Units
1. Introduction to Artificial Intelligence in the Medical Laboratory: Roles, Boundaries, Validation and Ethics 2. Result Interpretation Support and Reference Ranges: Validating the AI Sketch 3. Quality Control and Delta Check: Analytical Error Catching with AI 4. Panic/Critical Value Tracking and Notification: Security-Critical Process 5. LIS Integration and Data Flow: Connecting AI to the Right Place 6. Workflow Automation: From Sample Acceptance to Result Release 7. Microscopy and Image Pre-Assessment: Smear, Urine and Microbiology 8. Pre-Analytical Phase and Sample Compatibility: Hemolysis, Lipemia, Icterus 9. Reporting and Communication: Clear Result for Clinician and Patient 10. Data Privacy, KVKK, Ethics and Model Validation 11. End-to-End Workflow, Quality Management and Self-Audit
Unit 2 / 11

Result Interpretation Support and Reference Ranges: Validating the AI Sketch

Gains:

  • Understand that the reference range depends on age, gender, method and population and be able to verify the general ranges of artificial intelligence with the ranges of your own laboratory
  • Use artificial intelligence to produce a clear preliminary interpretation of test results and be able to provide expert clinical interpretation and final decision.
  • Ability to catch pitfalls such as panic value, unit inconsistency, and lack of clinical context before interpretation

A laboratory result is almost meaningless without a reference range underneath it. "Ferritin 18" alone is neither high nor low; It only makes sense with the information "18 ng/mL, female reference 15-150". Result interpretation is the art of putting a number into context: the right reference range, the right unit, the right age-gender group, and the patient's clinical condition. AI can be a powerful generator of this interpretation — but this is where the most dangerous errors occur, because AI is fluent in providing a “generally accepted” reference range, and that range may not be the same as your lab's validated range.

In this unit, you will learn how to safely use artificial intelligence as an assistant that pre-interprets the results in an understandable way; We'll walk you through how to catch reference range, unit, and clinical context pitfalls before interpretation. The basic principle remains unchanged: AI generates preliminary interpretation draft; The specialist makes the decision to release the clinical interpretation and conclusion.

Why can't the reference range be "generic"

Reference interval is the central 95% value distribution of an analyte in a reference population considered healthy. Many variables are hidden in this definition:

  • Age: The ranges of many analytes are different in newborns, children, adults and the elderly. Alkaline phosphatase (ALP) is much higher in the child than in the adult due to bone growth; Someone who does not know this may think that the child's normal ALP is "high".
  • Gender: Analytes such as hemoglobin, ferritin, creatinine, uric acid have different ranges in men and women.
  • Method and device: The same analyte may give different numerical ranges with different measurement methods. Therefore, each laboratory verifies the reference range according to its own device and population.
  • Physiological state: Pregnancy, hunger/satiety, time of day (such as cortisol) affect the result.

So the rule of thumb is this: the reference range given by the AI ​​is a preliminary information, while your own laboratory's certified range is the real one. When the two conflict, that of your own laboratory prevails.

Attention: Ask the AI ​​"is this value normal?" Asking and interpreting the report based on the answer produces systematic errors due to reference range differences. First give your laboratory's range to the prompt, then ask for comments.

Step by step: safe result interpretation flow

  1. Clarify the unit. The same analyte can be reported in different units (glucose mg/dL or mmol/L; calcium mg/dL or mmol/L). Explicitly give the AI ​​the unit and check back if it has converted.
  2. Provide your own reference range. Add age, gender, and your lab's range to the prompt. Don't let the AI ​​fit ranges from its own memory.
  3. Give clinical context. Information such as diagnosis, medications, pregnancy, fasting status can completely change the interpretation.
  4. Have panic/critical values ​​marked. Have critical thresholds checked before starting the comment; If there is a critical value, notification, not comment, takes priority.
  5. Filter the draft. Read the AI-generated comment, soften the language of certainty, remove diagnostic implications, and add the note “consult your physician.”

The following table summarizes common reference spacing pitfalls:

trap

example

right approach

age blindness

Mistaking high ALP in a child as "abnormal"

Use age specific range

gender blindness

Interpreting female ferritin with male range

Use gender specific range

Unit interference

Mistaking 5.5 mmol/L glucose as mg/dL

State the unit clearly

Method difference

Applying the general range to your own device

Verify lab range

lack of context

Interpreting TSH in pregnancy with general range

Add physiological state

Limit of interpretation: result is not diagnosis

The laboratory result is a piece of evidence, not the diagnosis itself. A high CRP (C-reactive protein, a marker of inflammation) does not mean "there is an infection"; Inflammation can have many causes. A positive screening test does not indicate definitive disease; Confirmatory testing and clinical evaluation are required. When explaining the result, AI can easily make precise and incorrect sentences such as "this value indicates disease X". Your job is to reduce this language to "this value may be compatible with X, but the diagnosis is made by clinical evaluation." The boundary between interpretation and diagnosis is central to laboratory ethics.

Weak prompt / Strong prompt

Weak prompt:

Hemoglobin 10, ferritin 8. What does this mean, what disease does the patient have?

This request is unitless, has no age-gender, no reference range, and directly asks for a diagnosis of “what disease?” — ​​pushing the AI ​​into the role of physician and inviting a definitive, wrong answer.

