0

**Introduction**

0

In 1998 Figge reported a strong linear relationship between anion gap and albumin concentration, which has led to the widespread recommendation to correct anion gap for albumin. It was proposed that since albumin is an anion, failing to correct for a low albumin level could allow an anion-gap acidosis to go undetected. This paper has been cited hundreds of times, with its results incorporated into many models of acid-base physiology (including the recent review shown above). However, I believe this paper is flawed due to circular logic. This post will discuss the Figge paper and other evidence regarding whether anion gap should be corrected for albumin.

0

**Figge 1998**** Anion gap and hypoalbuminemia**

0

This was an observational study based on simultaneous analysis of electrolytes and albumin levels in 152 critically ill patients and 9 healthy patients. Graphs were constructed relating the serum albumin level to the difference between the anion gap and unmeasured gap anions ([GA], figure below). A very strong correlation was observed (r=0.97). Based on the slope of this curve it was calculated that anion gap should vary by 2.5 mM for every 1 mg/dL difference in albumin.

0

0

Wait a second! How did the authors

*measure*the “*unmeasured*anions”? That’s impossible! What the authors actually did was a series of mathematical manipulations to equate the difference between the anion gap and unmeasured anions with the sum of the charges on albumin, calcium, and magnesium:0

0

They did not measure the negative charge on the albumin ([Albumin]), but instead used a formula to calculate this charge which was derived by the same research group (Figge 1992). Thus, the graph above would be more accurately labeled as follows:

0

0

The variation in calcium and magnesium concentration is small and contributes little. Ultimately the authors were comparing the measured albumin concentration with the calculated charge of the albumin based on that concentration. Their data is really just a reflection of the equation that they were using to calculate the charge on the albumin. This design did not allow their hypothesis to be

*tested*in any meaningful fashion.0

Durward 2003 replicated Figge's findings using a similar approach applied to critically ill children. These authors generated a plot of the calculated charge on albumin versus the albumin concentration (figure below). They used a formula that was a simplification of the formula used by Figge 1998, which was also derived in the publication by Figge 1992.

0

0

As in the Figge 1998 paper, this graph is little more than a reflection of the underlying equation for estimating the charge on albumin. In fact, given that the average pH of these patients was 7.35, the slope of the graph can be calculated to be 0.27 without the use of any additional patient data (see below). Therefore, rather than serving as a true validation of Figge 1998, this is a repetition of the same exercise in circular logic. Note that the slope derived by these authors is equal to 2.7 mM per mg/dL.

0

0

0

0

0

0

**Other studies relating anion gap to albumin**

0

Three other papers have reported the relationship between anion gap and albumin. These experiments had a similar design: a large set of patient data was used compare the anion gap and the albumin concentration.

0

Carvounis 2000 related anion gap to albumin among 432 patients admitted to a medical intensive care unit. These authors found that relationship between anion gap and albumin varied depending on the bicarbonate level. The ratio of the change in anion gap to the change in albumin was 1.45 mM per mg/dL for patients with a normal to high bicarbonate, and 1.89 mM per mg/dL for patients with a low bicarbonate. Overall there was a very weak correlation between albumin and anion gap (r=0.11).

0

Feldman 2005 related anion gap to albumin among 5328 patients in various departments throughout the hospital. They found a uniform correlation between anion gap and albumin, with a slope of 2.3 mM per mg/dL albumin. The strength of this correlation was intermediate (r=0.48).

0

Dinh 2006 calculated a regression equation relating anion gap to albumin in 639 sets of lab values (more on this study below). These authors found a slope of 1.2 mM per mg/dL albumin with a correlation coefficient of r=0.16. As a comparison, they also related anion gap to lactate concentration, which had a slope of 0.99 mM per mM (as expected, it should equal one) with a correlation coefficient of r=0.70.

0

Unlike the Figge and Durward studies, these studies used a non-circular design to relate anion gap to albumin. This relationship was weak and poorly generalizable between studies. As discussed further below, there are many determinants of the charge on albumin, which may cause this to vary between different patient populations and different disease states.

