The anion gap is a figure worked out from a routine blood test: sodium minus the sum of chloride and bicarbonate, in mmol/L. It has no single normal range. In a 2020 study at three hospitals, the intervals derived from patients whose electrolytes were otherwise normal were 10–17, 6–14 and 5–12 mmol/L. We placed a gap of 15 against each. It sits inside one hospital's interval and above the other two. Against that study's all-patient intervals, 9–19, 5–15 and 5–15, it is inside all three. Before ion-selective electrodes, the range in common use was 8–16 mmol/L.
Chloride shifted upward on one electrode analyser tested in Denver
The older continuous-flow analysers measured sodium by flame photometry and chloride by colorimetry. Lolekha and colleagues at Ramathibodi Hospital in Bangkok give the old analysers' ranges as 8–16 or 9–18 mmol/L, and report in 2001 that ion-selective electrodes lowered the reference range to 3–11 (PubMed 11369341). Their conclusion: "one must use an analyzer-specific reference range".
Winter and colleagues at Denver General Hospital had tested in 1990 whether 8–16 still held on a Beckman ASTRA analyser (PubMed 2302006). The range had shifted to 3–11 "in one of our laboratories", from 29 healthy volunteers and 120 blood donors, mainly because chloride values had moved up. Chloride is subtracted, so a higher reading leaves a smaller gap.
The fall was not uniform. A 2009 review by C. Higgins on acutecaretesting.org, a site run by the analyser maker Radiometer, says the electrode switch "cannot be the whole explanation" (Higgins 2009), and reports healthy-subject means from 5.9 to 12 mmol/L across eight electrode analysers. A Bangkok companion paper derived 5–12 from 124 healthy volunteers on a Synchron CX5 and found textbooks still using 8–16 (PubMed 11369334).
Hospital patients with normal electrolytes yield a different range at each of three laboratories
Pratumvinit and colleagues, writing from Bangkok, Singapore and California, studied the gap at three hospitals to compare instruments and Southeast Asian populations (doi:10.1016/j.cca.2019.10.012). Their samples came from consecutive adults whose blood was sent for electrolytes and creatinine, so these were hospital patients. One site ran a cobas 8000 analyser and two an Architect c16000. The abstract does not say which used which.
Across all patients the intervals were 9–19, 5–15 and 5–15 mmol/L; in the normal-electrolyte subgroup, 10–17, 6–14 and 5–12. Every pair of subgroup intervals differs, so the two hospitals sharing a model still arrived at different subgroup intervals. The authors conclude that "different instruments produce different AGs"; the shared-model pair suggests the instrument is not the only difference.
We counted the whole-number gaps from 0 to 30 on which the three subgroup intervals disagree, one placing the value inside while another places it outside. Ten of the 31 qualify. On the all-patient intervals, 8 do. These are points on a number line and say nothing about patients.
Potassium raises any gap by its own concentration
An alternative definition adds potassium, giving (sodium + potassium) − (chloride + bicarbonate). Higgins calls this the less frequently used definition and writes that the traditional 8–16 approximates to 12–20 mmol/L with potassium included. Potassium enters at whatever value was measured, so a sodium-only gap of 12 with potassium at 4.4 mmol/L becomes 16.4. Every interval quoted here is for the sodium-only formula.
Albumin accounts for almost all of the protein charge hidden in the gap
Part of the charge the measured ions leave unaccounted for sits on protein. Figge, Rossing and Fencl put it at about 12 mEq/L in 1991, at albumin 4.3 g/dL, and found that globulins play a negligible role, which leaves albumin carrying it (PubMed 2045713).
In 1998 Figge and colleagues, studying 9 normal subjects and 152 critically ill patients, found that each g/L drop in albumin made the gap understate the gap anions by 0.25 mEq/L, or 2.5 per g/dL (PubMed 9824071). The abstract gives no figure for normal albumin. Feldman, Soni and Dickson, across 5,328 patients in Dallas in 2005, measured a slope of 2.3 per g/dL, wrote the adjustment from 4 g/dL and recommended adjusting first (PubMed 16310513). Among the 1,578 with low or high albumin, adjustment changed whether the gap was assessed as increased, normal or decreased in 44%.
Figge's 1998 factor is written per g/L, the unit two Singapore public hospital laboratory pages use. Singapore General Hospital lists albumin at 40–51 g/L from age 13 (SGH Pathology), and KK Women's and Children's Hospital lists 35–50 g/L from age 19 (KKH eLab Book). Forty g/L is 4.0 g/dL, the reference albumin also built into the widely used corrected-calcium formula. Pratumvinit's group found the albumin–gap correlation weak, though low albumin went with lower gaps in two hospitals and high albumin with higher gaps in all three.
The reference albumin can move the corrected gap further than the coefficient does
On our arithmetic, the assumed normal albumin can matter more than the 2.3 against 2.5 question. Switching the coefficient from 2.5 to 2.3 moves the adjusted gap by 0.2 mmol/L per g/dL of deficit, or 0.4 at an albumin of 2.0 g/dL. Raising the assumed normal from 4.0 to 4.3 g/dL moves every adjusted gap by 0.75 mmol/L at the 2.5 factor, whatever the albumin. That 4.3 is the normal serum in Figge's 1991 example and not a published correction constant. Nothing here settles which reference is right; Feldman's 44% is the measurement in patients.
Sodium, chloride and bicarbonate are all the basic formula sees
A calculator can do the subtraction and the adjustments above, but cannot know which analyser produced the numbers or what interval that laboratory derived. Each ion here carries a single charge, so mmol/L equals mEq/L.
Until 21 October 2026 our anion gap calculator labelled each result low, normal or raised against a fixed 8–12 mmol/L, limits none of the intervals above gives, and moved them up by a fixed 4 for potassium. It pre-filled albumin as 4.0 g/dL and printed a corrected gap nobody asked for, ignored a g/L albumin without saying why and printed health advice beside its label. The revision made that day drops the label. It places the gap below, inside or above six attributed sodium-only intervals and any laboratory range entered, uses the measured potassium, and takes albumin in g/L or g/dL, refusing a value that only fits the other unit. With the page's default sodium 140, chloride 104 and bicarbonate 24 mmol/L, plus albumin 20 g/L, it gives a gap of 12.0, or 17.0 and 16.6 adjusted by Figge's and Feldman's factors. It places 12.0 inside five of the six intervals and 17.0 inside only the 10–17.
Unmeasured anions remain an estimate each laboratory calibrates for itself
Kraut and Madias wrote in 2007 that the gap's normal value "can vary widely", reflecting both measurement methods and variation between individuals (PubMed 17699401). Lolekha's group asks for an analyser-specific range, Higgins writes that each laboratory must determine its own reference interval, and Pratumvinit's authors say laboratorians should verify theirs. None of them treats one published interval as fitting every laboratory.