Gains:
- Ability to tailor AI recommendations to population-specific contraindications, stage and safety constraints
- Ability to recognize how generalization and hallucination risks of AI vary in different patient groups.
- Ability to accurately determine the threshold for verification and expert consultation for each specific group
There is no such thing as an "average patient" in physiotherapy. A patient who had a stroke, a 5-year-old child, a frail 85-year-old, an athlete who had knee surgery; each of these carries entirely different security constraints, objectives, and communication requirements. Artificial intelligence (AI) tends to default to the “general adult,” and these generalizations can be dangerous in special populations. In this unit, you will learn how to tailor AI recommendations to five key specific groups, how the risk of hallucinations and generalizations varies in each group, and how to determine your consultation threshold.
Why generalizations are dangerous
AI considers the most common situations in the training data to be "normal". But while a balance exercise is safe for a young adult, it may pose a risk of falls, spasticity (continuous involuntary contraction of muscles) or safety issues in a post-stroke patient. While stretching is appropriate for a young athlete, it may carry a risk of fracture in an elderly person with osteoporosis. While an increase in load is good for a healthy knee, it can damage the growth plate in a growing child. AI does not know these differences per se; If you don't specify it, it gives the general recommendation.
Caution: Never apply AI output to special populations with a "general adult" assumption. For each group, include population-specific contraindications, safety restrictions, and team consultation (physician, pediatrician, neurologist) when necessary.
Five groups, five attention points
population
Main risk
Be sure to tell AI
Consultation threshold
Neurological (stroke, MS, Parkinson's, SCI)
Falls, spasticity, fatigue, safety
Neurological status, fall history, need for supervision
Low — frequent team decision
Pediatric (child)
Growth plate, development, play-based, consent
Age, developmental stage, parental involvement
Collaboration with the pediatrician
Geriatric (elderly)
Osteoporosis, falls, fragility, polypharmacy
Bone density, fall history, comorbidity
Moderate — high risk of fracture/fall
Orthopedic/post-op
Surgical protocol, tissue healing phase
Surgery type, surgeon protocol, stage
Based on surgeon protocol
athlete
Overload, early return, re-injury
Sport, return criteria, season
Sports physician/team
Step by step: adaptation to special population
- Define the group and subprofile. Not just "old"; Like "82 years old, has osteoporosis, fell twice."
- Write population-specific constraints to the prompt. Clearly state contraindications and safety limits.
- Question AI's generalization. “Is this recommendation safe for this group?” Have it checked also.
- Apply consultation threshold. If the decision is in the field of another discipline, consult.
- Add surveillance and security plan. Especially in groups with fall/safety risk.
- Adapt, approve, monitor. As a physical therapist, give final approval and monitor closely.
Three mini cases (in numbers)
Case 1 — Neurological (stroke). 67 years old, right hemiparesis, history of 1 fall. The AI suggested a general “30-second balance on one leg” exercise. The physiotherapist changed this due to fall risk: parallel bar, supervised, assisted start. With safe progress in 4 weeks, unsupported standing time increased from 5 seconds to 22 seconds, and there were no falls. Lesson: safety always comes first in the neurological group.
Case 2 — Geriatric (osteoporosis). 78 years old, severe osteoporosis. AI recommended flexion-based exercises for the lower back. The physical therapist removed it because repeated lumbar flexion in osteoporosis increases the risk of vertebral fractures; replaced it with extension and posture exercises. Lesson: AI does not know population-specific contraindications; siz bilmelisiniz.
Vaka 3 — Pediatrik. 8 years old, knee pain after sports (Osgood-Schlatter). AI gave adult protocol (heavy lifting). The physiotherapist took into account the age of growth, reduced the load, changed it to play-based and pain-free exercises, and stayed in touch with the pediatrician. The parent is included in the program. Lesson: a child is not a little adult.
Weak prompt / Strong prompt
Weak: "Give balance exercises for an elderly patient."
Lower profile, no history of falls, no comorbidities, and no safety; AI generates risky recommendation assuming young adult.
Güçlü: "Produce a DRAFT of a safe balance exercise for an 82-year-old patient with osteoporosis and a history of 2 falls, using a walker. Avoid lumbar flexion due to osteoporosis. Add initial support and supervision advice to each exercise. Warn movements that will increase the risk of falling separately. The clinical decision and supervision plan is mine."
