What is the difference between a lab reference range and a decision threshold?

My report has two kinds of number on it and I have been treating them as the same kind of thing. Some analytes come with an interval — "3.5 to 5.3", with my value flagged if it falls outside. Others come with a single cut-off and a comment, like LDL-C with a note saying "target below 1.8 mmol/L if very high risk". My LDL-C is 2.6, which is not flagged as abnormal, and yet the comment implies it is above where it should be. My HbA1c is 45 mmol/mol, also not flagged, and yet 48 is apparently the diabetes threshold, so 45 is presumably not a comfortable place either.

So: where do reference intervals actually come from, where do the thresholds come from, and why do they disagree? I would like to know which of the numbers on my report are statements about a population and which are statements about my risk, because I have clearly been reading the first kind as though it were the second.

Secondary question, since the report also came in different units from my last one: are there conversions I should have memorised so I stop comparing numbers that are not comparable?

They are different kinds of claim entirely. A reference interval describes what a population looks like. A decision threshold describes where an outcome changes. Nothing guarantees they coincide, and for the analytes you have named they emphatically do not.

Where a reference interval comes from

A lab recruits a reference population — typically a few hundred people screened as apparently healthy — measures the analyte on its own platform, and reports the central 95% of the resulting distribution, usually partitioned by sex and sometimes by age. That is it. Three consequences follow directly:

• It is definitionally a 5% false-flag rate. Being outside the interval means you are unusual relative to that reference population, not that you are ill.
• It is method-specific. The interval belongs to that assay on that platform. Carrying a number from one lab to another lab's interval is a category error.
• It inherits the health of the reference population. If a quarter of the apparently healthy recruits had undiagnosed hepatic steatosis, the upper limit of ALT will be too generous. This is exactly what happened, and it is why the ALT upper limit is contested.

Where a decision threshold comes from

From outcomes, not distributions. Someone followed a large cohort or ran a trial, related the analyte to events, and picked a value at which the balance of benefit and harm from acting changes. So:

• HbA1c 48 mmol/mol (6.5%) is a diagnostic threshold chosen because retinopathy prevalence rises steeply above roughly that point in population data. Your 45 is not "normal" in a reassuring sense — it is high-risk non-diabetic hyperglycaemia, the range in which progression to type 2 diabetes is substantially elevated. The lab did not flag it because it is not outside a population interval. The threshold and the interval are answering different questions and only the threshold is about you.
• LDL-C 1.8 mmol/L is not a reference limit at all. Population LDL-C in a Western reference group centres well above it. It is a treatment target derived from trials in which lowering LDL-C reduced events, with the target set lower for higher baseline risk. Your 2.6 is unremarkable relative to the population and above target relative to your risk category, and both statements are true simultaneously.
• eGFR 60 mL/min/1.73 m² is a staging boundary, not a normal limit. A 78-year-old at 58 and a 31-year-old at 58 are in the same stage and in entirely different clinical situations.
• UACR 3 mg/mmol (30 mg/g) is an outcome-derived boundary for albuminuria, and risk rises continuously below it. There is no albumin excretion rate that is protective; less is simply better.

The general pattern: for anything cardiometabolic, thresholds are usually lower than the reference interval's upper limit, because the population being sampled is not healthy in the relevant sense. For anything hormonal or haematological, thresholds are usually derived from the interval itself, because there is no comparable outcome literature. Knowing which category an analyte falls into tells you how much weight the flag deserves.

The one that is neither

Lp(a) has a threshold of roughly 125 nmol/L quoted for elevated risk, but risk is continuous and the number is essentially fixed for life. It is not a monitoring target, it is a risk stratifier — you measure it once, and its only function is to move you into a category where other targets get tightened. Treating it as something to improve is a misunderstanding of what kind of number it is.

edited 19 May 2026 by Dr_Aoife_Brennan — tightened the wording; no substantive change

On units, since comparing across systems is where most of the avoidable errors come from. Here are the conversions worth actually knowing, with the arithmetic shown so you can check the direction rather than guess it.

