Characteristics of pulse oximetry and arterial blood gas in patients with fibrotic interstitial lung disease

Abstract

Background

Fibrotic interstitial lung disease (ILD) is frequently associated with abnormal oxygenation; however, little is known about the accuracy of oxygen saturation by pulse oximetry (SpO₂) compared with arterial blood gas (ABG) saturation (SaO₂), the factors that influence the partial pressure of carbon dioxide (PaCO₂) and the impact of PaCO₂ on outcomes in patients with fibrotic ILD.

Study design and methods

Patients with fibrotic ILD enrolled in a large prospective registry with a room air ABG were included. Prespecified analyses included testing the correlation between SaO₂ and SpO₂, the difference between SaO₂ and SpO₂, the association of baseline characteristics with both the difference between SaO₂ and SpO₂ and the PaCO₂, the association of baseline characteristics with acid-base category, and the association of PaCO₂ and acid-base category with time to death or transplant.

Results

A total of 532 patients with fibrotic ILD were included. Mean resting SaO₂ was 92±4% and SpO₂ was 95±3%. Mean PaCO₂ was 38±6 mmHg, with 135 patients having PaCO₂ <35 mmHg and 62 having PaCO₂ >45 mmHg. Correlation between SaO₂ and SpO₂ was mild to moderate (r=0.39), with SpO₂ on average 3.0% higher than SaO₂. No baseline characteristics were associated with the difference in SaO₂ and SpO₂. Variables associated with either elevated or abnormal (elevated or low) PaCO₂ included higher smoking pack-years and lower baseline forced vital capacity (FVC). Lower baseline lung function was associated with an increased risk of chronic respiratory acidosis. PaCO₂ and acid-base status were not associated with time to death or transplant.

Interpretation

SaO₂ and SpO₂ are weakly-to-moderately correlated in fibrotic ILD, with limited ability to accurately predict this difference. Abnormal PaCO₂ was associated with baseline FVC but was not associated with outcomes.

WHAT IS ALREADY KNOWN ON THIS TOPIC.

• Pulse oximetry (SpO₂) is commonly used as a surrogate for arterial blood oxygen saturation; however, the reliability of SpO₂ as a surrogate of SaO₂ in patients with fibrotic interstitial lung disease (ILD) remains unclear.

WHAT THIS STUDY ADDS

• To our knowledge, this is the first study to assess the characteristics of pulse oximetry and arterial blood gas (ABG) in patients with fibrotic ILD.

HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY

• Our findings raise important questions regarding the clinical utility of ABG and pulse oximetry in patients with ILD.

• These results may affect the way each measurement is used clinically and highlight the need for additional studies comparing and optimising these tests.

Introduction

Fibrotic interstitial lung disease (ILD) is frequently associated with clinically significant hypoxaemia and need for supplemental oxygen.¹ Arterial oxygen saturation (SaO₂) is typically estimated using pulse oximetry (SpO₂), which can be performed at rest, with ambulation and nocturnally. A minority of patients with fibrotic ILD also have abnormalities in the partial pressure of carbon dioxide (PaCO₂) as measured by arterial blood gas (ABG), which can occur secondary to advanced ILD or other causes (eg, respiratory muscle weakness, concomitant airways disease).

SpO₂ is an inexpensive and non-invasive method of estimating SaO₂; however, several factors affect the accuracy of SpO₂, including darker skin pigmentation, anaemia, acidosis, critical illness and systemic sclerosis.^{2 3} The accuracy of SpO₂ as a surrogate of SaO₂ has been assessed in multiple populations, including patients with critical illness, both stable and acute exacerbation of chronic obstructive pulmonary disease, and chronic thromboembolic pulmonary hypertension. However, there are limited data on the accuracy of SpO₂ in fibrotic ILD and on the potential to predict this difference using baseline characteristics.^4–7 There are similarly limited data on factors that influence PaCO₂ and the impact of PaCO₂ on disease outcome.