Powerful prompt:

Your role: assistant preparing a pre-interpretation DRAFT to the laboratory specialist. Diagnosis; Use “possible” language; add a note directing the patient to the physician.Patient: 34-year-old woman, not pregnant, regular menstruation.Results (with our laboratory reference):- Hemoglobin 10.2 g/dL (ref female 12.0-15.5)- Ferritin 8 ng/mL (ref female 15-150)- MCV 74 fL (ref 80-100)Task: explaining this picture to the patient in plain language write an outline; making a deduction; State what information requires clinical evaluation.

The strong prompt gives unit and reference, adds clinical context, delimits the role, and requests the language of “possible.”

three mini cases

Case 1 — Correct use. A specialist gives the values ​​of Hb 10.2 g/dL, ferritin 8 ng/mL, MCV 74 fL to YZ with the strong prompt above in a 34-year-old female patient. YZ produces a sober outline that “low ferritin and low MCV may be compatible with iron deficiency anemia, but the diagnosis is made by physician evaluation.” The expert verifies the reference ranges with those of his own laboratory and approves the draft. Duration: 2 minutes instead of 8 minutes.

Case 2 — Reference range trap. Another specialist asks YZ about the ALP 320 U/L value in a 5-year-old child, but does not specify the age. YZ assumes the adult range (approximately 40-130 U/L) and interprets the result as "significantly high, liver/bone pathology should be considered." However, in a 5-year-old child, 320 U/L may be close to normal due to ALP bone growth. The comment was incorrect because no age was given. Lesson: always put the age in the prompt.

Case 3 — Unit interference. A technician gives the glucose 5.5 value to AI without units. YZ assumes mg/dL and says "marked hypoglycemia"; whereas the value is mmol/L and 5.5 mmol/L (about 99 mg/dL) is completely normal. Since the unit was not specified, a serious misinterpretation arose. Lesson: without a unit, the number cannot be interpreted.

Copiable prompt templates

PRE-INTERPRETATION DRAFT TEMPLATERole: assistant preparing a pre-interpretation draft to the laboratory specialist.Guidelines: make a diagnosis; Use “possible/may be compatible” language; In each comment, indicate which reference range you assume; avoid definitive statement; Add a note at the end: "The diagnosis is made by clinical evaluation, consult your physician."Patient: [age, gender, relevant clinical information].Results (with our unit + laboratory reference): [list].

REFERENCE RANGE VERIFICATION TEMPLATEFor each analyte below: use the laboratory reference range I provided; do not add range from own memory. Indicate whether it is appropriate for age and gender. If there is an analyte for which no range is given, mark it as "[lab range required]", do not guess. Analytes: [list].

UNIT CONSISTENCY TEMPLATECheck the units of the results below. If a unit is missing, type "[unit required]". If you convert, show both the source and target unit and multiplier; write down the intermediate steps for me to recalculate the result. Results: [list].

PATIENT LANGUAGE SIMPLIFICATION TEMPLATETranslate the following technical comment into plain Turkish that a patient without medical training can understand. Don't use fear/panic language, don't imply a definitive diagnosis, end with "consult your physician for evaluation." Technical comment: [text].

Common mistakes

  • Relying on the AI's reference range. The overall range may not be the range for your device/population; Give your own value to the mother.
  • Skipping age and gender. Child, elderly, pregnant and gender differences completely change the range in many analytes.
  • Not specifying the unit. The same number carries completely different clinical meaning in different units.
  • Confirming the diagnostic language. Leaving the sentence "This value is disease X" uncorrected confuses interpretation and diagnosis.
  • Embed the panic value in the comment. If there is a critical value, the priority is rapid notification, not comment (Unit 4).
Tip: Do a “certainty hunt” as you read the commentary draft: look for words like “indicates,” “evidence,” “diagnoses,” and replace each with “may be compatible with,” “may suggest,” “requires evaluation.” A small linguistic difference is a big ethical difference.

In summary

Result interpretation is the task of matching the number with the correct reference range, correct unit, correct age-gender and clinical context. Artificial intelligence can quickly produce an outline of this meeting, but assuming the reference range is universal, omitting the unit, or using diagnostic language will lead to serious errors. Safe flow: clarify unit, provide self-reference, give clinical context, mark critical values, filter draft. The conclusion is a piece of evidence; The diagnosis is made by clinical evaluation and the decision to release the interpretation belongs to the specialist.

Application task

Select three abnormal parameters from your (or a sample) report. For each, write (1) your laboratory's validated reference range, (2) the unit, (3) age-gender and physiological status, if any. Then create a prompt with the "Pre-Interpretation Draft" template, print it out, and correct each sentence with a diagnostic implication by performing a "precision hunt". Finally, simplify the interpretation into patient language.

checklist

  • [ ] I have clearly stated the unit for each analyte.
  • [ ] I gave the validated reference range of my own laboratory to the prompt.
  • [ ] I added age, gender and physiological state (pregnancy, starvation, etc.).
  • [ ] I had it checked first to see if there was a critical/panic value.
  • [ ] I translated AI's diagnostic sentences into "possible/compatible" language.
  • [ ] I added the note "Consult your physician" and approved the comment as an expert.
Previous Next