Surgical Assessments Need Better Feedback, Not Just More Comments

In surgical training, a growing stream of workplace assessments is meant to show whether residents are progressing toward independent practice. Yet a larger pile of evaluations does not automatically create better learning. A commentary in Global Surgical Education – Journal of the Association for Surgical Education argues that Entrustable Professional Activities, or EPAs, have helped address the quantity of feedback while leaving a more difficult problem unresolved: whether the comments are specific, constructive and useful enough to change what a trainee does next. The distinction matters because feedback is not simply a record of performance. It is an educational signal that should help a resident understand what happened, why it mattered and how to improve during the next clinical encounter.

EPAs are defined professional tasks that a trainee may eventually be trusted to perform with decreasing supervision. In general surgery, they provide a structured way to document observations made during real patient care rather than relying exclusively on examinations or end-of-rotation impressions. The American Board of Surgery introduced EPAs in 2023, and the commentary describes their arrival as a major shift in workplace assessment. By prompting more evaluations during clinical encounters, the system has generated substantially more data about residents’ performance. That increase responds to a longstanding weakness in surgical education: faculty members often observe trainees frequently but record feedback inconsistently, leaving learners with limited information about their development.

More frequent assessments can improve visibility, but the educational value of each assessment depends heavily on its narrative content. Summative evaluations completed asynchronously, including end-of-rotation reports and competency committee reviews, may arrive too late to guide performance in the moment. They can also be broad, using general language that describes a resident as capable or progressing without identifying the particular behavior that produced that judgment. EPA forms offer more opportunities to capture comments close to the clinical event, but the commentary emphasizes that free-text feedback remains highly variable. A short statement of appreciation may be encouraging, yet it does not necessarily tell a trainee how to handle tissue, anticipate the next operative step, organize a case or respond to an unexpected finding.

The discussion draws on work by Moore and colleagues examining narrative feedback in general surgery through an established classification framework. That analysis found differences in the kind of feedback residents received according to practice readiness, case complexity and trainee and evaluator gender. Residents considered ready for practice and those involved in more complex cases were less likely to receive coaching or formative comments. Male-identifying residents received less evaluative feedback, while male faculty were less likely to provide narrative comments classified as specific or coaching. These findings do not establish why the differences occurred, but they show that assessment language is shaped by the context and people involved. Counting completed forms alone can therefore conceal meaningful variation in what residents are actually being told.

High-quality feedback is often described as forward-looking because it connects observation with a realistic next step. A useful comment might identify a concrete action, explain the clinical reasoning behind it and indicate how the learner can demonstrate improvement. It should be tailored to the task and the individual, constructive rather than merely punitive, and feasible within the pressures of clinical work. Reviews of formative feedback have also recommended validated approaches such as the Quality of Assessment of Learning system and guided workplace-based assessment tools. Such frameworks can help educators move beyond praise or criticism toward comments that support growth and inform future entrustment decisions. The goal is not to make every evaluation lengthy, but to make each one sufficiently precise to be acted upon.

Improving the comments will require faculty development, but the commentary cautions against treating all educators as though they need the same intervention. Some faculty may complete few narrative assessments and first need support understanding the purpose and mechanics of EPA documentation. Others may routinely write comments but rely mostly on evaluative language, such as whether a resident met expectations, without adding coaching that explains how to advance. A precision-based approach would use an educator’s feedback profile to identify the specific skill requiring attention. That could include the proportion of assessments containing narrative text, the frequency of coaching language or the specificity of recommendations. Such individualized development is practical in principle, although leaders still face familiar barriers involving faculty time, participation and the scale of implementation.

The same EPA records that describe residents could also reveal how faculty teach. Because evaluators generate portfolios of comments over time, institutions could analyze patterns across individuals, divisions or departments. These profiles might help identify educators who need assistance and provide leaders with evidence about engagement in assessment. Artificial intelligence and large language models could make this review more manageable by synthesizing hundreds of comments into longitudinal summaries. For a resident, such a summary might reveal repeated strengths in tissue handling alongside recurring suggestions to improve operative anticipation or case progression. For a program director, it could highlight performance trends earlier and support a more individualized learning plan. The proposed use is analytical and supportive: technology would organize information that is difficult for humans to review at scale, rather than replace clinical judgment.

