humalog pen.
The insulins are the same move at a larger scale, and here the announcement and the file can be laid side by side. Novo Nordisk said in December 2024 that "the list prices of Fiasp® will be reduced by 75%, matching the price of NovoLog®" and that Tresiba would be "reduced by 72.2%," both starting 1 January 2026. Eli Lilly's trade notice announced "70% reductions in Wholesale Acquisition Cost" on the Humalog U-200 pen and the Lyumjev presentations, effective 1 February 2026. This file, which is a survey of what pharmacies actually paid and never saw either notice, recorded −75.1%, −72.2% and −70.0%. Three announced percentages, matched to within two tenths of a point. NovoLog is the control: its own 75% cut landed two years earlier, from $34.36 to $8.59, and it did not move at all this year.
One line in Novo's notice is worth reading twice. Cutting Tresiba's list by 72.2% made it "20% lower than its unbranded biologic equivalent, insulin degludec, thereby rendering the latter redundant. As such, Novo Nordisk Pharma, Inc. will discontinue the unbranded biologic versions." The cheaper unbranded version came off the market. Here a falling price and a departing competitor arrived together, which is not how a falling index is usually read.
Then there is what a plan nets. When a manufacturer cuts list, the rebate that was funding the gap generally comes down with it. Gross falls. Net moves far less, and often barely at all. This file carries gross, so it cannot tell you what a plan ends up paying. Neither can the CPI. Whether your own contract passes a list price cut through, or absorbs it in a rebate guarantee, is a question about your contract, and it is the first thing I would go and read.
Table view: every brand product, sortable
| Drug | Category | Jul 2025 $/unit | Jul 2026 $/unit | Change | Generic listed |
|---|
Method and limitsHow to read this, and where it breaks: six things a hostile reader would go after, in the order they matter Tap to see more
1. It counts products, not dollars and not prescriptions. A brand filled a thousand times a year gets the same one vote as a blockbuster filled ten million times. Nothing here is an inflation rate, a spending trend, or a price index, and it should never be read next to the CPI as though it were a competing measure of the same thing.
2. NADAC only sees retail. CMS surveys "chain and independent retail community pharmacies," with specialty and closed door pharmacies "excluded from the surveys at this time." Keytruda, Ocrevus, Trikafta, Entyvio, Revlimid and Imbruvica all return zero rows. The fifty five products above two thousand dollars a unit are the most expensive brands retail pharmacies buy, not the most expensive brands in the country.
3. Brand rates are partly propagated, not observed. This is the one that surprises people. CMS states that "if the published price for a drug increases by 5%, then the NADAC rate for that drug is also increased by 5%," validated against later surveys. So a brand NADAC increase is not always an independent discovery that acquisition cost rose. Some of it is the published list price change being passed straight through and confirmed afterwards. Read the increases here as evidence about published brand pricing behavior, which is real and consequential, rather than as fifty five separate field observations. Where a survey cannot produce a rate, an existing brand rate stays on the file for up to twelve months, so "flat" can also mean "not repriced."
4. It is gross, not net. Discounts or rebates "not reflected on the invoice at the drug line item level are not factored into the NADAC calculation." Nothing here is what a plan nets, and no public file carries that.
5. The population is NADAC's own brand classification. The file carries B, B-ANDA, B-BIO and G. This page uses plain B. Adding the other two brand classes was tested and changes nothing above two thousand dollars a unit: of the sixty B-ANDA and B-BIO products priced on both dates, not one sits above two thousand. The three high cost biosimilars in the file, Steqeyma, Fulphila and Yesintek, carry no July 2025 rate at all, so a matched panel cannot see them either way. Generic products are excluded throughout.
6. Dollar bands mix pricing units. "Per unit" is a tablet, a milliliter or a gram depending on the product, so a price band is not a clean economic class. The result holds inside each dosage form separately: above two thousand dollars a unit, no declines among the thirty two milliliter priced products and none among the twenty three priced per unit.
