Clinical reasoning games: what each type trains

A diagnostic game trains you to reach a hypothesis; a management game trains the next step, deciding what to do with it. Both fit into the study of the clinical cycle, but they exercise different parts of reasoning, and knowing which part each one trains helps you choose where to put your time.

NEXO is a daily medical diagnosis game: one clinical case a day, six progressive clues, and ICD-10 proximity feedback across five levels.

This text organizes clinical reasoning games by the decision stage they exercise, summarizes what published studies with the Human Dx platform show, explains why feedback and confidence calibration matter, and closes with the limits of research on games. Game formats by course phase are in games for medical students, the general evidence on gamification in gamification in medical education and virtual patients in virtual patients.

Key points

• Clinical reasoning has stages: collecting data, forming hypotheses, choosing management. Each type of game exercises some more than others.
• Virtual patient tools more often improved specific measures, such as data collection, diagnostic ideas and management, than general measures of problem solving.
• Decision games under time pressure have already been tested in randomized trials with emergency physicians, with an effect on triage.
• Studies with the Human Dx platform show that short cases with open answers serve to measure diagnostic performance and that combined diagnoses from several participants are more often correct.
• Confidence and accuracy go together only loosely; immediate feedback is what allows one to be adjusted to the other.
• Research on games still has few trials, heterogeneous outcomes and many measures without validation.

Summary

• The stages of clinical decision
• The types of game by the stage they train
• What the Human Dx data show
• Feedback and calibration
• Limits of studies on games
• How to combine the types
• Frequently asked questions
• Checklist for choosing a reasoning game

The stages of clinical decision

Clinical reasoning is not a single skill. A scoping review on methods of assessing clinical reasoning, with 377 articles, concluded that there are many methods and that each one aligns with different components of this complex construct (Daniel et al., 2019). The authors recommend assessment programs that combine complementary methods to cover all components (Daniel et al., 2019).

The same goes for training. If assessment requires several methods, training also calls for more than one format. A game can be great for one stage and not have been made for another, and that is not a flaw: it is focus.

Collecting data

The first stage is deciding what to ask, what to examine and what information to seek. It appears strongly in virtual patients, in which you choose the questions and the case responds. The review on virtual patient tools found more consistent improvement in case-specific measures, such as data collection, than in general measures (Plackett et al., 2022).

Forming hypotheses

The second stage is moving from findings to hypotheses and ordering them. It is the stage of differential diagnosis, covered in detail in the text on differential diagnosis with clues. Diagnostic games, in which the case arrives gradually and you need to name the condition, concentrate the training here.

Deciding management

The third stage is deciding what to do: treat, observe, refer, transfer. The "key features" of a clinical problem, used in clinical decision exams, were defined as the critical steps in identifying and managing that problem (Bordage et al., 1995). In the validation done for the Canadian national exam, physicians outside the committee confirmed 92% of these critical steps in the retrospective study and 94% in the prospective one (Bordage et al., 1995).

Interpreting data under uncertainty

A fourth skill cuts across the others: interpreting a new piece of data when the picture is still ambiguous. The script concordance test was created to measure exactly that, and the review of its validity evidence supports its use to assess the interpretation of clinical data under uncertainty (Lubarsky et al., 2011). How this format relates to probability before and after a test is in the text on pretest probability.

The types of game by the stage they train

The text about games for medical students separates games by format. Here the cut is different: which decision the game asks you to make and what studies show about that decision.

Hypothesis games

These are games in which the case arrives in parts and the task is to reach the diagnosis. They train the move from findings to hypothesis and the revision of the hypothesis with each new piece of information. Presenting the case gradually, in serial clues, is the most used way of teaching clinical reasoning, perhaps because it directly imitates the diagnostic activity of physicians (Schmidt and Mamede, 2015).

The strong point is repetition: short cases allow you to see many presentations in a short time, and the feedback comes right after the guess. The text about clinical reasoning shows how this repetition helps build illness scripts.

Data gathering and management games

These are virtual patients in which you choose questions, tests and management, and the case responds to the choices. In the review of 19 studies with undergraduate students, the analyses of data gathering, diagnostic ideas and management showed improvement in 72% of cases, against 43% in those of knowledge application, thinking flexibility and problem solving (Plackett et al., 2022). The authors suggest that case-specific measures may be more sensitive to change, given that clinical reasoning depends heavily on context (Plackett et al., 2022).

Types, evidence and design of virtual patients are in virtual patients; here it is enough to keep in mind that they are the most studied format for the management stage.

