Gains:
- Ability to accurately position the 'pass/fail' logic of OAE and automatic ABR-based screening results and the difference between screening and diagnosis with the support of artificial intelligence
- Ability to use artificial intelligence as a pre-screener that prioritizes cases that require follow-up in screening lists and avoid the risk of false negatives
- Understanding the limits and notification obligations of artificial intelligence output in workplace noise screening and school screening
Hearing screening is a screening process that quickly identifies people who may have hearing problems in a society and directs them to diagnostic evaluation. Screening is not diagnosis; This is the most critical sentence of this unit. Screening says "there may be something wrong with this person, let it be examined"; The diagnosis says, "This person has this type and degree of loss." Artificial intelligence is a powerful pre-screener for pre-evaluating screening results — particularly in sorting through large numbers of results and prioritizing those that require follow-up. But it cannot completely prevent false negatives (missed real cases) and can never make the diagnostic decision.
In this unit, we will cover three common screening contexts: newborn hearing screening, school-age screening, and workplace noise screening. In each of them, we will clarify where you can safely use artificial intelligence and where you should stop.
The logic of screening tests
Let's get to know the two basic newborn/infant screening methods.
OAE (Otoacoustic Emission): Measures the very faint sounds produced by a healthy cochlea (hearing organ in the inner ear) in response to sound. A small probe is placed in the ear; If the cochlea gives an "echo" the test is considered "passed". OAE is fast but cannot detect problems behind the cochlea (auditory nerve).
Automatic ABR (Auditory Brainstem Response): Measures the electrical response of the auditory nerve and brainstem to sound. It is especially preferred in intensive care babies as it can detect conditions that OAE may miss, such as auditory neuropathy (a condition in which the cochlea works but nerve conduction is impaired).
The results of both tests are usually "pass" or "fail/refer". “Fail” is not a diagnosis; indicates the need for further evaluation. "Pass" indicates that there is no obvious problem in the current scan, and it does not guarantee lifetime hearing.
Caution: An auditory neuropathy-type loss may remain hidden in a baby with a "pass" result from OAE. The family is not told "it's over = it will never be a problem"; Symptom monitoring and follow-up, if necessary, are recommended for losses that may develop later.
Why is early diagnosis so critical?
The reason why newborn hearing screening is such a common and seriously taken program is that hearing is the basis of language and speech development. From the moment a baby is born, he learns language by hearing the sounds around him; If hearing loss goes unnoticed, this window of development can be silently missed. In case of early detected and appropriately supported hearing loss, the child's language development progresses significantly better than if it is detected late. That's why the "false negative" of screening—that is, missing a baby with real loss as "passed"—is the most costly mistake in the field: lost months can later translate into developmental differences that are difficult to compensate for.
Artificial intelligence is especially valuable in this context in ensuring that the chain of tracking is not broken. A common problem in screening programs is that the baby with a "fail" result never comes for further evaluation (family does not show up, the record is lost, the appointment is forgotten). AI can help reduce this “lost follow-up” rate by curating the list of cases requiring follow-up and generating reminder texts. But the decision about which baby is truly at risk and how the diagnostic process will proceed is always the clinical responsibility of the specialist and the program.
Balance of false negatives and false positives
Every screening test can make two types of errors. A false negative is missing the person who actually has the problem because they are "over it"—the most dangerous mistake in healthcare because the child remains unfollowed. A false positive is essentially marking the healthy person as a “fail” — creating unnecessary anxiety and additional testing, but is less dangerous.
You should strike this balance consciously when using AI in screening lists: in pre-screening, it is generally safer to keep the threshold towards "not missing" (i.e. to follow the suspect). The table below compares the three contexts.
Context
Typical method
Risk of false negative
Safe role of AI
newborn
OAE / automatic ABR
High (developmental delay)
Result list editing, follow-up reminder
school age
Pure voice scanning
Medium (training effect)
Class/list prioritization
workplace
Noise audiometry
Medium-high (occupational disease)
Trend/comparison plot
Step by step: pre-assessment of scanning with AI
- Anonymize and structure data. List results by code/sequence number instead of ID, pass/fail, and measurement value if applicable.
- Define the pre-qualification rule. Prioritize those who "fail" and those who are on the border; Set it so you don't miss the decision.
- Have the AI prioritize. Ask for a list that breaks down what needs to be monitored, what needs to be rescanned, and what is normal.
- Watch for false negatives. Even in the "Pass" group, if anyone describes symptoms (family suspicion, speech delay), follow up.
- Leave diagnostic guidance to the expert. The final diagnosis and further testing decision belongs to the audiologist/physician.
- Enforce notification obligation. Operate occupational health notification and protection processes, especially in cases of losses detected in the workplace.
