Accession Number Mismatches: Why Studies Land Unmatched in PACS

A CT tech finishes a trauma scan at 2:40 a.m. The worklist was timing out, so she typed the patient’s name and a best-guess accession number at the console and pushed the images. By morning the study sits in the PACS QC queue with no order attached, no prior comparison loaded, and a radiologist who can’t sign a report against it. Nobody did anything unusual. That’s the problem.

Unmatched studies rarely come from one dramatic failure. They come from small gaps between the RIS or EHR, the interface layer, the modality, and the archive, each of which treats the accession number slightly differently. This guide follows the number end to end, lists the specific ways it breaks, and lays out a QC and prevention routine that a PACS team can run without heroics.

What the Accession Number Actually Does

The accession number is the order’s identity inside the imaging department. The ordering system (RIS, or the radiology module of the EHR) assigns it when the exam is scheduled, and every downstream system uses it as the key that ties pixels back to a request, a patient encounter, a billing event, and eventually a signed report.

In the DICOM header it lives in tag (0008,0050), Accession Number. The IHE Radiology Technical Framework spells out the intended model in its appendix on the topic: the accession number identifies the order, and the Requested Procedure ID distinguishes individual procedures when one order contains more than one. Many sites collapse the two into the same value because most of their orders map to a single procedure. That shortcut works until the day it doesn’t.

Here’s why the number matters so much more than the patient name or date of birth. Your PACS can usually find the right patient by MRN. What it can’t do without the accession number is answer the questions the radiologist and the billing team actually ask:

• Which order does this study fulfill?
• Is the order still open, or was it already resulted?
• Which ordering physician gets the report?
• Which charge drops when the report is signed?

When the header carries a value the PACS can’t match to an open order, the study isn’t lost. It’s stranded. It sits as a valid set of images with no workflow attached, and stranded studies have a way of being found late, often after the referring clinician calls to ask where the result is.

How the Number Travels From Order to Report

The path is easier to troubleshoot once you write it down as a chain of handoffs. In a typical IHE Scheduled Workflow setup it looks like this:

1. Order placed. A clinician orders an exam in the EHR. The RIS (or the EHR’s radiology module) creates the order and assigns the accession number.
2. Order sent. The ordering system sends an HL7 order message, an ORM^O01 in older v2 interfaces or an OMI^O23 in sites that follow the newer IHE option, to the departmental systems. IHE calls this the Procedure Scheduled transaction.
3. Worklist entry created. A broker or worklist provider translates that HL7 message into a DICOM Modality Worklist entry: patient demographics, accession number, Requested Procedure ID, scheduled procedure step, modality, and station.
4. Modality queries the worklist. The tech selects the patient at the scanner. The modality copies the accession number and related keys from the worklist response into the header of every image it creates.
5. Images stored. The study arrives at the PACS, which looks up the accession number and patient ID against the orders it has received and marks the study as matched or unverified.
6. Report sent back. After the radiologist signs, the reporting system sends an ORU^R01 result message carrying the same accession number, so the EHR can file the report against the original order.

Every step is a translation, and every translation is a chance to change the value. In practice, the interface engine or hl7 integration software that turns order messages into worklist entries is where a surprising share of mismatches start. Mapping rules decide which HL7 field becomes (0008,0050), whether a site prefix gets added, whether leading zeros survive, and what happens when an order is updated or cancelled. If that mapping is wrong, the modality faithfully copies a wrong value into every image, and the PACS faithfully refuses to match it.

So when you investigate an orphan, don’t start at the scanner. Start by pulling the original HL7 order and the worklist entry side by side. If they already disagree, the modality was never going to get it right.

A few things to check on that comparison:

• The field mapping: which OBR or IPC field feeds the accession number, and is it the same field the RIS considers authoritative?
• Transformations: padding, prefixes, case changes, or character stripping applied by the broker.
• Update handling: does an order change (ORM with a change or replace control code) update the existing worklist entry, or create a second one?
• Timing: how long between the order message arriving and the worklist entry becoming queryable at the modality?

Eight Ways a Study Ends Up Unmatched

Most orphan studies fall into a handful of repeat patterns. Knowing which one you’re looking at tells you who has to fix it and whether it will happen again tomorrow.

1. Manual entry at the modality

The worklist is down, slow, or filtered so tightly that the patient doesn’t appear, so the tech types the demographics and accession number by hand. One transposed digit and the study won’t match. This is the single most common cause at most sites, and it clusters around downtime, night shifts, and emergency workflows.

2. The worklist was never queried

Some techs start a study from a previous patient’s entry or from a “new patient” template out of habit, even when the worklist works fine. The header ends up with an empty accession number or one copied from the wrong exam.