Decision games under time pressure

A research group developed serious games to train the triage of trauma patients in hospitals without a trauma center. In the first trial, with 320 physicians, those who played one of the two games left fewer severe patients without transfer in a validated simulation, with differences of 18 and 17 percentage points compared with the control (Mohan et al., 2018). The text educational material, in the same trial, had no significant effect (Mohan et al., 2018).

The two games included feedback on diagnostic errors, explaining the origin and the consequences of each one (Mohan et al., 2018). In a larger randomized trial, with 800 emergency physicians, training with the game reduced the proportion of older adults with severe trauma who were not transferred: 49% against 57% in the usual education group, an adjusted difference of 7 percentage points (Mohan et al., 2026).

It is one of the few examples of a game tested with an outcome measured in the following year's patient visits, and not only in a test (Mohan et al., 2026). These games are for trained physicians, but they show that specific management decisions can also be trained with games.

Interpretation games under uncertainty

Some formats present a case, a hypothesis and a new piece of data, and ask whether the data makes the hypothesis more or less likely. It is the logic of the script concordance test, whose scores correlate little with tests of pure factual knowledge (Lubarsky et al., 2011). This suggests that the format measures something beyond memory of facts, which makes it a good complement for those who already study with questions.

Case platforms with open response

There are also platforms in which you read a short case and write, without alternatives, an ordered list of diagnoses. Human Dx is the most studied of this type, and two studies published with its data are in the next section.

The types side by side

Type of game	Stage it trains most	What studies show
Hypothesis with clues	Formulating and revising hypotheses	Most used format in teaching reasoning (Schmidt and Mamede, 2015)
Virtual patient with choices	Data gathering and management	More frequent improvement in case-specific measures (Plackett et al., 2022)
Decision under time pressure	Specific management, such as triage	Randomized trials with emergency physicians (Mohan et al., 2018; Mohan et al., 2026)
Interpretation under uncertainty	Weight of a new piece of data	Supported validity for measuring interpretation under uncertainty (Lubarsky et al., 2011)
Cases with open response	Diagnostic list without alternatives	Validated as a measure of diagnostic performance (Chatterjee et al., 2019)

The column "stage it trains most" is editorial guidance; the third column summarizes the cited sources.

What the Human Dx data show

Two studies published in JAMA Network Open in 2019 used data from the Human Dx platform (Chatterjee et al., 2019; Barnett et al., 2019). The numbers below come only from these articles.

Measuring diagnostic performance

The first study analyzed 11,023 attempts to solve short, open case simulations, made by 1,738 practicing physicians, residents and students in the United States (Chatterjee et al., 2019). The goal was to validate an automated method of measuring diagnostic performance, with three measures: accuracy, efficiency and a combined score (Chatterjee et al., 2019).

Practicing physicians had higher average accuracy than students, with a difference of 8.1 points on the study scale (Chatterjee et al., 2019). They were also more efficient than residents, interns and students (Chatterjee et al., 2019). The authors conclude that diagnostic performance is higher in those with more training and that the score can be a valid measure, with immediate feedback to the user (Chatterjee et al., 2019).

An important point for those reading this study: it validates a way of measuring, and does not test whether playing improves performance. Short cases with open answers have already been used to measure diagnostic performance; the training effect is another question.

Several diagnoses together

The second study used data from 2,069 users, including residents, practicing physicians and students, who solved structured cases submitted by other users (Barnett et al., 2019). Accuracy, defined as the correct diagnosis among the first three on the list, was 62.5% for individual physicians and reached 85.6% in groups of nine (Barnett et al., 2019).

Groups of two to nine participants, with accuracy from 77.7% to 85.5%, outperformed individual specialists solving cases in their own subspecialty, who got 66.3% right (Barnett et al., 2019). The gain varied with the main complaint, from 17.3 percentage points in abdominal pain to 29.8 in fever (Barnett et al., 2019). The authors call for further studies in the clinical context (Barnett et al., 2019).

What this says for those who study

The group study is about collective intelligence applied to diagnosis, not about learning (Barnett et al., 2019). For the student, the practical lesson is modest and useful: comparing your list with that of other people who solved the same case shows hypotheses you did not consider. Discussing the case with the class is the classroom version of this idea.

Feedback and calibration

Calibration is the relationship between how often you are right and how often you think you are right. A study with general practice physicians defined it precisely that way, as the relationship between diagnostic accuracy and confidence in that accuracy (Meyer et al., 2013).

