Spaced repetition and flashcards for medicine

• A flashcard is only useful if it makes you pull the fact out of memory. Reading the answer first is not studying; it is sightseeing.
• Forgetting is steep in the first day or two and then slows. Reviewing just before you would have forgotten is what converts short-term memory into long-term memory.
• A scheduler such as Anki does the timing for you: a card you recall easily waits longer; a card you fail comes back sooner. The algorithm is useful, but the cards you write decide whether it works.
• One card, one fact. Cloze deletions, short answers and well-chosen images beat lists, paragraphs and multi-step questions. Cards built by other people teach you their organisation, not yours.
• Reviews come before new cards, every day, or the deck collapses under its own weight. Sleep and honest marking of "again" are part of the method, not optional extras.

By the end of this page you should be able to:

1. Explain the forgetting curve and the spacing effect in your own words, and say why testing beats re-reading.
2. Describe in outline how a spaced-repetition scheduler chooses the next interval for a card.
3. Write atomic, well-formed cards using basic, cloze and image formats, and recognise the bad patterns that produce "leech" cards.
4. Set a daily review routine that clears the queue before adding new cards, and estimate when a deck has become too large to sustain.
5. Summarise what the evidence in medical students does and does not show, and match your card content to what licencing examinations actually test.
6. Combine spacing with sleep, realistic workloads and honest self-assessment rather than with anxiety and card hoarding.

This is a pathway page about how to learn medicine, not about a disease. It sits in the wider group of pages on how to study, how to use question banks and how to learn on the wards. The mechanism is memory, and the tool is the flashcard, but the real subject is a discipline a student practises every day for years.

There are two common failures this page tries to prevent. The first is the student who reads a textbook three times, feels familiar with it and cannot recall any of it a week later. The second is the student who downloads a deck of 40,000 cards, reviews 800 every day for a fortnight, burns out and stops completely. Both are problems of method, not of effort.

Medicine is a large, slow-moving, heavily memorised subject with a long training period. A fact learned in the first week of a course may be tested in an examination four years later and used in practice twenty years later. Four features make spaced repetition fit that profile.

• The body of knowledge is too large to hold in working memory all at once, so it must be externalised into a system that brings each item back at the right time.
• Most of what a student must know is discrete and testable: a normal value, a drug dose, the first step of an algorithm, a named association. Those map cleanly onto cards.
• Feedback is immediate and unambiguous: either you recalled the answer or you did not. Guessing that you "sort of knew it" is itself the error the method catches.
• The payoff is delayed, which is exactly the condition under which short cramming fails and distributed practice wins. ^[1]

A note on scope. Spaced repetition is a study technique, not a treatment, a diagnosis or a guarantee of examination success. Nothing here replaces supervised clinical training, local protocols, or a teacher. Cards encode what you already understand; they do not substitute for understanding.

The forgetting curve is the observation, first described by Hermann Ebbinghaus in the 1880s and since replicated in many forms, that memory of new material drops steeply in the hours after learning and then levels off. The sharpest loss happens early: a large fraction of what is learned today is gone tomorrow if nothing is done. After that the curve flattens, and what survives the first day tends to survive the first week.

Three points matter for a medical student.

• The drop is not a personal failure. It is a property of memory. Brains discard what they have not used, on the sensible assumption that it may not matter.
• Re-reading stops the drop but does not rebuild the memory. Seeing the answer again feels like knowing it. That feeling is the illusion of competence, and it is the main trap in medical revision.
• The best moment to review is not immediately after learning, when the memory is still warm, but after some forgetting has happened, when pulling the answer back requires genuine effort. That effort is what strengthens the trace. ^[8]

The curve is not a number you can quote for a given topic. It varies with how meaningful the material is, how it was first encoded, how tired the learner is and how much related knowledge exists. A student with a good mental model of acid-base balance forgets a new buffer calculation more slowly than one who memorised the formula by rote. Spaced repetition does not remove that advantage; it builds on it. Every fact that gets connected to something else lasts longer, which is why the cards you write should give a reason wherever the source supplies one.

Two separate findings combine to make the method work.

The spacing effect is that the same total study time produces better long-term retention when it is distributed across several sessions than when it is massed into one. A meta-analytic review of classroom studies found that distributed practice outperformed massed practice overall, with larger effects when the material was more complex and when the interval between sessions was longer. ^[3] The effect is not mysterious. Each spaced encounter arrives after some forgetting, so the memory is rebuilt from a weaker state, and re-learning after a gap is more durable than re-learning immediately.