0

**Studies comparing the performance of anion gap vs. albumin corrected anion gap**

0

Anion gap is commonly used as a screening tool to detect the presence of an anion-gap metabolic acidosis. Three studies have compared the performance of anion gap (AG) versus albumin-corrected anion gap (ACAG) for detection of lactic acidosis. These studies had similar designs. They retrospectively obtained laboratory values of patients with and without lactic acidosis. Patients with other etiologies of elevated-AG metabolic acidosis were excluded (i.e., patients with diabetic ketoacidosis or poisoning). Using these two groups of patients, the sensitivity and specificity of AG and ACAG for detection of lactic acidosis were determined.

0

Dinh 2006 compared the ability of AG vs. ACAG to detect lactic acidosis (defined as lactate > 2.5 mM) in 639 sets of laboratory values from 356 patients at a tertiary care hospital. The area under the receiver-operator curve was nearly equal for AG and ACAG (0.757 vs. 0.750 respectively; figure below). Although ACAG tended to be about 5 mM higher than AG, both tests had the same performance.

0

0

0

Chawla 2008a compared AG and ACAG as predictors of elevated lactic acid (defined as >2 mM) among 285 patients admitted to a medical-surgical ICU. The area under the receiver-operator curve was nearly equal for AG and ACAG (0.55 vs. 0.57 respectively). The two tests had nearly identical performance if a cutoff value of ACAG set at 4 mM above the cutoff used for the AG (table below). One limitation of this study was that labs were not necessarily obtained simultaneously.

0

0

Chawla 2008b is a replication of their earlier study with the requirement that labs were drawn simultaneously. They compared the ability of AG vs. ACAG to detect elevated lactic acid (defined as >2.5 mM) among 497 new sets of laboratory values from 143 ICU patients. The area under the recover-operator curve was nearly identical for AG and ACAG (0.70 vs. 0.72, respectively). The two tests had very similar performance if a cutoff value of ACAG was set about 4 mM above the cutoff used for the AG:

0

0

**Why is there such a weak relationship between anion gap and albumin level?**

0

As discussed above, the concept that anion gap should change by 2.5 mEq/L for every mg/dL change in albumin was based on the equation that Figge 1998used to

*calculate*the charge on albumin. This equation, described in Figge 1992, is based on the properties of albumin*en vitro*. However, it does not take into account that*en vivo*, albumin may bind to various cations which decrease its effective negative charge to a variable extent:0

0

Furthermore, albumin may undergo a variety of oxidative modification

*en vivo*(Oettl 2007). pH may also effect albumin's charge. Thus,*en vivo*the effective charge on albumin may be lower than predicted and likely varies depending on multiple factors. This may defeat the ability of a simple equation to correct anion gap for albumin concentration.0

**A reminder to apply evidence-based medicine to acid-base physiology**

0

Recently several models based on Stewart's approach to acid-base physiology have been proposed for clinical use. The physical chemistry underlying these models is extremely rigorous, but nonetheless it is impossible to account for the entire complexity of

*en vivo*acid-base physiology. As impressive as these models may be on paper, they nonetheless require clinical validation. Care must be taken to avoid circular logic when using a model to calculate both components of a graph, which is a common error.0

**Take-home points**

- Current formulas that correct the anion gap for albumin are based on a publication by Figge 1998 which was flawed by circular logic.
- Clinical data sets relating anion gap to albumin have been unable to demonstrate a consistent or strong relationship between these variables.
- For detection of lactic acidosis, the uncorrected anion gap and the anion gap corrected for albumin have the same test performance.
- If the anion gap is corrected for albumin using conventional equations, the cutoff value must be increased by about 4 mM (compared to the cutoff value for the uncorrected anion gap).
- Figge 1998 has been cited over one hundred times and has been used to derive numerous models of acid-base physiology. These models may require revision.

### Josh Farkas

Josh is the creator of PulmCrit.org. He is an associate professor of Pulmonary and Critical Care Medicine at the University of Vermont.