Confidence and accuracy go poorly together

In this study, 118 physicians got 55.3% of the easier cases right and 5.8% of the harder ones (Meyer et al., 2013). Confidence, however, changed little: 7.2 versus 6.4 on a scale from 0 to 10 (Meyer et al., 2013). More confidence was associated with fewer requests for additional tests, and the authors warn that this mismatch can prevent the review of difficult cases (Meyer et al., 2013).

In an earlier study, with 72 students, 72 residents and 72 internal medicine professors, confidence and accuracy were only slightly aligned (Friedman et al., 2005). When they were not aligned, residents were overconfident in 41% of cases, professors in 36% and students in 25% (Friedman et al., 2005).

Overconfidence and error

A review on the topic argues that physicians in general underestimate the chance that their own diagnosis is wrong and that this tendency toward overconfidence has individual factors and factors reinforced by the system (Berner and Graber, 2008). Part of the problem is that, in practice, feedback on the diagnosis often does not arrive.

A recent review with specialists in clinical reasoning reminds us that many diagnostic errors go unnoticed by the physicians themselves, unless they cause relevant harm (Bradford et al., 2026). The experts consulted described calibration as something that fluctuates and depends on context, and recommended measuring it also in intermediate decisions, before the final diagnosis (Bradford et al., 2026).

Why feedback is the center of the game

In education in general, feedback is among the strongest influences on learning, but the effect can be positive or negative, depending on the type of feedback and how it is given (Hattie and Timperley, 2007). The authors also discuss the timing of feedback as one of the issues that weigh on the effect (Hattie and Timperley, 2007).

Reasoning games have a natural advantage here: each case ends with an answer, and feedback arrives right away. In a pilot study with 72 emergency physicians, game sessions accompanied by specialists who gave immediate and personalized feedback on reasoning were described as fun and valuable by 93% of the participants in the intervention group (Mohan et al., 2023).

How to train calibration

A simple way, as an editorial guideline: before seeing the answer to each case, write down the hypothesis and your confidence in it, from 0 to 10. After a few cases, compare. If confidence was high on errors, it is worth slowing down in cases where everything seems obvious. If it was low on correct answers, it is worth trusting a little more in the patterns you already master. With the record, the feedback from each case becomes learning about your own way of making mistakes.

Limitations of studies on games

Research on games in health education has grown, but there are still gaps that apply to all formats.

Few trials and varied outcomes

The Cochrane review on educational games for health professionals, with a search up to January 2012, included only two randomized trials (Akl et al., 2013). A later systematic review, with a search up to August 2017, gathered 30 trials with 3,634 participants and did not perform a meta-analysis because of the heterogeneity of populations, interventions and outcomes (Gentry et al., 2019). The same review concluded that there was not enough evidence to say whether one type of game is more effective than another (Gentry et al., 2019). The general picture of these reviews is in the text about gamification in medical education.

Formats little compared with each other

In the narrative review on teaching clinical reasoning, only 24 of the 48 articles had empirical data (Schmidt and Mamede, 2015). The authors note that studies on the effectiveness of the serial cues format are still lacking, that there is some evidence in favor of the whole case focused on knowledge, and that approaches centered on teaching "a way of thinking" proved largely ineffective (Schmidt and Mamede, 2015). They propose that students at different stages of training benefit from different approaches (Schmidt and Mamede, 2015).

Measures without validation

In the review on virtual patients and clinical reasoning, 42% of the studies did not report any measure of validity or reliability of the reasoning outcome (Plackett et al., 2022). Even well-studied assessment formats have gaps: in the script concordance test, the evidence on the response process and on educational consequences is still weak (Lubarsky et al., 2011).

Measuring is not training

The studies with Human Dx data validated a performance measure (Chatterjee et al., 2019) and tested the combination of diagnoses (Barnett et al., 2019). Neither was designed to measure how much someone learns by playing. This difference matters for reading any study on games: ask whether the outcome is performance in the game itself, on a test or in practice.

Effects in practice are still rare

The triage trial with 800 physicians is an exception, and even in it the effect appeared in the lack of transfer of severe patients, with no difference in excessive transfer nor in the combined outcome of mortality or 30-day readmission (Mohan et al., 2026). Results like these show the path and also how much remains to be measured.

How research can move forward

The gaps point to clear questions: comparing formats with each other, using validated measures, following up on performance months later and, when possible, measuring decisions in practice. Medical education researchers who want to study training with cases in clues can learn about NEXO Researcher. If you coordinate a medical school, clerkship or residency and want to use cases like these in teaching and track the class result, learn about NEXO Closer. Doctors who want to review cases can learn about NEXO Curator, and those who produce medical content, NEXO Creators.

