ABG Interpretation in 6 Steps

In depth

The one-page summary above is what the PDF prints. These sections go further.

Buffers and Henderson–Hasselbalch

pH = 6.1 + log [HCO₃⁻ ÷ (0.03 × PaCO₂)], where 6.1 is the pKa of carbonic acid and 0.03 converts PaCO₂ to dissolved CO₂ in mmol/L. At HCO₃⁻ 24 and PaCO₂ 40 the ratio is 20, log 20 = 1.30, so pH is 7.40. pH follows the ratio, not either value: double both and pH does not move, which is how a normal pH can hide two opposing disorders.

The Henderson form needs no logs: [H⁺] in nEq/L = 24 × PaCO₂ ÷ HCO₃⁻; 40 nEq/L is pH 7.40, and between 7.20 and 7.50 each 0.01 fall in pH adds about 1 nEq/L. Use it to check a gas for consistency. The analyzer calculates HCO₃⁻ from measured pH and PaCO₂; the chemistry panel's total CO₂ runs 1–2 mEq/L higher because it includes dissolved CO₂. A difference over about 4 in samples drawn together suggests a timing or sampling error.

Bicarbonate is the main extracellular buffer; its pKa is far from 7.4, but the system is open: the lungs remove CO₂ and the kidneys regenerate HCO₃⁻. Hemoglobin buffers CO₂ inside red cells, deoxygenated hemoglobin better (the Haldane effect). The kidney reclaims filtered HCO₃⁻, mostly in the proximal tubule, and makes new HCO₃⁻ for each H⁺ it excretes as titratable acid or ammonium; ammonium excretion is what rises over days in chronic acidosis.

Acidemia below about 7.20 depresses contractility, blunts the response to catecholamines and dilates arterioles. Alkalemia above about 7.55 lowers ionized calcium and K⁺, shifts the oxyhemoglobin curve left and raises the risk of arrhythmia and seizure.

Compensation: formulas, limits and timing

Compensation moves the other value in the same direction as the primary change and never overshoots, so a pH on the far side of 7.40 from what the primary disorder predicts means a second disorder. These are the Boston rules most US references use (Berend 2014; Merck). They are cohort averages, so judge small misses.

Shortcut for metabolic acidosis down to pH about 7.10: PaCO₂ is close to the last two digits of the pH. For a respiratory disorder, work out what both the acute and the chronic rule predict. A measured value between the two means acute-on-chronic disease, or a single respiratory disorder plus a metabolic one; only the history separates them.

Timing explains most exam traps. Buffering takes seconds; the respiratory response starts in minutes and is complete in 12–24 hours; renal compensation needs 3–5 days. So a high HCO₃⁻ on day 1 of a hypercapnic event was there before it, and a chronically hypercapnic patient ventilated quickly to a PaCO₂ of 40 keeps a high HCO₃⁻ for days (post-hypercapnic alkalosis). Chronic respiratory alkalosis is the one disorder whose compensation can bring pH near normal.

Anion gap, delta-delta and other gaps

Normal anion gap was 12 ± 4 on older analyzers and is often lower (about 3–11) on modern ion-selective electrodes, so use your lab's range. Albumin is the largest unmeasured anion: each 1 g/dL fall lowers the gap about 2.5 (Figge 1998, against a normal albumin of 4.4). A septic patient with albumin 2.0 and a gap of 12 has a corrected gap of 17, a hidden lactate or ketone load.

When the gap is high, compare its rise with the fall in HCO₃⁻: delta ratio = (AG − 12) ÷ (24 − HCO₃⁻). The delta gap asks the same thing another way: measured HCO₃⁻ + (AG − 12) is the HCO₃⁻ the patient would have without the gap acid. Over 26 means a coexisting metabolic alkalosis; under 22, a coexisting normal-gap acidosis. Rastegar 2007 calls both rough guides, because acid is partly buffered in cells and anions are excreted at different rates.

The osmolar gap finds uncharged toxins. Calculated osmolality = 2 × Na⁺ + glucose ÷ 18 + BUN ÷ 2.8 (+ ethanol ÷ 3.7); the gap, measured minus calculated, is normally under 10 mOsm/kg (Berend 2014). Na⁺ 140, glucose 90 and BUN 14 give 290; a measured 320 is a gap of 30. Early after methanol or ethylene glycol the osmolar gap is high and the anion gap normal; as they are metabolized to formate, glycolate and oxalate the osmolar gap falls and the anion gap rises, so a normal osmolar gap late does not exclude poisoning. Mannitol, propylene glycol (IV lorazepam vehicle) and severe ketoacidosis also raise it.

