The association of arterial partial oxygen pressure with mortality in patients with severe acute pancreatitis: a retrospective cohort study Authors: Yiji Chen, Jianhua Wan, Wenqing Shu, Xiaoyu Yang, Huajing Ke, Wenhua He, Yin Zhu, Nonghua Lu, Liang Xia Published: 2025 ABSTRACT BackgroundPatients with severe acute pancreatitis (SAP) frequently develop hypoxic acute respiratory failure (AHRF), with a mortality rate as high as 37%. However, the optimal partial pressure of oxygen (PaO2) for SAP patients remains unclear to date. This study aims to investigate whether partial pressure of oxygen is associated with mortality in SAP patients.MethodsA retrospective cohort study was conducted on patients with severe acute pancreatitis (SAP) admitted to the First Affiliated Hospital of Nanchang University from 2015 to 2024. Propensity score matching (based on whether arterial oxygen partial pressure PaO2 ≥ 80 mmHg during the first 3 days after ICU admission, assigning patients to the liberal PaO2 group or conservative PaO2 group), univariate logistic regression analysis, Cox regression analysis, subgroup analysis, Kaplan–Meier (K–M curve) survival analysis, and sensitivity analysis were employed to thoroughly evaluate the association between PaO2 and mortality in SAP patients. The primary outcome was 28-day mortality.ResultsThe study included 1585 patients. We found that higher PaO2 was associated with lower 28-day mortality rates. In logistic regression analysis after propensity score matching, the incidence rates of adverse outcomes such as persistent circulatory failure (OR 0.50; 95% CI 0.35–0.69; P < 0.001) and persistent multiple organ failure (OR 0.60; 95% CI 0.47–0.78; P < 0.001) significantly decreased. The K–M curve demonstrated significant reductions in 28-day mortality (P = 0.02), 90-day mortality (P = 0.0079), and overall mortality (P = 0.008) in the liberal PaO2 group, with all P values showing statistical significance. Subgroup analysis revealed that the association between higher PaO2 and mortality in SAP patients varied across different age groups, BMI values, SIRS and APACHE II scores, and smoking status. Sensitivity analysis demonstrated stable results after excluding specific populations. On the third day of ICU admission (P = 0.016), higher PaO2 correlated with improved outcomes compared to the conservative group, particularly when PaO2 stabilized around 100 mmHg.ConclusionsEarly maintenance of higher PaO2 (≥80 mmHg) during the initial ICU period was associated with lower mortality.Supplementary InformationThe online version contains supplementary material available at 10.1186/s40635-025-00843-8. Introduction Approximately 20–30% of patients with acute pancreatitis (AP) progress to severe acute pancreatitis (SAP). We typically confirm diagnoses based on the 2012 revised International Atlanta classification Criteria. The prognosis of SAP depends not only on pancreatic injury itself but is also closely associated with persistent organ failure, among which acute hypoxemic respiratory failure (AHRF) represents a key factor contributing to early mortality [1]. The pathological mechanisms primarily involve microcirculatory dysfunction, impaired mitochondrial oxygen utilization, and damage to the alveolar–capillary barrier due to an excessive inflammatory response. Current guidelines recommend that during respiratory support in the intensive care unit (ICU), a partial pressure of PaO2 around 60 mmHg—corresponding to an oxygen saturation above 90%—should be maintained, which appears clinically reasonable [2, 3]. However, in SAP patients complicated by AHRF, this target may be insufficient to meet their high metabolic oxygen demand, potentially leading to serious complications and poor clinical outcomes [4, 5]. Previous studies have shown heterogeneous oxygenation targets across different diseases. For instance, de Jonge et al. reported a linear relationship between FiO2 and ICU mortality, along with a U-shaped association for PaO2 (i.e., both low and high PaO2 values were associated with increased mortality) [6]. Subsequent studies have further explored and refined these targets. Girardis et al. demonstrated that a conservative oxygenation strategy (maintaining PaO2 between 70 and 100 mmHg) was associated with reduced ICU mortality compared with conventional therapy; in contrast, Barrot et al. found that among patients with acute respiratory distress syndrome (ARDS), a conservative oxygenation strategy targeting PaO2 between 55 and 70 mmHg did not improve 28-day survival [7–9]. Although these studies provide a theoretical basis for oxygen management, the optimal PaO2 target specifically for patients with severe acute pancreatitis remains undetermined. Therefore, this study aims to investigate whether higher PaO2 (≥80 mmHg) is associated with lower mortality in patients with SAP. Methods Study design and participants The study was approved by the ethics committee of The First Affiliated Hospital of Nanchang University (No. 2011001). Informed consent by verbal was obtained from all participants approved by the ethics committee. This study adopted a retrospective cohort study design. The research data were derived from patients with severe acute pancreatitis (n = 2625) admitted to the Pancreatic ICU Ward of the Department of Gastroenterology, the First Affiliated Hospital of Nanchang University, from 2015 to 2024. The exclusion criteria were as follows: (1) excluding patients under 18 years (n = 33); (2) excluding pregnant patients (n = 69); (3) excluding patients with end-stage liver and kidney diseases (n = 11); (4) excluding patients admitted to the ICU for less than 3 days (n = 112); and (5) excluding patients without PaO2 data for 3 consecutive days (n = 815). Finally, 1585 SAP patients were included. This manuscript was prepared according to the STROBE guidelines [10]. Data collection and oxygen partial pressure grouping Data were extracted from the electronic medical record system. Demographic variables included age, sex, and body mass index (BMI). Disease-related information encompassed smoking status, alcohol use, and medical history, including hypertension, diabetes, and chronic obstructive pulmonary disease (COPD). Clinical signs and symptoms included body temperature, heart rate, respiratory rate, and mean arterial pressure (MAP). Disease severity was assessed using the Systemic Inflammatory Response Syndrome (SIRS) criteria and Acute Physiology and Chronic Health Evaluation II (APACHE II) score. Laboratory parameters consisted of white blood cell count (WBC), hematocrit (HCT), albumin (ALB), triglycerides (TG), blood urea nitrogen (BUN), creatinine (Cr), serum calcium (Ca), pH, Inspired Oxygen concentration (FiO2). Complications recorded included acute peripancreatic fluid collection (APFC), pancreatic pseudocyst, walled-off necrosis (WON), acute necrotic collection (ANC), infected pancreatic necrosis (IPN), other local complications of pancreatitis, persistent renal failure, respiratory failure, and multiple organ failure (MOF). Outcomes assessed included 28-day and 90-day mortality, overall mortality, hospital length of stay (LOS), ICU length of stay, and use of renal replacement therapy (CRRT), mechanical ventilation, and percutaneous catheter drainage. Definition of oxygen partial pressure groups Patients were classified based on their lowest recorded PaO2 values during the first 3 days following admission to the pancreatic ICU. If multiple arterial blood gas measurements were obtained on the same day, the lowest PaO2 value was selected. Patients who maintained PaO2 ≥ 80 mmHg on all 3 days were assigned to the liberal PaO2 group; all others were classified into the conservative PaO2 group [11]. The first ICU day was defined as the period from ICU admission until the first midnight. The second and third ICU days were defined as the subsequent 24-h intervals beginning at midnight [12]. The detection of arterial partial pressure of oxygen originates from blood gas analysis, and the Pancreatic ICU uses the GEM Premier 5000 Werfen blood gas analyzer. The manufacturer is located in the United States. Statistical analysis Continuous variables are presented as mean ± standard deviation (SD) if normally distributed, or as median with interquartile range (IQR) if skewed. For missing values in the data, we first conducted a randomization test for missing data, and then excluded data sets with more than 20% missing values. Meanwhile, we used the K-nearest neighbor imputation algorithm (Knn) and multiple imputation methods to impute the data. When the baseline data of patients were unbalanced, we matched and balanced the baseline data of the two groups of patients through a propensity score matching (PSM) cohort, and calculated the absolute standardized mean difference (SMD) corresponding to each variable to check the degree of PSM and evaluate the imbalance between the two groups before and after matching. We also used inverse probability of treatment weighting (IPTW), standardized mortality ratio weighting (SMRW), pairing algorithm (PA), and overlap weighting (OW) for weighting. Finally, when the SMD value is ≤0.1, it indicates that all variables between the two groups are in a balanced state. After balancing baseline characteristics, univariable logistic regression was performed to compare clinical outcomes—including persistent circulatory failure, multiple organ dysfunction, infected pancreatic necrosis, persistent renal failure, and persistent respiratory failure—as well as the use of mechanical ventilation and continuous renal replacement therapy (CRRT). Odds ratios (ORs) with 95% confidence intervals (CIs) were calculated to evaluate whether the incidence of organ failure was reduced. In addition, after further processing with Bonferroni correction, all P values are statistically significant. The proportional hazards (PH) assumption was tested and satisfied (all P > 0.05). Cox proportional hazards regression was used to analyze