Are we assessing medical students' recall or readiness for practice?
Frontiers in Medicine · 09.04.2026 · 13
DOI: 10.3389/fmed.2026.1801030 ↗Kurz gefasst
Fachartikel in der Zeitschrift Frontiers in Medicine, online erschienen am 09.04.2026. Verfasst von Waseem Jerjes und 2 weiteren Autor:innen. Beteiligte Einrichtung: Department of Primary Care and Public Health, Imperial College London, London, United Kingdom. Die Arbeit wurde bisher 1-mal zitiert (Crossref) und nennt 25 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
Medical education is full of apparent certainties. Pre-clinical students can memorise enzyme pathways, rare syndromes, and lists of adverse drug reactions well enough to perform strongly in single-best-answer (SBA) examinations. Yet the transition from student to newly qualified doctor often exposes a more uncomfortable reality: clinical work is not a rehearsal of tidy facts but an ongoing negotiation with uncertainty, time pressure, competing risks, incomplete information, and the needs of other people in a system. The problem is not that knowledge is unimportant. The problem is that much of what we currently reward and amplify is a narrow and unreliable proxy for readiness to practise. When assessment culture predominantly prizes rapid recall of decontextualised details, learners logically optimise for what is rewarded. Students build their weeks around question banks, flashcards, and revision strategies that primarily target short-term exam performance. Surveys of learner assessment preferences repeatedly show that the methods learners perceive as "fair" or "efficient" are often those that best align with predictable recall demands rather than authentic practice performance (1)(2)(3)(4)(5). That alignment is reinforced by an expanding industry of algorithmic learning platforms that personalise drills and predict forgetting, further intensifying the message that the main task is to memorise more, faster.The educational consequences are visible. When learners rely heavily on rehearsal of disclosed items or narrow patterns of remembered answers, their conceptual grasp becomes fragile-especially when they face new presentations that share surface features with familiar cases but differ in clinically important ways. Interventions that shift learners away from rote memorisation towards conceptual understanding can produce measurable improvements, precisely because memorisation-based performance can mask deeper misunderstandings until clinical complexity forces them to the surface (2,(6)(7)(8)(9)(10)(11)(12). The move from "exam-ready" to "practice-ready" should not feel like cognitive whiplash; yet it often does. This mismatch is not simply pedagogical. It reflects the collective choices medical schools and regulators make about what they count, what they tolerate, and what they expect learners to become. If assessment drives learning, then assessment design is not a technical afterthought-it is the curriculum.-The assessment culture problem: what we reward becomes what we teach SBA examinations are widely used because they are scalable, reliable, and accepted. They are also relatively inexpensive compared with many performance-based assessments. But reliability and convenience are not enough. The central question is whether our dominant assessment methods validly sample the capabilities that constitute safe, effective clinical work.A useful lens is van der Vleuten's utility formula, which frames assessment quality as the product of reliability, validity, educational impact, acceptability, cost, and feasibility (3). SBA formats can score highly on reliability, acceptability, and feasibility, and well-constructed SBA items can assess applied reasoning when they are carefully blueprinted and written to test interpretation rather than simple recognition. However, validity and educational impact still depend on item quality and the degree to which questions require meaningful application rather than recognition of rehearsed fragments. Reviews applying utility-based approaches to SBA item sets highlight the practical challenge: many assessments drift towards what is easiest to construct and mark, rather than what best captures real clinical judgement (4). When that drift occurs, assessments may still look rigorous but function educationally as memorisation incentives.Even when we add oral examinations, vivas, and OSCEs, we do not automatically solve the validity problem. Performance assessments can introduce their own distortions when scoring is influenced by factors unrelated to the construct being measured. Constructirrelevant variance is well described in high-stakes testing, including effects of examiner strictness, interactional style, and other factors that can bias outcomes while appearing "objective" (5). Evidence from OSCE settings suggests that examiner seniority and experience can be associated with bias in certain domains such as communication scoring (6). Language and accent effects also matter: if examiner judgements shift with a candidate's accent, fluency, or linguistic style, then the assessment is partly measuring communication presentation rather than clinical competence (7,(13)(14)(15)(16)(17). In practice, these effects can reward confident delivery and penalise cautious reasoning, even when the latter is safer.In short, the problem is not the existence of SBA assessment itself, but a culture in which recall-heavy written formats and performance under artificial conditions are allowed to stand in disproportionately for readiness. This is especially problematic when the transition to practice requires not just knowing, but knowing how to act when knowledge is incomplete.