Look at any pathology report and you’ll see numbers in brackets next to each result. It’s tempting to read them all the same way: inside is normal, outside is bad. But those numbers aren’t always the same kind of number. Some describe what’s typical in healthy people. Others are lines drawn from clinical evidence that tell a doctor to act. Mixing them up is an easy mistake to make.
Reference intervals: describing healthy people
A reference interval (RI) is usually defined as the central 95% of results from a carefully selected reference population of healthy people, running from the 2.5th to the 97.5th percentile. By definition, then, 2.5% of healthy people sit below the interval and 2.5% sit above it. An RI describes where most healthy people sit. It doesn’t tell you where disease starts, or where treatment helps.
How a reference interval is made
The main international guideline is CLSI EP28-A3c. It recommends collecting samples from screened healthy volunteers to give at least 120 results for each group you want to report separately, such as each sex or age band.
The limits are then calculated in one of two ways (Ozarda, 2016):
- Nonparametric: sort the results and read off the 2.5th and 97.5th percentiles. This makes no assumption about the shape of the distribution, and it’s the guideline’s recommended method. With 120 results you can also put 90% confidence intervals around each limit.
- Parametric: assume the results (or a transformation of them, such as their logarithms) follow a Gaussian “bell curve”, then use the mean and standard deviation. Many analytes are skewed, so transformation is often needed. A large multicentre comparison found the two approaches give very similar limits.
Either way, outliers need to be found and removed first.
Transference and verification
Recruiting 120 volunteers per group is expensive, and the guideline admits that few laboratories, or even manufacturers, do it. Instead, it encourages labs to transfer an interval established elsewhere and verify that it suits their own method and population. The simplest verification uses just 20 samples from local healthy people. If no more than 2 fall outside the proposed limits, the interval can be adopted, at least provisionally. If 3 or more fall outside, you test another 20. If that set also fails, look for a method or population difference. Indirect methods (“data mining”) instead apply statistics to large sets of routine patient results. They help with hard-to-recruit groups like neonates, children and older people, but rely on statistics rather than screening to exclude unwell people.
Harmonised reference intervals in Australia
Not long ago, Australian labs using the same methods quoted different intervals. A 2010 survey found sodium lower limits ranging from 132 to 137 mmol/L, and potassium upper limits from 4.8 to 5.5 mmol/L, so the same result could be flagged at one lab and not another.
Between 2012 and 2014, the AACB ran a series of workshops, working with the RCPA, to fix this (Tate et al, 2014). The evidence included a formal reference interval study in Canberra (“Aussie Normals”), analysis of around 5 million outpatient results, a bias study across 24 labs and 8 analyser platforms, and paediatric data from 15 laboratories. The result was a first panel of common intervals, endorsed by the AACB and the RCPA. Adult examples include:
- Sodium: 135–145 mmol/L
- Potassium: 3.5–5.2 mmol/L
- Calcium: 2.10–2.60 mmol/L
They assume methods traceable to recognised reference materials, and labs still verify them locally.
Harmonisation is ongoing. The RCPA’s current table of harmonised chemical pathology intervals now also covers bilirubin, creatine kinase, liver enzymes and lipase, and existing intervals get reviewed. In August 2025, the AACB harmonisation committee released a draft update for adult ALP, proposing sex- and age-specific intervals (including one for women aged 50 and over) to reduce over-flagging in postmenopausal women.
Partitioning: one size doesn’t fit all
Sometimes one interval for everyone would be misleading, so the reference population is split, or partitioned, usually by age and sex. Statistical rules help decide when subgroups differ enough to justify separate limits. Two Australasian examples:
- Creatinine depends on muscle mass, so adult intervals differ by sex: 60–110 µmol/L for males and 45–90 µmol/L for females (ages 19 to under 60).
- Alkaline phosphatase (ALP) rises during bone growth. The harmonised table has separate bands from birth through the teenage years, with separate intervals for boys and girls from age 10.
Decision limits: drawing a line for action
A decision limit (also called a cut-off or threshold) is set from clinical evidence, such as outcome studies, the risk of complications or diagnostic accuracy studies. It answers a different question: not “is this typical?”, but “at what level should we diagnose, treat, or worry?”
Decision limits can sit inside, outside or right on the edge of a reference interval, and they’re often specific to a context, such as diagnosis versus monitoring. For some tests, decision limits have replaced reference intervals altogether. The international guideline mentioned earlier gives cholesterol and HbA1c as examples.
Example 1: HbA1c and diabetes diagnosis
The RCPA Manual lists three bands for HbA1c:
- 15–42 mmol/mol (3.5–6.0%): reference interval
- 43–47 mmol/mol (6.1–6.4%): indeterminate
- ≥48 mmol/mol (≥6.5%): diagnostic of diabetes
The 48 mmol/mol line isn’t the top of the healthy distribution. Diabetes thresholds have long been based on the glucose levels above which diabetic retinopathy appears, and pooled data from eight studies showed HbA1c is at least as strongly linked to retinopathy as blood glucose (d’Emden et al, 2012).
The threshold comes with rules, too. The Australian Diabetes Society says asymptomatic people need two tests on separate days, both in the diabetes range, for a clinical diagnosis, and people with symptoms should have diabetes confirmed by blood glucose. Iron or B12/folate treatment, haemolytic anaemia and some haemoglobinopathies can make HbA1c misleadingly low. A threshold is only as good as the test underneath it.
There’s even a second decision limit on the same test: a general target for glycaemic control of <53 mmol/mol (7%) in people with diabetes. Same analyte, different question, different number.
