Agreement and clinical utility of non-invasive SpCO versus arterial COHb in acute carbon monoxide poisoning: a prospective observational study Authors: Hee Young Lee, Da Kyung Kim, Yoonsuk Lee Published: 2026 ABSTRACT Rapid and accurate measurement of carboxyhemoglobin (COHb) is essential in managing acute carbon monoxide poisoning (ACOP). Non-invasive pulse CO-oximetry (SpCO) offers a convenient alternative to arterial blood gas analysis (ABGA), but its clinical accuracy and reliability in low COHb severity remain uncertain. This study evaluated the agreement and predictive performance of SpCO compared to arterial COHb in ACOP patients. In this prospective observational study, 81 ACOP patients underwent paired measurements of SpCO and arterial COHb. Patients were categorized into three groups based on COHb severity: low (< 10%), moderate (10–20%), and high (> 20%). Agreement was assessed using Deming and Passing–Bablok regression, Lin’s concordance correlation coefficient (CCC), Bland–Altman analysis, and accuracy metrics (R2, RMSE, MAE). A linear mixed-effects model (LMEM) was constructed to identify predictors of COHb, incorporating SpCO, treatment status, and oxygenation parameters. At baseline, SpCO showed only limited agreement with arterial COHb, with moderate concordance in the moderate COHb range but substantial positive bias and wide limits of agreement, particularly at higher concentrations. After HBOT, agreement deteriorated markedly, with unstable regression slopes, reduced concordance, and persistent heteroscedasticity across all severity strata. Similar deterioration was observed at follow-up in the non-HBOT group. In mixed-effects modeling, SpCO, COHb severity, and treatment time point were significant predictors of arterial COHb, whereas age, sex, and oxygenation parameters were not. The model demonstrated strong explanatory performance (marginal R2 = 0.904; RMSE = 2.61). Agreement between non‑invasive SpCO and arterial COHb is highly dependent on COHb concentration and measurement timing and declines markedly after HBOT, limiting the interchangeability of these methods in post‑treatment settings. SpCO may provide adjunctive information at baseline, particularly in moderate COHb ranges, but does not achieve sufficient agreement or precision to replace arterial COHb measurement, especially after treatment. Arterial COHb remains essential for accurate post-treatment assessment in ACOP. Introduction Carbon monoxide (CO) is a colorless, tasteless, odorless gas that results from the incomplete combustion of various carbon compounds. CO can also accumulate in enclosed spaces through exhaust gas1,2. CO poisoning remains a leading cause of accidental poisoning worldwide, resulting in thousands of emergency department visits and hospitalizations each year. CO exposure occurs due to the incomplete combustion of fuels (such as wood, coal, oil, paraffin, propane, and natural gas) in enclosed or poorly ventilated spaces. Since CO binds to hemoglobin with an affinity over 200 times that of oxygen, it significantly reduces the blood’s oxygen-carrying capacity, leading to hypoxic injury in critical organs such as the brain and heart. The rapid and accurate detection of carboxyhemoglobin (COHb) levels is crucial in both emergency and occupational health settings. In the United States, approximately 40,000 patients with carbon monoxide cons to hospitals every year, and reported mortality rates in clinical series range from approximately 1% to 31% across different cohorts and exposure contexts, reflecting the heterogeneity of clinical presentations and study populations3. More than 500 patients lose their lives every year due to accidents, and the mortality rate increases by 5–10 times in cases of intentional CO poisoning4,5. In addition, 25–50% of survivors of poisoning suffer from various and permanent neurocognitive complications6. According to the National Emergency Department Information System (NEDIS) data, a total of 7821 patients with acute CO poisoning visited emergency departments in Korea in 20187. Hyperbaric oxygen therapy (HBOT) is the treatment of choice for acute carbon monoxide poisoning (ACOP)8. COHb is not only a diagnostic marker but also plays a pivotal role in the initial assessment and clinical decision-making for CO poisoning9,10. Elevated COHb levels reflect the severity of exposure and hypoxic burden, and can be used to stratify patients according to risk. Furthermore, higher initial COHb concentrations have been associated with poorer short-term outcomes and a higher risk of delayed neurologic sequelae11. While arterial blood gas analysis (ABGA) remains the gold standard for measuring COHb, non-invasive pulse CO-oximeter devices allow monitoring of peripheral carboxyhemoglobin saturation (SpCO) using multi-wavelength pulse co-oximetry. However, their accuracy and clinical utility remain controversial, particularly across varying COHb concentrations and physiological conditions. This study aims to compare COHb levels measured by a non-invasive SpCO and ABGA in ACOP patients who visited the emergency department. In particular, it evaluates the agreement between the two methods across clinically relevant COHb severity levels and examines the temporal changes following HBOT to determine their clinical utility. Methods Ethical approval This study was a prospective observational investigation conducted at the emergency department of Yonsei University Wonju Severance Christian Hospital. Consecutive patients with a clinical diagnosis of acute carbon monoxide poisoning (ACOP) were screened for enrollment. Initially, 90 patients were recruited; however, nine were excluded due to incomplete paired measurements, technical errors, or violation of eligibility criteria, resulting in 81 patients included in the final analysis. The sample size of 90 was predetermined using PASS software (version 14.0, NCSS, East Kaysville, UT, USA), assuming a standard deviation of 4% for COHb, a non-inferiority margin of − 