-Knowledge mastery is not memory performance: designing "desirable difficulties" for durable, usable knowledge If we accept that knowledge matters, we must become more precise about what kind of knowledge we want learners to have. Clinical knowledge is not simply stored; it must be retrieved under pressure, adapted to the person in front of you, and integrated with constraints and values. That means medical education should differentiate between shortterm performance and durable learning that transfers.The learning science concept of "desirable difficulties" offers a pragmatic route forward. Strategies such as spaced retrieval, interleaving, and effortful recall can improve long-term retention and transfer by strengthening memory in ways that mimic real use (8,(18)(19). Importantly, these strategies are not just about remembering more; they are about remembering in a form that supports discrimination between similar presentations, adaptation to context, and resistance to the illusion of competence created by cramming.In practical terms, this suggests an assessment culture that privileges repeated, low-stakes retrieval over isolated, high-stakes recall events. It also suggests that the most meaningful learning signal is not whether a student can reproduce information once, but whether they can retrieve and apply it weeks later, in slightly altered circumstances, with competing priorities.However, learning strategies alone cannot overcome an assessment regime that continues to pay primarily for short-term recall. If we want students to engage in spaced, cumulative learning, we must build a system where cumulative evidence matters more than a single performance spike.-Programmatic assessment: shifting from snapshots to longitudinal judgements of competence Programmatic assessment aims to replace single, high-stakes "snapshots" with many smaller data points collected over time, triangulated with narrative feedback, and used to support both learning and judgement (9). The promise is not that every micro-assessment is perfect, but that a programme of assessment-properly designed-produces a more valid picture of competence than any single tool.International experience of implementation shows both the appeal and the difficulty. Programmatic assessment requires not only cultural change but operational infrastructure: protected faculty time for observation and feedback, administrative support for scheduling and record-keeping, agreed formats for narrative comments, and reliable processes for collating evidence across hospital and community placements. Institutions also need clear governance structures, including who reviews aggregated evidence, how disagreements between assessors are handled, how concerns are escalated, and what level of evidence is sufficient for progression decisions. These requirements increase workload during implementation, especially in distributed programmes where multiple supervisors contribute data under service pressure, but they are necessary if longitudinal judgement is to be credible, fair, and educationally useful (10,(20)(21)(22). A feasible starting model may therefore be phased rather than wholesale, using a limited number of recurring assessment activities, shared rubrics, scheduled review points, and a trained decision-making group before broader expansion. Yet those challenges are also the point. Readiness for practice is inherently longitudinal. No single exam sitting can credibly represent a learner's ability to practise safely across contexts, teams, and uncertainty levels.A programmatic approach also creates space to reward the "invisible work" of clinical practice: how learners prioritise, how they recognise risk, how they use resources, how they communicate uncertainty, and how they modify their plan when the situation changes. These are precisely the capabilities that traditional recall-heavy assessments under-sample. Done well, programmatic assessment can also improve fairness. It reduces the influence of one-off performance fluctuations and shifts decision-making towards patterns, trends, and consistency across contexts. It can reduce the dominance of construct-irrelevant variance by relying less on a single oral performance and more on repeated sampling of comparable constructs. Crucially, it can support the message that looking things up and asking for help are not admissions of incompetence but elements of safe practice.Modern clinical practice is resource-rich. Clinicians consult guidelines, formularies, pathways, colleagues, and increasingly decision-support tools. Yet many assessments still act as if safe practice is primarily an internal memory performance.A more defensible model is to explicitly assess resource use: how learners locate information, judge its trustworthiness, apply it to the patient's context, and document their reasoning. National assessments have already demonstrated that large-scale evaluation of applied prescribing