Gains:
- Ability to understand the prescription formula (NAL-NL2, DSL), gain, compression and fitting steps and verify the artificial intelligence recommendation with real ear measurement (REM)
- Ability to use artificial intelligence in fitting troubleshooting, translating patient complaints into adjustment language and generating checklists
- Ability to apply that the fitting decision must be confirmed by the device software and REM data, and that artificial intelligence does not replace calibrated measurement
After the device is selected, comes the most technical and repetitive phase of the work: fitting (programming the hearing aid according to the patient's hearing loss and ear). Fitting is determining how much gain (amount of sound amplification) the device will give at which frequency, how it will compress the sounds, and how it will limit disturbing levels. In this unit, we will learn how to use AI as a troubleshooting and communication assistant in the fitting process, but why you should always confirm the accuracy of the settings with a real ear measurement. Main principle: AI can translate patient complaint into setting language and generate checklist; However, the fitting decision is verified by the device software and real ear measurement (REM) data, it does not replace the calibrated measurement.
Prescription formula, gain and compression
The basis of fitting is a prescription formula. The prescription formula is a method that calculates how much gain should be given at each frequency according to the patient's hearing thresholds. The two most common formulas are NAL-NL2 (a frequently used prescription method in adults) and DSL (a method especially preferred in children). These formulas give an initial target; The device software proposes an initial setting based on this target.
Compression is the compression of different sounds (between whispering and shouting) into a range that the patient can hear comfortably. While soft sounds cannot be heard in an ear with hearing loss, very loud sounds can still be disturbing; compression manages this narrowed hearing field.
Caution: The "target reached" graphic displayed by the device software on the screen represents an estimate, not the actual output produced by the device at the ear. Real ear acoustics vary from person to person. The fitting is not considered verified just because the software meets the target.
Why real ear measurement (REM) is indispensable
REM (Real-Ear Measurement) is the measurement of the sound level actually produced by the device in the patient's own ear by placing a thin probe microphone in the ear canal. The canal volume, shape and acoustics of each ear are different; The same device produces different output in two different ears at the same setting. REM is the only objective method of indicating whether the actual output of the device is reaching the prescription target. The answer to Fitting's question, "Is he okay?" is not the patient's subjective expression, but the REM data.
AI cannot do REM; It is a physical measurement that requires a calibrated device and probe. AI can be most helpful in sketching out the interpretation of REM results or outlining areas where the goal is not being met — but the expert does the measurement and final adjustment.
Fitting steps
step
what to do
Role of AI
1. Loss data
The audiogram is entered into the device software
None (calibrated data)
2. Recipe selection
NAL-NL2/DSL destination is determined
Reminder of formula difference
3. Initial setting
The software generates the recommendation
None
4. REM verification
Actual output is measured relative to the target
REM summary interpretation draft
5. Fine tuning
Adjustments are made according to the complaint
Translating the complaint into the setting language
6. Tracking
Data logging + patient feedback
Generate a checklist
Step by step: fitting troubleshooting with AI
- Collect the complaint. Anonymously record the patient's statement ("my own voice sounds weird", "the letters s are squeaky").
- Have AI translate. Ask for an outline that translates the complaint into possible tuning directions (e.g. which frequency band, gain or compression).
- Preserve possible language. Say "Don't give exact settings; list possible aspects and things to check."
- Verify with REM. Confirm the effect of each setting on the target with real ear measurement.
- Try it with the patient. Combine with real-life sounds and subjective feedback.
- Close with checklist. Complete the steps with the fitting checklist produced by YZ.
Weak prompt / Strong prompt
Weak prompt:
The patient hears his own voice strangely. How many dB should I reduce on the device?
It requires precise numerical tuning (the AI cannot know without measuring in the ear), and ignores REM and clinical context.
Powerful prompt:
Your role: fitting troubleshooting assistant. DO NOT give exact dB value; list only possible adjustment DIRECTIONS and points to check with REM. Complaint (anonymous): the patient hears his own voice "hollow/strange" (may have a feeling of occlusion). Device: RIC, closed mold. Prescription: NAL-NL2.Task: list possible causes of this complaint and adjustments/acoustics that could be tried; Write the REM or clinical verification step for each direction. State that the final setting is with the expert.
