The anion gap; its merits and demerits

Before allowing the reader to submerge into this swamp of digression, one offers a summary of this topic from the "Required Reading" section hidden among the CICM Fellowship Exam preparation material.

For actual education, the exam candidates are directed to the LITFL anion gap page, and to the excellent online works of Kerry Brandis.

The Gap in brief

This one of the "scanning" tests one performs on the available electrolyte values to determine the cause of a metabolic acidosis. It is defined as the sum of serum anion concentrations subtracted from the serum cation concentrations. Specifically, by convention the cations are sodium and potassium, and the anions are bicarbonate and chloride. These measured anions account for only about 85% of the total anionic charge of the extracellular fluid. The gap affords us an estimate of the concentration of this unmeasured 15%, the "miscellaneous" anionic electrolytes which are present in the bloodstream. The "normal" expected value is around 12, though it is adjusted for the serum albumin, which - being a negatively charged protein - contributes to the anionic charge of the extracellular fluid.

The anion gap is one of those concepts which lends itself well to explanation by Gamblegram:

A brief history of this concept

It is difficult to trace the exact origins of the anion gap. Even as early as 1969 this method was in widespread use. An early article by Emmett and Narins (1977) presents us with a mature method, clearly well practiced and recognised as an important tool. Prior to 1969, the term "anion gap" appears in the literature with quotation marks, suggesting that perhaps the authors viewed its acceptance as incomplete. The earliest mention I can find is in an article concerning the electrolytes of urine (1960).

Advantages of the anion gap as a diagnostic tool

The main reason we use this thing is for convenience. It helpfully stratifies metabolic acidosis for diagnostic purposes, and it is easy to calculate at the bedside.

Including potassium in the anion gap equation

There are several different permutations of the anion gap equation:

1. (Na + K) - (Cl + HCO₃)
2. (Na) - (Cl + HCO₃) 
3. (Na + K + 12) -  (Cl + HCO₃ + 24)

Equation 3 is somewhat dated (Brackett et al, 1965) and therefore there are really only two equations to choose from, which differ in whether or not theu use potassium. In other words, in some circumstances it is reasonable to omit potassium from the anion gap calculation, and the practice has become "widely accepted". The reason given for this is that its concentration is typically quite low, and therefore it can be omitted safely, usually without affecting the outcome of the calculations. The advantage of doing so is presumably because it safeguards the clinician from an unnecessary mental effort ("to simplify the equation and make it clinically useful", according to Salem et al, 1992).  A direct counterargument is that occasionally it does affect calculations. For instance, in Question 20.2 from the second paper of 2017, the delta ratio ends up 0.8 without the potassium and 1.1 with the potassium, giving rise to different interpretations of the acid-base disorder. Obviously, the higher the potassium, the more influence it will have on the difference between equations. 

The disappearance of potassium from anion gap calculations is not a recent phenomenon. The disagreement seems to have deep historical roots. For instance, in the 1976 article by Witte et al, the existence of the full equation is recognised, but the abbreviated (Na-Cl-HCO₃) formula is used. In contrast, Bleich et al (1969) used the full equation when they programmed their steampunk archaeotech to automate the interpretation of acid-base disturbances.  In general the literature is full of inconsistency, and nobody has made any attempt to demonstrate an outcome benefit or even any positive or negative influence on any clinically irrelevant outcome.   LITFL's CCC entry for the anion gap also discards potassium as "in practice offers little advantage". The definitive resource on the gap by Kraut and Madias (2007) acknowledges this as a commonplace shortcut ("many clinicians omit this variable") but does not comment on how it affects the validity of the results. 

The CICM examiners variably use potassium - an as a result, frequently one might arrive at a slightly different delta ratio value to the college SAQ answer. It is unclear whether anybody has ever been penalised for using potassium in the exam AG calculations, and the college marking rubrics being a jealously guarded secret, we may never find out about this.