capability is feasible. The UK Prescribing Safety Assessment provides an instructive example of assessing prescribing competence at scale, including decision-making that mirrors real prescribing tasks rather than pure recall (11). The broader implication is straightforward: if we can assess safe application in prescribing, we can design similar resource-permitted tasks in other domains, including diagnostic reasoning, escalation decisions, and safety-netting. This is also where the realities of AI must be handled with maturity. Learners already use AIenabled tools for study, summarisation, and question generation. The appropriate educational response is not to pretend this is not happening, but to define explicitly what is being assessed: verification, contextualisation, justification, and safe decision-making. In other words, the assessed skill becomes not "producing an answer unaided" but "arriving at a safe decision through appropriate methods." -Tolerance of "not knowing": uncertainty as a competency rather than a defect A distinctive feature of clinical practice is that uncertainty is not the exception; it is often the default. A curriculum that trains learners to behave as if certainty is always available can inadvertently teach two unsafe behaviours: false certainty (acting as if the answer is known) and paralysing uncertainty (not acting when action is needed).There is a growing argument for making uncertainty an explicit curricular and assessment domain, including teaching learners how to work with ambiguity, weigh probabilities, and plan next steps when information is incomplete (12). This is not about celebrating indecision. It is about teaching safe uncertainty: a stance where the clinician can articulate what is unknown, identify what information would change the plan, and implement safetynetting and escalation strategies.The idea of "safe uncertainty" has long been described as a practical clinical positionneither rigid certainty nor anxious indecision, but a disciplined and communicable way of working with what cannot be fully known (13). Translating that into assessment means designing tasks that deliberately include incomplete data, evolving information, or competing risks, and then scoring not only the final diagnosis or plan but the quality of uncertainty management. At an earlier stage, a learner might simply identify what remains uncertain; at a more advanced stage, they should be able to state a working diagnosis, assign a confidence level, identify what information would change the plan, and specify clear review or escalation triggers. That includes stating confidence levels, setting review triggers, documenting thresholds for escalation, and communicating the plan to patients and team members.A key equity advantage also emerges here. If we normalise "looking it up" and "here is how I will close the gap," we reduce the unfair advantage that may accrue to those who can perform fluent recall under observation, and we align assessment with what clinicians actually do.-Adaptive expertise: assessing the ability to adjust when the pattern does not fit If the future clinician's main task were only to reproduce learned responses, memorisation would be enough. But clinical work frequently demands adaptation: to atypical presentations, multimorbidity, polypharmacy, resource constraints, and patient preferences. This is where the concept of adaptive expertise becomes essential. Educational approaches that promote adaptive expertise aim to develop clinicians who can apply routine solutions efficiently when appropriate, but also innovate, reframe, and adjust when routine approaches fail (14).Assessments should therefore sample adaptation, not just reproduction. The cognitive psychology distinction between fast, intuitive pattern recognition and slower, analytical reasoning provides a useful structure for designing such sampling. Clinical reasoning often moves between rapid impressions and deliberative checking; safe practice depends on knowing when to shift mode, especially when initial patterns are misleading (15). Learners should be assessed on their ability to recognise when a case does not fit the expected pattern, to slow down, to test alternative hypotheses, and to revise decisions as new information emerges. This is not an argument against pattern recognition; it is an argument for making the switching behaviour visible and assessable. Key-features problems, evolving-data cases, and short "reasoning notes" can capture whether a learner is anchored to a first impression or can adjust appropriately under constraint.Even when knowledge is robust and uncertainty is managed safely, clinical decisions are rarely "purely technical." They take place within context: the patient's goals, the family's capacity, the service's constraints, and the ethical trade-offs involved.Educational interventions are more likely to create meaningful impact when they recognise that outcomes emerge through layered mechanisms, not simple input-output relationships. The concept of "principled adaptation" is helpful here: rather than judging whether a learner replicates an idealised model answer, we can judge whether they adapt decisions responsibly while staying anchored to sound principles (16). This matters because many real clinical decisions involve