TL;DR. You practice difficult patient conversations by running the six SPIKES steps in short role-plays and scoring each run against a fixed observation sheet, not by explaining the protocol again. Three components decide whether it works: a specific scenario, a defined standard, and feedback that points at a sentence someone actually said. In a randomized study of 68 participants, one hour of theory plus three hours of role-play raised SPIKES competence by 33.3 percent (Servotte et al., 2019). The limiting factor is rarely knowledge. It is the number of repetitions one named person gets. This article gives you the protocol: roles, timing, scenario order, observation sheet, the options when no partner is available, and how to keep the habit alive on a ward.
Key Takeaways
- With identical content, simulation beats lecture. Among 267 medical students, the role-play group improved more on recognizing emotional reactions (p = 0.005) and on handling emotional cues (p below 0.001), and 70.9 percent rated the training very helpful against 29.1 percent after the lecture (Alghzawi et al., 2026, BMC Medical Education).
- Three hours of role-play after one hour of theory raised the share of participants reaching SPIKES competence from 45.2 to 73 percent (Servotte et al., 2019, Western Journal of Emergency Medicine).
- SPIKES is a six-step protocol from oncology (Baile, Buckman et al., 2000, The Oncologist). It describes the shape of a conversation, not a way to rehearse one, so the practice is yours to organize.
- Repetition has a measurable effect, so far documented mainly outside medicine: scores rose by 14.1 points between the first and the second practice conversation at Schwäbisch Hall. Transferring that to a clinical disclosure is an inference, not a finding.
- Four contractual points decide whether a practice environment is usable in a hospital: EU hosting, a data processing agreement under Art. 28 GDPR, no model training on the data produced, and defined access rights to transcripts and evaluations.
Ask a ward team what the six letters of SPIKES stand for and you will get an answer within seconds. Ask when one named registrar last walked through all six of them with nobody's real diagnosis in the room, and the question changes shape. A pilot rehearses engine failure in a simulator, not at the first real engine failure. In clinical communication, the first repetition is usually the disclosure itself. The reason was never a lack of insight: rotas, no available partner, no free room, and the reluctance to fail in front of the person who writes your evaluation make regular practice structurally unlikely.
This article is not about what SPIKES means. It is about how to rehearse it. It gives you a 20-minute role-play protocol, an order to work through the scenarios in, an observation sheet keyed to the six steps, the options when neither a partner nor a slot exists, and the three decisions that keep the habit alive once the novelty wears off. It is written for the people who have to organize practice inside a hospital: ward leads, education supervisors, clinical L&D. One qualification belongs up front, because vendor content tends to leave it out: a role-play is not supervision and not a case conference. What it produces is repetition, and repetition is the part missing from clinical life.
Why does knowing SPIKES not change the conversation?
Knowing the protocol leaves behavior untouched, because SPIKES describes the structure of a conversation rather than teaching a skill. Baile, Buckman and colleagues published it in 2000 as six steps for breaking bad news to a patient with cancer (Baile et al., 2000, The Oncologist): Setting, Perception, Invitation, Knowledge, Emotions, Strategy and Summary. The sequence is easy to memorize, and that is exactly the trap. Anyone who can recite it feels prepared.
The gap between explaining and rehearsing is measurable. A study of 267 medical students delivered the same content in two formats, lecture against simulated role-play. Both groups improved significantly, but the role-play group gained more on recognizing emotional reactions (p = 0.005) and on handling emotional cues (p below 0.001), and 70.9 percent of them called the training very helpful against 29.1 percent after the lecture (Alghzawi et al., 2026, BMC Medical Education).
Competence data points the same way. In a randomized emergency-department study, 68 participants received one hour of theory on the SPIKES protocol followed by three hours of role-play across six scenarios with structured debriefing. SPIKES competence rose 33.3 percent, self-efficacy 55 percent, and 73 percent of the training group reached the defined competence threshold against 45.2 percent of controls (Servotte et al., 2019, Western Journal of Emergency Medicine). The ratio in that design is the real finding: one hour of explanation to three hours of practice.
What turns a role-play into actual practice?
A role-play becomes practice when a scenario, a scoring standard and evidence-based feedback are all present. Remove any one of the three and it turns into a conversation about conversations. All three can be prepared in a few minutes, and together they separate a rehearsal from a pleasant group discussion.
First, the scenario: a specific situation with a diagnosis, the patient's existing understanding, and a clear task for the person practicing. "Let's do a difficult conversation" is not a scenario. "Mrs M., 54, suspected diagnosis now confirmed, expects a routine result, arrives alone" is one. Second, the standard: a list of observable behaviors, agreed before the run, that the run is judged against. Third, the feedback: a response that points at specific sentences from the run rather than at a general impression.