AI could also intervene closer to the moment when feedback is written. A language model might flag a comment that lacks a specific observation or prompt an educator to add an actionable coaching recommendation while preserving the original intent. Related work has explored whether AI can help faculty convert fixed-mindset wording into growth-mindset language, and emerging research is examining whether naturally occurring intraoperative teaching conversations can be analyzed and summarized. Such tools could capture educational guidance that currently disappears after an operation, when both teacher and learner are working under intense cognitive demands. But the potential benefits come with substantial limits. Algorithms can reproduce bias, generate inaccurate or fabricated interpretations and create an appearance of objectivity that their human-designed rules do not warrant. Any system would need oversight, privacy protections and careful validation against expert judgment.

The central test for EPAs is therefore not how many assessments a program collects, but whether the resulting feedback helps residents become better surgeons. More observations can provide a stronger foundation for detecting patterns, but quantity without specificity may simply increase documentation burden for educators and review burden for trainees. Surgical education leaders will need to pair structured assessments with clear expectations for narrative quality, targeted faculty development and systems that make useful feedback easier to produce and interpret. AI may eventually reduce some of the administrative work, but it cannot substitute for meaningful observation or the relationship between a resident and an educator. The promise of EPAs will be realized only when the growing volume of assessment data is matched by comments that are timely, equitable, technically grounded and actionable in the operating room.

One implication of this discussion is that feedback quality should be treated as a property of the assessment system, not merely as an individual writing skill. An EPA can record a supervision decision and still provide little explanation of the behaviors that led to it. Conversely, a brief comment may be educationally valuable if it identifies a precise observation and connects it to a feasible next action. This distinction is important for program review because completion rates and entrustment ratings describe whether documentation occurred, whereas narrative analysis asks whether the documentation can support learning.

Classification frameworks offer a way to make that second question more visible. By distinguishing evaluative language from coaching, and general statements from specific comments, educators can examine patterns that would otherwise remain hidden in a large assessment database. Such categories should not be mistaken for a complete measure of educational value: a comment’s usefulness also depends on clinical context, timing, the learner’s prior experience and whether the suggested action is realistic. Still, a shared vocabulary can help faculty discuss feedback more consistently and make development goals more concrete.

The reported differences by case complexity and practice readiness also suggest that feedback systems should not assume that the most advanced or challenging encounters automatically generate the richest coaching. Faculty may interpret a near-independent resident’s performance as requiring less explanation, even when complex cases contain important opportunities to discuss judgment, anticipation and prioritization. Similarly, an assessment indicating readiness for practice does not eliminate the need for developmental guidance. Ongoing coaching can clarify how a resident should extend performance, manage variation and prepare for responsibilities that exceed the specific encounter being evaluated.

Attention to evaluator and trainee characteristics adds an equity dimension to EPA design. If narrative specificity or evaluative language varies across groups, the resulting record may provide some residents with clearer developmental information than others. The source evidence identifies differences but does not establish their causes, so corrective efforts should avoid assuming that bias is the only explanation. Programs could instead use patterns as prompts for review, examine assessment contexts and ensure that faculty receive guidance on observing and documenting performance in behavior-based terms. Monitoring should support improvement rather than turn isolated comments into judgments about individual educators.

For artificial intelligence to assist responsibly, its role would need to be defined around transparency and human review. A system might identify that a comment contains praise without an observable behavior or a recommendation without a clear next step, but that signal would be a prompt for the faculty member, not a final rating. Any generated summary should remain traceable to the underlying comments so residents and program directors can check whether recurring themes are genuine. The caution is especially important when language models analyze intraoperative conversations, where context, speakers and intended meaning may be difficult to determine.

Evaluation of an AI-supported feedback process should therefore include more than efficiency. Educational leaders would need to ask whether comments become more specific and actionable, whether faculty preserve clinical nuance, and whether summaries accurately represent the record across different learners and evaluators. The source commentary frames these tools as augmentation because observation, judgment and the educator–learner relationship remain central. A successful implementation would reduce avoidable documentation and synthesis work while preserving accountability for the feedback that ultimately informs resident development and decisions about entrustment.

Subject of Research: Quality of narrative feedback in surgical education

Article Title: EPAs gave us feedback quantity, but what about quality?

Article References: Jou, K., & Holmstrom, A. L. (2026). EPAs gave us feedback quantity, but what about quality?. Global Surgical Education – Journal of the Association for Surgical Education, 5(1), Article 172. https://doi.org/10.1007/s44186-026-00576-6

Image Credits: AI Generated

DOI: 10.1007/s44186-026-00576-6

Keywords: surgical education, feedback, Entrustable Professional Activities, workplace assessment, resident training, faculty development, competency-based education, artificial intelligence, EPAs, gave, quantity, quality