Anyway. In a series of acidosis scenarios, there will be a range of electrolyte values where the use or non-use of potassium will give rise to slightly different interpretations of the acid-base disturbance, because the conventional thresholds for interpretations remain the same. To model the influence of using different anion gap equations, one might plot delta ratios in a scenario where all electrolytes other than chloride are stable. In the graph below, the sodium potassium and bicarbonate remain 145, 5 and 10, respectively. The delta ratios calculated using both possible anion gap equations are plotted. The delta ratio calculations are also altered for this experiment; logic dictates that when you omit potassium, you should also change the expected AG value. According to the RCPA, the anion gap range WITH potassium is 8-16, i.e. you'd take 12 as the reference value for delta ratio calculations. Without potassium the range is 4-13, i.e. the reference value would be 8.5. 

From this, it is apparent that there will be a tiny range of results where the equations disagree in a way which might be clinically significant. At the lower end of the delta ratio range, there is a range of values over which the presence of a possible normal anion gap metabolic acidosis might not be recognised if you added potassium. At the higher range of delta ratios, there will be a value range where the omission of potassium will result in a failure to recognise a concomitant metabolic alkalosis

The magnitude of the range over which the equations will disagree and the magnitude of the difference appears laughably tiny, provided you use a modified "normal" value for the anion gap in your delta ratio calculations. For instance, when the "avec K" equation yields a value of 1.00 suggesting a HAGMA, the "sans K" equation yields 0.89 if the potassium level was 5.0 mmol/L. This discrepancy would make one clinician think there is a minor NAGMA mixed in with the raging HAGMA, whereas the clinician who omitted the potassium might not recognise the second acid-base disorder. This becomes amplified if the potassium is higher. For instance, the graph below was calculated with a biologically plausible K+ value of 8.0 mmol/L:

Sure, with a K+ of 8.0 this sort of acid-base wankery doesn't seem like it would be at the top of your problems list, but the fact remains that with higher K+ values the choice of anion gap equation might produce results sufficiently different to influence decisionmaking. Whether or not those differences lead to different outcomes remains to be determined, and the pragmatic intensivist might strongly argue that this level of microscopic attention to detail is completely pointless at the coalface of clinical medicine. However, in at least one past CICM Part II exam question, the decision to include or not include the potassium had led to two different possible answers (Question 18.1 from the second paper of 2018).

So, which equation do I use to pass the exam?

The CICM examiners seem to use the abbreviated equation, at least 75% of the time. Of the candidates who insist on using the full equation, none have been obviously discriminated against on this basis, but we really don't know. As the literature (including Oh's Manual) recognises both equations, one is forced to conclude that the choice of equation does not matter, and that the college should also allow for a difference of opinion on this matter. Having said that, in July of 2019 Jeremy Cohen the Chair of Second Part Exam held forth that "unless otherwise specified in the question stem, anion gap calculations should be made without inclusion of potassium or correction to albumin". 

Adjustment of the anion gap for albumin

Albumin plays a role in the anion gap because conventionally it accounts for much of it. Each albumin molecule has a significant number of negatively charged amino acid residues which contributes to the net anionic charge of the body fluids; even though the albumin may be present in a miniscule molar concentration (around 0.6 mmol/L), the fact that 10 or more residues may be ionised results in a factor-of-ten multiplication of its mEq value.

So, the "normal" anion gap value is expected to be 12. Most laboratories report it as a range from 10 to 14, but for the purposes of delta ratio calculation we need to settle on a solid figure, and 12 is the generally accepted compromise. LITFL remark that the anion gap has become 12 ever since we have moved on to using ion-selective electrodes, and that in old days of flame photometry the normal anion gap value was actually 8-16. Anyway, of this 12mEq gap, 10mEq is contributed by albumin, when the albumin concentration is normal (i.e. around 40g/L).

