Gains:
- Ability to summarize and prioritize hemogram, biochemistry and urinalysis results with species/race reference range, supported by AI
- Ability to cross-validate AI interpretation with reference range, sample quality and clinical picture
- Ability to understand that reference ranges vary depending on device and type, AI does not produce numbers, and the final decision lies with the physician.
The laboratory is an interpreter that converts the invisible into numbers. A hemogram (complete blood count; the number and ratio of blood cells), a biochemistry panel (blood tests that show organ function), or a urinalysis may reveal kidney failure, an infection, or a metabolic disease that the exam cannot detect. But a laboratory result alone is not a diagnosis; is a piece of a puzzle. Artificial intelligence (AI) is a quick help in organizing these pieces, highlighting out-of-reference values, and summarizing trends. But there is a critical rule here: reference ranges vary by species, breed, age and device. If AI assumes a general range, it will produce incorrect interpretation. In this unit, you will learn how to summarize and prioritize laboratory results safely with AI. It is the physician who sees the clinical picture that turns the result into a diagnosis.
Reference range: the context of everything
Whether a value is "high" or "low" depends on which reference range it is viewed from. These ranges are different in cats and dogs, kittens and elderly people, and even in different analyzer devices. For example, the normal hematocrit value of a greyhound is higher than other dog breeds; It appears "high" relative to a general range but is normal for the breed. When interpreting values to AI, be sure to give the reference range and type of the device you are using. Otherwise, the AI may attach the wrong "abnormal" label based on an average range it learned from the internet.
Caution: AI does not perform numerical analysis itself; It edits the numbers and reference range you provide. It may make an error if asked to "calculate" a value. Use AI only as a summary, prioritization and organization tool.
Sample quality: garbage in, garbage out
The reliability of a laboratory result depends on sample quality. Hemolysis (breakdown of red cells) falsely increases potassium and some enzyme values; lipemia (excess fat in the blood) disrupts some measurements; clot indicates low platelet count; Wrong storage shifts values. AI does not know these pre-analytical errors; It processes the number put in front of it as it is. Therefore, the physician should evaluate the sample quality before interpreting the result.
Step by step: Lab summary with AI
- Collect context. Species, race, age, gender, clinical complaint.
- Add device reference. For each parameter, give the device's own reference range.
- Check sample quality. Is there hemolysis, lipemia, clot?
- Highlight non-reference values. Just ask the AI to flag anomalies.
- Match with clinical picture. Read the values together with the inspection and complaint.
- Watch the trend. Compare with previous results and determine the direction.
- The doctor makes the interpretation. Diagnosis and treatment decisions belong to the physician.
three mini cases
Case 1. A 14-year-old, 3.8 kg cat has creatinine 3.4 mg/dL (device reference 0.8–2.4), urea 78 mg/dL (reference 16–36), SDMA high. AI highlighted these three values as "above reference, may be related to kidney function" and made a trend chart with previous values. The physician evaluated the urine density and clinical picture and made the staging himself. AI edited; the doctor commented.
Case 2. Potassium in a dog came to 7.1 mmol/L and the AI marked "severe hyperkalemia". The physician noted significant hemolysis in the sample; In the sample taken again, the value was 4.6 mmol/L (normal). The AI only looked at the number; The physician caught the pre-analytical error.
Case 3. A greyhound had a hematocrit of 58%; The AI is "high" relative to the general canine reference, he said. The physician remembered the race-specific reference and evaluated the value as normal. This case shows that AI will mislead if device and race are not referenced.
A comparison chart
input
AI's output
Risk
No device reference given
Comment by overall range
False abnormal/normal
Species/breed not specified
Mean type assumption
Mislabeling
Sample quality unknown
Processes the number as it is
Pre-analytical error remains hidden
Clinical picture not given
Isolated number interpretation
Without context, misleading
Four copyable templates
Role: Laboratory summary assistant (you don't diagnose, you edit).Context: Species [...], race [...], age [...], complaint [...].Task: Arrange the results below according to the instrument reference ranges I PROVIDE; only mark and group NON-reference values (e.g. kidney, liver, electrolyte). Keep the comment in "possible" language. Results (value + reference): [...]