Retrieval practice, or the active recall of information from memory, strengthens memory and comprehension. ^[1] Testing yourself is not merely an assessment of what you know; it is a learning event. In the classic test-enhanced learning paradigm, taking a memory test improved long-term retention more than spending the same time restudying, even when the students felt the restudy had helped more. ^[9] The implication for cards is direct: a card must demand recall, not recognition.

These are among the most robust findings in cognitive psychology and medical education reviews summarise them for the health professions. ^[1] They do not promise that spacing cures poor prior understanding or that retrieval fixes bad notes. They promise that given decent input, spaced retrieval produces more durable memory than cramming and re-reading, and that gap is large enough to matter for examinations.

A spaced-repetition program such as Anki keeps a record of every card: how many times it has been reviewed, how long the last interval was, how easy it was to recall and how many times it has been failed. The algorithm it applies is conceptually simple even where the detail is proprietary.

Three terms are worth knowing.

• Interval: the number of days until the card is next shown.
• Ease factor: a number that rises when you recall a card easily and falls when you fail it. It is the slope of your personal forgetting for that card.
• Lapse: a failed review. Each lapse resets the interval to near zero and reduces ease. A card that lapses repeatedly is a "leech": either the card is badly written or the knowledge behind it has not been learned properly.

The scheduler is a servant, not a strategy. It cannot fix a bad card, a card that asks two questions at once, or a student who presses "good" on cards they half-knew. The programs are genuinely effective at delivering the right item at the right time, and randomised trials of spaced education in health professionals show gains in knowledge and skill, with the largest effects where retrieval is built into the design. ^[3] But observational data in medical students also shows mixed results for course examinations, so the scheduler is not a substitute for good input and good sleep. ^[2]

The quality of a deck is the quality of its worst card. The rules follow from the mechanisms above: one retrieval attempt per card, at a sensible level of difficulty, with unambiguous answers.

The practical rules.

• One fact per card. If the answer has "and", you probably have two cards. Split them.
• Write in your own words, from your own notes, not by pasting a sentence from a source. The recall of writing the card is itself a learning event, and the gate that rejects copied text exists for a reason.
• Ask for the smallest useful unit. "What is the first-line antibiotic for X in pregnancy?" is a good card. "Everything about antibiotic choice for X" is not a card at all.
• Make the front unambiguous. If two answers could be correct, the card is broken. Either narrow the wording ("the most specific test") or split the card.
• Keep the answer short. A single phrase or sentence is ideal; a paragraph means the card is doing the wrong job.
• Use cloze deletions for definitions and associations, because the surrounding text keeps the answer meaningful. Emptycloze questions ("What is X?") teach guessing, not knowledge.
• Add images when the clinical sign is visual: a rash, an ECG, a fundus, a gross specimen, a labelled diagram. Image occlusion cards test recognition of parts, which is a different skill from naming them.
• Avoid "orphan" cards that do not connect to anything: isolated mnemonics, obscure eponyms, and trivia. If you cannot explain why a fact matters, do not make a card for it.
• Keep formatting plain. Markdown, tables or HTML inside a card render unpredictably. Numbers and units are fine; raw angle brackets before numbers are not: write "less than 60" or use the proper symbol, never a stray less-than sign followed by a space.

A short example of a bad card and its repair.

Bad front: "List the causes of microcytic anaemia." Bad back: "Iron deficiency, thalassaemia, anaemia of chronic disease, sideroblastic anaemia, lead poisoning." This card tests a list, cannot be graded cleanly and gives no clinical context. Better: five separate cards, each pairing one cause with its mechanism or its most useful test, for example "Microcytic anaemia with a high red-cell count and normal ferritin: which hereditary disorder?" answered "Thalassaemia trait (high HbA2 on haemoglobin electrophoresis)."

There is no correct number of cards for everyone, because it depends on how fast you review, how dense the cards are and how much time you have. What is controllable is the heuristic that keeps the system alive.

• Clear reviews before adding new cards. The review queue is the debt; new cards are new borrowing. An empty review queue at the end of each day is the sign that the system is healthy.
• Add new cards only as fast as you clear them the next day. If today's reviews already fill the time you have, adding another batch multiplies tomorrow's load. Adjust daily, not weekly.
• Measure time, not cards. Size the daily new-card batch to fit inside the time you can spare, and let the queue, not a quota, decide.
• Prefer consistency over volume. Fifty cards every day for six weeks beats three hundred cards on Saturday. The spacing depends on regularity.
• Suspend, do not delete and do not tolerate, cards that are wrong, ambiguous or no longer relevant. A deck you do not trust becomes a deck you do not open.
• Watch for overload signals: the review backlog growing for more than three days, average time per card rising, and the urge to mark "good" on cards you half-knew. Those mean the new-card rate is too high or the cards are too long. ^[2]

A seeded deck from a senior student can be a good starting point, but treat it as a library, not a syllabus. Import cards only in the topics you are currently studying, and only after you have edited them into your own words. A card you did not write is a card whose reasoning you have not checked. Copying a deck at full speed is one of the commonest ways to turn spaced repetition into efficient time-wasting.