2.0%, a one-sided α of 0.025, and 90% power. The protocol received approval from the Institutional Review Board of Yonsei University Wonju Severance Christian Hospital (CR322127). Written informed consent was obtained either directly from patients with preserved decision-making capacity or from legally authorized representatives when patients were unable to provide consent. When consent was initially obtained from a legally authorized representative, patients were subsequently informed about study participation and provided written informed consent once decision-making capacity was recovered. This clinical study adhered to the International Conference on Harmonization (ICH) Guidelines and the principles of the Declaration of Helsinki. Clinical data collection To evaluate the performance of a commercial non-invasive COHb monitoring device, a clinical study was conducted with 81 participants who underwent COHb level measurements before and after hyperbaric oxygen therapy (HBOT). COHb levels were measured using both a reference standard and a commercial device for comparison. As the reference standard, a point-of-care blood gas analyzer (GEM Premier 5000, Instrumentation Laboratory, Bedford, MA, USA) was used to analyze COHb levels from arterial blood samples directly. In contrast, the commercial device, a non-invasive COHb monitor (Rad-97, Masimo Corporation, Irvine, CA, USA), estimated COHb levels using a multi-wavelength pulse co-oximetry method. COHb levels were categorized into three groups based on their distribution: low (< 10%), moderate (10–20%), and high (> 20%). Study design Patients were not pre-classified as having confirmed carbon monoxide poisoning at presentation. Instead, patients with suspected carbon monoxide exposure were identified based on the clinical context at emergency department arrival. After informed consent was obtained, non-invasive SpCO measurements were performed during the initial emergency department assessment. The diagnosis and assessment of severity of carbon monoxide poisoning were determined based on an integrated clinical evaluation, including carboxyhemoglobin levels from arterial blood gas analysis, clinical symptoms and neurologic status, exposure history, information from emergency medical services and bystanders, and environmental findings suggestive of carbon monoxide exposure. Arterial blood samples were collected to get baseline measurements, and HBOT was performed. After completing HBOT, the non-invasive monitoring device was reapplied, and additional arterial blood samples were collected to obtain the second set of data. Likewise, the non-HBOT group was monitored within 30 min of obtaining consent, and arterial blood samples were collected to establish baseline measurements. After three hours, the non-invasive monitoring device was reapplied, and additional arterial blood samples were collected to obtain the second set of data (Fig. 1). Because the actual completion time of hyperbaric oxygen therapy may vary due to compression/decompression and clinical workflow, follow-up SpCO and arterial blood gas measurements were uniformly performed approximately 3 h after treatment completion to ensure consistency. The same time interval was applied to the non-HBOT group.Fig. 1Flowchart of clinical study; (a) HBOT group, (b) Non-HBOT group. Flowchart of clinical study; (a) HBOT group, (b) Non-HBOT group. The estimated carbon monoxide exposure time was determined using a hierarchical approach. When available, patient self-reports after recovery of consciousness were prioritized, followed by caregiver or family reports. Emergency medical service documentation, including scene arrival and rescue timelines, was additionally reviewed. In cases where direct temporal information was unavailable, exposure duration was estimated based on scene characteristics, including fuel type, degree of enclosure, and ventilation conditions. Such estimates represent clinically informed approximations rather than direct measurements. Patients in the non-HBOT group either declined HBOT or were clinically assessed as having mild or asymptomatic carbon monoxide poisoning and were managed with standard conservative treatment, including high-flow supplemental oxygen and clinical observation. Inclusion and exclusion criteria A total of 81 patients with a clinical diagnosis of ACOP who presented to the emergency department were prospectively enrolled in this study. Eligible patients were adults (aged 18 years or older) with documented exposure to carbon monoxide. All included patients had paired measurements of non-invasive pulse CO-oximetry (SpCO) and arterial COHb, and provided informed consent. Patients were excluded if they had incomplete or missing paired COHb or SpCO measurements, underlying hemoglobinopathies, or SpCO measurements with insufficient signal quality due to technical factors such as poor peripheral perfusion, patient motion artifacts, unstable sensor placement, or transient signal instability. Among the final 81 enrolled patients, 60 received HBOT while 21 did not. Based on initial COHb levels, patients were further classified into three severity groups: low (< 10%, n = 39), moderate (10–20%, n = 24), and high (> 20%, n = 18) (Fig. 2).Fig. 2Flowchart of classifying the data; ACOP: Acute carbon monoxide poisoning, COHb: Carboxyhemoglobin, HBOT: Hyperbaric oxygen therapy. Flowchart of classifying the data; ACOP: Acute carbon monoxide poisoning, COHb: Carboxyhemoglobin, HBOT: Hyperbaric oxygen therapy. Statistical analysis All statistical analyses were performed using SPSS Statistics (version 28.0; IBM Corp., Armonk, NY, USA). A two-sided p-value < 0.05 was considered statistically significant. Baseline characteristics were summarized by treatment group (HBOT vs. non-HBOT). Continuous variables were expressed as mean ± standard deviation (SD) or median with interquartile range (IQR), as appropriate, and categorical variables as counts and percentages. Between-group comparisons at baseline were