How to combine the types

Each format covers part of the steps, and the review on assessment methods recommends exactly combining complementary methods (Daniel et al., 2019). A possible script, as editorial guidance:

1. A guess every day: a short diagnostic case, with the guess and confidence noted before the answer.
2. Management a few times a week: a virtual patient or a case discussion in which you have to decide what to do.
3. Data interpretation: questions that ask for the weight of a new finding on a guess.
4. Review with the class: comparing lists of guesses for the same case and discussing the differences.
5. Record of mistakes: keeping the finding that separated the right disease from the one you chose.

Editorial guidance; adjust to the stage of the course and the stage you are at.

Common mistakes when training with games

Some habits reduce what a game can teach. The items below are editorial guidance, linked to the sources when there is one.

• Looking only at the score: getting it right on the day says little; what teaches is understanding why the wrong guess seemed right. The effect of feedback depends on the type of feedback and on how it is used (Hattie and Timperley, 2007).
• Always training the same step: those who only do diagnostic cases train management little, and the opposite is also true. Each method covers different parts of reasoning (Daniel et al., 2019).
• Ignoring your own confidence: without noting how confident you were, feedback does not help calibrate. Confidence and getting it right usually go together little (Meyer et al., 2013).
• Playing too fast on hard cases: haste helps train patterns, but the cases you got wrong deserve a second reading, finding by finding.
• Reading the measure as learning: doing well in a game shows performance in that format; the real test is reasoning in front of the patient, with supervision.

Training the guess in NEXO

In NEXO, each case arrives in up to six clues and the task is to reach the diagnosis, which concentrates the training on the step of forming and reviewing guesses. Each guess shows the proximity in ICD-10 on five levels, immediate feedback that says whether you are in the right group of diseases and how far you were. On the third clue the area of the case appears and, on the fifth, the ICD range; you can skip a clue whenever you want to move forward.

This feedback is good material for training calibration: note your confidence before each guess and compare it with the proximity the game shows. The Daily mode brings one case a day, the same for everyone, which allows comparing lists with the class, as in the study of combined diagnoses. The Archive gathers previous cases, Multimedia brings cases with image and audio, Continents, cases linked to Africa, the Americas, Asia, Europe and Oceania, and Adrenaline, with the dose chosen beforehand, trains the guess with time.

The Anki button creates the card straight from NEXO and opens it in the Anki app, searches the cards you already have and, when the case has an image, takes the image to the card. The ranking has the Global, Universities and Group tabs, the latter with an invite code. To understand the limits of each mode, see the NEXO economy page, and to learn about the game, read what NEXO is.

Frequently asked questions

What are clinical reasoning games?

They are games with clinical cases in which you have to make a reasoning decision: form a guess, choose which data to seek, decide on management or interpret new information. Each type concentrates the training on one of these steps.

Do clinical reasoning games work?

There are positive results, especially in measures specific to the case, such as data gathering, diagnostic ideas and management (Plackett et al., 2022). The reviews still do not allow saying which type of game is most effective (Gentry et al., 2019).

What is the difference between a diagnosis game and a management game?

The diagnosis game trains you to reach the hypothesis from the findings. The management game trains the next step, deciding what to do, as in virtual patients with choices and in triage games tested with emergency physicians (Mohan et al., 2026).

What is Human Dx and what did the studies show?

It is an online platform of short cases with open answers. Studies with its data showed that the score can measure diagnostic performance (Chatterjee et al., 2019) and that combined lists from several participants were more accurate than individual ones, from 62.5% to 85.6% (Barnett et al., 2019).

What is diagnostic calibration?

It is the relationship between how often you are right and how confident you are in being right (Meyer et al., 2013). Physicians and students often have confidence poorly aligned with accuracy (Friedman et al., 2005), and the feedback from each case helps adjust one to the other.

How does NEXO help train clinical reasoning?

Each case arrives in up to six clues, and the guess shows the proximity in ICD-10, which focuses the training on formulating and reviewing hypotheses, with immediate feedback. The daily case, the same for everyone, allows you to compare your reasoning with your class.

What type of game should be used in the internship?

In the internship, the management stage weighs more, because you start deciding what to do with the hypothesis. It is worth keeping the daily hypothesis training and adding cases in which you must choose the management, such as virtual patients with choices and discussions with the preceptor. The combination follows the recommendation to use complementary methods (Daniel et al., 2019).

Do games replace the clerkship?

They complement the clerkship: they expand the number of cases you see and give quick feedback, while contact with the patient, supervision and discussion with the preceptor remain at the center of training. The text on clinical reasoning shows how to combine the two things.