3. Cancelled or rescheduled orders that get renumbered

An order is cancelled and re-entered, or rescheduled in a way that the RIS treats as a new order. The tech scans against the old worklist entry, which still showed on the console, and the images carry an accession number that no longer exists as an open order.

4. One order, several studies (or the reverse)

A CT chest, abdomen, and pelvis may be ordered as one exam and acquired as one study, or ordered as three exams and acquired in one continuous scan. If the order structure and the acquisition don’t line up, you get either one study with three orders competing for it or three studies pointing at one order. The IHE split between accession number and Requested Procedure ID exists for this case, but only if both systems use it.

5. Patient merges after acquisition

The patient arrives as a trauma alias, gets scanned, and is later merged into their real record with an ADT^A40. If the merge reaches the RIS but not the PACS (or reaches them in a different order), the study’s patient ID no longer matches the order’s patient ID, even though the accession number is correct. The IHE Patient Information Reconciliation profile describes how these updates should flow.

6. Outside studies with foreign accession numbers

A CD or a cloud share from another hospital carries that hospital’s accession numbers. Import it as is and the PACS sees a valid number that belongs to nobody in your RIS. Some of those values will even collide with your own numbering by coincidence.

7. Truncation from length limits

DICOM defines Accession Number with the SH (Short String) value representation, which allows 16 characters maximum. If the RIS generates a longer value, or the broker adds a prefix that pushes it past 16, something downstream truncates it. Different systems truncate differently, some from the right and some by rejecting the value outright, so the header no longer matches the order.

8. Format differences between sites

Multi-site groups and newly acquired imaging centers often run different numbering formats: one site uses a numeric sequence, another prefixes with a facility code, a third pads with zeros. When studies move between sites or into a shared archive, “00451873” and “451873” look like two different orders.

Quick comparison

Cause	Where it starts	Typical symptom in PACS	Who usually fixes it
Manual entry	Modality console	Accession number off by one or two characters	PACS admin with tech confirmation
Worklist not queried	Modality workflow	Blank or copied accession number	PACS admin; training issue
Cancel/renumber	RIS order management	Number valid in history, not open	RIS team plus PACS admin
One-to-many orders	Order design	Study matches one of several orders	RIS analyst and radiology lead
Late patient merge	ADT feed	Accession matches, patient ID doesn’t	HIM or interface team
Outside import	Import station	Number unknown or colliding	Import staff at intake
Truncation	Interface mapping	Shortened value	Interface analyst
Site format differences	Numbering standards	Same exam, different string	Enterprise imaging governance

Finding Orphans Before Radiologists Do

An orphan you catch at 7 a.m. is a five-minute correction. One you find when a clinician escalates is an incident. The difference is mostly about whether anyone looks at the unverified queue on a schedule.

Watch the unverified or QC queue

Most PACS platforms hold studies that fail to match in some form of exception list: unverified, QC, or “not reconciled.” Assign an owner by shift. A queue that belongs to everyone gets checked by no one, and the older a study gets in that list, the harder it is to reconstruct what happened at the scanner.

Run a daily reconciliation report

The queue only shows studies the PACS knows are unmatched. A reconciliation report catches the other direction: orders marked complete in the RIS with no study in the archive, and studies matched to orders that were later cancelled. A useful daily report compares:

• Orders with an “exam completed” status versus studies received for that accession number
• Studies received in the last 24 hours with no matching open order
• Studies where the accession number matches but the patient ID or exam code doesn’t
• Image counts per study against what the modality reported sending, if your modalities send performed procedure step messages

Trend the counts by modality and by shift. If one CT scanner produces most of your orphans, you’re probably looking at a worklist configuration or a training issue on that unit, not a system-wide problem.

Ask the techs

Reports tell you what broke. The techs can tell you why. A two-minute conversation about why a worklist entry didn’t appear often surfaces a filter setting, a station AE title mismatch, or a scheduling habit that no report will show.

Fixing Them Without Breaking the Audit Trail

Correcting an orphan means changing DICOM header values, and that is a clinical data change, not housekeeping. Treat it with the same discipline as any edit to the medical record.

Decide who can edit headers

Limit header editing to a small, named group: PACS administrators and trained QC technologists. Front-line techs should flag problems, not fix them in the archive. Broad edit rights produce well-meant corrections that quietly attach images to the wrong patient, which is a far worse outcome than an unmatched study.

Follow a fixed QC sequence

A consistent routine keeps corrections defensible:

1. Confirm the patient’s identity from at least two identifiers against the RIS and EHR.
2. Find the correct open order and accession number. If no order exists, get one created rather than inventing a value.
3. Check for a duplicate study already matched to that order.
4. Update the header fields through the PACS QC tool (accession number, and patient ID if needed), never by re-sending edited files from a workstation.
5. Confirm the study now matches, priors load, and the worklist in the reading environment shows it.
6. If the study was already sent to other systems (a VNA, an AI service, a teleradiology partner), make sure the correction propagates there too.
7. Record the reason code for the correction.

