Can Formula-Assisted Venous Sampling Reliably Estimate Arterial Blood Gases?

Arterial Blood Gas (ABG) analysis is accepted as the gold standard method for demonstrating acid-base balance in the evaluation of metabolic and respiratory diseases [1]. However, the procedure requires experience, is painful and carries risks of various complications, making its routine application challenging. In clinical practice, the radial, brachial and femoral arteries are most frequently selected; arterial injury, thrombosis that may cause distal ischemia, hemorrhage and aneurysm formation are among the most common complications [2]. The likelihood of these complications increases further with repeated sampling attempts [3]. In diseases such as COPD, pulmonary embolism, and congestive heart failure that present with hypoxemia and dyspnea, arterial blood gas analysis is frequently performed for differential diagnosis and treatment monitoring, exposing patients to similar risks each time. In recent years, Venous Blood Gas (VBG) analysis has been increasingly investigated in clinical practice because of its potential to predict arterial parameters. It is being considered as an alternative or complementary method to invasive arterial sampling, particularly in emergency departments, intensive care units, and home healthcare settings. Moreover, the storage of these data in cloud-based systems via telemedicine applications and their processing with artificial intelligence algorithms now enables prediction of arterial values from venous parameters through learning from large datasets; this contributes substantially to clinical decision support systems.

Numerous studies have been published on this topic. In their emergency department study comparing arterial and venous blood gas parameters, Bakoğlu et al. [1] reported a strong correlation between arterial and venous pH, pCO2 and HCO3⁻ values, while the correlation between pO2 and sO2 was weak. In a multicenter prospective study, Boulain et al. [4] examined the effectiveness of samples obtained from the central venous system in estimating arterial blood gas and lactate levels; they stated that venous pH, pCO2, sO2 and lactate values could be used-via various formulasto estimate arterial blood gas in cardiac patients, although the results were not perfectly concordant. Ak et al. [5] demonstrated that pH, pCO2, and HCO3⁻ values obtained from venous blood gas analysis reliably predicted arterial results in patients presenting with COPD exacerbation. Lemoel et al. [6] developed an approach to predict arterial values from venous blood gas using the SpvO2 parameter and reported a marked improvement in detecting arterial blood gas abnormalities when SpvO2 adjustment was added to raw data. In light of these findings, the aim of our study was to collect simultaneous arterial and venous blood gas samples from patients presenting to the emergency department with an indication for arterial blood gas analysis and to compare the pCO2, pO2, bicarbonate, and pH parameters; additionally, we sought to investigate the predictability of arterial values from venous parameters using mathematical formulas previously described in the literature.

Study design and ethical approval

This prospective clinical study was conducted in the Emergency Department of Ufuk University School of Medicine between 01 May 2016 and 30 September 2016. Patients for whom arterial blood gas analysis was clinically indicated and who agreed to participate were informed about the study and provided written informed consent. Ethical approval was obtained from the Ufuk university non‑interventional scientific research evaluation commission. The principles of the world medical association declaration of Helsinki were observed throughout the study.

Sample and sampling method

A total of 105 patients who met the inclusion criterion of having a clinical indication for arterial blood gas analysis and who consented to participate were enrolled. For each patient, arterial and venous blood gas samples were drawn simultaneously. Venous blood gas samples were obtained through the intravenous line placed in the antecubital region and collected into heparinized negative‑pressure vacuum tubes. For arterial blood gas sampling, the radial artery was the first choice; in cases where this was not suitable, the femoral artery was punctured. Prefilled blood gas syringes (Plasti-med Blood Gas Syringe, 2mL, Turkey) containing 72 IU lithium heparin and fitted with a 22-gauge needle were used for arterial sampling.

Analytical method

All blood gas samples were analyzed on the blood gas analyzer located in the resuscitation room of the emergency department (Radiometer ABL800 BASIC, Denmark). The arterial and venous parameters obtained were compared with the acceptable error limits reported by the Clinical Laboratory Improvement Amendments (CLIA) [7].