An improved understanding of the interpretation and prognostic significance of ABG findings would enhance clinical decision-making. We therefore sought to compare SpO₂ and SaO₂ in patients with fibrotic ILD, identify clinical variables associated with discordance between these measurements and identify variables and outcomes associated with abnormal PaCO₂. We hypothesised that SaO₂ and SpO₂ would be within 2% for 95% of patients with fibrotic ILD at rest when SpO₂ was ≥80% as previously seen in the general population,² and that patients with fibrotic ILD with resting hypoxaemia would have higher levels of discordance between SaO₂ and SpO₂. Lastly, we hypothesised that clinical variables associated with abnormally high PaCO₂ in patients with fibrotic ILD would include high body mass index (BMI), active smoking status and lower forced expiratory volume in one second (FEV₁).

Methods

Study design, population and data collection

A retrospective chart review was completed on patients with fibrotic ILD enrolled in the prospective multicentre Canadian Registry for Pulmonary Fibrosis (CARE-PF).^{8 9} Eligible patients for this substudy were adults with any fibrotic ILD subtype who had an outpatient ABG performed while breathing room air. There were no exclusion criteria.

ABG measurements

The most recent room air ABG collected on any date after diagnosis or within 90 days prior to identification of fibrotic ILD by chest CT was recorded to provide the greatest spectrum of data that would maximise generalisability (ie, some patients provided an ABG early in their disease course, and some patients later in their disease course). Extracted data included pH, arterial partial pressure of carbon dioxide (PaCO₂), arterial partial pressure of oxygen (PaO₂), bicarbonate (HCO₃^-) and arterial oxygen saturation (SaO₂). PaCO₂ was assessed as both a continuous and a categorical variable that grouped patients as abnormal (<35 or >45 mmHg) or normal (35–45 mmHg).¹⁰ Each ABG was assessed for acid-base abnormalities and further categorised into their primary abnormality (referred to as ‘acid-base category’), including acute respiratory acidosis or alkalosis, chronic respiratory acidosis or alkalosis, or metabolic acidosis or alkalosis. Given the absence of clear standards on ABG interpretation, we created the following definitions for each acid-base abnormality:

• Acute respiratory acidosis was defined as pH <7.35 and PaCO₂ >45 mmHg.

• Acute respiratory alkalosis was defined as pH >7.45 and PaCO₂<35 mmHg.

• Metabolic acidosis was defined as pH <7.35 and HCO₃^- <22 mEq/L.

• Metabolic alkalosis was defined as pH >7.45 and HCO₃^- >26 mEq/L.

• Chronic respiratory acidosis was defined as pH <7.4, PaCO₂ >45 mmHg, HCO₃^- > 24 mEq/L with PaCO₂ and HCO₃^- increased above their respective baselines (PaCO₂= 40 mmHg, HCO₃^- = 24 mEq/L) in a ratio of 10:2 or greater.

• Chronic respiratory alkalosis was defined as pH >7.4, PaCO₂<35 mmHg, HCO₃^- <24 mEq/L with PaCO₂ and HCO₃^- decreased below their respective normal values (PaCO₂ = 40 mmHg, HCO₃^- = 24 mEq/L) in a ratio of 10:3 or greater.

Additional measurements

Baseline characteristics used in the correlation analyses included age, sex, self-reported race, BMI, smoking pack-years and baseline lung function (percent-predicted values for forced vital capacity (FVC), forced expiratory volume in one second (FEV₁) and diffusion capacity of the lung for carbon monoxide (DLCO)). Pulmonary function tests (PFTs) were included for correlational analysis if patients had a PFT completed within 6 months of the ABG collection date. All PFTs were performed in accredited hospital-based laboratories according to standardised technical specifications.^{11 12} Physiological obstruction was defined as FEV₁/FVC <0.7. Patients with a room air ABG and room air SpO₂ taken within 30 days of each other were included in the analyses assessing correlation of and difference between SaO₂ and SpO₂, with a secondary analysis considering only same-day tests. ILD diagnoses were assigned based on multidisciplinary assessment according to previous guidelines where available, grouping patients into five major categories of idiopathic pulmonary fibrosis (IPF), connective tissue disease-associated ILD (CTD-ILD), fibrotic hypersensitivity pneumonitis (HP), unclassifiable ILD and other ILD.

