White paper

The case log gap

What the published evidence says about resident and fellow procedure logging — and what has actually been shown to fix it.

Nudges fail. Structure works. Across every study we could find, telling residents to log better changed nothing, and changing the conditions of the task changed the number.

Compiled 7 September 2026 · Written for program directors, program coordinators, and the residents and fellows who keep the log.

How to read this

Every identifier on this page was resolved against Europe PMC or the publisher of record before publication, and each one links to the record so any sentence here can be checked in a click. Almost every study below is single-center, and each treats a different source as truth — billing records, an operating-room schedule, an anesthesia information system, an extraction pipeline. None of those is a gold standard. What makes the finding credible is not any single study; it is that six research groups, in six specialties, using six different comparators, all point the same way.

We make software in this category, so read section six first if you would rather start with what argues against us.

The short version

Nine findings, each verified against its record.

Cases in the operating-room record that reach the ACGME logOrthopedics 91%, general surgery 69%, neurosurgery 58% PMID 32950429: 72%
Anesthesiology: automatically generated total vs self-reportedMedian anesthetics, P=.0014. Emergent cases: 151 vs 54.5 PMID 36545371: 1226.5 vs 1134.5
Emergency medicine: extracted procedures vs manually loggedManual logging captured roughly 56% of what was found PMID 38265802: 7,617 vs 4,291
General surgery interns: captured volume vs ACGME-logged volumeCases per intern-year — a mean gap of 50.2, P<.001 PMID 29055744: 176.5 vs 126.3
Neurosurgery: ACGME log against billed procedures2,150 billed, 1,749 logged PMID 28966813: −18.65%
Cardiology fellows: weekly procedures, manual then automatedP=.007. Procedural elements 20.9 → 40.2, P<.001 PMID 33680308: 17.1 → 24.2
Residents reporting a logging backlog of three months or moreAcross seven surgical specialties PMID 27049679: 51.4%
Protected administrative time: cases logged within 30 daysP<.001. Within 7 days: 28.7% → 37.2% PMID 32147467: 52.7% → 69.9%
An intraoperative reminder to logLag time increased for PGY-3 and PGY-4 residents PMID 34786618: No improvement

The gap is real, large, and consistent in direction

Six independent research groups, six specialties, six different reference standards. They agree: the procedure log a trainee submits is not an accurate record of the procedures they performed, and the error runs one way. It undercounts.

Surgery, across three specialties

Graduated residents' ACGME logs were compared against their institution's own operating-room records across 11,923 cases and 46 residents. Overall logging accuracy was 72%, and it varied enormously by program: orthopedics 91%, general surgery 69%, neurosurgery 58%. PMID 32950429

Single academic center, 2009–10.

Neurosurgery, measured independently

Against 2,150 billed procedures, the ACGME log undercounted by 18.65% and a separately maintained departmental log by 12.88%. Thirteen of 22 categories were under-logged in both systems at once, which argues against a single measurement artifact. PMID 28966813

14 residents, single center.

General surgery interns

An institutional capture system recorded a mean of 176.5 operative cases per intern-year against 126.3 in the ACGME log — a mean difference of 50.2 cases, P<.001. PMID 29055744

Five cohorts, single center.

Anesthesiology

Case logs generated from the clinical record found a median 1226.5 anesthetics against 1134.5 self-reported, P=.0014. The category-level differences are starker than the total: emergent cases 151 vs 54.5, intracerebral nonvascular open 64 vs 30. An earlier study at a different institution found more than half of residents under- or over-reported their total by at least 5% against an anesthesia information management system. PMID 36545371, 21156981

42 residents across three graduating classes at one center; the earlier study is a separate single center.

Emergency medicine

A silent extraction system captured 7,617 procedures where residents had manually logged 4,291 over the same period — a 78% increase, meaning manual logging accounted for roughly 56% of what was found. The pipeline itself was 99.5% accurate against ground truth, with sensitivity 97.4% and specificity 100%. PMID 38265802

47 residents, one year, single center.

Cardiology fellows

Weekly documented procedures rose from 17.1 ± 6.8 to 24.2 ± 6.1, P=.007, on moving from manual to automatically generated logs. Total procedural elements nearly doubled, from 20.9 ± 12.5 to 40.2 ± 11.4, P<.001. PMID 33680308

Single fellowship program.

It is not a diligence problem

The obvious explanation is that residents do not care enough. The evidence says otherwise: they know the log is wrong, they say so when asked, and the conditions they work in are hostile to the task in ways that have been measured.