Without the standard, feedback becomes a matter of taste and the learner mostly discovers whose opinion carries weight in the room. Without the scenario, everyone rehearses the easy version. Without evidence-based feedback, the run leaves behind a mood and no learning point. A rehearsal with no defined standard changes nothing.
How do you practice a difficult patient conversation in 20 minutes?
One full run needs three people, 20 minutes and a scenario card, and nothing else. The three roles are the clinician who runs the conversation, the patient or relative who answers from the scenario card without helping, and the observer who stays silent and fills in the sheet. Rotate the roles after each run and a lunch break produces three repetitions per person per week.
| Phase | Time | What happens |
|---|---|---|
| Setup | 3 min | Hand out the scenario card, the patient role reads it silently, the observer takes the sheet, the clinician names one learning goal for this run |
| The conversation | 8 min | Run it without interruption, including when it goes wrong. Stopping protects exactly the moments worth practicing |
| Self-assessment | 2 min | The clinician says first which of the six steps worked and which did not |
| Observer feedback | 5 min | The observer walks the six steps and quotes one sentence as evidence for each |
| Repeat | 2 min | Replay only the step that failed, not the whole conversation |
The last row matters most and gets dropped most often. A run without an immediate replay of the weak step produces an insight and no changed behavior. Two minutes on a second attempt at the same transition are worth more than a second full conversation next week.
Which scenario should you practice first?
Start with the scenario that loads the step your team keeps failing, not with the hardest conversation on the ward. Every clinical situation stresses a different one of the six steps. Knowing which is which turns general practice into targeted practice.
| Scenario | Step under load | Typical failure |
|---|---|---|
| First disclosure of a cancer diagnosis | Knowledge | Technical language, and too much information in one block |
| Prognosis and remaining time | Invitation | Answering a question the patient never asked |
| Family conversation after sudden deterioration | Emotions | Correcting the facts instead of naming the feeling |
| Shift to a palliative treatment goal | Strategy | Sentences along the lines of "there is nothing more we can do" |
| Angry relative on the ward phone | Setting | A conversation in the corridor, with no time frame and nowhere to sit |
Read the table as a practice aid, not as a diagnosis. Its job is to tell you where to start: if the complaints coming back from your wards are that families feel talked past, rehearse the Emotions step rather than the whole protocol. Phrases to avoid, "there is nothing more we can do" among them, are well documented in the palliative literature, and they leave a team's vocabulary only after a better sentence has been said out loud several times.
What does a good conversation look like on the observation sheet?
A conversation is good when every one of the six steps has an observable behavior attached to it, not when it felt empathetic. The observation sheet is therefore not a grading scale. It is a list of behaviors that were either shown or not shown.
- Setting: an undisturbed room, seated at eye level, the available time named at the start.
- Perception: one open question about what the patient already knows, asked before any information is given, such as "what have you been told so far?".
- Invitation: explicit agreement on how much detail the patient wants right now.
- Knowledge: a warning shot, then one core statement in everyday language, then a pause.
- Emotions: the feeling named rather than immediately explained, and silence left standing rather than filled.
- Strategy and Summary: a next step with a date and a named contact, and a summary the patient produces themselves.
Two kinds of feedback belong out of the debrief. "That was empathetic" is not feedback, because nobody can repeat it on purpose. "I would have done it differently" is not feedback either, because it swaps the standard for a preference. The usable third form sounds like this: "the treatment plan followed the diagnosis with no pause in between." That points at a moment, names the step, and can be corrected in the next run.
Isn't role-play awkward and unrealistic?
It is awkward the first time, and it becomes unrealistic precisely when nobody has defined what the conversation is judged against. The objection is fair, and senior clinicians state it most bluntly: a colleague is not a patient, they do not really cry, they do not really get angry, and acting it out in front of others feels undignified. In a hospital, where competence is the currency, rehearsing in public costs standing.
The learning effect, though, does not rest on acting quality. It comes from isolating one step, repeating it, and having an observer point at a specific sentence. The emergency-department study used trained relative actors, and its effect came from three hours of repetition with structured debriefing rather than from dramatic intensity (Servotte et al., 2019). Against the awkwardness, reduce the audience: three people, a closed door, no line manager present, and self-assessment always before external feedback.
What the classical format still does better stays true. A trained simulated patient produces physical reactions and unpredictable turns that a colleague cannot, and a facilitated seminar generates collegial reflection that a pair exercise does not replace. Both belong in the curriculum. And conversations that touch a colleague personally, after a loss of their own for instance, belong in supervision rather than in a role-play.