The mechanical rule is simple: reviews first, new cards second, and if there is no time for both, reviews only. The reason is in the curve. A review missed today is not postponed; it is forgotten, and the card collapses back to short-term memory. A new card postponed by a day costs one day. A review postponed by a day can cost the card.

This is why the evidence on medical students rewards regularity rather than heroic bursts. In a quasi-experimental study of final-year students over six months, the group using digital flashcards at intervals of 1, 3, 7, 14 and 28 days scored 16.24 on the post-test against 11.89 for the group using traditional methods, with the pre-test scores the same (11.42 against 11.58); over 90 percent of the flashcard group reported better retention, engagement and confidence. ^[4] The useful lesson is not the particular interval but the discipline of returning on schedule. A personalised scheduler achieves the same without a fixed table.

These are the failure modes that the evidence and experience describe.

• Card hoarding. A huge deck feels productive and is not. If you will not review it, you are collecting, not learning. Suspend whole sub-decks for rotations or examinations you are not preparing for.
• Copying other people's decks. You inherit their ambiguities and their level of simplification. Worse, you skip the act of writing, which is where the understanding forms.
• The illusion of knowing. Pressing "good" because the answer looks familiar is false data; the scheduler assumes it is true. Mark honestly: if you would not have written the answer unaided, it is "again" or "hard".
• Asking too much per card. Lists, tables and multi-step clinical questions create cards that cannot be graded and that rot.
• Ignoring leeches. A card failed six times is telling you something. Either the card is badly written or the underlying concept is not yet understood. Stop and fix the cause, not the count.
• Studying instead of sleeping. Memory consolidation happens during sleep. A review done at one in the morning encodes worse and consolidates worse, and the hours cost the next day.
• Confusing familiarity with fluency. Cards test recognition of your own wording. Examinations test application to new vignettes. Cards are necessary but not sufficient; keep doing question banks and clinical reasoning alongside them.
• Abandoning the method after one bad day. A missed day produces a backlog, which produces avoidance, which produces failure. The repair is to lower the new-card rate and clear the backlog a little each day, not to reset the deck.

The honest summary is that active retrieval and spacing are well supported in the laboratory and in health professions, that the effect in real medical courses is positive but not magic, and that the evidence for any particular Anki workflow is still thin.

Three conclusions to draw, and one caution.

• The classics hold. Dunlosky's 2013 review of study techniques, Roediger and Karpicke on test-enhanced learning and Cepeda's work on distributed practice are the intellectual basis of most of what this page recommends, and medical education reviews cite them. ^[8][9][10] Kerfoot's spaced-education trials and Cook's syntheses of technology-enhanced learning are the main lines of evidence inside medical education itself. ^[3][11]
• The effect exists and is decent. Where researchers measure knowledge, gains show. Step 1-specific work suggests a point range of 4 to 13 points for heavy Anki users, which is life-changing and not small. ^[2]
• Most of the evidence is observational and imperfect. Studies are small, rely on self-reported use and carry risk of bias. No randomised trial compares "well-written personal cards" with "cramming from a textbook" in the way that would settle the argument. Treat the method as well-supported and worth trying, not as law.
• What helps most is not the app. It is the behaviour the app demands: doing the work, every day, honestly. Two different students using the same scheduler will not get the same result if one of them writes bad cards and presses "good" on guesses.

There is also a regional note. In the United Kingdom and Ireland, the same principles apply, but the examination blueprints (PLAB and UKMLA under the General Medical Council, and the medical schools' own progress tests) weight clinical reasoning and professional judgement heavily. In North America the licencing examinations (USMLE and, historically in Canada, the MCCEE) reward a large, well-organised factual base. Match your card content to the examination you will actually sit: vignette-shaped cards for reasoning, fact cards for the base, and always cards that encode why, not just what.

The boards publish blueprints and practice material; they do not prescribe study tools. Learn the blueprint and use it to decide what deserves a card.

The National Board of Medical Examiners states that "Self-assessments are a robust learning tool that help you define a study plan, gauge progress and more." ^[6] Its self-assessments and subject exams are the closest thing to the real examination most students get, and the standard advice is to use them diagnostically rather than as a score to chase.