The urine anion gap (UAG = urine Na⁺ + K⁺ − Cl⁻) separates normal-gap acidoses. Ammonium leaves with chloride, so a kidney responding properly excretes a lot of Cl⁻ and the gap turns negative (about −20 to −50): the loss is from the gut ('neGUTive'). A zero or positive gap means the kidney cannot excrete ammonium: distal or type 4 RTA, or kidney failure. It misleads when urine Na⁺ is under about 20 or ketones are in the urine.

Strong ion difference (Stewart), briefly

Stewart's approach sets pH from three independent variables: PaCO₂, the strong ion difference (SID) and weak acids (mostly albumin and phosphate). Apparent SID = (Na⁺ + K⁺ + Ca²⁺ + Mg²⁺) − (Cl⁻ + lactate), normally about 40–42 mEq/L. A fall in SID (chloride gain, lactate) causes acidosis, a rise (chloride loss from vomiting) causes alkalosis, and low albumin causes alkalosis, which is why hypoalbuminemia masks acidosis. A quick screen is Na⁺ − Cl⁻, normally about 32–36; lower suggests a hyperchloremic component.

Normal saline has a SID of 0, so large volumes cause hyperchloremic acidosis; lactated Ringer's (effective SID about 28) and Plasma-Lyte (about 50) do so less. SMART (2018, about 15,800 ICU adults) found major adverse kidney events in 14.3% with balanced fluids versus 15.4% with saline; BaSICS (2021) and PLUS (2022) found no mortality difference.

Causes in depth

High-gap acidosis: MUDPILES is the classic list (methanol, uremia, DKA, propylene glycol, isoniazid and iron, lactate, ethylene glycol, salicylates). GOLDMARK (Mehta 2008) fits the ICU better: glycols, 5-oxoproline (regular acetaminophen in malnourished or septic patients), L-lactate, D-lactate (short bowel), methanol, aspirin, renal failure and ketoacidosis (diabetic, alcoholic, starvation, euglycemic on SGLT2 inhibitors). Type A lactic acidosis is poor O₂ delivery (shock, arrest, seizures, ischemic bowel, CO); type B occurs without it (metformin, liver failure, epinephrine, linezolid, propofol infusion syndrome, thiamine deficiency).

Normal-gap acidosis: HARDASS is hyperalimentation, acetazolamide, renal tubular acidosis, diarrhea, Addison's, spironolactone and saline. Add ureteral diversions, pancreatic fistulas and recovering DKA, when ketoanions lost in urine are replaced by chloride from saline and a high-gap acidosis becomes a normal-gap one.

Metabolic alkalosis needs a generator (H⁺ loss from stomach or kidney, alkali gain, or contraction) and a maintainer that stops the kidney excreting the extra HCO₃⁻: volume depletion, chloride depletion, hypokalemia or aldosterone. In contraction alkalosis, loop or thiazide diuresis removes chloride-rich, bicarbonate-poor fluid around an unchanged amount of HCO₃⁻. Urine chloride, not urine sodium, sorts the causes, because bicarbonaturia drags sodium into the urine even when the patient is dry.

Oxygenation and oxygen delivery

PAO₂ = FiO₂ × (PB − 47) − PaCO₂ ÷ 0.8: about 100 on room air at sea level with a PaCO₂ of 40, about 663 on FiO₂ 1.0. Of the five mechanisms of hypoxemia (Sarkar 2017), low inspired O₂ and hypoventilation leave the A–a gradient normal; V/Q mismatch, shunt and diffusion limitation widen it. V/Q mismatch responds to O₂; a large true shunt (over about 30%) barely does and needs recruitment, PEEP or treatment of the cause. The A–a gradient rises with FiO₂, so compare it only at the same FiO₂. P/F moves with PEEP and FiO₂, so report it with both; above 1,000 m, Berlin 2012 multiplies it by PB ÷ 760.

Above the knee, large PaO₂ changes barely move saturation; below it, saturation falls steeply. A right shift lowers affinity and unloads O₂ to tissue: acidosis (the Bohr effect), hypercapnia, fever, raised 2,3-DPG (chronic hypoxemia, anemia, altitude). A left shift holds O₂: alkalosis, hypothermia, low 2,3-DPG (banked blood), CO, methemoglobin, fetal hemoglobin.