Residents already know

In a survey across seven surgical specialties, 36.3% considered the log system inaccurate, 48.1% used the wrong metric for logging surgeon-versus-assistant role, 51.4% reported backlogs of three months or more, 64.2% found it hard to navigate and 68.8% found it burdensome. Orthopedic residents nationally reported that only 22% routinely logged clinic procedures, 20% joint injections and 13% office casts or splints. PMID 27049679, 24500594

The conditions are hostile to the task

Emergency medicine residents are interrupted constantly: 37% of interruptions fall during order entry and 41.5% during charting, and residents fail to return to the interrupted task 17% of the time. Vascular surgery residents spend roughly a fifth of their working time in the chart, much of it after hours. Time-motion work on internal medicine interns documents the same squeeze. PMID 40552315, 31010521, 30985861

So the task is deferred, and deferral is where the record decays

Nobody opens the ACGME Case Log System in the room. The case is written down somewhere and entered later — and the 51.4% three-month backlog is that gap made visible. Recall across three months is precisely what the accuracy studies above are measuring. PMID 27049679

Entry cost is measurable, and it moves

A proof-of-concept logbook built on iOS Shortcuts cut mean entry time from 135 seconds to 65 seconds for the minimum data set, P<.001, 95% CI 61.6–77.7. A mobile logbook improved documented compliance for central line insertions in an intensive care unit. PMID 33983132, 35051900

What has been tried, and the pattern in the results

This is the most useful thing in the literature and it is rarely stated plainly. Sort the interventions by whether they asked the resident to try harder or changed the conditions of the task, and the results sort themselves.

Passive
An intraoperative reminder to log (anesthesiology)
No improvement. Lag time increased for PGY-3 and PGY-4 residents. PMID 34786618
Structural
A protected weekly administrative hour (surgery)
Cases logged within 7 days rose 28.7% → 37.2%; within 30 days, 52.7% → 69.9%, P<.001. PMID 32147467
Structural
A machine-learning-assisted logging platform (three general surgery programs)
Logging volume rose — and fell again when the assistance was reduced. PMID 35193831
Structural
Silent extraction from the clinical record (emergency medicine)
78% more procedures captured than residents logged by hand. PMID 38265802
Structural
Automatically generated procedure logs (cardiology fellows)
Weekly procedures 17.1 → 24.2, P=.007. PMID 33680308
Structural
Automated notification (obstetric anesthesiology fellows)
Median cases per quarter 0 → 3, P=.007; share of antenatal cases 14% → 52%, P<.001. PMID 34041735
Structural
A case log fed by the department's own record (ophthalmology)
Cataract capture 85% → 91% — a real gain that still left 9% behind. PMID 35991093
Structuralupstream
Evening out case distribution by assignment (anesthesiology)
More even distribution across residents, and residents perceived greater fairness. PMID 33447648

Nudges fail. Structure works.

The single intervention that asked residents to try harder is the single intervention that did nothing — and it made senior residents worse. Every intervention that changed the conditions of the task moved the number.

The obstetric-anesthesiology result is narrow: one subspecialty, one center, one under-captured category, and it measures case participation rather than logging accuracy.
The machine-learning platform study is vendor-linked. We note it because we are also a vendor, and you should weigh it accordingly.

Automation is the answer, and it is not available

A 2026 narrative review in the Journal of Graduate Medical Education identified eight semiautomated case-log systems and zero fully automated end-to-end systems, and reported that the semiautomated ones consistently improved logging frequency, accuracy and completeness against manual self-report. A 2026 feasibility study found local and commercial language models reaching F1 ≈ 86.7 on case-log annotation, with an estimated 35+ hours of annotation saved per resident per year — radiology, single institution, not a deployed tool. PMID 42005891, 42220338

Automation works, and still does not close the gap

The ophthalmology platform moved cataract capture from 85% to 91% and stopped there. A residual gap survives even a case log wired into the institution's own systems. PMID 35991093

It is an institutional project, not a resident's option

Every automated system above needed record access, informatics staff and institutional sponsorship. A resident who wants their log to be right this year cannot build one. Neither, realistically, can most programs.

Which leaves the intervention a resident can actually reach

If reminders do nothing and automation is real but out of reach, the remaining lever is the cost of capture itself: make the record cheap to create at the point of care, and carry context forward so the same fact is never entered twice. The entry-time work and the assisted-logging work point at that from opposite directions. PMID 33983132, 35193831

To be explicit about our own position: Capnolog does not connect to a hospital record system, and is not building one. Nothing described in this section is something our software does.

This is, transparently, an argument that favors the category of product we make. We think the evidence genuinely supports it. We also think the next section is the one to read hardest.

What the evidence does not show

We went looking for the studies that would undercut this document. These are they.

