Industries / Health / Healthcare / Pharmacy
Inventory
Inventory is how a pharmacy buys its stock, stores it, counts it and accounts for every pack, from the wholesaler's delivery to the shelf, the returns bin and, for controlled drugs, the legal record.
7
Milestones achieved, from 1972 to 2023.
1
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1972
A number for every pack
The Drug Listing Act of 1972, approved on 16 August 1972, made drug makers give the FDA a current list of the drugs they make for sale. The FDA identifies each listed product by its National Drug Code (NDC), a 10-digit, three-part number that names the labeler, the product and the package size; pharmacies order, count and bill by it, and since a 2000 rule under HIPAA, the US health data privacy and transactions law, it is the required drug code in US electronic health care transactions. In March 2026 the FDA ruled that every NDC moves to one 12-digit format from 7 March 2033, because it expects to run out of 5-digit labeler codes in 10 to 15 years.
Lessona code that every system stores is slow to change, so the FDA delayed the new format by seven years to give systems time to handle it.
Credit
US Congress; US Food and Drug Administration
2001
Shortages counted nationally
The University of Utah Drug Information Service (UUDIS) began collecting US drug shortage data in January 2001 and publishes it on a public website run by ASHP, the American Society of Health-System Pharmacists. Outside researchers use the series to study shortages over two decades, for example of lead-poisoning antidotes from 2001 to 2022. ASHP and UUDIS counted 227 active national shortages in the second quarter of 2026, the third quarterly rise in a row.
Credit
University of Utah Drug Information Service; ASHP
2005
Controlled orders go electronic
US law requires an official DEA order form for every sale of Schedule I and II drugs, the two most tightly controlled classes. A Drug Enforcement Administration rule effective 31 May 2005 let pharmacies and their suppliers place these orders electronically, signed with a digital certificate, and keep the records electronically, through the Controlled Substance Ordering System (CSOS). From 3 November 2025 all CSOS enrollment goes through DEA's secure online portal.
Credit
Drug Enforcement Administration
2015
Radio tags check trays
Hospital pharmacies restock trays of drugs for operating rooms and resuscitation carts by hand, checking each item, count and expiry date. North York General Hospital in Toronto put RFID tags, radio labels that a scanner reads all at once, on its tray drugs from October 2014. In a test with planted errors, 6 of 16 trays restocked by hand were approved with 1 to 8 errors still in them; none of the 16 restocked with the scanner were, and each run took 4.4 minutes less on average.
Lessonthe technicians who restocked fastest by hand also left the most errors.
Credit
Edith Rolko and Thomas Chan, North York General Hospital; system by MEPS Real-Time (Intelliguard)
2019
Every opioid shipment, public
A DEA database called ARCOS (Automation of Reports and Consolidated Orders System) tracks every pain pill sold in the US from maker and distributor to pharmacy. Data from it, released in July 2019 in the federal opioid lawsuits in Cleveland, showed by The Washington Post's analysis that 76 billion oxycodone and hydrocodone pills were shipped from 2006 through 2012, and that six companies (McKesson, Walgreens, Cardinal Health, AmerisourceBergen, CVS and Walmart) distributed 75 percent of them.
Lessonrecords kept for a regulator showed the scale of oversupply only once the public could read them.
Credit
Scott Higham, Sari Horwitz and Steven Rich, The Washington Post
2019
Swaps allowed in a shortage
In February 2019 the UK changed the Human Medicines Regulations 2012 to create serious shortage protocols (SSPs), with NHS rules for England following in July 2019. When the Department of Health and Social Care declares a serious shortage of a medicine, a pharmacist may, on professional judgment and with the patient's agreement, supply the alternative the protocol names in place of what was prescribed. The NHS Business Services Authority, which processes pharmacies' NHS prescriptions and payments, had listed protocols up to SSP089, for lansoprazole tablets, by September 2026.
Credit
UK Department of Health and Social Care; NHS Business Services Authority
2023
Returned cancer pills reused
Unused oral cancer drugs that patients bring back are normally destroyed. At the outpatient pharmacies of four Dutch hospitals, from February 2021 to February 2023, drugs went out in sealed packs with a time-temperature indicator, and returned packs that passed checks of authenticity, appearance, remaining shelf life and storage temperature were dispensed to other patients. Of 1,071 patients, 171 returned 335 packs; 228 were reused, cutting waste by 68.1% and saving at least US$680 (€576) per patient per year.