Statistical analysis

Statistical analyses were performed using IBM SPSS statistics for windows, Version 20.0 (IBM Corp, Armonk, NY, USA). The normality of distribution of variables was assessed using visual methods (histograms and probability plots) and analytical tests (Kolmogorov-Smirnov, Shapiro-Wilk). Descriptive statistics for normally distributed variables were expressed as mean±standard deviation. Paired t‑tests were used to compare simultaneous arterial and venous measurements. Agreement between the two methods was examined with Bland-Altman plots. Additionally, regression analysis was conducted to investigate the ability of venous parameters to predict arterial values. A p value<0.05 was considered statistically significant.

pH

The mean arterial pH was 7.42±0.04 (range: 7.31-7.55), while the mean venous pH was 7.40±0.05 (range: 7.25-7.51). The difference between the two values was statistically significant (paired t test, p<0.001). A strong correlation was observed (Pearson’s r=0.820, p<0.001). Linear regression analysis was performed with arterial pH as the dependent variable, resulting in the following predictive model:
aV1pH = 1.73 + 0.77 × VpH (this study)
Other models reported in the literature were as follows:
aV2pH = 0.138 + 0.9851 × VpH [8]
aV3pH = 1.004 × VpH [5]
aV4pH = 0.763 × VpH + 1.786 [9]

Scatter plot analysis showed a linear relationship (R²=0.675), and Bland-Altman plots indicated minimal bias (Figure 1 & Table 1). The study formula (aV1pH) yielded an absolute reduction of approximately 10 percentage points (from 17% to 7%) in out-oflimit venous pH results versus unadjusted values. All previously published formulas (aV2pH-aV4pH) likewise produced clinically meaningful decreases compared with raw venous values (Table 1).

Figure 1:Bland-Altman plots of venous and arterial pH.

Table 1:Comparison of arterial and venous blood gas parameters between groups.

pCO₂

The mean arterial pCO₂ was 35.7±7.9mmHg (range: 16.7-64.3), whereas the mean venous pCO₂ was 40.2±8.3mmHg (range: 22.9- 66.0). The difference between arterial and venous pCO₂ values was statistically significant (paired t test, p<0.001). A strong correlation was observed (Pearson’s r=0.878, p<0.001). Linear regression analysis generated the following predictive model:
aV1pCO₂ = 0.834 × VpCO₂ + 2.253 (this study)
Comparison with previously published models yielded:
aV2pCO₂ = 0.806 × VpCO₂ + 1.457 [8]
aV3pCO₂ = 0.873 × VpCO₂ [5]
aV4pCO₂ = 0.611 × VpCO₂ + 9.521 [9]

Scatter plot analysis demonstrated a high linear fit (R²=0.77). Bland-Altman plots showed that most values were within the limits of agreement (Figure 1 & Table 1). Unadjusted venous pCO₂ exceeded CLIA limits in 51/106 (48%) pairs. Arterialization markedly reduced discordant results. The best-performing model for pCO₂ (aV3pCO₂) achieved an absolute reduction of 34.8 percentage points (48.0%→13.2%), corresponding to a relative reduction of about 72.5% in out-of-limit results compared with raw venous values. The study model (aV1pCO₂) performed similarly (14.2%). These findings indicate a clinically meaningful improvement in meeting CLIA acceptability through formula-based arterialization (Table 1).

Correlation between arterialized venous pO₂ and arterial pO₂ was higher (r=0.791, p<0.001). Scatter plot analysis showed a weak linear fit (R²=0.189). Bland-Altman plots indicated wide limits of agreement (Figure 1).

HCO₃⁻

The mean arterial HCO₃⁻ value was 22.5±3.7mmol/L, while the venous value was 23.9±3.9mmol/L. The difference between arterial and venous HCO₃⁻ values was statistically significant (p<0.001). A very strong correlation was observed (Pearson’s r=0.935, p<0.001). Regression analysis showed that venous bicarbonate was a significant predictor of arterial bicarbonate, while venous pCO₂ was not. The following model was obtained:
aVHCO₃⁻ = 0.9 × VHCO₃⁻ + 1 (this study)

Scatter plot analysis demonstrated a strong linear fit (R²=0.873). Bland-Altman plots showed minimal bias between measurements (Figure 1).