performed using the Wilcoxon rank-sum test for continuous variables and Fisher’s exact test or the chi-square test for categorical variables, as appropriate. For analyses involving repeated measurements, only paired observations with both pre- and post-treatment values available were included. Non-invasive measurements of carboxyhemoglobin (SpCO) and peripheral oxygen saturation (SpO2), along with blood levels of carboxyhemoglobin (COHb), oxyhemoglobin (O2Hb), and methemoglobin (MetHb), were analyzed before and after HBOT. Normality of within-subject change scores was assessed using the Shapiro–Wilk test and visual inspection of Q–Q plots. Because several key variables demonstrated non-normal distributions, non-parametric tests were used as the primary analytical approach. Within-group comparisons of pre- and post-treatment values were conducted using the Wilcoxon signed-rank test, and between-group comparisons of change scores were performed using the Wilcoxon rank-sum test. To further evaluate treatment effects while accounting for repeated measurements within individuals, linear mixed-effects models were fitted. Fixed effects included treatment group (HBOT vs. non-HBOT), time (pre vs. post), and their interaction. Age and sex were included as fixed covariates to adjust for potential confounding, and subject identification number was included as a random intercept. Agreement between non-invasive SpCO measurements and arterial COHb levels was assessed using multiple complementary approaches. To evaluate linear relationships while accounting for measurement error in both variables, Deming regression was performed. Because Deming regression can become unstable in the presence of non‑linearity or extreme values, a Passing–Bablok regression, a robust non‑parametric alternative that does not assume normality or homoscedasticity, was additionally conducted to provide a distribution‑free estimate of proportional and constant bias. Correlation analysis (Pearson or Spearman, as appropriate) was used to describe the strength of linear association but not as a measure of agreement. Concordance between methods was evaluated using Lin’s concordance correlation coefficient (CCC). Bland–Altman analysis was performed to assess systematic bias, limits of agreement, and potential heteroscedasticity across the range of measurements. Agreement analyses were conducted at baseline and follow-up in both the HBOT and non-HBOT groups, and subgroup analyses were performed according to predefined COHb severity categories to explore concentration-dependent variability. In addition, the coefficient of determination (R2), root mean square error (RMSE), and mean absolute error (MAE) were calculated to quantify predictive accuracy. To further identify factors associated with arterial COHb levels while accounting for repeated measurements within individuals, a linear mixed-effects model (LMEM) was constructed. Fixed effects included Age, Sex, SpCO, COHb severity (low, moderate, and high), treatment group (HBOT vs. non-HBOT), treatment time point (pre- vs. post-HBOT), peripheral oxygen saturation (SpO2), oxyhemoglobin (O2Hb), and methemoglobin (MetHb), with subject identification number included as a random effect. Model performance was evaluated using marginal and conditional R2 values, and residual variance was reported. Results General characteristics of the study population Baseline characteristics of the study population are summarized in Table 1, stratified by HBOT status. In terms of sex distribution, both groups showed a predominance of female patients. In the HBOT group, males accounted for 36.7% of patients, while females comprised 63.3%, whereas in the non-HBOT group, males and females accounted for 33.3% and 66.7%, respectively. The mean age was similar between the two groups, with a mean age of 51.0 years in the HBOT group and 50.0 years in the non-HBOT group. Baseline vital signs were generally comparable between groups. The HBOT group had a mean systolic blood pressure of 134.1 mmHg and a mean diastolic blood pressure of 83.7 mmHg, while the non-HBOT group showed corresponding values of 127.3 mmHg and 79.2 mmHg. Pulse rate, respiratory rate, and body temperature were also similar between groups, with no marked differences in overall distributions. Regarding mental status at presentation, the non-HBOT group showed a higher proportion of patients who were alert (61.9%) compared with the HBOT group (53.3%). In the non-HBOT group, 19.0% of patients exhibited verbal responses and 14.3% exhibited painful responses, with one patient (4.8%) being unresponsive. Consistent with these findings, the mean Glasgow Coma Scale (GCS) score was 12.0 in the HBOT group and 11.2 in the non-HBOT group. Clear differences were observed in the source of carbon monoxide exposure. In the HBOT group, the most common source was charcoal (83.3%), followed by gas exposure (10.0%), firewood (5.0%), and fire-related exposure (1.7%). Similarly, charcoal was the predominant source in the non-HBOT group (85.7%), although a small proportion of patients were exposed through firewood (4.8%) or fire (9.5%), with no cases related to gas exposure in this group. The maximum exposure time was comparable between groups. The HBOT group had a mean exposure duration of 5.8 h, while the non-HBOT group had a mean exposure time of 5.7 h, with similar distributions across both groups. There were no statistically significant differences between the HBOT and non-HBOT groups in age, sex distribution, initial vital signs (SBP, DBP, pulse rate, respiratory rate, and body temperature), level of consciousness category, or GCS score at presentation (all p > 0.05). The proportions of intentional exposure, the distribution of exposure sources (charcoal, firewood, gas, and fire), and the maximum estimated exposure time also did not differ significantly between groups (all p > 0.05).Table 1Baseline characteristics of study participants (n = 81).CharacteristicHBOT (n = 60)Non-HBOT (n = 21)p valueSex, n (%) > 0.990 Male22 (36.7%)7 (33.3%) Female38 (63.3%)14 (66.7%)Age (years)0.838 Average (mean ± SD)51.0 ± 15.250.0 ± 16.1 Median ([IQR])50.0 [41.8–60.0]49.0 [35.0–61.0]Vital signs (mean ± SD) Systolic blood pressure (mmHg)134.1 ± 20.4127.3 ± 19.80.117 Diastolic blood pressure (mmHg)83.7 ± 14.879.2 ± 15.40.113 Pulse rate (bpm)97.3 ± 19.590.0 ± 13.20.126 Respiratory rate (bpm)15.8 ± 6.616.1 ± 6.20.978 Body temperature (°C)36.6 ± 0.936.6 ± 0.90.621Mental status, n (%)0.270 Alert32 (53.3%)13 (61.9%) Verbal response19 (31.7%)4 (19.0%) Painful response9 (15.0%)3 (14.3%) Unresponsive0 (0%)1 (4.8%)Glasgow Coma Scale score (point)0.466 Average (mean ± SD)12.0 ± 3.511.2 ± 4.1 Median ([IQR])14 [9–15]15 [8–15]CO source, n (%)0.190 Charcoal50 (83.3%)18 (85.7%) Firewood3 (5.0%)1 (4.8%) Gas6 (10.0%)0 (0.0%) Fire1 (1.7%)2 (9.5%)Max exposure time (hours)0.888 Average (mean ± SD)5.8 ± 5.65.7 ± 4.8 Median ([IQR])3.1 [1.6–8.2]3.0 [1.8–8.0]*HBOT: hyperbaric oxygen therapy; SD: standard deviation; IQR: interquartile range; mmHg: millimeters of mercury; bpm: beats per minute; °C: degrees Celsius; CO: carbon monoxide. Table 1Baseline characteristics of study participants (n = 81).CharacteristicHBOT (n = 60)Non-HBOT (n = 21)p valueSex, n (%) > 0.990 Male22 (36.7%)7 (33.3%) Female38 (63.3%)14 (66.7%)Age (years)0.838 Average (mean ± SD)51.0 ± 15.250.0 ± 16.1 Median ([IQR])50.0 [41.8–60.0]49.0 [35.0–61.0]Vital signs (mean ± SD) Systolic blood pressure (mmHg)134.1 ± 20.4127.3 ± 19.80.117 Diastolic blood pressure (mmHg)83.7 ± 14.879.2 ± 15.40.113 Pulse rate (bpm)97.3 ± 19.590.0 ± 13.20.126 Respiratory rate (bpm)15.8 ± 6.616.1 ± 6.20.978 Body temperature (°C)36.6 ± 0.936.6 ± 0.90.621Mental status, n (%)0.270 Alert32 (53.3%)13 (61.9%) Verbal response19 (31.7%)4 (19.0%) Painful response9 (15.0%)3 (14.3%) Unresponsive0 (0%)1 (4.8%)Glasgow Coma Scale score (point)0.466 Average (mean ± SD)12.0 ± 3.511.2 ± 4.1 Median ([IQR])14 [9–15]15 [8–15]CO source, n (%)0.190 Charcoal50 (83.3%)18 (85.7%) Firewood3 (5.0%)1 (4.8%) Gas6 (10.0%)0 (0.0%) Fire1 (1.7%)2 (9.5%)Max exposure time (hours)0.888 Average (mean ± SD)5.8 ± 5.65.7 ± 4.8 Median ([IQR])3.1 [1.6–8.2]3.0 [1.8–8.0]*HBOT: hyperbaric oxygen therapy; SD: standard deviation; IQR: interquartile range; mmHg: millimeters of mercury; bpm: beats per minute; °C: degrees Celsius; CO: carbon monoxide. Baseline characteristics of study participants (n = 81). *HBOT: hyperbaric oxygen therapy; SD: standard deviation; IQR: interquartile range; mmHg: millimeters of mercury; bpm: beats per minute; °C: degrees Celsius; CO: carbon monoxide. Comparison of before and after HBOT changes Table 2 summarizes pre- and post-treatment changes in SpCO, SpO2, COHb, O2Hb, and MetHb in the HBOT group (n = 60) and the non-HBOT group (n = 21). Within-group analyses demonstrated significant pre-to-post changes in several parameters in both groups. In the HBOT group, non-invasive SpCO and arterial COHb levels decreased significantly, while oxyhemoglobin (O₂Hb) levels increased significantly over time (all p < 0.001). A small but statistically significant decrease in SpO₂ was also observed (p = 0.014). Changes in methemoglobin (MetHb) were not statistically significant. Similarly, in the non-HBOT group, significant reductions in SpCO and COHb levels were observed, along with significant increases in O₂Hb and SpO₂ over time (all p < 0.05). Changes in MetHb were not statistically significant.Table 2Changes in physiological parameters before and after HBOT.VariableHBOT group (n = 60)Non-HBOT group (n = 21)Between-group p valueMean difference95% CI[IQR]Within-group p valueMean difference95% CI[IQR]Within-group p valueSpCO (%)− 7.29[− 9.05, − 5.53] < 0.001− 9.73[− 13.34, − 6.13] < 0.0010.217SpO2 (%)− 0.83[− 1.98, 0.33]0.0140.71[0.06, 1.37]0.0340.019COHb (%)− 12.57[− 14.75, − 10.39] < 0.001− 9.84[− 12.75, − 6.93] < 0.0010.129O₂Hb (%)12.77[7.35, 18.19] < 0.00116.35[3.79, 28.92]0.0240.592MetHb (%)− 0.07[− 0.22, 0.07]0.686− 0.07[− 0.48, 0.35]0.9720.666*HBOT: hyperbaric oxygen therapy; CI: confidence interval; IQR: interquartile range; SpCO: peripheral carboxyhemoglobin saturation; SpO2: peripheral oxygen saturation; COHb: carboxyhemoglobin; O2Hb: oxyhemoglobin; MetHb: methemoglobin. Table 2Changes in physiological parameters before and after HBOT.VariableHBOT group (n = 60)Non-HBOT group (n = 21)Between-group p valueMean difference95% CI[IQR]Within-group p valueMean difference95% CI[IQR]Within-group p valueSpCO (%)− 7.29[− 9.05, − 5.53] < 0.001− 9.73[− 13.34, − 6.13] < 0.0010.217SpO2 (%)− 0.83[− 1.98, 0.33]0.0140.71[0.06, 1.37]0.0340.019COHb (%)− 12.57[− 14.75, − 10.39] < 0.001− 9.84[− 12.75, − 6.93] < 0.0010.129O₂Hb (%)12.77[7.35, 18.19] < 0.00116.35[3.79, 28.92]0.0240.592MetHb (%)− 0.07[− 0.22, 0.07]0.686− 0.07[− 0.48, 0.35]0.9720.666*HBOT: hyperbaric oxygen therapy; CI: confidence interval; IQR: interquartile range; SpCO: peripheral carboxyhemoglobin saturation; SpO2: peripheral oxygen saturation; COHb: carboxyhemoglobin; O2Hb: oxyhemoglobin; MetHb: methemoglobin. Changes in physiological parameters before and after HBOT. *HBOT: hyperbaric oxygen therapy; CI: confidence interval; IQR: interquartile range; SpCO: peripheral carboxyhemoglobin saturation; SpO2: peripheral oxygen saturation; COHb: carboxyhemoglobin; O2Hb: oxyhemoglobin; MetHb: methemoglobin. When comparing the magnitude of pre-to-post changes between groups, no statistically significant differences were observed. Although the reduction in COHb was numerically greater in the HBOT group, this difference did not reach statistical significance. The mean change in COHb was − 12.57% in the HBOT group and − 9.84% in the non-HBOT group, corresponding to a between-group difference of − 2.73% (95% CI − 6.3 to 0.8; p = 0.129). A statistically significant between-group difference was observed only for SpO2 (p = 0.019), reflecting opposite directions of change between the two groups. Agreement between SpCO and COHb across HBOT and non-HBOT conditions As shown in Table 3, the agreement between