Example 2: lipids, where “normal” isn’t the point
Lipids show the difference most clearly. The AACB’s 2018 harmonised lipid reporting guideline recommends that labs flag results against these limits:
- Total cholesterol >5.5 mmol/L
- Triglycerides >2.0 mmol/L
- LDL cholesterol >3.0 mmol/L
- HDL cholesterol <1.0 mmol/L (males) or <1.2 mmol/L (females)
- Non-HDL cholesterol >4.0 mmol/L
The guideline says plainly that these limits “are neither population reference intervals nor clinical action limits”. A flag is a prompt to review the patient’s absolute cardiovascular risk.
Treatment targets are different again. The 2018 document also lists National Heart Foundation targets for high-risk patients, such as LDL cholesterol <2.5 mmol/L (<1.8 mmol/L for very high risk). The 2023 Australian cardiovascular disease risk guideline instead bases treatment on estimated 5-year risk (Nelson et al, 2024). Lipid-lowering medicine is recommended at high risk (≥10%) and considered at intermediate risk (5% to <10%), and a total cholesterol above 7.5 mmol/L warrants treatment regardless. Its summary sets no LDL target, referring to specific lipid guidelines instead. Internationally, the 2019 ESC/EAS guidelines set an LDL cholesterol goal for very-high-risk patients of below 1.4 mmol/L and at least a 50% reduction from baseline.
Example 3: troponin and the 99th percentile
Troponin is a neat twist: its key decision limit comes from a reference population, but it’s used to make decisions, not to describe health.
The Fourth Universal Definition of Myocardial Infarction (2018) defines myocardial injury as a cardiac troponin above the 99th percentile upper reference limit (URL). Note that’s the 99th percentile, not the 97.5th of a conventional RI. And injury alone isn’t a heart attack. Acute MI needs a rise and/or fall in troponin with at least one value above the 99th percentile, plus clinical evidence of ischaemia, such as symptoms, new ECG changes or imaging findings.
A few more details matter:
- The 99th percentile is assay-specific, so you can’t swap numbers between manufacturers.
- The IFCC recommends sex-specific 99th percentiles.
- Rapid “rule-out” pathways often use different, assay-specific cut-offs and change values (“deltas”), validated for that purpose.
Example 4: eGFR and CKD stages
For kidney function, outcome data, not healthy-population percentiles, set the lines. In Kidney Health Australia’s handbook, chronic kidney disease (CKD) means an eGFR below 60 mL/min/1.73m², or evidence of kidney damage such as albuminuria, for at least 3 months. Stages follow eGFR: 1 (≥90), 2 (60–89), 3a (45–59), 3b (30–44), 4 (15–29) and 5 (<15 or on dialysis).
Here’s the interesting bit. An eGFR under 60 is very common in older people, so you might expect an age-adjusted “normal range”. But large meta-analyses showed it predicts higher risks of death, kidney failure and other harms, without consistent differences by age. So the Australasian Creatinine Consensus Working Group recommends against age-related decision points for eGFR in adults, and says a low eGFR shouldn’t be considered a normal part of ageing (Johnson et al, 2012).
Example 5: vitamin D
For vitamin D (serum 25-hydroxyvitamin D), an Australian and New Zealand position statement says a level of at least 50 nmol/L at the end of winter (10–20 nmol/L higher at the end of summer) is needed for optimal musculoskeletal health (Nowson et al, 2012). It’s a target based on bone and falls evidence, not the edge of a healthy distribution. The same statement warns that many automated assays are less precise below 50 nmol/L, exactly where the decision is made.
Critical limits: pick up the phone
A third kind of line is the critical or alert limit. It doesn’t mean “unusual” or “risky in the long run”. It means “this result could signal serious harm, so tell a clinician now”, usually by phone.
An RCPA–AACB working party has developed a harmonised starter alert list for Australia and New Zealand from outcome evidence, risk analysis, workload and stakeholder input (Campbell et al, 2019). Its worked example recommends an upper potassium alert threshold of >6.2 mmol/L for anyone older than 6 months, well above the reference interval’s upper limit of 5.2 mmol/L. A potassium of 5.4 mmol/L is outside the RI, but nobody needs a 2 am phone call about it.
Why “just outside” is often not abnormal
Because an RI covers 95% of healthy people, one in 20 perfectly healthy people will have a result outside the reference interval for any given test. That adds up quickly. If tests were independent, the chance of all 10 results landing inside their intervals would be 0.95¹⁰, or about 60%, and for 20 tests about 36%. Tests aren’t fully independent, so these are approximations, but the point stands. Electrolytes, liver function tests and a full blood count together cover around 20 analytes, making it statistically likely that at least one will be flagged (Smellie, 2006).
Add measurement uncertainty and day-to-day biological variation, and borderline flags are common. Many return inside the interval when repeated (regression to the mean). So a slightly abnormal result in someone who is otherwise well usually calls for a repeat test after an interval, not a cascade of investigations. Further work-up makes more sense if the result stays abnormal, gets worse, or fits with symptoms or other results.
What this means for patients
If you’re looking at your own report:
- A flag isn’t a diagnosis. Results just outside the interval are common in healthy people. Your doctor will interpret yours alongside your history, symptoms and other tests.
- Use the interval printed on your report. Intervals can differ between labs, so compare your result with the one from the lab that did the test (Pathology Tests Explained).
- Some lines are about risk, not how you feel. A cholesterol or HbA1c flag can appear when you feel completely well, because those limits aim to prevent future problems.
- Trends matter. A change in your own results over time can mean more than one number compared with a population.
- Read the report comments. Labs often note which kind of limit applies, such as a diagnostic threshold, treatment target or risk flag.