How do you practice with no partner and no slot?
With no partner, two options remain: talk the conversation through out loud alone, or hold it with a virtual counterpart that answers and then scores the run. The first trains phrasing and beats doing nothing. What it cannot supply is the reaction the Emotions step depends on.
The real bottleneck is not willingness anyway, it is the calendar. A three-person run requires three people free at the same time, which in shift work is the scarcest resource in the building. That is the constraint a simulated counterpart removes, because it takes the scheduling out of the loop.
At Sleak this part is called Training Mode: the learner picks a Training Scenario, a configured practice situation with a counterpart and context, speaks by voice with a virtual persona, and afterwards receives an evaluation from the AI Coach, the always-available coaching instance, against a Scorecard, the standard defined in advance for that conversation. The evaluation quotes moments from the transcript instead of offering general praise. The same SPIKES scene can be repeated in voice-based AI role-plays as many times as it takes, and the six-step observation sheet becomes the Scorecard.
One note on choosing between tools. Most simulation products aimed at medicine are built for examinations, for the OSCE and for licensing assessments, and they are designed around a student passing a test. That is a different job from the one described here. A hospital buying practice for people who already hold the post needs scenarios drawn from its own wards, an observation standard its own supervisors wrote, and a legal basis that survives a works-council review.
For hospital use, that legal basis is the deciding factor rather than conversational quality. Four points need checking: EU hosting, a data processing agreement under Art. 28 GDPR, a contractual exclusion of model training on the data produced, and defined access rights to transcripts and evaluations. The distinction that matters is which data is sensitive in the first place: the personas are fictional and no real patient exists, so what needs protecting is the learner, whose transcripts and scores are the actual record.
In practice: how do you keep this alive on a ward?
Practice survives a roster only when one named person owns the cadence and a full run is shorter than a handover. Three decisions are enough: who maintains the scenario cards, when the practice happens, and who knows the observation sheet. In real use, a fixed 20-minute slot inside a meeting that already exists beats a new appointment, because it never opens a negotiation about the calendar. And whoever gives the feedback should have no say in the learner's appraisal, or nobody will play a run deliberately hard.
What repetition volume does is easier to quantify outside medicine. In the rollout at Schwäbisch Hall, scores rose by 14.1 points between a participant's first and second practice conversation, the steepest part of the whole learning curve, and the programme has now produced more than 1,800 completed practice conversations.
Those numbers come from sales conversations, not from a ward. What they establish is the mechanism: applied competence grows with repetition measured against a defined standard, and the largest increment arrives early. Transferring that to a disclosure conversation is a plausible inference and not a finding, and it should be tested against your own setting. Where the clinical and non-clinical evidence agrees is unambiguous all the same: two or three scored repetitions move more than a fourth explanation of the protocol.
FAQ
How often should you practice a difficult patient conversation?
Two or three scored runs per scenario capture most of the effect, after which the curve flattens. Moving to the next scenario once all six steps have been shown cleanly once beats repeating the same conversation many times. Regularity matters more than volume: one short run per week does more than one full-day seminar per year.
Can you practice SPIKES alone?
Partly. Saying the core sentences out loud improves phrasing and sequence, and the warning shot before a diagnosis can be trained alone. What cannot be done alone is the Emotions step, because it needs a reaction to respond to. That requires either a practice partner or a simulated counterpart.
Does an AI role-play replace the simulated patient?
No. A trained simulated patient produces physical reactions and unpredictable turns that a simulated counterpart does not reproduce. The difference is availability: the simulated patient arrives for the scheduled seminar, the simulation is there for the twentieth repetition at 10pm. The two formats solve different problems.
Can AI-based conversation training be GDPR-compliant in a hospital?
Yes, provided four points are settled contractually: EU hosting, a data processing agreement under Art. 28 GDPR, exclusion of model training on the data produced, and defined access rights to transcripts and evaluations. Note that the sensitive data is the staff practice record, not the fictional patient data in the scenario. Involve the works council early where one exists.
Who should give the observer feedback?
Ideally someone who knows the observation sheet and has no say in the learner's appraisal. A direct line manager in the room lowers the willingness to play a run deliberately hard. Peer pairs at the same level work most reliably in day-to-day use.
Related reading
- Practicing Difficult Employee Conversations: 8 Scenarios Where AI Makes Leaders More Confident
- Scorecard-Based Coaching: How Structured Feedback Transforms Sales Training
- GDPR-Compliant AI Coaching: What to Check When Choosing a Platform
- AI-Powered Training in the Enterprise: How Companies Roll Out AI Coaching Successfully