choosing among imperfect options, balancing competing harms, and making time-sensitive compromises.Contextualised care is not a soft add-on; it is a measurable clinical competency. The ability to integrate contextual factors-such as socioeconomic constraints, health literacy, competing priorities, and system barriers-can be assessed and should be viewed as part of competence, not merely professionalism (17). This has direct implications for assessment design. If we continue to score primarily for recall and "standard answers," we risk training clinicians who can pass exams but struggle to deliver appropriate care when context alters risk, feasibility, or acceptability.A more authentic assessment culture would therefore reward (a) explicit reasoning, (b) safe uncertainty practices, (c) appropriate resource use, and (d) contextualisation of care decisions. These elements align with what patients actually experience as quality and safety.This concern is also supported by the broader preparedness-for-practice literature. Reviews of newly qualified doctors consistently identify gaps not in all domains, but in precisely those areas that depend on applied, situated performance: prescribing, clinical reasoning, acute care, prioritisation, handover, and navigating the ward environment (18,19). Seen in that light, the issue is not that graduates know nothing; it is that our assessment systems can over-certify recall while under-sample the work of early practice. The solution is not to "abolish memorisation," discard SBA examinations altogether, or lower standards. Clinical knowledge is foundational, and well-designed knowledge-based formats still have an important place. The issue is and we must rigorous for knowledge while that the assessments that learning also sample the capabilities that make knowledge usable and safe in shift would include we should durable, usable knowledge as the not short-term recall performance. That means spaced retrieval, interleaving, and cumulative low-stakes as educational with the evidence on high-stakes assessments they should be by longitudinal that and reward we should programmatic assessment not as a but as an explicit for of prescribing reasoning clinical and be to produce a judgement of competence than any single exam sitting (9). is and experience shows that feasibility depends on more than educational need faculty protected time, and practical systems for scheduling evidence, and it consistently across settings (10,(20)(21)(22). evidence must also be low-stakes data can support but progression decisions should with a trained group that reviews aggregated evidence over time, clear decision and deliberately to bias and is single that every can but the operational questions are which assessment tasks are how often they will be who will review and what support will make the Yet the is by the we are readiness for we should explicitly assess safe resource use and verification, rather than the that memory performance is the of clinical readiness. Prescribing assessments show that applied competence can be assessed at (11). can be for use, risk escalation and safety-netting. In a where information is the key skill is not facts but using them we should make uncertainty and contextualisation explicit assessment a place in and "safe uncertainty" should be as a clinical stance that can be and assessed (13). Assessment should therefore reward not only the answer when certainty is but the ability to and safely what remains learners to a and then to can also into a learning signal rather than to care should be as an with ways to whether learners can integrate context into decision-making (17). This is not about being it is about making decisions that are actually safe, and feasible for real practical way to these elements is to use a of recurring assessment that comparable data over time and them in a or learner of learner performance in medical with of use longitudinal and learning that their not in but in making multiple low-stakes data visible in place for longitudinal review by learners and faculty In practice, be to a of competence, and by both a and a short narrative a retrieval a reasoning a prescribing or an uncertainty during routine data then be at by the learner and for and by a trained group when progression decisions are with to trends, recurring evidence, and between The is not it is to make patterns across time that judgement on evidence rather than Seen this the can be as simple as a or what mat
Schlagwörter
Stichwörter der Autor:innen
Autor:innen und Institutionen
- Department of Primary Care and Public Health, Imperial College London, London, United Kingdom.
Zitieren
APA
Jerjes, W., Chan, S. C. C., & Majeed, A. (2026). Are we assessing medical students' recall or readiness for practice?. Frontiers in Medicine, 13. https://doi.org/10.3389/fmed.2026.1801030
Vancouver
Jerjes W, Chan SCC, Majeed A. Are we assessing medical students' recall or readiness for practice?. Frontiers in Medicine. 2026;13. doi:10.3389/fmed.2026.1801030
BibTeX
@article{Jerjes2026,
author = {Jerjes, Waseem and Chan, See Chai Carol and Majeed, Azeem},
title = {Are we assessing medical students' recall or readiness for practice?},
journal = {Frontiers in Medicine},
year = {2026},
volume = {13},
doi = {10.3389/fmed.2026.1801030}
}
EUVIMED