#### Latest posts by Josh Farkas (see all)

- IBCC chapter – Hyperthermia & heat stroke - July 19, 2019
- IBCC chapter & cast – Alcohol Withdrawal - July 18, 2019
- Pulmcrit- Evidence Update: Phenobarbital in Alcohol Withdrawal - July 15, 2019

Josh,

I think this post would benefit from screening by a peer reviewer. Some of the assertions may not be supported by the papers you quote.

I'm interested in your thoughts. Perhaps you could provide peer review. The last time we did this (http://www.pulmcrit.org/2014/07/preoxygenation-apneic-oxygenation-using.html) I thought it was very informative. Doing peer review in the light of day can be a bit awkward, but my guess is that anyone reading this blog would be interested in a good exchange of ideas. The intention of this blog has always been to explore different opinions on controversial topics.

Very interesting post. Thank you for the analysis, particularly the question on clinical relevance and whether calculating for albumin level has any clinical utility. Just a comment on the 'circular logic' argument, which is a pretty devastating accusation and would, as you say, require a complete reworking of the idea behind the contribution of Albumin as a weak organic acid. I think there may be a simpler, less 'fatal' explanation: The formula for calculating albumin contribution to acid/base is based on the 1992 Figge article, which directly calculated the contribution of Albumin to the acid/base milieu. Unfortunately, because of the complex interaction (you elude to above) of albumin en vivo, the results were too unwieldy to use at the bedside. The 1998 article merely provides a 'best fit' formula, based on the 1992 data, that can be used at the bedside. The 'fit' is near linear within physiological pH range.The graph from the Figge 1998 article above shows how much 'space' is left over for 'missing anions' once albumin concentration falls. I don't think it is intended to demonstrate the accuracy of the 0.25(42-[Alb]) equation, but it does graphically illustrate where potentially pathological missing anions could hide with a normal… Read more »

Yes, Daniel has started the process very well. If you follow the Figge work from paper to paper, their is a clear and lovely mathematical proof given for every move towards simplicity.

However, the most confusing statement in the post is:

"The two tests had nearly identical performance if a cutoff value of ACAG set at 4 mM above the cutoff used for the AG (table below)."

Why would we increase the cut-off? I am not seeing any possible reason why that would be the case and this move eliminates the sensitivity of albumin-correction with no explanation for why that would be a good idea.

Further, when our goal is to get an accurate picture of acid-base milieu, isn't sensitivity to discover hyperlactatemia not really the yardstick we should be using? If your

Daniel and Scott, Thanks for your thoughtful comments. Scott's last sentence was cut off, but I e-mailed him and he didn't think any important points were lost. Perhaps we could all agree that Figge 1992 contains a rigorous mathematical prediction of the behavior of albumin. Figge 1998 illustrates how these formula could be simplified to the point where it could be applied at the bedside. However, neither of these papers actually tests the 2.5(4.2-[Alb]) equation in patients. As discussed in the post, most clinical data suggests that this equation does not work en vivo. Considering the complexity of albumin's modifications and interactions with a wide variety of substrates en vivo, it may simply be impossible to derive an equation to predict albumin's charge en vivo. Corrected anion gap is more sensitive for the detection of lactic acidosis than uncorrected anion gap using the same cutoff. Agreed. However, the corrected anion gap achieves this higher sensitivity at the cost of a lower specificity. One rigorous way to compare two tests with varying sensitivity and specificity is by looking at the area under the ROC curve. For example Chawla 2008a found both tests had essentially the same area under the ROC curve.… Read more »

There doesn't seem to be much oint in optimizing spec at the expense of sens. We are using the anion gap as a screen; if it is false positive then all you have burned is a lactate level and perhaps a blood gas.

This article by gunnerson [crit care 2005;9:508] should have all of the SIG refs and a non-labor intensive SIG calculation

Read with interest.

See http://www.sciencedirect.com/science/article/pii/S0022214305003811

HC

Interested to know if this changes your approach to evaluating anion gap acidosis with respect to the model proposed in “Acid Base in the Critically Ill” (EMCrit Episodes 44-50). Ref: https://emcrit.org/wp-content/uploads/acid_base_sheet_2-2011.pdf

Thanks!