non-invasive SpCO and arterial COHb across HBOT and non-HBOT conditions. In the HBOT group, pre-HBOT measurements demonstrated moderate agreement, with a regression slope of 0.755 and an intercept close to zero (− 0.599). Concordance indices supported this relationship (CCC = 0.680; r = 0.794; R2 = 0.630), although Bland–Altman analysis revealed a positive systematic bias of 12.0% and wide limits of agreement (− 3.4–27.5%), accompanied by heteroscedasticity. These findings indicate that SpCO tended to overestimate COHb at higher concentrations, despite a modest linear association before treatment. Following HBOT, agreement deteriorated markedly. The regression slope became negative (− 1.021), and concordance was poor (CCC = − 0.115), with minimal linear correlation (r = − 0.171; R2 = 0.029). Although the mean bias decreased to 2.0% and the limits of agreement became narrower, heteroscedasticity persisted and concordance deteriorated, indicating that post-treatment SpCO measurements did not reliably track arterial COHb across individuals. Given the violation of key Deming regression assumptions, as evidenced by numerical instability and heteroscedastic error structure, Passing–Bablok regression was employed as a more robust approach for estimating proportional and constant bias. In the non-HBOT group, baseline measurements demonstrated strong agreement, with a slope of 0.955, high concordance (CCC = 0.860), strong correlation (r = 0.881; R2 = 0.776), and no evidence of heteroscedasticity, indicating stable measurement behavior in the absence of treatment. Despite a similar positive bias of 12.0%, concordance and regression metrics suggested more stable measurement behavior under baseline conditions. However, agreement declined substantially at follow-up. The regression slope increased to 3.226, while concordance (CCC = 0.219) and correlation (r = 0.259; R2 = 0.067) were low, suggesting that the steep slope was driven by influential observations rather than a consistent linear trend. Although the mean bias was reduced to 2.2% with narrower limits of agreement, the overall interchangeability between SpCO and COHb remained limited.Table 3Agreement and regression analysis between SpCO and arterial COHb measurements across HBOT and non-HBOT conditions.IndicatorsHBOT group (n = 60)Non-HBOT group (n = 21)Pre-HBOTPost-HBOTBaselineFollow-upSlope0.755− 1.021†0.955†3.226Intercept− 0.5993.327†− 0.764†− 6.984Lin’s CCC0.680− 0.1150.8600.219Pearson’s r0.794− 0.1710.8810.259R20.6300.0290.7760.067RMSE6.4963.4954.1802.476MAE4.6192.4193.0571.943Bland–Altman bias (%)12.0162.03512.0102.22495% LoA (%)− 3.418–27.450− 2.396–6.467− 3.625–27.644− 1.587–6.035HeteroscedasticityPresentPresentAbsentAbsent*HBOT: hyperbaric oxygen therapy; CCC: concordance correlation coefficient; RMSE: root mean square error; MAE: mean absolute error; LoA: limits of agreement. †Passing–Bablok regression was applied in place of Deming regression when violations of Deming regression assumptions were identified, including numerical instability and the presence of heteroscedastic measurement error. Passing–Bablok regression, a nonparametric and robust method that does not require assumptions of normality or homoscedasticity, was therefore used to provide more reliable estimates of proportional and constant bias between SpCO and arterial COHb measurements. Table 3Agreement and regression analysis between SpCO and arterial COHb measurements across HBOT and non-HBOT conditions.IndicatorsHBOT group (n = 60)Non-HBOT group (n = 21)Pre-HBOTPost-HBOTBaselineFollow-upSlope0.755− 1.021†0.955†3.226Intercept− 0.5993.327†− 0.764†− 6.984Lin’s CCC0.680− 0.1150.8600.219Pearson’s r0.794− 0.1710.8810.259R20.6300.0290.7760.067RMSE6.4963.4954.1802.476MAE4.6192.4193.0571.943Bland–Altman bias (%)12.0162.03512.0102.22495% LoA (%)− 3.418–27.450− 2.396–6.467− 3.625–27.644− 1.587–6.035HeteroscedasticityPresentPresentAbsentAbsent*HBOT: hyperbaric oxygen therapy; CCC: concordance correlation coefficient; RMSE: root mean square error; MAE: mean absolute error; LoA: limits of agreement. †Passing–Bablok regression was applied in place of Deming regression when violations of Deming regression assumptions were identified, including numerical instability and the presence of heteroscedastic measurement error. Passing–Bablok regression, a nonparametric and robust method that does not require assumptions of normality or homoscedasticity, was therefore used to provide more reliable estimates of proportional and constant bias between SpCO and arterial COHb measurements. Agreement and regression analysis between SpCO and arterial COHb measurements across HBOT and non-HBOT conditions. *HBOT: hyperbaric oxygen therapy; CCC: concordance correlation coefficient; RMSE: root mean square error; MAE: mean absolute error; LoA: limits of agreement. †Passing–Bablok regression was applied in place of Deming regression when violations of Deming regression assumptions were identified, including numerical instability and the presence of heteroscedastic measurement error. Passing–Bablok regression, a nonparametric and robust method that does not require assumptions of normality or homoscedasticity, was therefore used to provide more reliable estimates of proportional and constant bias between SpCO and arterial COHb measurements. Overall, SpCO demonstrated moderate agreement with arterial COHb under baseline conditions but showed markedly reduced reliability following HBOT and during follow-up in the non-HBOT group, underscoring the limitations of SpCO as a surrogate measure in post-treatment or low-concentration settings. Agreement between SpCO and COHb measurements by COHb severity category In Table 4, the agreement between SpCO and arterial COHb after stratifying patients in the HBOT group by COHb level (low, moderate, and high). Across all strata, agreement patterns differed markedly by both COHb concentration and treatment time point. In the low COHb group, pre-HBOT measurements demonstrated poor agreement, with negligible