The strong prompt does not require numerical decisions, it gives the complaint in context, and links REM verification to each step.
three mini cases
Case 1 — Occlusion complaint. A patient wearing the new device hears his own voice "as if he were talking from a well." The audiologist gives the complaint to the AI; YZ lists that this may be related to the occlusion effect (the closed pattern trapping its own sound), controlling the ventilation/pattern and low frequency setting. The expert reviews the pattern and low-frequency gain and verifies it with REM; complaints decrease. AI gave possible directions, measurement and decision remained with the expert.
Case 2 — “S” crackle. The patient says that high-frequency sounds ("s", "sh") are disturbing. The expert translates the complaint to artificial intelligence; The draft recommends control of high frequency gain and maximum output limit. The expert sees that the REM and high frequency region exceed the target, corrects the setting and measures again. Subjective complaint is resolved when met with objective measurement.
Case 3 — Error without REM. An expert does not rely solely on the software target and do REM due to lack of time; The patient returns a few days later saying "everything is too loud." This time, the expert performs REM and finds that the actual output is significantly above the target — the software's prediction did not work in this ear. The setting is corrected. Lesson: software target is not a substitute for REM.
Copiable prompt templates
TEMPLATE FOR TRANSLATING COMPLAINT INTO SETTING LANGUAGEYour role: fitting assistant. Don't give exact dB. Translate the following patient complaint into possible causes and tuning/acoustic aspects to check; Write the REM/clinical verification step for each direction. Complaint: [complaint] Device/mould: [information]
REM RESULT COMMENT DRAFT TEMPLATEIn which frequency regions is the following REM measurement above/below the prescription target? Summarize and mark which area may require adjustment. The final decision belongs to the expert. Data: [REM values + target]
FITTING CHECKLIST TEMPLATEProduce a step-by-step checklist for a hearing aid fitting session: loss data, prescription selection, initial fit, REM verification, fine-tuning, patient trial, follow-up plan. Add a "verified" box to each step.
OCCLUSION/FEEDBACK DISCRIMINATION PATTERNPatient describes the following complaint: [complaint]. Could this be a feeling of occlusion, acoustic feedback, or excessive gain? List the distinctive questions and check steps for each possibility. The decision belongs to the expert.
Common mistakes
- Skipping REM. Do not rely solely on the software target and make real ear measurements.
- Mistaking subjective expression for measurement. Accepting the patient's "fine" as fitting verification.
- Requesting exact dB from AI. Allowing the model to make numerical adjustments without measuring at the ear.
- Attributing the complaint to a single reason. Adjusting without distinguishing between occlusion, feedback and overgain.
- Neglecting to follow up. Thinking that the first session is enough; skipping data logging and fine-tuning with feedback.
In summary
Fitting; The prescription formula is that gain and compression are programmed according to the patient, and its accuracy is only confirmed by real ear measurement (REM). AI can translate patient complaint into possible adjustment directions, summarize REM results and produce checklist; But the exact setting, measurement and final decision are up to the expert. The software target and the patient's subjective "good" are not substitutes for a calibrated real-ear measurement.
Application task
Set up an anonymous fitting scenario: a device type, prescription formula, and a patient complaint (e.g., occlusion or "s" squeal). Get the possible directions with the "translating the complaint into setting language" template, then break down the possibilities with the "occlusion/feedback discrimination" template. For each possible setting direction, write down which REM or clinical step you will verify with. Finally, produce a “fitting checklist” and compare it with your own routine.
checklist
- [ ] I recorded the patient complaint anonymously and in context.
- [ ] I had the AI ask for possible directions instead of exact dB.
- [ ] I connected each tuning aspect to the REM verification scheme.
- [ ] I distinguished occlusion/feedback/overgain.
- [ ] I combined subjective feedback with objective measurement.
- [ ] I closed the follow-up and fine-tuning plan with the checklist.