What the anion gap actually measures
Plasma is electrically neutral, so the total charge carried by cations equals the total charge carried by anions. A chemistry panel measures only one cation seriously — sodium — and two anions, chloride and bicarbonate. Subtract the two measured anions from the measured cation and you are left with a positive number: not a real gap in the charge balance, but the amount of anionic charge the panel never sees. That is the anion gap.
The unmeasured anions in a healthy person are dominated by albumin, which at physiological pH carries a large net negative charge, along with phosphate, sulphate and organic acids. Set against them are unmeasured cations — potassium, calcium, magnesium, and immunoglobulins — which are smaller in aggregate. The residue is the 8 to 12 mmol/L most laboratories quote.
The gap earns its keep in one situation: a patient with a low bicarbonate. Bicarbonate can fall for exactly two reasons. Either an acid has been added to the blood and bicarbonate has been consumed buffering it, in which case the acid's conjugate base stays behind as an unmeasured anion and the gap rises; or bicarbonate has been lost directly through the gut or the kidney, in which case chloride is retained to preserve neutrality and the gap does not move. High-gap acidosis and normal-gap acidosis have almost entirely different causes and different treatments, and the gap is the single number that tells them apart.
The formula, and why albumin has to be corrected for
The gap is AG = [Na⁺] − ([Cl⁻] + [HCO₃⁻]). Some centres use the four-electrolyte form, [Na⁺] + [K⁺] − [Cl⁻] − [HCO₃⁻]. Neither is more correct; they simply reference different intervals, because adding potassium adds its entire serum concentration to the answer. If you include potassium you must raise the reference by roughly 4 mmol/L, which is why the traditional threshold of 12 becomes about 16. Mixing the two is a common source of a spuriously high gap.
Albumin is where most bedside errors come from. It is the largest single unmeasured anion, and critically ill patients are frequently hypoalbuminaemic. Figge and colleagues showed that each 1 g/dL fall in albumin below 4.0 g/dL removes about 2.5 mmol/L of anionic charge from the plasma, and therefore shrinks the measured gap by the same amount. The correction restores it:
AGcorrected = AG + 2.5 × (4.0 − albumin in g/dL). In SI units, add 0.25 × (40 − albumin in g/L).
An ICU patient with an albumin of 2.0 g/dL has a gap that reads 5 mmol/L low. A measured gap of 12 in that patient is really 17 — a clearly raised gap that an uncorrected reading calls normal. This is not a rare edge case; it is the usual state of affairs in a critical care unit, and it is why the corrected figure is the one to act on.
Two further numbers finish the assessment. The delta gap is how far the corrected gap has risen above its reference. The delta ratio compares that rise with the fall in bicarbonate from 24 mmol/L. In a pure high-gap acidosis each unmeasured anion added should consume roughly one bicarbonate, so the ratio sits near 1. A ratio well below 1 means bicarbonate fell further than the gap rose, so something else is also consuming bicarbonate. A ratio well above 2 means the gap rose further than bicarbonate fell, so something else is propping bicarbonate up.
Worked example: a 24-year-old in diabetic ketoacidosis
The panel reads sodium 137, chloride 100, bicarbonate 12 mmol/L, albumin 3.0 g/dL. Work it through on paper.
- Measured gap. 137 − (100 + 12) = 137 − 112 = 25 mmol/L.
- Albumin correction. Albumin is 1.0 g/dL below 4.0, so add 2.5 × 1.0 = 2.5 mmol/L.
- Corrected gap. 25 + 2.5 = 27.5 mmol/L, unmistakably raised against a reference of 12.
- Delta gap. 27.5 − 12 = 15.5 mmol/L of unmeasured anion — here, β-hydroxybutyrate and acetoacetate.
- Fall in bicarbonate. 24 − 12 = 12 mmol/L.
- Delta ratio. 15.5 ÷ 12 = 1.29, inside the 1 to 2 band, so this behaves as a single, pure high-gap acidosis.
- Corrected bicarbonate. 12 + 15.5 = 27.5 mmol/L. Above 26, which is the same message the ratio gave: once you mentally remove the ketoacids, bicarbonate is at the top of normal rather than in the middle, consistent with the vomiting that so often accompanies ketoacidosis.
- Expected PaCO₂. Winter's formula gives 1.5 × 12 + 8 = 26 mmHg, ± 2. If the measured PaCO₂ came back at 40, the patient is not compensating and has a superimposed respiratory acidosis — a genuine emergency in this setting.
Notice how much the last three steps added. The gap alone said “ketoacidosis”. The delta ratio, corrected bicarbonate and Winter's prediction said “ketoacidosis, plus a chloride-losing process, and check the ventilation”.
How to read the number you get
Start with the corrected gap against the reference your laboratory quotes. Older flame-photometry analysers gave a normal gap of 8 to 16 mmol/L with a midpoint of 12; ion-selective electrodes read chloride slightly higher and have pulled many modern reference intervals down to 3 to 11 mmol/L. Using 12 as your threshold on a modern analyser will make you miss modest elevations, so change the reference field to whatever your own laboratory reports.
A raised corrected gap sends you looking for the anion. The classic memory aid is GOLDMARK — glycols (ethylene and propylene), oxoproline (from chronic paracetamol), L-lactate, D-lactate, methanol, aspirin, renal failure, and ketoacidosis. In practice lactate, ketones and uraemia account for the overwhelming majority; measure a lactate and ketones before reaching for anything exotic. If those are unrevealing and the gap is large, calculate the osmolar gap as well, because a toxic alcohol produces a raised osmolar gap early and a raised anion gap only later, once it has been metabolised.