The USMLE site offers content outlines and free preparation materials, including interactive experiences and practice questions, and defines Step 1 as an assessment of "important concepts of the sciences basic to the practice of medicine, with special emphasis on principles and mechanisms underlying health, disease, and modes of therapy." ^[7] That wording is a direct guide to what a card should encode: principles and mechanisms, not isolated mnemonics. Step 2 CK then tests application to patient care, so cards for later years should be vignette-shaped.

The General Medical Council, which sets the UKMLA, and the Medical Council of Canada, whose examination (the MCCEE) was discontinued as a single high-stakes test and whose licencing assessments now sit within a wider system, similarly publish frameworks rather than methods. ^[12][13] The lesson is the same everywhere: the examination tests what you can use, not what app you used. Cards are preparation; they are not evidence of readiness by themselves.

Memory consolidation is sleep-dependent, and sleep loss harms the very thing the cards are trying to build. A scoping review of sleep and academic achievement in first-year university students found that shorter sleep, particularly below 6 to 7 hours per night, was generally associated with poorer academic achievement, and that poor sleep quality and chronic sleep deprivation also linked to lower performance. ^[5] That is a population finding, not a personal prescription, but it is strong enough to justify treating sleep as part of the study plan rather than as the leftover.

Practical points.

• Protect a fixed wake time and count backwards for the study window. Irregular sleep times waste more learning than they save.
• Review early in the day when attention is highest. Cards done with half-closed eyes encode badly and still get recorded as learned, which corrupts the scheduler's data.
• Stop adding new cards in the week before a high-stakes examination, and move to maintenance mode: only reviews, plus question banks and targeted repair. New material in the last week mainly adds anxiety.
• Leave at least one day a week with no new cards, so errors can be absorbed and the backlog can be cleared. Sustainable beats optimal.
• Use light and routine to anchor the clock, and follow national sleep guidance for your age group. If you are routinely getting less than the 6 to 7 hour range the review singles out, address that before adding more cards.

Spaced repetition does not require a laptop or a subscription.

• Anki is free on desktop and Android, and runs offline once downloaded.
• Where bandwidth is limited, sync occasionally and keep the card set small and local. A deck of 2,000 well-chosen cards is more useful than a deck of 40,000 imported ones.
• A paper Leitner box works the same way: five compartments, moved forward one step on recall and back to the first on failure. The scheduling is coarser, but the principle is identical.
• If no printer or device is available, use a simple exercise book with cards cut into strips and review in your own hand. The memory benefit comes from the act of retrieval, not from the screen.
• Write cards from the textbooks and guidelines you already have, in your own words, citing them. Never transcribe another student's deck by hand: you inherit their mistakes without their context.

• Forgetting is steep early and then flattens. The best review is the one done after some forgetting, not immediately after reading.
• Testing yourself improves retention; re-reading creates an illusion of knowledge. Retrieval practice is one of the most robust findings in learning research. ^[1][9]
• Distributed practice beats massed practice, especially for complex material with longer intervals. ^[3][10]
• A scheduler chooses the next interval from your recall, ease and lapse record. It solves timing; it does not solve bad cards or dishonest marking.
• One fact per card. Basic questions, cloze deletions and image occlusion cards. Avoid lists, paragraphs and multi-step questions on one card.
• Reviews first, new cards second, every day. If there is no time for both, reviews only.
• Card quality is set by the worst card in the deck. Write in your own words and split, suspend or rewrite anything you do not trust.
• Evidence in medical students is positive but observational. Step 1 data suggest a 4 to 13 point advantage for heavy users. ^[2]
• Match cards to the examination blueprint: principles and mechanisms, and vignette-shaped reasoning, not trivia. ^[6][7]
• Spaced repetition and good sleep are part of the same system. Below roughly 6 to 7 hours per night is associated with poorer academic achievement. ^[5]
• Start small, stay honest, and stop adding cards when the queue stops being cleared. Card hoarding and shared decks are the commonest failure modes.

• Study skills: how to study medicine as a whole, how to read, take notes and organise a revision year.
• Question banks: how to use timed, mixed practice alongside flashcards, and how to learn from wrong answers.
• Clinical examination and communication pages: facts alone do not pass; procedural and interpersonal skills need supervised practice.
• Sleep and fatigue in students: the physiology and practical management of a short night, and when lack of sleep becomes a health problem rather than a study choice.
• Evidence-based medicine: how to appraise the reviews cited here, including heterogeneity and risk of bias.
• Statistics and reasoning: positive predictive value, sensitivity, specificity and the difference between association and causation, which matters for interpreting the studies above.