Statistical analysis

Data shown are mean±SD, median (IQR) or number (%). All continuous variables included in multivariable analyses were scaled by a factor of 10. A two-sided p value <0.05 was considered statistically significant for all analyses. All analyses were performed using Stata BE V.17.0 (StataCorp, College Station, Texas, USA).

The correlation between SaO₂ and SpO₂ was assessed using a Spearman correlation coefficient (r value), with the difference between SaO₂ and SpO₂ assessed using a paired t-test. The association of the difference between SaO₂ and SpO₂ with age, sex, race and baseline lung function (FVC and DLCO) was tested on unadjusted analysis using a Spearman’s rank correlation, followed by multivariable linear regression with the same baseline variables.

The associations of PaCO₂ with each of age, sex, BMI, smoking pack-years and baseline lung function were tested on unadjusted analysis using linear and logistic regression for continuous and categorical PaCO₂, respectively, as defined above. This was followed by multivariable linear regression and multivariable logistic regression, respectively, adjusting for age, sex, BMI, smoking pack-years and baseline lung function. Three separate adjusted models were completed with different baseline lung function measurements included due to collinearity between FVC and FEV₁. The primary model included FVC and DLCO, with additional models including FEV₁ and DLCO, and another with all three measurements. This was completed to assess whether FEV₁, used as a surrogate for concomitant airways disease, may affect our results. Unadjusted and adjusted Cox proportional hazard regressions assessed whether PaCO₂ was associated with time to death or transplant, with the multivariable model adjusted for age, sex, BMI, smoking pack-years and baseline lung function (FVC and DLCO).

The association of acid-base category with age, sex, BMI, smoking pack-years and baseline lung function was tested on both unadjusted and adjusted logistic regression as well as with a multinomial model that assessed all acid-base categories. The same three adjusted models used in the PaCO₂ regression analyses were used for each acid-base category regression analysis and the multinomial model. Regression analyses were completed if the acid-base category had ≥25 patients, with acute respiratory alkalosis, metabolic alkalosis, chronic respiratory acidosis and chronic respiratory alkalosis reaching this threshold. Unadjusted and adjusted Cox proportional hazard regressions assessed whether each acid-base category was associated with time to death or transplant, with the multivariable model adjusted for the same baseline variables used in the regression analyses.

Results

Study population and baseline characteristics

A total of 532 patients with an outpatient room air ABG were included (table 1). Median time from ILD diagnosis to performance of the ABG was 4 years (IQR 0.9–6.1), with 151 patients (28%) having the ABG drawn within the first year of ILD diagnosis. ABG characteristics are shown in figures 1 and 2. Mean SaO₂ was 91±5% for the full cohort. Mean SaO₂ was 92±4% and SpO₂ was 95±3% for 105 patients with both an ABG and a corresponding room air SpO₂. Mean PaCO₂ was 38±6 mmHg, with 135 patients (25%) having PaCO₂<35 mmHg and 62 patients (12%) having PaCO₂ >45 mmHg. Of patients with elevated PaCO₂, 7 patients (16%) had evidence of physiological obstruction compared with 40 patients (11%) without an elevated PaCO₂. Of the 151 patients who had an ABG drawn within the first year of diagnosis, 7% had an elevated PaCO₂ vs 9% of those who had the ABG drawn within the first 5 years of diagnosis and 11% for patients who had the ABG taken any time after 10 years of diagnosis. Mean pH was 7.43±0.05, with 155 patients (29%) having an abnormal pH, 27 of whom were acidotic (17%) and 128 alkalotic (83%).