The panel. Retrieval is complete on both dates. The 15 July 2026 file carries 2,293 brand NDCs and the 16 July 2025 file carries 2,459, and both pulls match the counts the API reports. Of those, 2,065 carry a price on both dates and collapse to 1,539 products, taking the median where several NDCs share a description. Two descriptions carrying more than one pricing unit were dropped outright. NDCs present on only one date are excluded, as a matched panel requires, and those exclusions skew expensive: a median of $17.28 against $12.75, and 6.6% above two thousand dollars against 3.2%. They are new products entering at high prices, which makes what is left conservative. Nothing you type is stored, sent or logged.
Sources
- U.S. Bureau of Labor Statistics, Consumer Price Index, July 2026 news release and Table 2; series CUUR0000SEMF01, CUUR0000SEMD01, CUUR0000SEMC01, extracted 28 August 2026.
- U.S. Bureau of Labor Statistics, Measuring Price Change in the CPI: Medical care, factsheet.
- Hicks AL, Berndt ER, Frank RG. Auditing the prescription drug consumer price index in a changing marketplace. Health Economics, 2024. doi 10.1002/hec.4836.
- Centers for Medicare and Medicaid Services, Medicare Drug Price Negotiation Program: Negotiated Prices for Initial Price Applicability Year 2026, fact sheet.
- Centers for Medicare and Medicaid Services and Medicaid.gov, National Average Drug Acquisition Cost, weekly files dated 16 July 2025 and 15 July 2026, retrieved 29 August 2026.
- Business Group on Health, 2026 Employer Health Care Strategy Survey, Executive Summary.
- IQVIA Institute, U.S. Medicine Use Trends 2026, published 28 April 2026.
- Catlin AC, Cowan CA. History of Health Spending in the United States, 1960-2013. Centers for Medicare and Medicaid Services, Office of the Actuary, November 2015.
- Medicare Payment Advisory Commission, Payment Basics: Part B Drugs Payment Systems, November 2021.
- On the AWP to WAC multiplier and its reduction from 1.25 to 1.20 by the major publishers in September 2009 following litigation: Journal of Managed Care Pharmacy, What Is the Price Benchmark to Replace Average Wholesale Price (AWP)?, 2010;16(7):492.
- U.S. Department of Health and Human Services, Office of Inspector General, Replacing Average Wholesale Price: Medicaid Drug Payment Policy, OEI-03-11-00060.
- Boehringer Ingelheim, Important Notice: Limited Rebate Program for Glyxambi, Jardiance, Synjardy, Synjardy XR and Trijardy XR, November 2025, distributed to the trade and carrying the January 1, 2026 WAC reduction.
- National Community Pharmacists Association, WAC Decreases Coming: Take Action Now, alert to members dated 10 December 2025, carrying the effective dates for the Imbruvica, Eliquis, Farxiga, Jardiance, Boehringer Ingelheim and Eli Lilly insulin reductions.
- Centers for Medicare and Medicaid Services, Medicare Transaction Facilitator (MTF) Overview for Dispensing Entities, fact sheet.
- Centers for Medicare and Medicaid Services, Medicare Drug Price Negotiation Program, Selected Drugs and Negotiated Prices, listing the ten drugs for initial price applicability year 2026 and the fifteen for 2027.
- Novo Nordisk, Novo Nordisk to lower US list prices for Tresiba and Fiasp by more than 70%, press release, Plainsboro NJ, 5 December 2024.
- Novo Nordisk, Novo Nordisk to lower US prices of several pre-filled insulin pens and vials up to 75% for people living with diabetes in January 2024, press release, 14 March 2023, carrying the NovoLog reduction.
- Eli Lilly and Company, February 2026 Humalog (insulin lispro) U-200 and Lyumjev WAC reduction, trade announcement, trade.lilly.com.
What this tool does not account for. Rebates, discounts and any other price concession, so every figure here is gross rather than net. Dispensing fees, pharmacy contract terms and member cost share. Drugs dispensed anywhere other than a retail community pharmacy. Utilization, which is the other term in what a plan actually spends. It prices products, not a plan.
Fringe Theory is independent and unaffiliated. Views expressed are my own and do not represent those of my employer. Nothing here is legal, tax, medical, or investment advice.
Figures described as modeled or derived are my own arithmetic from the public sources listed, not disclosed data.