proportional association (slope ≈ 0), low concordance (CCC = 0.221), and weak correlation (r = 0.354). Bland–Altman analysis revealed a modest positive bias (5.9%) with limited dispersion of differences and no evidence of heteroscedasticity. After HBOT, agreement further deteriorated, with loss of correlation (r = − 0.070; R2 = 0.005), minimal concordance (CCC = − 0.038), and the emergence of heteroscedasticity, despite a reduction in mean bias to 1.6%. In the moderate COHb group, pre‑HBOT measurements showed only modest concordance (CCC = 0.403) and moderate correlation (r = 0.550; R2 = 0.302) compared with the low COHb group, but this was accompanied by a substantial positive bias (14.1%) and wide limits of agreement, indicating that SpCO and arterial COHb are not directly interchangeable even in this range. Following HBOT, agreement collapsed, with near-zero correlation (r = 0.024; R2 = 0.001), negligible concordance (CCC = 0.013), and persistent heteroscedasticity, despite a marked reduction in average bias (2.3%). In the high COHb group, pre-HBOT measurements demonstrated the strongest linear association among all strata, with a slope of 1.554, moderate concordance (CCC = 0.282), and relatively strong correlation (r = 0.606; R2 = 0.367). However, this group also exhibited the largest systematic bias (21.6%) and the widest limits of agreement, indicating substantial overestimation by SpCO at higher COHb levels. After HBOT, agreement declined markedly, with reversal of correlation (r = − 0.527), reduced explanatory power (R2 = 0.278), and persistent heteroscedasticity, although mean bias decreased to 2.6% with narrower limits of agreement.Table 4Agreement and regression analysis between SpCO and arterial COHb stratified by COHb level in the HBOT group.COHb levelIndicatorsLow (n = 28)Moderate (n = 19)High (n = 13)Pre-HBOTPost-HBOTPre-HBOTPost-HBOTPre-HBOTPost-HBOTSlope0.000†4.000†0.044†− 2.727†1.554− 0.305†Intercept5.700†− 1.000†11.121†5.846†− 24.8152.011†Lin’s CCC0.221− 0.0380.4030.0130.282− 0.421Pearson’s r0.354− 0.0700.5500.0240.606− 0.527R20.1250.0050.3020.0010.3670.278RMSE3.0342.4914.6343.4729.2323.702MAE2.3411.7893.6952.2146.2712.167Bland–Altman Bias5.9481.56114.1052.34221.5652.59695% LoA1.257–10.369− 1.700–4.8225.786–22.424− 2.548–7.2324.965–38.165− 2.942–8.134HeteroscedasticityAbsentPresentAbsentPresentAbsentPresent*COHb: carboxyhemoglobin; HBOT: hyperbaric oxygen therapy; CCC: concordance correlation coefficient; RMSE: root mean square error; MAE: mean absolute error; LoA: limits of agreement. †COHb levels were stratified into low, moderate, and high categories based on pre-HBOT arterial COHb concentrations. Agreement between SpCO and arterial COHb was evaluated using Deming regression and Bland–Altman analyses before and after HBOT within each COHb stratum. When Deming regression yielded unstable or implausible estimates due to restricted data range, heteroscedasticity, or influential observations, Passing–Bablok regression was additionally applied as a robust non-parametric alternative. Table 4Agreement and regression analysis between SpCO and arterial COHb stratified by COHb level in the HBOT group.COHb levelIndicatorsLow (n = 28)Moderate (n = 19)High (n = 13)Pre-HBOTPost-HBOTPre-HBOTPost-HBOTPre-HBOTPost-HBOTSlope0.000†4.000†0.044†− 2.727†1.554− 0.305†Intercept5.700†− 1.000†11.121†5.846†− 24.8152.011†Lin’s CCC0.221− 0.0380.4030.0130.282− 0.421Pearson’s r0.354− 0.0700.5500.0240.606− 0.527R20.1250.0050.3020.0010.3670.278RMSE3.0342.4914.6343.4729.2323.702MAE2.3411.7893.6952.2146.2712.167Bland–Altman Bias5.9481.56114.1052.34221.5652.59695% LoA1.257–10.369− 1.700–4.8225.786–22.424− 2.548–7.2324.965–38.165− 2.942–8.134HeteroscedasticityAbsentPresentAbsentPresentAbsentPresent*COHb: carboxyhemoglobin; HBOT: hyperbaric oxygen therapy; CCC: concordance correlation coefficient; RMSE: root mean square error; MAE: mean absolute error; LoA: limits of agreement. †COHb levels were stratified into low, moderate, and high categories based on pre-HBOT arterial COHb concentrations. Agreement between SpCO and arterial COHb was evaluated using Deming regression and Bland–Altman analyses before and after HBOT within each COHb stratum. When Deming regression yielded unstable or implausible estimates due to restricted data range, heteroscedasticity, or influential observations, Passing–Bablok regression was additionally applied as a robust non-parametric alternative. Agreement and regression analysis between SpCO and arterial COHb stratified by COHb level in the HBOT group. *COHb: carboxyhemoglobin; HBOT: hyperbaric oxygen therapy; CCC: concordance correlation coefficient; RMSE: root mean square error; MAE: mean absolute error; LoA: limits of agreement. †COHb levels were stratified into low, moderate, and high categories based on pre-HBOT arterial COHb concentrations. Agreement between SpCO and arterial COHb was evaluated using Deming regression and Bland–Altman analyses before and after HBOT within each COHb stratum. When Deming regression yielded unstable or implausible estimates due to restricted data range, heteroscedasticity, or influential observations, Passing–Bablok regression was additionally applied as a robust non-parametric alternative. Overall, stratified analyses showed that the apparent agreement between SpCO and arterial COHb varied by COHb concentration, with relatively stronger linear association at higher pre‑HBOT levels but consistently large bias and wide limits of agreement. After HBOT, agreement weakened further, with unstable regression slopes, reduced concordance, and emerging heteroscedasticity, indicating that SpCO cannot be regarded as a reliable stand‑alone surrogate for arterial COHb, particularly in post‑treatment or low‑concentration settings. Linear mixed-effects model predicting arterial COHb levels Table 5 shows that a linear mixed-effects model was fitted to examine the associations between arterial COHb levels and clinical covariates, including age, sex, SpCO, COHb severity (low