Table 1.

Baseline characteristics of entire cohort (N=532)

Variable	Value
Age, years	68 (59–74)
Sex
Male	299 (56)
Female	233 (44)
Race
Asian	37 (7)
Black	12 (1)
Indigenous	24 (5)
White	445 (85)
Other/missing†	13 (2)
Smoking status
Ever smoker	372 (70)
Pack-years, smokers	22 (9–38)
Pack-years, all	11 (0–31)
Body mass index, kg/m2	29 (25–34)
ILD diagnosis category
Idiopathic pulmonary fibrosis	165 (31)
CTD-associated ILD	149 (28)
Hypersensitivity pneumonitis	67 (13)
Unclassifiable ILD	100 (19)
Other ILD*	51 (9)
Baseline lung function
FVC %-predicted	65±19
FEV1 %-predicted	69±19
FEV1/FVC <0.7	48 (11)
DLCO %-predicted	47±17
Comorbidities
Pulmonary hypertension	12 (2)
Coronary artery disease	77 (14)
Congestive heart failure	15 (3)

Data shown are mean±SD, median (IQR) or n (%).

*Other ILD included drug-related ILD (n=3), vasculitis (n=5), cystic lung disease (n=8), infiltrative lung disease (n=3), acute respiratory distress syndrome (n=1), asbestosis (n=3), aspiration-related ILD (n=1), organising pneumonia (n=8), idiopathic pleuroparenchymal fibroelastosis (n=1), respiratory-bronchiolitis ILD (n=2), radiation pneumonitis/ILD (n=1), other (n=15).

†Other race included Pacific Islander and Metis.

CTD, connective tissue disease; DLCO, diffusion capacity of the lung for carbon monoxide; FEV1, forced expiratory volume in one second; FVC, forced vital capacity; ILD, interstitial lung disease.

Figure 1.

(A) Histogram of SaO₂ values with a reference line at the boundary of normal. (B) Histogram of SpO₂ values with a reference line at the boundary of normal. (C) Histogram of PaCO₂ values with reference lines at boundaries of normal (35 and 45 mmHg). (D) Histogram of pH values with reference lines at boundaries of normal (7.35 and 7.45). Abnormal values are shown in light blue.

Figure 2.

(A) Scatter plot of room air SaO₂ and paired room air SpO₂ measured within 30 days of each other. (B) Scatter plot comparing paired SaO₂ from room air arterial blood gas (ABG) and room air SpO₂ taken within 30 days of each other. Red lines indicate mean difference and the corresponding 95% CI. (C) Scatter plot comparing paired SaO₂ and PaO₂ from room air ABG. Red line indicates the curve of best fit. A minor jitter function was added to all graphs to better display overlapping data points.

Association of SaO₂ and SpO₂

Room air SaO₂ and SpO₂ values were documented within 30 days of each other for 105 patients. SaO₂ and SpO₂ were moderately correlated (r=0.39, p<0.001; figure 2A), with SpO₂ on average 3.0% higher than SaO₂ for those with paired measurements (95% CI 2.3 to 3.7, p<0.001). For patients with an SpO₂ ≥80%, 36% had an SpO₂ that was within 2% of the SaO₂. There was greater discordance between SaO₂ and SpO₂ with lower SaO₂ values (r=–0.60 for association of SaO₂ with the difference between SaO₂ and SpO₂, p<0.001) (figure 2B). Correlation between SaO₂ and SpO₂ was stronger (r=0.55) when limiting to measurements from the same day (online supplemental figure 1), with SpO₂ on average 1.8% higher than SaO₂ for those with same-day paired measurements (95% CI –3.5 to –0.2, p=0.02). On both unadjusted and adjusted analysis, none of age, sex, race, baseline FVC or baseline DLCO were associated with the difference between SaO₂ and SpO₂. Results were similar when restricting analyses to same-day measurements. As a surrogate for elevation, study site had no impact on study results. The non-linear association of SaO₂ and PaO₂ is shown in figure 2C.