Weak prompt / Strong prompt
Weak prompt:
The following babies have OAE results, tell me which ones are deaf: [list]
It calls for precise diagnostic language like “deaf,” ignores the screening-diagnosis distinction, and ignores the risk of false negatives.
Powerful prompt:
Your role: pre-screening assistant who organizes screening results. DON'T MAKING A DIAGNOSIS; Remind me that a "pass/fail" screening is not a diagnosis. Data (anonymous, with sequence number): [pass/fail list + notes if any]Task: (1) prioritize the "fail" ones, (2) mark separately the "pass" but not noted (family suspicion, etc.), (3) write the recommended next step for each group. State that the final diagnosis and further testing decision is up to the audiologist/physician.
The strong request is anonymous, preserves the screening-diagnosis distinction, takes into account the false negative, and leaves the decision to the expert.
three mini cases
Case 1 — Prioritization. One screening unit has 120 newborn results; 8 "left". The audiologist gives the anonymous list to the artificial intelligence and has the tracking list edited. AI separates 8 "fail" cases and 3 cases with a note of "pass, but there is a family history of hearing loss". The specialist follows 11 babies. AI edited 120 lines in minutes; The risk of kidnapping has decreased.
Case 2 — False negative capture. In the school screening, a 9-year-old child received a "pass" in the pure tone screening, but the teacher noted that "he repeats it all the time and does not hear it when called from behind." The audiologist scans this note into the artificial intelligence with the filter "Are there any symptoms in the pass group?" The child is taken for further evaluation and a slightly wavy loss is detected. Numerical “passed” did not override the clinical symptom.
Case 3 — Workplace bias. There is annual noise audiometry of 50 workers in a factory. The expert asks for a sketch comparing the anonymous values with last year. YZ marks 5 workers showing significant threshold shift at 4000 Hz. The specialist refers them to diagnostic evaluation and personal protective equipment review; Initiates the occupational health notification process.
Copiable prompt templates
SCREEN LIST PRIORITIZATION TEMPLATEDivide the following anonymous screening results into three groups: (1) follow-up priority ("fail"), (2) review due to grade/doubt ("pass" + indication), (3) routine. Screening is not diagnosis, it is not a definitive diagnosis. Data: [list]
FALSE NEGATIVE FILTER TEMPLATERecords with a "Pass" result but with a symptom/family suspicion/development note appear as a separate list and write why you recommend follow-up. Data: [pass group + notes]
WORKPLACE TREND COMPARISON TEMPLATECompare this year's and last year's noise audiometry thresholds for the following anonymous worker codes; Mark those that show significant deterioration (especially 3000-6000 Hz). The decision and notification belongs to the expert. Data: [values]
FAMILY/RELATED INFORMATION TEMPLATE Write an informational paragraph in plain Turkish with the message "Screening is not a diagnosis, passing does not give a lifetime guarantee, staying does not mean deaf." Don't create anxiety, be realistic, emphasize the importance of follow-up.
Common mistakes
- Mistaking a scan for diagnosis. Presenting a "pass/fail" result as a definitive diagnosis.
- Forgetting the false negative. Overlooking symptomatic cases in the "Passed" group.
- False reassurance to family. Saying "Pass = no problem"; Ignoring possibilities such as auditory neuropathy.
- Bypassing the notification obligation. Not reporting the loss detected in the workplace to the occupational health process.
- Entering identification data into the vehicle. Giving the baby/worker identity to artificial intelligence without anonymizing it.
In summary
Screening is an elimination that eliminates potential problems and directs them to diagnostic evaluation; It is not a diagnosis. AI is a powerful pre-screener for sorting through large numbers of results and prioritizing those that need follow-up, but it cannot completely prevent false negatives. If there are symptoms despite "Pass", it is the specialist's decision to follow up, and if "Fail", further evaluation is the decision of the specialist; Notification and protection obligations must be enforced in cases of losses detected in the workplace.
Application task
Prepare an anonymous, made-up screening list (at least 15 lines, pass/fail, and symptom notes on some). Give the AI two lists with the “prioritization” and “false negative filter” templates. Then review it manually: is there a line the AI missed or overmarked? Produce an information paragraph for the family using the “family information” template and correct it for exaggeration/under-assurance.
checklist
- [ ] I provided the scan results anonymously and structured.
- [ ] "Failed" and I prioritized following those at the border.
- [ ] I reviewed the symptomatic cases in the "Passed" group separately.
- [ ] I left the final diagnosis and further testing decision to the specialist.
- [ ] I have observed the notification/protection obligation in the workplace context.
- [ ] I wrote the family information in a realistic and non-anxious language.