vs. moderate vs. high), HBOT group (treated vs. untreated), HBOT session (pre- vs. post-treatment), SpO₂, O₂Hb, and MetHb, with subject ID included as a random effect.Table 5Results of linear mixed-effects model predicting arterial COHb levels and clinical covariates.EffectEstimateStd. errort-valuep value95% CIFixed effects Intercept− 9.3547.678− 1.2180.223[− 24.40, 5.69] Age (per year)0.0230.0141.6200.105[− 0.00, 0.05] Sex (male vs. female)0.2490.4600.5410.588[− 0.65, 1.15] SpCO0.2590.0564.620 < 0.001[0.15, 0.37] COHb severity (moderate vs low)5.5090.8416.551 < 0.001[3.86, 7.16] COHb severity (high vs low)14.7221.08213.601 < 0.001[12.60, 16.84] HBOT group (treated vs untreated)− 0.1930.489− 0.3940.693[− 1.15, 0.77] HBOT session (post vs pre)− 4.9180.532− 9.238 < 0.001[− 5.96, − 3.87] SpO20.1430.0781.8460.065[− 0.01, 0.30] O2Hb− 0.0020.012− 0.1560.876[− 0.03, 0.02] MetHb0.3350.4920.6800.496[− 0.63, 1.30]Random effects Subject ID (intercept variance)0.000–––– Residual variance6.788–––– Model fit & accuracy AIC767.962–––– BIC773.957–––– Marginal R20.904–––– Conditional R20.904–––– RMSE2.605––––*Std.: standard deviation; CI: confidence interval; SpCO: peripheral carboxyhemoglobin saturation; COHb: carboxyhemoglobin; SpO2: peripheral oxygen saturation; O2Hb: oxyhemoglobin; MetHb: methemoglobin; AIC: Akaike information criterion; BIC: Bayesian information criterion; RMSE: root mean square error. Table 5Results of linear mixed-effects model predicting arterial COHb levels and clinical covariates.EffectEstimateStd. errort-valuep value95% CIFixed effects Intercept− 9.3547.678− 1.2180.223[− 24.40, 5.69] Age (per year)0.0230.0141.6200.105[− 0.00, 0.05] Sex (male vs. female)0.2490.4600.5410.588[− 0.65, 1.15] SpCO0.2590.0564.620 < 0.001[0.15, 0.37] COHb severity (moderate vs low)5.5090.8416.551 < 0.001[3.86, 7.16] COHb severity (high vs low)14.7221.08213.601 < 0.001[12.60, 16.84] HBOT group (treated vs untreated)− 0.1930.489− 0.3940.693[− 1.15, 0.77] HBOT session (post vs pre)− 4.9180.532− 9.238 < 0.001[− 5.96, − 3.87] SpO20.1430.0781.8460.065[− 0.01, 0.30] O2Hb− 0.0020.012− 0.1560.876[− 0.03, 0.02] MetHb0.3350.4920.6800.496[− 0.63, 1.30]Random effects Subject ID (intercept variance)0.000–––– Residual variance6.788–––– Model fit & accuracy AIC767.962–––– BIC773.957–––– Marginal R20.904–––– Conditional R20.904–––– RMSE2.605––––*Std.: standard deviation; CI: confidence interval; SpCO: peripheral carboxyhemoglobin saturation; COHb: carboxyhemoglobin; SpO2: peripheral oxygen saturation; O2Hb: oxyhemoglobin; MetHb: methemoglobin; AIC: Akaike information criterion; BIC: Bayesian information criterion; RMSE: root mean square error. Results of linear mixed-effects model predicting arterial COHb levels and clinical covariates. *Std.: standard deviation; CI: confidence interval; SpCO: peripheral carboxyhemoglobin saturation; COHb: carboxyhemoglobin; SpO2: peripheral oxygen saturation; O2Hb: oxyhemoglobin; MetHb: methemoglobin; AIC: Akaike information criterion; BIC: Bayesian information criterion; RMSE: root mean square error. The analysis revealed that SpCO, COHb severity, and HBOT session were significant predictors of arterial COHb levels. Higher SpCO values were associated with higher arterial COHb concentrations after multivariable adjustment, although the magnitude of this association was modest. COHb severity demonstrated a strong graded effect, with the moderate-severity group exhibiting COHb levels approximately 5.5 percentage points higher than the low-severity reference group, and the high-severity group exhibiting COHb levels approximately 14.7 percentage points higher than the low-severity group (both p < 0.001). Independent of treatment group, arterial COHb levels were significantly lower at the post-treatment time point compared with baseline, indicating a marked temporal reduction in COHb following treatment. In contrast, age, sex, HBOT group assignment, SpO₂, O₂Hb, and MetHb were not significantly associated with arterial COHb levels in the adjusted model (all p > 0.05). Random-effects estimation indicated that the subject-level intercept variance was negligible, suggesting minimal residual between-subject variability in baseline COHb after accounting for fixed effects. Model fit indices demonstrated strong explanatory performance (AIC and BIC values shown in Table 5), with high marginal and conditional R2 values, indicating that the fixed effects explained the majority of variance in arterial COHb levels. The RMSE reflected the remaining within-subject variability. Discussion This prospective observational study evaluated the agreement between non-invasive SpCO measurements and arterial blood gas–derived COHb levels in patients with acute carbon monoxide poisoning, with particular attention to the effects of concentration range and treatment status. Using the Rad-97 pulse CO-oximeter, we found that agreement between SpCO and arterial COHb was highest before treatment in the moderate COHb range, whereas SpCO tended to overestimate arterial COHb at lower concentrations and demonstrated increasing variability at higher levels. After HBOT, both correlation and concordance declined substantially, indicating that SpCO did not reliably reflect arterial COHb under post-treatment conditions. These findings were consistent across multiple agreement metrics and were reinforced by severity-stratified analyses. Previous studies have reported high sensitivity and specificity of SpCO for detecting carbon monoxide exposure, supporting its role as a rapid, non-invasive screening tool at presentation12. In our cohort, this pattern was most evident in the moderate COHb range, where concordance was highest. However, consistent with prior reports, agreement was limited at lower concentrations and became increasingly unstable at higher COHb levels, with wide limits of agreement and heteroscedasticity observed. Meta-analyses and systematic reviews have similarly shown that although average bias may be modest, the variability between SpCO and arterial