Associations and prognostic significance of PaCO₂

The association of PaCO₂ with baseline variables is shown in table 2 (panels A, B). PaCO₂ as a continuous variable was associated with FVC, FEV₁ and smoking pack-years on unadjusted analysis (figure 3). On the primary adjusted analysis, only FVC and DLCO were independently associated with PaCO₂ as a continuous variable. Compared with normal PaCO₂, abnormal PaCO₂ (either high or low) was associated with lower baseline FVC and FEV₁ on unadjusted analysis, but only with lower FVC on the primary adjusted analysis. Results were not significantly different on the additional adjusted analyses in which FEV₁ was either substituted for FVC or added to the primary adjusted model. PaCO₂ as either a continuous or categorical variable was not associated with time to death or lung transplant on multivariable analysis.

Table 2.

(Panels A–B) Unadjusted analyses and the primary adjusted analysis of the association between baseline variables and PaCO₂ as both a continuous and a categorical variable; (panels C–F) unadjusted analyses and the primary adjusted analysis of the association between baseline variables and each acid-base abnormality with ≥25 patients in each category.

(A)
Variable	PaCO2 continuous
	Unadjusted analyses			Adjusted analysis
	Coefficient	95% CI	P value	Coefficient	95% CI	P value
Age	–0.29	–0.73 to 0.15	0.20	0.12	–0.45 to 0.69	0.68
BMI	0.28	–0.62 to 1.19	0.54	0.25	–0.77 to 1.27	0.63
Pack-years	–0.19	–0.41 to 0.04	0.10	–0.02	–0.32 to 0.27	0.87
Male sex	–0.28	–1.31 to 0.75	0.59	–0.64	–1.94 to 0.66	0.33
FVC	–1.07	–1.35 to –0.79	<0.001	–1.28	–1.67 to –0.88	<0.001
DLCO	0.01	–0.37 to 0.35	0.97	0.56	0.14 to 0.97	0.008

(B)
Variable	Abnormal PaCO2 (high or low)
	Unadjusted analyses			Adjusted analysis
	OR	95% CI	P value	OR	95% CI	P value
Age	1.06	0.91 to 1.24	0.43	0.99	0.79 to 1.23	0.91
BMI	0.77	0.56 to 1.07	0.12	1.02	0.68 to 1.51	0.94
Pack-years	0.95	0.88 to 1.03	0.21	0.93	0.83 to 1.05	0.25
Male sex	1.24	0.87 to 1.77	0.24	1.31	0.79 to 2.16	0.29
FVC	1.15	1.04 to 1.28	0.008	1.22	1.05 to 1.43	0.01
DLCO	0.98	0.86 to 1.12	0.77	0.88	0.75 to 1.03	0.11

(C)
Variable	Acute respiratory alkalosis
	Unadjusted analyses			Adjusted analysis
	OR	95% CI	P value	OR	95% CI	P value
Age	1.12	0.88 to 1.41	0.36	0.94	0.65 to 1.37	0.75
BMI	0.84	0.51 to 1.38	0.49	1.52	0.80 to 2.90	0.20
Pack-years	1.00	0.99 to 1.01	0.57	0.99	0.83 to 1.18	0.91
Male sex	0.87	0.52 to 1.46	0.60	0.87	0.40 to 1.89	0.73
FVC	1.27	1.10 to 1.47	0.001	1.49	1.19 to 1.86	0.001
DLCO	1.06	0.88 to 1.29	0.54	0.89	0.70 to 1.13	0.34