SpO₂ reads carboxyhemoglobin as oxyhemoglobin, so CO poisoning shows a normal SpO₂, and methemoglobin pulls SpO₂ toward 85%. Many analyzers calculate SaO₂ from PaO₂, which is just as blind; ask for co-oximetry. In Sjoding 2020 (NEJM), occult hypoxemia (SaO₂ under 88% with SpO₂ 92–96%) occurred in 11.7% of paired readings in Black patients versus 3.6% in White patients; get a gas when SpO₂ does not fit the patient.

CaO₂ = (1.34 × Hb × SaO₂) + (0.003 × PaO₂): Hb 15, SaO₂ 0.98 and PaO₂ 100 give 19.7 + 0.3 = 20 mL/dL (sources use 1.34 to 1.39). DO₂ = cardiac output × CaO₂ × 10, about 1,000 mL/min, against a VO₂ of about 250, so extraction is about 25% and SvO₂ 65–75%. With Hb 8, SaO₂ 0.90 and PaO₂ 60, CaO₂ is 9.8. Raising PaO₂ to 100 (SaO₂ 0.98) gives 10.8, a 10% gain; raising Hb to 10 gives 12.2, a 25% gain; a 20% rise in cardiac output raises DO₂ by 20%.

Venous gases and sampling errors

In stable patients, peripheral venous pH runs about 0.03 lower than arterial, HCO₃⁻ about 1–2 mEq/L higher and PCO₂ about 4–6 mm Hg higher on average (Byrne 2014 meta-analysis; Lee 2025). PCO₂ agreement is too loose for individual decisions, but a venous PCO₂ under 45 reliably rules out arterial hypercapnia. In shock the differences widen; a central venous-to-arterial PCO₂ gap over 6 mm Hg suggests cardiac output too low for need.

Bubble error grows with time and agitation, including pneumatic tube transport. Baird 2013 advises analysis within 15 minutes when PO₂ matters and within 30 minutes for plastic syringes at room temperature, which should not be iced. Leukocyte larceny occurs with extreme leukocytosis or thrombocytosis (leukemia, often WBC over 100,000): cells in the syringe consume O₂ within minutes; treat the patient, not the number. Baird 2013 cites advice to wait at least 30 minutes after a ventilator change.

Mixed disorders: worked examples

1. Aspirin overdose, tinnitus, RR 32: pH 7.48, PaCO₂ 22, HCO₃⁻ 16, Na⁺ 140, Cl⁻ 104, albumin 4.0

1. Alkalemia with a low PaCO₂: primary respiratory alkalosis.
2. The overdose is acute, so use the acute rule: HCO₃⁻ about 20 expected. Measured 16 is lower: a metabolic acidosis too.
3. Gap 140 − 120 = 20; delta ratio 8 ÷ 8 = 1.0, a pure high-gap process.
4. Answer: respiratory alkalosis plus high-gap acidosis, the salicylate pattern. If intubated, match the high minute ventilation; a rising PaCO₂ moves salicylate into the brain.

2. Type 1 diabetes, two days of vomiting: pH 7.40, PaCO₂ 38, HCO₃⁻ 23, Na⁺ 140, Cl⁻ 90, glucose 450, albumin 4.0

1. pH, PaCO₂ and HCO₃⁻ all look normal.
2. Gap 140 − 113 = 27: a high-gap acidosis is present whatever the pH.
3. Corrected HCO₃⁻ 23 + 15 = 38: a metabolic alkalosis too.
4. Answer: DKA plus vomiting alkalosis; the low Cl⁻ is the clue. Expect large K⁺ and Cl⁻ deficits.

3. COPD, baseline PaCO₂ 55, now drowsy with pneumonia: pH 7.25, PaCO₂ 80, HCO₃⁻ 34

1. Acidemia with a high PaCO₂: primary respiratory acidosis.
2. PaCO₂ up 40: the acute rule predicts HCO₃⁻ 28 and pH 7.08; the chronic rule, HCO₃⁻ 40 and pH 7.28.
3. Measured values fall between: acute-on-chronic respiratory acidosis.
4. Answer: acute-on-chronic hypercapnic failure. On NIV or the ventilator, aim for the baseline PaCO₂, not 40.