Credit
E. M. Smale and colleagues, published in JAMA Oncology
Watch for
Demand sensing at a wholesaler
Cencora, a drug wholesaler, says its AI compares past forecasts with incoming orders to flag shifts in demand early, and reports forecasting gains of 10%. It is also linking data on manufacturers, FDA actions and other risk factors to spot patterns that may lead to shortages.
Achieved whenan outside party shows fewer stock-outs or shorter shortages at pharmacies supplied with these forecasts than at a comparison group.
Claimed by
Matt Sample, Cencora, October 2026
Open
01
A test for diversion alarms
A public, synthetic record of six months of drug movements in a made-up 300-bed hospital (withdrawals from cabinets, doses given, waste, returns and counts), with diversion, the theft of controlled drugs by staff, planted in known places. Add a scoring script that measures how many days each detection method takes to flag each planted case and how many false alarms it raises.
The briefCredit · unclaimed02
A public stock forecast benchmark
A standing benchmark that asks forecasters to predict, three months ahead, how many units of each medicine will be dispensed in each area of England, and scores them each month against the newly published figures, so that tools that help pharmacies order stock can be compared on the same public data.
The briefCredit · unclaimed
Open 01 · the brief
A test for diversion alarms
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public, synthetic record of six months of drug movements in a made-up 300-bed hospital (withdrawals from cabinets, doses given, waste, returns and counts), with diversion, the theft of controlled drugs by staff, planted in known places. Add a scoring script that measures how many days each detection method takes to flag each planted case and how many false alarms it raises.
Why it is still open
Published detection methods are trained and tested on private data. A 2022 study by the software maker Invistics and 10 hospitals in 4 health systems used 27.9 million transactions with 22 known cases, and its model flagged them a median of 74 days sooner than existing methods, but the data are not public. Bluesight's 2025 report draws on more than 266 million transactions from over 1,100 hospitals, also private. A search of Hugging Face in November 2026 found no public diversion dataset.
Built with what exists
Python with SimPy to simulate ward workflows; Synthea for synthetic patients and orders; scikit-learn, XGBoost and isolation forests for baseline detectors; Hugging Face Datasets to publish the set.
First step
write the event schema and five diversion patterns from the cases described in the 2022 study and in Carollo and colleagues' 2024 operating-room study, then generate one simulated month.
Done looks like
a published set of at least 1 million synthetic transactions with 20 planted cases of five patterns, three baseline detectors scored by median days to detection and false alarms per 1,000 staff, and results from at least one outside team.
Open 02 · the brief
A public stock forecast benchmark
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A standing benchmark that asks forecasters to predict, three months ahead, how many units of each medicine will be dispensed in each area of England, and scores them each month against the newly published figures, so that tools that help pharmacies order stock can be compared on the same public data.
Why it is still open
Published work uses one-off or private sets. The 2023 Isfahan AI competition used drug sales from 12 pharmacies of one Iranian medical university. A 2025 preprint by M. Menebo used the English Prescribing Dataset, but only for immunological products and vaccines, and reports R² of 0.80 for an XGBoost model with lagged weather data. In 2021 J. S. Frazer and G. R. Frazer fitted one forecasting method (Holt-Winters) to 7.5 billion English prescriptions from 2014 to 2020, to measure how far the pandemic pushed prescribing away from the forecast. None offers a leaderboard that scores new forecasts as new months arrive.
Built with what exists
the NHS Business Services Authority's English Prescribing Dataset (monthly items and quantities for each medicine, by prescribing practice); Python with Nixtla's statsforecast for seasonal baselines; LightGBM; open time-series forecasting models such as Amazon's Chronos and Google's TimesFM.
First step
download 36 months of the dataset for one integrated care board and publish the seasonal-naive baseline's errors.
Done looks like
a leaderboard covering the 500 most-dispensed medicines across every integrated care board in England, scored monthly for six months by mean absolute scaled error against a seasonal-naive baseline, with at least three entrants.