Thus, our expectations of the "normal" anion gap need to be adjusted for albumin. The general rule is that for every 4 g/L decrease in the albumin level, the "normal" anion gap decreases by 1. Another (mathematically equivalent) approach is to instead add the albumin correction to the measured anion gap result, keeping the "normal" value the same - this is sometimes referred to as the Figge formula:

anion gap + [ 2.5 × (4 - albumin, g/dL) ]

This yields the same sort of results, at a practical level (i.e. there should be no penalty for using this in the CICM Second Part Exam, as the conclusions you reach will still be the same). However, the reader is reminded that much of what we do for the exam is a kind of performance art. The actual bedside use of this correction, as a means of improving the sensitivity of the anion gap as a diagnostic screening tool, is questionable in the era of easily available lactate and ketone measurements (for example), and the whole method by which the albumin correction formula was derived has been called into question. 

Problems with the albumin-adjusted anion gap

Using the aforementioned equation, at an albumin value of 0 one has an "expected" anion gap of 2.

This is probably somewhat silly. Consider the "misc" group in the above diagram. It consists of lactate, phosphate, urate, sulfate, and a few others which in normal states of health probably have a concentration of 0.5-2mmol/L each. These may not be fully dissociated, but still the "baseline" anion gap probably should never be less than 5mEq/L or so.

Perhaps to account for this sillyness, many authors adjust their albumin correction factor down from the usual 4g-1mmol relationship. This 1:4 correction factor varies with authors. For instance, Brandis reports a slightly down-revised correction factor of 3.5-4g; though it is unclear which source this is quoting, that value is a little more believable. Other sources - for example, this excellent review of the issue from 2005 - give numbers closer to 5-6g, which would give us a realistic-sounding minimum "albumin=0" gap of around 5-6mmol/L. Similarly,  LITFL give an estimate as 5mmol for every 10g of albumin, and they also correct for phosphate.  They offer Kraut's 2007 article as a reference. However, a detailed reading of Kraut reveals that the correct factors is somewhere around 2.3-2.5mmol for every 10g/L  of albumin. This figure was arrived at by Feldman et al, after auditing blood results and calculating anion gap values for about 5000 patients.  

Online calculators and generally the gap-using public persist in using the 4g rule, and so this convention is propagated throughout this site, and in the blood gas interpretations offered by this author.

Sources of error and bewilderment in anion gap measurements

Anion gap in pseudohyponatremia

Isosmolar hyponatremia (eg. the sort you see with ridiculously high serum protein, around 100g/L) leads to a spuriously decreased serum sodium level, and therefore to an error in the calculation of the anion gap.

Of course, if the sodium is affected by this, then the other ions are as well, including the anions, Unfortunately these concentration are given as molar values, rather than ratios. If they were given as ratios there would be no problem, because the ratio of cations to anions is preserved in this situation.

Generally, one can get around this problem by using a direct-measuring ion selective electrode to measure the electrolytes - such a device should ignore all the extra protein.

The issue is complicated further by the influence of the charge of all that extra protein. Like albumin, some might be negatively charged, and others may be positively charged. Exactly which is which is difficult to establish, and can vary considerably even within a family of very similar proteins. For instance, a study of anion gaps among multiple myeloma patients has revealed that IgM does not contribute much charge, IgG decreases the anion gap, IgA increases it, and furthermore IgA kappa chains and IgA lambda chains behave differently (kappa is more cationic). So, overall, this is a confusing mess.

Anion gap in pseudohyperchloraemia due to high serum lipids

One does not hear of it often, as one does not often see patients with absurd amounts of serum lipid. However, as it turns out, the lipids confuse the colorimetric method of chloride measurement by interfering with light scattering.

The effect is well documented, and again it is one of those things one can try to prevent by measuring chloride with an ISE rather than a spectrophotometer.

Unfortunately, the ion-selective chloride electrode still has its limitations.

Anion gap in the pseudohyperchloraemia due to halide and salicylate toxicity

The chloride-selective ISE is reasonably selective for chloride, but it tends to lose its selectivity in the presence of extreme concentrations of other, non-chloride anions. Specifically halide anions such as fluoride bromide and iodide have been implicated in spuriously high chloride measurements.

These days we don't tend to use much iodide (even the ionic contrast media have toned it down). And nobody, ever, uses therapeutic fluoride in concentrations which might cause it to alter the anion gap (your patient would be either on fire, exploding, dissolving, or some combination of all three).