(D)
Variable	Metabolic alkalosis
	Unadjusted analyses			Adjusted analysis
	OR	95% CI	P value	OR	95% CI	P value
Age	1.35	0.98 to 1.85	0.07	1.38	0.81 to 2.36	0.24
BMI	1.15	0.63 to 2.12	0.65	1.62	0.74 to 3.56	0.23
Pack-years	1.01	0.99 to 1.02	0.25	1.11	0.90 to 1.37	0.34
Male sex	0.57	0.30 to 1.10	0.10	0.18	0.06 to 0.57	0.004
FVC	0.75	0.60 to 0.94	0.01	0.67	0.45 to 0.99	0.04
DLCO	0.73	0.53 to 1.00	0.05	0.69	0.45 to 1.04	0.08

(E)
Variable	Chronic respiratory acidosis
	Unadjusted analyses			Adjusted analysis
	OR	95% CI	P value	OR	95% CI	P value
Age	1.11	0.77 to 1.60	0.56	0.93	0.56 to 1.54	0.77
BMI	1.38	0.65 to 2.90	0.40	2.32	0.97 to 5.55	0.06
Pack-years	0.97	0.94 to 1.00	0.07	0.74	0.50 to 1.08	0.12
Male sex	0.73	0.33 to 1.64	0.44	1.05	0.33 to 3.36	0.94
FVC	0.70	0.53 to 0.92	0.01	0.78	0.52 to 1.17	0.23
DLCO	1.02	0.75 to 1.35	0.91	1.06	0.74 to 1.54	0.74

(F)
Variable	Chronic respiratory alkalosis
	Unadjusted analyses			Adjusted analysis
	OR	95% CI	P value	OR	95% CI	P value
Age	1.00	0.71 to 1.40	0.99	0.77	0.48 to 1.24	0.28
BMI	0.61	0.27 to 1.35	0.22	1.01	0.35 to 2.89	0.98
Pack-years	1.00	0.99 to 1.02	0.67	1.03	0.81 to 1.31	0.79
Male sex	1.13	0.51 to 2.49	0.76	2.19	0.62 to 7.64	0.22
FVC	1.35	1.07 to 1.71	0.01	1.61	1.17 to 2.20	0.003
DLCO	0.91	0.667 to 1.23	0.54	0.80	0.56 to 1.12	0.19

Linear regression was used with PaCO₂ as a continuous variable, and logistic regression with PaCO₂ as a categorical variable (abnormal (<35 or >45 mmHg) and normal (35–45 mmHg)). Logistic regression was used for all analyses with the acid-base categories. Significant associations are bolded. All continuous variables were scaled by a factor of 10.

BMI, body mass index; DLCO, diffusion capacity of the lung for carbon monoxide; FVC, forced vital capacity.

Figure 3.

(A) Scatter plot of PaCO₂ and baseline FVC percent. (B) Scatter plot of PaCO₂ and baseline FEV₁ percent. (C) Scatter plot of PaCO₂ and smoking pack years. FEV₁, forced expiratory volume in one second; FVC, forced vital capacity.

Acid-base analyses

The association of each acid-base category with baseline variables is shown in table 2 (panels C–F). A total of 15 patients had an acute respiratory acidosis, 7 patients had a metabolic acidosis, 70 patients had an acute respiratory alkalosis, 41 patients had a metabolic alkalosis, 26 patients had a chronic respiratory acidosis and 28 patients had a chronic respiratory alkalosis.

On unadjusted analysis, both higher FVC and FEV₁ were associated with an increased risk of an acute respiratory alkalosis (eg, for every 10% increase in FVC, there was a 1.12-fold increased likelihood of having an acute respiratory alkalosis). Both FVC and FEV₁ remained significantly associated with acute respiratory alkalosis on their respective adjusted analyses.

On unadjusted analysis, lower FVC, FEV₁ and DLCO were associated with an increased risk of metabolic alkalosis. In the primary adjusted analysis, FVC remained independently associated with an increased risk of metabolic alkalosis; however, no baseline lung function measurements retained statistical significance in the adjusted analysis that included all baseline lung function measurements.