COHb is substantial, particularly when COHb levels are low or fluctuating, limiting interchangeability across the full concentration spectrum13,14. A consistent finding in the present study was deterioration in agreement after treatment. Although mean bias decreased and limits of agreement narrowed following HBOT, correlation and concordance declined, indicating that reduced bias did not translate into improved reliability. Similar patterns during follow-up in the non-HBOT group suggest that this behavior reflects measurement under post-treatment or low-concentration conditions rather than an HBOT-specific effect. Both the HBOT and non-HBOT groups demonstrated significant within-group reductions in COHb and related blood gas parameters over time. However, formal comparisons of change magnitudes between the two groups did not reveal statistically significant differences. Accordingly, the observed reductions in COHb are best interpreted as time-dependent physiological resolution rather than evidence of treatment group superiority. The linear mixed-effects analysis supports this interpretation. After adjustment, SpCO remained associated with arterial COHb, but COHb severity and treatment time point exerted stronger effects, with lower post-treatment COHb levels reflecting expected temporal CO elimination rather than treatment-group assignment. Significant associations with O₂Hb and MetHb further indicate that changes in oxygenation and dyshemoglobins influence non-invasive CO estimation, consistent with prior reports describing reduced SpCO accuracy during rapid CO elimination and variable physiological or technical conditions15,16. Accordingly, while SpCO provides useful non-invasive information, post-treatment assessment and HBOT decision-making should remain symptom-driven and should not rely solely on COHb values17. Carbon monoxide exposure time was estimated based on available clinical information and reflects a clinically reasoned estimate derived from available objective and contextual information. In patients with altered mental status, recall and estimation bias are unavoidable, and exposure duration may therefore be imprecise. Importantly, exposure time was used only as a descriptive baseline characteristic and was not included as a predictor in the statistical models; therefore, the primary conclusions of this study are not dependent on the estimation of exposure duration. Mechanistic considerations help explain the deterioration after treatment. Rapid oxygen loading during HBOT alters hemoglobin absorption characteristics across visible and infrared wavelengths, which complicates multi-wavelength separation of COHb and O2Hb and reduces the dynamic range of actual COHb values shortly after therapy. Under stable conditions, Rad-97 performance has been described with a SpCO bias of approximately ± 2% within a range of roughly 0–15% COHb, with accuracy declining outside this range and in states of dyshemoglobinemia18. HBOT also changes peripheral perfusion through systemic vasodilation and temporary redistribution of blood flow. Therefore, pulse oximetry, which depends on stable pulsatile flow, becomes more variable under such hemodynamic shifts19. In our cohort, MetHb increased modestly yet significantly by 0.18% in the HBOT group, while changes were not significant in the non-HBOT group. Even small shifts in MetHb or redox state can interfere with spectral decomposition, as current algorithms do not perfectly differentiate the overlapping signatures of COHb, O2Hb, and MetHb20. These effects align with the reduced agreement observed after therapy. Placed alongside previous research, the focus shifts from device development to context and study design. Earlier accuracy studies often used older platforms and consistently reported wide limits of agreement at low concentrations; more recent work has also employed the Rad-97, frequently in retrospective designs21. The present dataset differs in that it pairs pre- and post-HBOT measurements within the same patients, presents complementary agreement analyses across predefined severity bands, and integrates SpCO with oxygenation indices in the analysis. The practical takeaway is deliberately selective. SpCO can assist with rapid screening and initial monitoring at presentation, particularly in moderate exposures, while arterial sampling is obtained. After HBOT, arterial COHb should remain the basis for clinical decision-making. Although COHb is not routinely rechecked after HBOT, when reassessment is clinically indicated such as in patients with persistent symptoms, an incongruent clinical course, or disposition decisions, non-invasive surrogates are insufficient. Pulse oximetry itself may be inaccurate in the presence of dyshemoglobins and after hyperbaric exposure, and SpCO demonstrated reduced agreement post-HBOT in our cohort. In such situations, arterial blood gas analysis with co-oximetry should be obtained. Conclusion In this prospective observational study, agreement between non-invasive SpCO measurements and arterial COHb levels declined after hyperbaric oxygen therapy. While SpCO provided reasonably consistent estimates at moderate COHb levels before treatment, its reliability decreased following HBOT. These changes are likely related to physiological alterations induced by hyperbaric oxygen therapy, which can affect non-invasive CO measurement accuracy. Clinically, our findings indicate that SpCO should not be used as the sole method for assessing carbon monoxide exposure or guiding clinical decisions after HBOT, and arterial COHb measurement remains essential for post-treatment evaluation. Although future improvements in sensor technology may enhance non-invasive monitoring, caution is warranted when interpreting SpCO values in dynamic treatment settings. Limitations Generalizability of the findings is limited by the single-center design, the relatively small size of the high-COHb subgroup, and the evaluation of only a single pulse CO-oximetry platform.