Bromide medicines are also largely absent from modern formularies, making this a rare cause of anion gap abnormalities. Fortunately, we now live in an age when one can easily consume an unspeakable amount of cola beverage, developing a gain-impairing toxicity from brominated vegetable oil (which is widely used to emulsify citrus-flavoured soft drinks, and is generaly recognised as safe). And one might one day be lucky enough to meet an individual sufficiently deranged to have recreationally consumed a titanic quantity of dextromethorphan bromide, thereby causing clinically significant bromism.

Additionally, the salicylate anion can occasionally interfere with chloride measurement and cause a falsely elevated chloride reading, though this is not the case for all ISEs, and probably has something to do with the gradual loss of selectivity which occurs over the age of an even well-maintained electrode.

Anion gap in the setting of sodium correction for hyperglycaemia

Let us consider a common scenario - the HONK patient. Sure, the primary disturbance is that of hyperosmolarity due to glucose, but in addition to it there are usually ketones and lactate, which one might expect to increase the anion gap.

The HONK patient's sodium is usually corrected for hyperglycaemia; a correction factor of around 2.4mmol/L per every 100mg/dL is usually applied (that ends up being a 1.6mmo/L decrease in sodium for every 5.6mmol/L increase in glucose).

So, the end result of such calculations is the "corrected" sodium. That, to the gullible, might give the impression of "correctness" (its the corrected sodium, right, so the "uncorrected" value must be "incorrect" somehow).

However, there is nothing incorrect about the laboratory sodium. The corrected value is used purely to stop people from giving these seemingly hyponatremic patients vast quantities of hypertonic saline.

The serum sodium may appear low purely because the glucose has osmotically attracted vast amounts of water into the ECF, thereby diluting all the electrolytes. The corrected sodium, therefore, is the value one's sodium will return to once the glucose has been eliminated by insulin therapy. This should not influence your anion gap calculations. In fact, if you were going to use the "corrected" sodium, you should also use a "corrected" potassium and a "corrected" chloride value, because all the electrolytes are diluted equally in HONK. This is illustrated in LITFL's "Metabolic Muddle 005"

Anion gap in the presence of an excess of unmeasured cations

The presence of unmeasured cations can lead to an abnormally low or even negative anion gap.

Of course, generating a negative anion gap its not as easy as getting a mouthful of some random cation.

To get a nice negative anion gap going, you need to also be mindful of what that cation you're eating is conjugated with. If it is also an unmeasured strong anion, the anion gap will remain normal.

Behold:

Thus, the pregnant woman overdosed on magnesium sulfate will have a normal anion gap, because sulfate is not measured, and the magnesium - though in this case it might be measured, perhaps even subjected to obsessive hourly scrutiny- is not included in the anion gap calculation. Thus, the anion gap will remain the same.

Not so for lithium. Lithium preparations these days are either lithium carbonate or lithium citrate. In either case the conjugate anion is essentially a fluffy white cloud, easily metabolised or somehow otherwise obliterated by grinding in the great biochemical mills of the human organism. The remaining lithium is a strong unmeasured cation, and it will influence the anion gap.

Lithium chloride and lithium bromide were also available back in them days, and would have given the intensivists of the 1970s endless hours of biochemical amusement. Unfortunately, the halide salts of lithium turned out to be hideously toxic, and manufacturers withdrew them. These days we only have preparations conjugated with organic anions (eg. citrate, orotate, and so forth).

Other than monovalent and divalent elemental cations, one can occasionally encounter a negative anion gap in places which use parenteral Polymyxin B, which is strongly polycationic. THAM, the exogenous buffer, will do something similar.

In summary, the anion gap has numerous disadvantages, but as a screening tool it has the specific utility in helping us classify metabolic acidosis into handy diagnostic categories, which influence our management. Furthermore, comparing the change in anion gap to the change in bicarbonate can help us identify the presence of a "mixed" metabolic acid-base disorder, i.e. a coexisting normal anion gap metabolic acidosis, or a metabolic alkalosis. This topic is explored in the chapter on the delta ratio and its various equivalents.