Checklist for choosing a reasoning game

1. Which stage does the game train: hypothesis, data collection, management or interpretation?
2. Is the feedback immediate and does it explain the error?
3. Is the answer open or only multiple choice?
4. Can you record your confidence before the answer?
5. Are the cases reviewed by someone who knows the subject?
6. Are there enough cases to repeat the same stage many times?
7. Can you compare your reasoning with your class?
8. Does the game fit into your routine without taking time away from the clerkship and the questions?
9. Is there a published study on the format, and is the outcome learning or just a measurement?
10. Do you combine formats to cover the other stages?

Want to train today? Play the case of the day on NEXO.

References

1. Daniel M, Rencic J, Durning SJ, et al. Clinical Reasoning Assessment Methods: A Scoping Review and Practical Guidance. Acad Med. 2019. https://pubmed.ncbi.nlm.nih.gov/30720527/
2. Plackett R, Kassianos AP, Mylan S, et al. The effectiveness of using virtual patient educational tools to improve medical students' clinical reasoning skills: a systematic review. BMC Med Educ. 2022. https://pubmed.ncbi.nlm.nih.gov/35550085/
3. Bordage G, Brailovsky C, Carretier H, Page G. Content validation of key features on a national examination of clinical decision-making skills. Acad Med. 1995. https://pubmed.ncbi.nlm.nih.gov/7718059/
4. Lubarsky S, Charlin B, Cook DA, et al. Script concordance testing: a review of published validity evidence. Med Educ. 2011. https://pubmed.ncbi.nlm.nih.gov/21401680/
5. Schmidt HG, Mamede S. How to improve the teaching of clinical reasoning: a narrative review and a proposal. Med Educ. 2015. https://pubmed.ncbi.nlm.nih.gov/26383068/
6. Mohan D, Fischhoff B, Angus DC, et al. Serious games may improve physician heuristics in trauma triage. Proc Natl Acad Sci U S A. 2018. https://pubmed.ncbi.nlm.nih.gov/30150397/
7. Mohan D, Chang CH, Fischhoff B, et al. Using Serious Games to Increase the Implementation of Trauma Triage Guidelines: A Randomized Clinical Trial. JAMA. 2026. https://pubmed.ncbi.nlm.nih.gov/42008277/
8. Chatterjee S, Desai S, Manesh R, et al. Assessment of a Simulated Case-Based Measurement of Physician Diagnostic Performance. JAMA Netw Open. 2019. https://pubmed.ncbi.nlm.nih.gov/30646211/
9. Barnett ML, Boddupalli D, Nundy S, et al. Comparative Accuracy of Diagnosis by Collective Intelligence of Multiple Physicians vs Individual Physicians. JAMA Netw Open. 2019. https://pubmed.ncbi.nlm.nih.gov/30821822/
10. Meyer AN, Payne VL, Meeks DW, et al. Physicians' diagnostic accuracy, confidence, and resource requests: a vignette study. JAMA Intern Med. 2013. https://pubmed.ncbi.nlm.nih.gov/23979070/
11. Friedman CP, Gatti GG, Franz TM, et al. Do physicians know when their diagnoses are correct? Implications for decision support and error reduction. J Gen Intern Med. 2005. https://pubmed.ncbi.nlm.nih.gov/15857490/
12. Berner ES, Graber ML. Overconfidence as a cause of diagnostic error in medicine. Am J Med. 2008. https://pubmed.ncbi.nlm.nih.gov/18440350/
13. Bradford A, Shahid U, Cholankeril R, et al. Methods to measure clinicians' diagnostic calibration: an appraisal for real world application. Diagnosis (Berl). 2026. https://pubmed.ncbi.nlm.nih.gov/42678279/
14. Hattie J, Timperley H. The Power of Feedback. Rev Educ Res. 2007. https://doi.org/10.3102/003465430298487
15. Mohan D, Elmer J, Arnold RM, et al. Testing a Novel Deliberate Practice Intervention to Improve Diagnostic Reasoning in Trauma Triage: A Pilot Randomized Clinical Trial. JAMA Netw Open. 2023. https://pubmed.ncbi.nlm.nih.gov/37195666/
16. Akl EA, Kairouz VF, Sackett KM, et al. Educational games for health professionals. Cochrane Database Syst Rev. 2013. https://pubmed.ncbi.nlm.nih.gov/23543543/
17. Gentry SV, Gauthier A, L'Estrade Ehrstrom B, et al. Serious Gaming and Gamification Education in Health Professions: Systematic Review. J Med Internet Res. 2019. https://pubmed.ncbi.nlm.nih.gov/30920375/