Decreased FVC and FEV₁ were associated with a greater frequency of a chronic respiratory acidosis on unadjusted analysis. On the primary adjusted analysis, no variables retained statistically significant association with a chronic respiratory acidosis. However, both FEV₁ and BMI were associated with a chronic respiratory acidosis in the model with FEV₁ substituted for FVC and in the model with all three baseline lung function measurements. Higher FVC was associated with an increased risk of a chronic respiratory alkalosis on both unadjusted and multivariable analyses.

In the primary multinomial model, FVC (OR 1.4, p=0.005) was associated with acute respiratory alkalosis, male sex (OR 0.18, p=0.003) and FVC (OR 0.65, p=0.04) were associated with metabolic alkalosis, no variables were associated with a chronic respiratory acidosis, and FVC (OR 1.61, p=0.002) was associated with a chronic respiratory alkalosis. In the multinomial model that substituted FEV₁ for FVC, FEV₁ (OR 1.39, p=0.009) was associated with acute respiratory alkalosis, male sex (OR 0.22, p=0.007) and DLCO (OR 0.63, p=0.03) were associated with metabolic alkalosis, BMI (OR 2.64, p=0.03) and FEV₁ (OR 0.62, p=0.02) were associated with a chronic respiratory acidosis, and FEV₁ (OR 1.45, p=0.03) was associated with a chronic respiratory alkalosis. In the multinomial model that included all three baseline lung function measurements, both BMI (OR 2.84, p=0.02) and FEV₁ (OR 0.48, p=0.02) remained associated with a chronic respiratory acidosis, and only FVC (OR 2.02, p=0.03) was associated with a chronic respiratory alkalosis. None of the acid-base categories were associated with time to death or transplant on either unadjusted or multivariable analysis.

Discussion

We completed a retrospective chart review on a large prospective cohort to assess the clinical utility and potential limitations of pulse oximetry compared with ABG in patients with fibrotic ILD. To our knowledge, this is the first study to assess the reliability of SpO₂ as a surrogate of SaO₂ in patients with fibrotic ILD, showing frequent but mostly minor differences in these values. This is also the first study to assess the clinical significance of abnormal PaCO₂ in patients with fibrotic ILD, showing some association of abnormal PaCO₂ with reduced lung function but without any substantial prognostic significance.

Contrary to our hypothesis, most patients (64%) had a difference between SaO₂ and SpO₂ of 3% or more. The correlation of SaO₂ and SpO₂ was 0.39 (considered weak to moderate) when these were measured within 30 days of each other, improving to only 0.55 when measured on the same day. This correlation is lower than other studies completed in other respiratory populations (typically r>0.88).² However, some of the recent studies in COVID-19 have mixed results, with one study showing only a 1% discrepancy between SpO₂ and SaO₂ but others having much weaker correlation,^13–15 suggesting the importance of evaluating specific diseases and subpopulations in future studies.¹⁶ In the general population, pulse oximetry is considered accurate to within 2% (±1 SD).² Some factors that decrease reliability include hypoxaemia, low perfusion states, anaemia and darker skin pigmentation,3 7 17–20 with the potential that ILD-specific factors contribute to this greater discordance. We found that SpO₂ tended to overestimate SaO₂, and the variance was larger at lower levels of arterial oxygenation, which is consistent with most previous studies,² but with no baseline characteristics other than SpO₂ itself that were sufficiently associated with the magnitude of difference between SaO₂ and SpO₂ to be clinically useful. Another study showed that SpO₂ ≤91% identified nearly all patients with fibrotic ILD who required long-term oxygen therapy,²¹ suggesting that patients with SpO₂ >91% do not need testing for hypoxaemia even though there may still be substantial discordance with SaO₂. Lastly, quality of life and depressive symptoms are worse in patients with IPF who have more advanced physiological abnormality, functional impairment, exercise tolerance and greater oxygen needs, thus suggesting the importance of appropriate management of hypoxaemia in patients with severe disease.^{22 23}