Task: Create a trend table and direction summary of [parameter] values from the following different dated results (increasing/decreasing/stable). Making clinical interpretation and diagnosis. Only numerical trend.Results: [...]
Task: In this result set, mark the values that give rise to suspicion of pre-analytical error (hemolysis, lipemia, clot, starvation) and write "sample quality must be confirmed". Results: [...]
Task: Match the following abnormal values with the clinical findings and list in "possible" language which differential diagnosis they support. Definite testimony; The doctor will make the decision through examination and examination. Values + findings: [...]
Weak prompt / Strong prompt
Weak: “Interpret these blood results, what is the patient?”
Strong: "13 year old, neutered female cat, complaining of weight loss. Arrange the following values according to the instrument references I GIVE, group only those that are not references, flag if sample quality is in doubt. Don't make a definitive diagnosis; I will interpret with the clinical picture. Creatinine 3.4 (0.8–2.4), T4 6.8 (0.8–4.0) ..."
Type, reference range and limit are given in the powerful prompt; AI's work is limited to regulation.
Common mistakes
- Not providing the device reference. AI labels incorrectly based on general range.
- Omitting the species/race context. Breed variations such as greyhound hematocrit escape.
- Ignoring sample quality. Hemolysis/lipemia values show spurious.
- Attributing diagnosis to isolated number. The value gains meaning with the clinical picture.
- Making the AI "calculate" numbers. AI makes errors in calculation; He organizes, you calculate.
Panic values and timely intervention
Some laboratory values are so critical that they require intervention as soon as they are seen; these are called panic value (critical value). For example, too high potassium (hyperkalemia) can disrupt the heart rhythm and lead to sudden death; too low blood sugar (hypoglycemia) can lead to seizures and coma; A severe hypercalcemia or a very low hematocrit requires immediate evaluation. When summarizing a result set, AI can highlight such values and mark them as “urgent attention”; This is a helpful reminder to avoid missing out on a busy day.
However, there are two traps here. First, the panic value may not always be real: a pre-analytical error (hemolysis, wrong tube, delayed processing) can produce a spurious critical value; therefore, the physician must quickly confirm the value by sample quality and clinical presentation. Second, just because the AI sees a value as “within the normal range” does not mean that that patient is not urgent; Some patients may deteriorate rapidly even with normal laboratory values because the picture is not yet reflected in the blood. Therefore, the panic value is not a call to action and the normal value is not an assurance. When faced with a critical result, the decision chain is always the same: verify sample quality, match with the clinical picture, repeat if necessary and manage the intervention as a physician. AI increases speed; The weight of the decision rests with the doctor.
In summary
The laboratory result is not a diagnosis, but a clue. AI is quick at organizing these cues, highlighting off-reference, and trend-tracking; but the reference range varies according to species, breed and device and must be given. The physician evaluates the sample quality; Pre-analytical errors are traps that AI cannot see. Read the values together with the clinical picture. It is the physician who translates the result into a diagnosis and bears the responsibility.
Application task
Get your last three lab results. Have AI organize it by giving the device reference range and type for each. Then: (1) verify whether the abnormalities flagged by the AI are indeed abnormal for the type/device, (2) look for suspicion of pre-analytical error in a sample, (3) trend chart for one of your chronic patients and interpret the direction. Match the comments to the clinical picture.
checklist
- [ ] I gave the species, race, age and complaint.
- [ ] I entered the device reference range for each parameter.
- [ ] I checked the sample quality (hemolysis, lipemia, clot).
- [ ] I didn't make the AI calculate numbers, I just arranged them.
- [ ] I matched the values with the clinical picture.
- [ ] I compared the trend with previous results.
- [ ] As a physician, I made the final interpretation and diagnosis.