HbA1c: percent (NGSP) versus mmol/mol (IFCC)

mmol/mol = (A1c% − 2.15) × 10.929
A1c% = (mmol/mol ÷ 10.929) + 2.15

Worked, for your 45 mmol/mol: 45 ÷ 10.929 = 4.117; + 2.15 = 6.27%. And the diagnostic threshold: (6.5 − 2.15) × 10.929 = 4.35 × 10.929 = 47.5, which is why the UK threshold is quoted as 48. Your 45 corresponds to 6.3%, comfortably inside the prediabetic band under either system.

Glucose: mg/dL versus mmol/L

mmol/L = mg/dL ÷ 18.0
mg/dL = mmol/L × 18.0

Worked: 100 mg/dL ÷ 18 = 5.56 mmol/L. 7.0 mmol/L × 18 = 126 mg/dL, which is the fasting diagnostic threshold in both systems.

Cholesterol and LDL-C: mg/dL versus mmol/L

mmol/L = mg/dL ÷ 38.67

Worked: LDL-C 70 mg/dL ÷ 38.67 = 1.81 mmol/L — the same target expressed twice. And 100 mg/dL ÷ 38.67 = 2.59, so your 2.6 mmol/L is about 100 mg/dL, which is a number you will see quoted constantly in American sources.

Triglycerides: a different divisor

mmol/L = mg/dL ÷ 88.57

Triglycerides use a different factor from cholesterol because the molar mass differs. Using 38.67 for triglycerides is a common and large error — it inflates the mmol/L figure by a factor of 2.3.

Creatinine: mg/dL versus µmol/L

µmol/L = mg/dL × 88.4

Worked: 1.05 mg/dL × 88.4 = 92.8 µmol/L. Note the numerical coincidence with the triglyceride factor; they are unrelated and it is a trap.

Urine albumin:creatinine: mg/g versus mg/mmol

mg/mmol = (mg/g) ÷ 8.84

Worked: 30 mg/g ÷ 8.84 = 3.39, hence the "3 mg/mmol" boundary being described elsewhere as "30 mg/g". These are the same threshold and people routinely treat them as two.

Apolipoprotein B: g/L versus mg/dL

mg/dL = g/L × 100

So ApoB 0.90 g/L is 90 mg/dL. Simple, but worth checking, because a factor of 100 error here produces a number so absurd that people assume the lab is broken.

Lp(a): the one you cannot convert

Lp(a) is reported either as mass (mg/dL) or as particle concentration (nmol/L), and there is no valid universal conversion between them, because Lp(a) particles vary in apolipoprotein(a) isoform size between individuals. Multiplying by 2 or 2.5 is a widely used approximation and it is wrong by a variable amount in a person-specific direction. If your result is in mass units and the threshold you are comparing against is molar, the honest answer is that you cannot make the comparison and should ask for the molar assay.

Keep a note of which unit system each of your historical results used. A silent unit change between labs is the single most common cause of a "my numbers collapsed" panic, and it is fixed by multiplication.

A short note on the phrase "within normal limits", because it is doing two jobs and only one of them is honest.

Used precisely, it means the value falls inside a stated interval derived from a stated population on a stated assay. Used loosely — which is how it reaches patients — it means "nothing here requires my attention today", which is a clinical judgement about the whole person and not a property of the number.

The gap between the two shows up in three recurring situations that are all relevant to anyone tracking their own labs during weight loss:

• Trending within range. A creatinine moving from 68 to 96 µmol/L across four draws is inside most reference intervals at every point and is a 41% rise, comfortably beyond the reference change value for creatinine. The flag never fires. The trend is the finding.
• Two values that are individually fine and jointly odd. A normal creatinine-based eGFR alongside a cystatin C-based eGFR fifteen points lower is two unflagged results and one real discordance. Same for a normal LDL-C alongside a clearly raised ApoB.
• A value that is normal for the population and wrong for you. A TSH of 4.1 mIU/L is inside almost every interval and may be the wrong number for someone on levothyroxine whose dose was set for a body 20 kg heavier.

The practical upshot is that the flag column is the least informative part of the report. Read the numbers, keep the series, and note that "not flagged" is a statement about a reference population you were never a member of.