When a controlled-substance discrepancy turns out to be more than a miscount, someone has to investigate it, and the first hours decide how much of the evidence is still available. This guide lists what an investigation actually asks for, in the order the requests tend to come: the records, the interviews, and the filings with regulators. It is built from a peer-reviewed description of how ten hospitals investigate, from DEA's rules, and from the matters the Department of Justice has described in public settlements. It is not a substitute for your own policy or counsel, and it is about internal investigations; for the separate subject of a DEA inspection, see what inspectors ask first.

Step 1: The first reconciliation

A 2022 study in the American Journal of Health-System Pharmacy recorded how ten hospitals in four health systems handled a suspected case. They reported similar methods. First, the investigator "manually reconciled ADC and electronic medical record (EMR) records," which required between 4 and 20 hours depending on the number of transactions by the clinician in question (Knight et al., AJHP 2022). ADC is the automated dispensing cabinet.

That first reconciliation is a three-way comparison for one person over a defined period: what the cabinet says was dispensed, what the medication administration record says was given, and what the cabinet says was wasted. A dose that was withdrawn but has no administration entry, or a waste entry without a witness, is where the questions begin. The practical request to whoever owns the data is: every transaction for this user, this drug class, this window, from each system, exported at once, with timestamps.

The study also warns that the hours add up. The same authors report that with a consolidated dataset, users described the same audit as taking 10 to 30 minutes. Whether or not you buy that product, the finding shows where the delay lives: in pulling and lining up records from separate systems. Know in advance who can export each one.

Step 2: The upstream records

If the first reconciliation does not explain the gap, the investigation moves upstream. The same paper lists the five kinds of data its team had to consolidate, which doubles as a list of the records an investigator may need: medication administration records, cabinet dispense and waste transactions, internal inventory transactions into and within the central pharmacy, wholesaler purchases, and employee time-clock records (Knight et al.). It also names the cross-checks they enable:

  • Wholesaler shipments against receipts in the inventory system, to see whether a full or partial shipment was diverted in transit or on receipt.
  • Inventory moves from the central pharmacy against the restocking of the cabinet.
  • Cabinet dispenses against documented administration and any later waste.
  • Clock-in and clock-out times against the times of medication transactions, to see whether someone accessed drugs while not working.
  • Pain scores in the record against dispensing patterns, to see unusually high dispensing in relation to episodes of high or low pain scores.

A pharmacy without cabinets has the same shape in different records: wholesaler invoices and receiving logs, the perpetual inventory and the last physical count, dispensing records for the drug in question, and the schedule of who was on shift. Under 21 CFR 1304.04 these records must be kept for at least two years, so the request is only as good as your retention (eCFR, 21 CFR 1304.04).

Step 3: The interview

The study's authors note that, once suspected diversion has been detected, "the value of an interview is key," and that the Mayo Clinic estimates approximately 75% of suspected diversion investigations have been brought to closure by a confession by the person diverting (Knight et al.). That figure is the authors' citation of another institution's estimate, so treat it as indicative rather than as a benchmark. They also stress that hospitals need investigations conducted by trained drug diversion professionals and a leadership team of departmental heads, both to protect innocent workers and to confirm the software's alerts.

In the hospitals studied, the interview was of the clinician suspected of diversion or their supervisor, and it came after the record audit had shown further evidence. The order matters: you interview with the reconciliation in hand, so you are asking about specific transactions, not about suspicion. Our investigation playbook sets out the sequence, and the red flags guide covers what prompted it.

Step 4: The filings, which run from discovery

Federal law adds clocks that the investigation cannot pause. Under 21 CFR 1301.76(b) the registrant must notify the DEA Field Division Office in writing of a theft or significant loss within one business day of discovery, and file a complete DEA Form 106 within 45 days after discovery (eCFR, 21 CFR 1301.76). The rule lists factors for deciding whether a loss is significant, including the quantity lost in relation to the type of business and the specific substances lost. Document why you decided a discrepancy was or was not significant; see our Form 106 guide.

States add their own. These are examples from the state pages we have read and linked to the rule text, and they differ in when the clock starts:

  • Ohio: the Board of Pharmacy immediately on discovery, by telephone or another method it determines, and a written report within 45 days (Ohio).
  • Michigan: notice to the department within 45 days after completion of the investigation (Michigan).
  • Georgia: a report to the Georgia Drugs and Narcotics Agency within three days of discovery, for pharmacies (Georgia).
  • New Jersey: notification of the Drug Control Unit upon discovery, with the DDC-52 form (New Jersey).
  • Illinois: a copy of the DEA Form 106 sent to the Department's Drug Compliance Unit concurrently with the federal filing (Illinois).

The Michigan rule is worth a second look: because its clock runs from the end of the investigation, a short investigation starts that state's clock sooner than a long one, while the federal and several other state clocks have already started at discovery. Each state page links the rule so you can confirm the current text.

What regulators look at afterward

Public Department of Justice settlements show what DEA reviewed once a matter reached it. After a hospital reported that 400 tablets of oxycodone 30 mg had been lost or stolen from an inpatient pharmacy, DEA diversion investigators found, according to the government's contentions, that the health system failed to keep an accurate inventory, failed to keep the controlled substances in a secure location, and failed to properly supervise employees; it agreed to pay $150,000 and there was no determination of liability (DOJ, District of New Jersey). In a 2026 settlement of $3.2 million, the government alleged at least 628 violations of recordkeeping and security requirements, among them failing to notify DEA of theft or loss, keep accurate records, complete biennial inventories, maintain complete order records, and provide effective controls (DOJ, Eastern District of California). A methadone clinic's 2022 settlement followed a DEA inspection that covered physical and operational security, required recordkeeping, receipt and dispensation records, and a recordkeeping accountability audit (DOJ, Southern District of Ohio).

Read together, the lists are the investigator's checklist turned around: an accurate inventory, secure storage, supervision, timely notification of theft or loss, order records, and biennial inventories. Whatever your own investigation finds about an individual, assume the regulator will also ask about those.

A request list you can prepare now

Before you need it, write down who can export each of the records above, how far back each system keeps data, and how long an export takes. Test it once by pulling one user's last 30 days from every system and timing the effort. If it takes a day and a half, you have learned that your investigations will take that long too. Keep the interview checklist, the Form 106 template and the state notification contacts in one place.

To check where your program stands, run the hospital self-assessment.

Sources: Knight T, May B, Tyson D, McAuley S, Letzkus P, Enright SM. Detecting drug diversion in health-system data using machine learning and advanced analytics. Am J Health Syst Pharm 2022 (PMC9353695, open access; the first two authors list an analytics vendor as their affiliation); 21 CFR 1301.76 and 1304.04 (eCFR, current); U.S. Attorney's Office press releases, District of New Jersey, Eastern District of California and Southern District of Ohio; the state pages linked above. Educational content, not legal advice.