A normal corrected gap in the presence of a low bicarbonate points to bicarbonate loss: diarrhoea, a proximal or distal renal tubular acidosis, a ureteric diversion, or large-volume saline. A genuinely low corrected gap is uncommon and usually artefactual, but a cationic paraprotein in myeloma, lithium or bromide toxicity, and severe hypercalcaemia can all produce one.
Finally, treat the gap as a screening number, not a diagnosis. It has poor sensitivity for modest lactate elevations: a lactate of 4 mmol/L raises the gap by 4, which can easily disappear inside the reference interval. A normal gap does not exclude a lactic acidosis, and in a shocked patient you measure the lactate directly.
A failure mode that has nothing to do with arithmetic: mismatched draws
The three electrolytes behind the gap are usually drawn together, but they are not always read together. A ward panel and a blood-gas syringe can be minutes to hours apart, and in a patient who is being actively treated that gap in time is not a footnote — it is the difference between measuring the disease and measuring the treatment.
The clearest version of this happens during fluid resuscitation for diabetic ketoacidosis. Large volumes of 0.9% saline carry far more chloride than plasma does, so chloride rises and bicarbonate is diluted between one draw and the next even while the ketoacidosis itself is unchanged. The corrected gap on the second panel can look like it is “closing” toward normal, when what has actually happened is a normal-gap process — iatrogenic hyperchloraemia — being added on top of the original high-gap one. The delta ratio calculated from that second panel will read low, which is the correct arithmetic answer to the wrong question if the two draws are compared as though nothing but the acidosis had changed.
The practical check is simple: read the timestamp before you read the number. If a trending gap has moved because bicarbonate rose while chloride also rose, ask what was infused between the two draws before concluding the acidosis is resolving. A repeat gap once boluses have stopped, or a directly measured ketone level, settles the question that electrolytes alone cannot.
Delta ratio bands and what each one implies
| Delta ratio | Corrected bicarbonate | Interpretation | Typical setting |
|---|---|---|---|
| < 0.4 | Well below 22 | Mostly a normal-gap acidosis | Diarrhoea or renal tubular acidosis with a small added anion |
| 0.4 – 0.8 | Below 22 | Mixed high-gap and normal-gap acidosis | DKA part-treated with large-volume saline |
| 0.8 – 2.0 | 22 – 26 | Pure high anion gap acidosis | Lactic acidosis, early ketoacidosis |
| > 2.0 | Above 26 | High-gap acidosis plus metabolic alkalosis or chronic respiratory acidosis | Ketoacidosis with vomiting; lactic acidosis in chronic CO₂ retention |
The two columns agree by construction: corrected bicarbonate equals measured bicarbonate plus the delta gap, so a ratio above 2 and a corrected bicarbonate above 26 are two statements of the same arithmetic.
Mistakes that make an anion gap misleading
- Not correcting for albumin. The single commonest error in critical care. At an albumin of 2.0 g/dL the gap reads 5 mmol/L low, which is enough to hide a lactic acidosis entirely.
- Mixing conventions. Adding potassium to the gap but comparing against a reference of 12 manufactures a raised gap in a normal patient.
- Using 12 when your laboratory says 8. Ion-selective electrodes shifted reference intervals downward. Read your own laboratory's interval off the report.
- Calculating the delta ratio when the gap is normal. The ratio answers the question “does the rise in gap account for the fall in bicarbonate?”, which is meaningless when the gap has not risen.
- Forgetting that bicarbonate on a chemistry panel is total CO₂. It runs about 1 to 2 mmol/L above the blood-gas calculated bicarbonate, and the two are drawn at different times.
- Ignoring hyperglycaemia's effect on sodium. A very high glucose pulls water into the plasma and dilutes sodium, which lowers the gap. Check the corrected sodium before concluding a diabetic patient has no gap.
- Treating a normal gap as reassurance. Modest lactate elevations hide inside the reference interval. Measure lactate when the clinical picture demands it.
Where this sits among the acid-base tools
The anion gap is the bedside shortcut of the bicarbonate-centred (Henderson-Hasselbalch) approach. The Stewart physicochemical approach reaches the same conclusions through the strong ion difference and the strong ion gap, which handle albumin and phosphate explicitly rather than by correction. For routine work the corrected gap and the delta ratio give the same answers with far less arithmetic, and the albumin correction closes most of the distance between the two methods.
Related bedside numbers on this site: serum osmolality and the osmolar gap for suspected toxic alcohol ingestion, fractional excretion of sodium when the acidosis accompanies acute kidney injury, and eGFR when you suspect uraemic acidosis.
Key terms
- High anion gap metabolic acidosis (HAGMA)
- An acidosis in which an acid has been added to the blood; the conjugate base remains as an unmeasured anion and raises the gap.
- Normal anion gap (hyperchloraemic) acidosis
- An acidosis caused by bicarbonate loss, with chloride retained to preserve electroneutrality, so the gap is unchanged.
- Delta gap
- The rise in the corrected anion gap above its reference value — an estimate of the concentration of unmeasured anion present.
- Corrected bicarbonate
- Measured bicarbonate plus the delta gap: what the bicarbonate would be if the high-gap acid were removed. Below 22 implies a second, normal-gap acidosis; above 26 implies a metabolic alkalosis.
- Winter's formula
- The expected respiratory compensation for a metabolic acidosis: PaCO₂ = 1.5 × [HCO₃⁻] + 8, accurate to about ± 2 mmHg.
- Total CO₂
- What a chemistry analyser reports as bicarbonate. It includes dissolved CO₂ and carbamino compounds, so it runs slightly above the blood-gas bicarbonate.