There are limited data on PaCO₂ levels and the predictors and clinical significance of abnormal PaCO₂ in patients with fibrotic ILD. In our study, only 7% of patients early in their disease course (within 1 year of diagnosis) had elevated PaCO₂ compared with 11% of patients when measured ≥10 years after diagnosis. This is consistent with a previous study showing that patients with fibrotic ILD are typically eucapneic, even with poor lung function, and that changes in PaCO₂ are not related to spirometric values.²⁴ This is contrary to other respiratory populations such as chronic obstructive pulmonary disease (COPD), which has an inverse relationship between FEV₁ and PaCO₂.^25–29 In our population, only decreased baseline FVC was a reliable predictor of elevated or abnormal PaCO₂ and chronic respiratory acidosis in adjusted models.

Neither abnormal PaCO₂ nor specific acid-base categories were associated with time to death or transplant in our analysis. This is in contrast to patients with obstructive lung disease, and particularly COPD in which higher PaCO₂ is associated with worse disease severity.²⁵ Additionally, chronic respiratory acidosis has several clinical consequences in COPD, including myocardial depression, arrhythmias, decreased peripheral vascular resistance and hypotension, respiratory muscle weakness and an increase in proinflammatory cytokines.³⁰ The severity of acidosis is also associated with poor prognosis in COPD.^31–38 For patients with restrictive lung disease, PaCO₂ at time of lung transplant is associated with increased 1-year post-transplant mortality, and it is therefore used as part of the Lung Allocation Score.³⁹ However, there are no data to our knowledge that confirm the clinical significance of abnormal PaCO₂ values prior to transplant, suggesting that there is currently insufficient evidence to justify routine performance of ABG for assessment of acid-base status in patients with fibrotic ILD.

Due to our limited sample size, we analysed SaO₂ and SpO₂ values that were taken within 30 days of each other, rather than requiring these to be acquired on the same day. However, our results were similar when limiting to same-day measurements in patients who had concurrent measurements. A larger study is needed to further evaluate the correlation and difference between SaO₂ and SpO₂. We did not have any data on the type of pulse oximeter used, and other literature suggests variability in SpO₂ between different brands of pulse oximeter. We did not have haemoglobin measured concurrent to the ABG and SpO₂ values, so this variable was therefore excluded from the multivariable model. Future studies should evaluate whether anaemia alters the reliability of SpO₂. We also lacked reliable data on skin colour and were further limited by small numbers of patients in some race subgroups. More careful assessment of skin colour would be helpful in future studies to better assess how skin pigmentation affects SpO₂ accuracy. All our measurements were completed in a static environment; however, it would be of interest to test correlation of these values under both static and dynamic circumstances, particularly as SpO₂ is often used in an exercise setting to determine which patients qualify for long-term oxygen therapy. Finally, we captured patients at variable durations since diagnosis. Although this increases heterogeneity, this also improves generalisability of findings.

In summary, we show only a weak-to-moderate correlation between SaO₂ and SpO₂ in patients with fibrotic ILD and frequent minor differences in these values, without any clinically useful predictors of a greater difference. These findings raise questions regarding the clinical utility of ABG and pulse oximetry in patients with ILD. As the field of ILD moves towards a precision medicine approach,⁴⁰ this discordance of SpO₂ and SaO₂ highlights a potential pitfall in exclusively using SpO₂ in the assessment of the treatable trait of hypoxaemia. This further emphasises the need for additional studies comparing and optimising these tests. Abnormal PaCO₂ was associated with baseline FVC but was not prognostically informative. Collectively, these findings call attention to some of the limitations of ABGs in patients with fibrotic ILD, indicating the need for additional high-quality prospective studies to better characterise the potential clinical utility and performance characteristics of ABG in this patient population.

Data availability statement

No data are available.

Ethics statements

Patient consent for publication

Not applicable.

Ethics approval

This study involves human participants and was approved by University of British Columbia Research Ethics Board approval #H22-00780. Participants gave informed consent to participate in the study before taking part.