In this study we examined how closely peripheral venous blood gas results (pH, pCO2, pO2, HCO3-) reflect arterial values and what clinical value “Arterialized” (aV) prediction formulas derived from venous measurements add compared with raw venous results. The findings show that the degree of similarity differs by parameter and a single, uniform substitution strategy is not appropriate. Venous and arterial values are quite similar for pH and HCO3-, raw venous values for pCO2 are not consistently reliable on their own, and for pO2 deriving a dependable arterial estimate from venous data does not appear practical. We found an average arterial-venous pH difference of about 0.03 with a strong relationship; our formula reduced the mean difference below 0.01 and lowered the out-oflimit rate from 17%to7%. Published formulas (Ak, Khan, Kim) produced similar improvements, though the formula developed on our own dataset understandably performed slightly better [1,5,8,9]. Remaining within commonly cited acceptable difference limits (0.04-0.05) supports the use of venous pH aided by an adjustment [10-13]. For HCO3- the arterial-venous difference was small and clinically negligible, in line with prior reports [1,9,14], making venous measurement a practical option for initial metabolic assessment (Figure 2).

Figure 2:Scatter plots of venous versus arterial pH.

The situation is more problematic for pCO2. Although the mean arterial-venous difference fell within ranges reported in earlier studies [11-13], nearly half (≈48%) of raw venous pCO2 values lay outside acceptable error limits, weakening stand‑alone substitution. Our conversion formula reduced this proportion to 14%; other formulas also improved but to a lesser extent [5,8,9]. Variation in patient profiles and hypercapnia prevalence (e.g. 19-33% across series) can affect specificity, so broad external validation is needed before routine “replacement” claims are justified [11-13]. For pO2 the venous-arterial relationship was only moderate; modelling improved correlation, yet not to a clinically dependable level. Other studies likewise conclude venous pO2 and saturation do not reliably mirror arterial oxygenation [1,3,5,15]. Consequently, in scenarios where assessing hypoxemia is critical, arterial sampling remains necessary.

Overall, our results outline a three-tier picture: (1) pH and HCO3- : Venous values are generally suitable for most clinical contexts and become even more dependable with modest adjustments. (2) pCO2: Raw venous measurement is variable; conversion formulas reduce error but broad multicenter confirmation is required. (3) pO2: Current evidence does not support deriving clinically reliable arterial oxygen status from venous measurements. Superior performance of models built on our own dataset (notably for pH and pCO2) is expected and does not imply superiority until externally tested. Multicenter studies encompassing diverse patient groups (COPD exacerbation, sepsis, heart failure, ICU, emergency presentations) and different device types with adequate sample sizes are needed to determine generalizability of arterialization formulas. This study has several limitations. First, its single-center design and relatively small sample size limit the generalizability of the findings, particularly in patients with severe hypoxemia or hemodynamic instability. Second, as only one blood gas analyzer was employed, potential device-specific variability could not be assessed. Third, the conversion formulas were developed and validated using the same dataset, highlighting the need for external validation in broader and more diverse populations [16,17].

From telemedicine and AI Perspective, in the future, systematically accumulating both venous and arterial blood gas results-together with timestamp, primary diagnosis, oxygen/ ventilation support, key laboratory values, and clinical follow-up data-within secure digital platforms will be highly valuable. Large, well-structured data repositories can underpin development of AI-assisted, more accurate, patient-tailored prediction models. Practical tools could then estimate arterial values from venous measurements while transparently displaying error and uncertainty; this may reduce unnecessary arterial sampling, accelerate decision-making and strengthen remote monitoring for chronic conditions. Patient privacy, ethical approval, data security, continuous performance surveillance and timely model updates (e.g. when devices change) are critical safeguards. A stepwise buildout of such an ecosystem could enable broader, safer clinical use of venous blood gas analysis.

This prospective study indicates that simultaneously obtained venous blood gas samples can reflect arterial pH, pCO2 and HCO3⁻ with clinically acceptable agreement, whereas venous pO2 remains unreliable for estimating arterial oxygenation; thus, venous blood gas analysis may serve as a practical approach to reduce the need for invasive arterial puncture in serial monitoring of acid-base status and ventilation but cannot fully replace arterial sampling when precise assessment of hypoxemia or oxygenation is required. Mathematical conversion formulas from the literature provided only partial improvement in predicting arterial values and did not overcome the limitations related to oxygen parameters. The single-center design, limited sample size and underrepresentation of patients with severe hemodynamic instability constrain generalizability. Larger, multicenter studies integrating advanced predictive (including AI-based) models are warranted to refine venous-to-arterial estimation strategies.

AI-assisted tools were used only for language polishing/ translation and minor table formatting. No AI tool was used for data generation, statistical analysis, interpretation or drawing conclusions. The authors take full responsibility for the content.