Four copyable templates
Role: Physiotherapy assistant (special population aware).Task: Produce exercise DRAFT for [population + subprofile].Specify mandatory: age, diagnosis, comorbidity, fall/safety history, need for supervision.Rule: Warn of contraindications specific to this population in a separate heading.For each recommendation, put safety margin and supervision/support note if necessary.
Task: Safety-check this general recommendation for [population].Recommendation: [Overall output of AI]Profile: [sub-profile + contraindications]I want: Check every item that is risky/contraindicated for this group + reason +safe alternative. Don't decide, show risk.
Task: Protocol compliance check for post-op patient. Program outline: [exercises] Surgical protocol: [stage restrictions given by the surgeon] I want: Mark any item that conflicts with the surgical protocol in the program. Protocol is essential; Warn me to remove conflicting suggestion.
Task: Play-based, pain-free adaptation for the pediatric patient. Exercise: [adult version], child: [age, diagnosis]. Rule: Growing age safety, play/motivation language, parental involvement, pain-free limit. Do not use heavy loads or adult dosages.
Tip: Prepare yourself a "red line list" for each specific population (e.g. lumbar flexion in osteoporosis, unsupervised balance in stroke, heavy lifting in a child). Quickly scan the AI output against this list; Generalization errors are mostly caught here.
Another important dimension in special populations is communication and goal adaptation. While the goal for an athlete may be a return to performance, for a frail senior the goal may be to maintain independent living and fall prevention; For a child, motivation is based on games and entertainment. The general motivational or educational language that AI produces does not reflect these differences; For example, telling a child "you will increase your muscle strength" in adult language is ineffective, but turning the same movement into a game is useful. When adapting the target and language to the population, give the AI this context explicitly (“in gamified language that an 8-year-old will understand”) and check that the output is both safe and age-appropriate. The right target and the right language determine the success of the treatment as much as the right exercise.
Comorbidity and polypharmacy layer
In special populations, a single diagnosis rarely comes alone. An elderly patient may have osteoporosis, hypertension, diabetes and balance problems simultaneously; each of these affects the exercise decision. Comorbidity (more than one co-existing disease) and polypharmacy (use of multiple medications) are the layers most overlooked by AI, as AI often focuses on a single diagnosis. For example, the risk of falling in a patient using blood thinners has more serious consequences; Heart rate may not be a reliable indicator of exercise intensity in a patient taking beta blockers. If you don't explicitly write these interactions into the prompt, the AI won't take them into account. Therefore, give the complete profile, especially in multi-diagnosed patients, and contact the prescribing physician if you are in doubt about drug-exercise interactions.
Tip: In complex patients, ask the AI a separate question “list how comorbidities and medications in this profile may affect exercise safety”; but use this list as a reminder and verify each item clinically and with a physician if necessary.
Common mistakes
- “General adult” assumption. Applying AI output without adapting it to the population is risky.
- Alt profili belirtmemek. "Yaşlı" yetmez; comorbidity, fall history and restraint are required.
- Bypassing population-specific contraindication. Pitfalls such as flexion are overlooked in osteoporosis.
- Ignoring the post-op protocol. The surgeon's protocol always takes precedence.
- Considering the child as a small adult. Growth, development and play aspects should not be neglected.
- Raising the consultation threshold. When in doubt, making team decisions increases safety.
In summary
AI generalizes by default across specific populations, and these generalizations can be dangerous. Each group (neurological, pediatric, geriatric, orthopedic/post-op, athlete) carries its own safety restrictions, contraindications and consultation thresholds. Clearly define the subprofile, write population-specific constraints on the prompt, check the output against a “red line list,” prioritize the surgical protocol in post-op, and obtain team consultation when in doubt. Final approval always belongs to the physiotherapist.
Application task
Choose two different specific populations. Write a detailed subprofile for each and produce an exercise outline with the first template. Then security-check each draft with the second template and capture at least one population-specific contraindication. Create your own "red line list" with at least five items.
checklist
- [ ] I defined the group and the detailed subprofile (comorbidity, history, limitation).
- [ ] I wrote population-specific contraindications on the prompt.
- [ ] I have security-checked the AI output.
- [ ] If it was post-op, I prioritized the surgeon's protocol.
- [ ] I applied the required consultation threshold (physician/specialist collaboration).
- [ ] I added a supervision/safety plan and approved it as a physical therapist.