Keep the audit trail complete

Every correction should log who changed what, from which value to which value, and why. That record protects the department when a report is challenged, and it doubles as your best source of data for prevention. If the reason codes say “manual entry, worklist unavailable” forty times a month, you don’t have a QC problem. You have a worklist uptime problem.

Handle the edge cases deliberately

Some corrections need more than a header edit. If the wrong patient’s images were reported, the report itself has to be addressed through your incident and amendment process. If a study belongs to two orders, decide with the radiology lead whether to link it to both, split it, or create a combined order. Write those decisions down once so the next person doesn’t improvise.

Prevention: Closing the Gaps at the Source

Cleaning the queue every morning is necessary, but the goal is a queue that barely fills. Prevention lives upstream, in configuration and governance.

Enforce the worklist

Configure modalities so that worklist selection is the default path and manual entry requires a deliberate step. Some sites require a reason when a tech bypasses the worklist. Pair that with a documented downtime procedure: a printed or exported schedule, a block of reserved accession numbers, and a reconciliation step when systems come back.

Monitor the interface, not just the servers

An interface engine can be “up” while silently queuing messages or rejecting a malformed field. Set alerts on:

• Message queue depth and age for the order feed into the worklist
• Negative acknowledgments (AR or AE) from the receiving system
• Gaps in message flow during hours when orders should be arriving
• Worklist entries created with blank or out-of-format accession numbers

Standardize the accession format

Pick one format across every site that shares an archive, and keep it comfortably under the 16-character DICOM limit after any prefix is added. Document whether leading zeros are significant, which characters are allowed, and how site codes are encoded. For outside studies, decide on a single rule: either reassign a local accession number at import, or preserve the original in a separate field and map the study to a local order.

Test mismatches before go-live

New modalities, RIS upgrades, and interface changes are where many orphan patterns begin. Add test cases that target the failure modes above, not just the happy path:

• An order placed, then cancelled and re-entered before the scan
• An order with three procedures under one accession number
• A trauma alias patient merged after acquisition
• An accession number at exactly 16 characters, and one that would exceed it after prefixing
• An outside study imported with a colliding number
• Worklist downtime with manual entry and later reconciliation

Run each one through to a signed report and confirm the ORU lands on the right order in the EHR. If you only test until the images arrive in PACS, you’ve tested half the chain.

Review the numbers monthly

Bring the reconciliation trends, correction reason codes, and interface alerts to a short monthly review with the RIS, PACS, and interface owners in the same room. Most mismatch patterns only become obvious when someone sees all three data sets together.

FAQ

Is an unmatched study a data loss risk?

Not usually. The images are stored and retrievable. The risk is clinical and operational: a delayed report, missing prior comparisons, a charge that never drops, or, in the worst case, a correction that attaches images to the wrong patient.

Should techs be allowed to fix accession numbers at the modality?

Before sending, yes, if they caught a typo against the worklist or order. After the study reaches the PACS, corrections should go through the QC workflow so they’re logged and propagated.

What’s the difference between the accession number and the Study Instance UID?

The accession number identifies the order and is assigned by the ordering side. The Study Instance UID is a globally unique DICOM identifier for the study itself, usually generated by the modality or provided in the worklist. A study can be perfectly valid by UID and still be orphaned by accession number.

How should we handle outside studies?

Pick one policy and apply it at the import station every time. Either assign a local accession number tied to a local order, or keep the outside number in a separate attribute and reconcile manually. Mixing both approaches is how collisions happen.

Why do some orphans show the right accession number?

Usually because the patient ID changed after acquisition, through a merge or an identity correction, and the PACS didn’t receive or apply the update. Check the ADT history before editing the study.

Where to Start This Week

If your QC queue already feels like a second worklist, don’t try to fix everything at once. Start with three moves:

• Assign a named owner to the unverified queue on every shift and track the count daily.
• Pull ten recent orphans and compare the HL7 order, the worklist entry, and the DICOM header for each. The pattern usually shows up by the fifth.
• Check your accession format against the 16-character SH limit after every prefix and transformation the broker applies.

Sources

• DICOM PS3.5, Data Structures and Encoding, Section 6.2 Value Representation (NEMA)
• DICOM PS3.6, Data Dictionary, tag (0008,0050) Accession Number (NEMA)
• IHE Radiology Technical Framework, Volume 1, Appendix A: Clarification of Accession Number and Requested Procedure ID
• IHE Wiki, Patient Information Reconciliation profile