Logged volume is not a clean proxy for training quality

When ACGME case minimums were introduced in orthopedic sports medicine, mean logged shoulder arthroscopies fell from 109.8 to 82.0 (P=.025) and knee arthroscopies from 178.6 to 124.8 (P=.006), with total logged cases dropping from 2045.5 to 1699.3 (P=.038). Residents at the 10th percentile still met the minimum every year. Whether that reflects less operating or less logging cannot be separated — but either reading should make anyone cautious about reading competence off a count. PMID 34195633

Not all variation between residents is a logging artifact

Coefficients of variation across residents at the same PGY level ran 6.0–34.1%, then 11.2–66.8% across two years, and the variability persisted when ACGME data was compared against independent ACS NSQIP data. A meaningful share of the spread is real difference in experience. Better tools will not equalize it. PMID 30195662

Residents judge their own role well, even when they count badly

Resident self-assessment of operative involvement showed substantial agreement with independent staff assessment: ICC 0.80, kappa 0.67. The failure is in tallying and coding over time, not in judgment about a given case. PMID 22854152

Program directors and case logs disagree about what matters

In otolaryngology, 47 of 111 procedures were rated essential by at least 75% of program directors — and some ACGME key-indicator procedures were not among them, while nearly half of the procedures directors called essential were logged fewer than ten times by recent graduates. PMID 41724906

41% director response rate; otolaryngology, not anesthesiology.

The consequence nobody has measured

We could not find a single peer-reviewed study, in any specialty, tracing the path from a resident being short of case log minimums to a documented remediation or extended-training outcome. Reasons for data-prompted site visits are documented, and the most common trigger there was survey noncompliance rather than case log data. This is a genuine gap in the literature, and anyone who tells you what a shortfall costs a resident is extrapolating. PMID 34178287

And the accreditor knows about the friction

For regional anesthesiology and acute pain medicine fellowships, the ACGME made case logging optional for two full years and deferred citations for procedural-experience shortfalls — a phased posture that suggests the compliance friction described here is well understood by the body that requires the logs.

What this means if you run a program

The coordinator carries this

A national survey of 6,372 program and institutional coordinators found 38.4% screening positive for burnout, 23.9% with moderate-to-severe depression, 30.4% with moderate-to-severe anxiety, and 27.9% saying they were likely or very likely to leave their job within a year — with chasing residents among the named dissatisfaction drivers. A separate analysis of coordinator job descriptions found only 31.2% covered all nine ACGME-defined responsibility domains. The chasing is real work, and it is not in most job descriptions. PMID 39705400, 41694770

The requirements are likely to move

A 60-member national Delphi panel of anesthesiology program directors and practicing anesthesiologists reached consensus on new case log minimums ahead of anticipated 2027 ACGME revisions: 40 arterial lines, 20 central lines, 10 fiberoptic intubations, 10 one-lung ventilation, 10 neuromonitoring, 10 cardiac and 10 lung POCUS, 20 non-operating-room anesthesia cases, peripheral nerve blocks raised 40 → 60 and cardiac cases 20 → 25. These are consensus recommendations, not adopted requirements. But a program whose residents log at 72% accuracy against a rising bar has an arithmetic problem, not a motivational one. PMID 42610591

Independence

The ACGME's published policy is explicit, and we quote it rather than paraphrase it because it constrains us:

“does not endorse, approve, or recommend any specific data collection product”
“strictly prohibits vendor access to its systems”

It further states that a vendor claiming ACGME endorsement, or claiming to meet ACGME standards, is making a false representation.

Capnolog makes no such claim. We are not endorsed by, approved by, affiliated with or partnered with the ACGME or any specialty board. Our backend never contacts acgme.org; where our software helps with the ACGME Case Log form it runs inside the resident's own browser session, under the resident's own login, and the resident presses Submit. We describe a mechanism. We claim no relationship.

References

Every identifier above was resolved against Europe PMC on 7 September 2026. Each links to its PubMed record.

Primary sources

ACGME Case Log System
ACGME Non-Endorsement Policy on Data Collection Software
ACGME Program Requirements for Graduate Medical Education in Anesthesiology (2025, reformatted)
ACGME Review Committee for Anesthesiology FAQs
American Board of Anesthesiology certification requirements

The reachable half of the problem

A resident cannot buy protected administrative time and cannot commission an extraction pipeline. What is left is the cost of writing the case down in the first place, and the context that gets carried forward so nothing is typed twice. That is the part of this we work on.

See what this looks like for your program Get Capnolog

The full working document — including the studies that did not make this page, and the method used to resolve each identifier — is available on request.