4. ROSC after 15 minutes, being bagged: pH 7.03, PaCO₂ 55, HCO₃⁻ 14, Na⁺ 138, Cl⁻ 98, lactate 12, albumin 4.0

1. Acidemia; high PaCO₂ and low HCO₃⁻ both lower pH, so both are primary.
2. Winter's 1.5 × 14 + 8 = 29 (27–31); PaCO₂ 55 is far above it.
3. Gap 138 − 112 = 26; delta ratio 14 ÷ 10 = 1.4, pure lactic acidosis.
4. Answer: combined lactic and respiratory acidosis. Ventilate toward normocapnia; bicarbonate without more ventilation adds CO₂.

5. Profuse diarrhea for three days: pH 7.29, PaCO₂ 30, HCO₃⁻ 14, Na⁺ 136, K⁺ 2.9, Cl⁻ 112, albumin 4.0; urine Na⁺ 30, K⁺ 20, Cl⁻ 80

1. Metabolic acidosis; Winter's 29 (27–31), PaCO₂ 30 fits.
2. Gap 136 − 126 = 10: normal gap.
3. UAG 30 + 20 − 80 = −30: the kidney is excreting ammonium, so the loss is from the gut.
4. Answer: normal-gap acidosis from diarrhea with hypokalemia; a positive UAG would point to distal RTA.

6. Day 3 after surgery, NG suction 2 L/day: pH 7.52, PaCO₂ 48, HCO₃⁻ 38, K⁺ 3.0, urine Cl⁻ 8

1. Metabolic alkalosis; expected PaCO₂ 40 + (0.6–0.7 × 14) = 48–50, so PaCO₂ 48 is compensation.
2. Answer: chloride-responsive alkalosis from gastric losses; replace volume, chloride and K⁺, and do not chase the PaCO₂.

Adjusting the ventilator to the gas

PaCO₂ varies inversely with alveolar ventilation. With CO₂ production and dead space constant, new rate = current rate × current PaCO₂ ÷ target PaCO₂: rate 14, PaCO₂ 60, target 45 gives about 19. Lung-protective ventilation keeps VT at 4–8 mL/kg predicted body weight with plateau 30 cm H₂O or less (ARDSNet 2000: 6 vs 12 mL/kg, mortality 31.0% vs 39.8%), so rate is the usual lever, up to 35. Change one setting at a time and recheck.

Permissive hypercapnia: the ARDSNet protocol targets pH 7.30–7.45, raises the rate for pH under 7.30, and allows bicarbonate or a larger VT if pH stays under 7.15; avoid it with raised intracranial pressure or RV failure. Its oxygenation goal is PaO₂ 55–80 or SpO₂ 88–95%; HOT-ICU (2021, PaO₂ 60 vs 90) found no 90-day mortality difference, and most ICUs aim for SpO₂ about 92–96%.

At intubation for metabolic acidosis (DKA, salicylates), match the patient's own minute ventilation, often 15–25 L/min, and recheck a gas within about 30 minutes; standard settings can drop pH below 7.0. With raised intracranial pressure keep PaCO₂ 35–45: cerebral blood flow changes roughly 2–4% per mm Hg, and the Brain Trauma Foundation (2016) advises against prophylactic hyperventilation to 25 or less. After cardiac arrest, AHA targets normocapnia and SpO₂ 92–98%.

A PaCO₂ rising on unchanged settings means more CO₂ production (fever, shivering, overfeeding, bicarbonate) or more dead space (PE, low cardiac output, auto-PEEP, worsening ARDS); the PaCO₂–EtCO₂ gap, normally 2–5 mm Hg, widens with dead space. Bicarbonate remains contested: BICAR-ICU (2018, 389 patients with pH 7.20 or less) found no overall benefit, and in the acute kidney injury stratum 28-day survival was 63% versus 46%, a hypothesis-generating subgroup.

Exam pearls

• Compensation never overshoots: a pH under 7.40 points to the acidosis as primary.
• Winter's: PaCO₂ = 1.5 × HCO₃⁻ + 8 (±2); above the range means respiratory acidosis too.
• Correct the gap for albumin: add 2.5 per 1 g/dL below 4.
• MUDPILES or GOLDMARK for high-gap acidosis; HARDASS for normal-gap.
• Vomiting and NG suction: hypochloremic, hypokalemic alkalosis; urine Cl⁻ under 20 means saline-responsive.
• Salicylates: respiratory alkalosis plus high-gap acidosis.
• Hypoxemia with a normal A–a gradient: hypoventilation or low inspired O₂.
• Right shift (acid, CO₂, heat, 2,3-DPG) unloads O₂; CO poisoning gives a normal SpO₂.
• An air bubble pulls PO₂ toward 150 and lowers PCO₂; delay lowers PO₂ and raises PCO₂.