mortality outcomes, yielding hazard ratios (HRs) with 95% CIs to assess the association between higher PaO2 levels and mortality. Subgroup analyses were conducted based on smoking history, age, BMI, SIRS, and APACHE II score to evaluate which groups of people benefit. Kaplan–Meier survival curves were generated to compare 28-day, 90-day, and overall mortality between groups. Sensitivity analyses were performed to test the robustness of the findings under different assumptions, including exclusion of patients with recurrent acute pancreatitis, those with chronic obstructive pulmonary disease (COPD), and inclusion only of patients admitted within 3 days of symptom onset. All statistical analyses were conducted using R software, version 4.2.2. Results Data collection and baseline characteristics In this study, a total of 2625 SAP patients from the First Affiliated Hospital of Nanchang University were selected. After applying the exclusion criteria, 1585 patients were screened and included in the final study (Fig. 1). Meanwhile, after propensity score matching, they were divided into the conservative PaO2 group (1069 cases) and the liberal PaO2 group (516 cases). Table S1 shows a comparison of baseline data between the conservative PaO2 group and the liberal PaO2 group. In terms of age, the median age in the conservative group was 51.0 (39.0–64.0), while that in the liberal group was 46.0 (35.0–63.0). The number of females (392 vs. 181), as well as the numbers of people with a smoking history (345 vs. 177) in the conservative group, were all higher than those in the liberal group. There was no significant difference in BMI between the two groups [24.5 (22.6–27.1) vs. 24.6 (22.7–27.7)]. Except for age, TG, Ca, and pH, the other baseline data were similar. After propensity score matching and IPTW method (Table 1), the baseline data were more balanced than before matching, and the SMD value was still ≤0.1.Fig. 1Flowchart of the screening and enrollment of eligible subjects. SAP, severe acute pancreatitis; ICU, intensive care unitTable 1Baseline characteristics of patients in conservative and liberal PaO2 groups after the IPTW matchingCharacteristicUnmatchedAfter IPTW matchingConservative PaO2 (n = 1069)Liberal PaO2 (n = 516)SMDConservative PaO2 (n = 1069)Liberal PaO2 (n = 516)SMDN10695161069516Sex (male), n (%)677 (63.3)335 (64.9)0.033683.1 (63.8)327.7 (63.7)0.003<0.1Age (years)51.71 (16.05)49.39 (17.82)0.13751.00 (16.05)51.20 (18.05)0.012<0.1BMI (kg/m2)25.11 (3.82)25.23 (4.09)0.03125.14 (3.87)25.10 (4.11)0.011<0.1Smoking, n (%)345 (32.3)177 (34.3)0.043354.3 (33.1)172.0 (33.4)0.007<0.1Drinking, n (%)381 (35.6)197 (38.2)0.053390.7 (36.5)190.0 (36.9)0.009<0.1Hypertension, n (%)293 (27.4)129 (25.0)0.055285.0 (26.6)139.3 (27.1)0.01<0.1COPD, n (%)6 (0.6)5 (1.0)0.0477.6 (0.7)3.6 (0.7)<0.001<0.1Diabetes, n (%)161 (15.1)94 (18.2)0.085171.9 (16.1)82.9 (16.1)0.001<0.1Temperature (℃)37.26 (0.81)37.20 (0.71)0.08437.23 (0.81)37.22 (0.73)0.014<0.1Pulse (bpm)106.83 (21.58)107.94 (22.03)0.051107.30 (21.76)107.41 (22.23)0.005<0.1Respirations (bpm)26.83 (7.72)26.28 (7.68)0.07126.65 (7.67)26.60 (7.67)0.006<0.1Mean arterial pressure (mmHg)99.70 (16.19)101.17 (16.99)0.089100.21 (16.21)100.13 (16.97)0.005<0.1WBC (×109/L)14.43 (6.25)14.21 (6.28)0.03414.34 (6.21)14.26 (6.29)0.013<0.1HCT (%)39.94 (9.25)40.24 (10.07)0.0340.03 (9.41)40.02 (9.79)0.001<0.1ALB (g/L)34.18 (5.94)34.82 (6.43)0.10234.40 (6.14)34.42 (6.24)0.003<0.1TG (mmol/L)6.58 (9.38)7.73 (9.91)0.1197.00 (10.00)7.00 (9.34)0.001<0.1BUN (mmol/L)8.40 (6.53)8.05 (6.38)0.0548.29 (6.41)8.30 (6.51)0.001<0.1Cr (mmol/L)123.85 (121.43)118.75 (103.75)0.045121.99 (117.96)120.94 (104.27)0.009<0.1Ca (mmol/L)1.92 (0.33)1.96 (0.34)0.1171.93 (0.32)1.93 (0.35)0.003<0.1PH7.40 (0.08)7.38 (0.09)0.2157.39 (0.08)7.39 (0.08)0.008<0.1SIRS2.32 (0.91)2.28 (0.91)0.0422.31 (0.91)2.30 (0.92)0.011<0.1APACHE II12.53 (4.94)12.92 (5.20)0.07712.69 (5.06)12.72 (5.15)0.005<0.1Data are presented as mean ± standard deviation (continuous variables) or number (percentage) (categorical variables); patients in the two groups were matched using inverse probability of treatment weighting (IPTW); a standardized mean difference (SMD) < 0.1 indicates good balance between groupsBMI body mass index, COPD chronic obstructive pulmonary disease, Bpm beats per minute, MAP mean arterial pressure, WBC white blood cell count, HCT hematocrit, ALB albumin, TG triglycerides, BUN blood urea nitrogen, Cr creatinine, Ca calcium, SIRS Systemic Inflammatory Response Syndrome, APACHEII Acute Physiology and Chronic Health Evaluation II, IPTW the inverse probability of treatment weighting Flowchart of the screening and enrollment of eligible subjects. SAP, severe acute pancreatitis; ICU, intensive care unit Table 1Baseline characteristics of patients in conservative and liberal PaO2 groups after the IPTW matchingCharacteristicUnmatchedAfter IPTW matchingConservative PaO2 (n = 1069)Liberal PaO2 (n = 516)SMDConservative PaO2 (n = 1069)Liberal PaO2 (n = 516)SMDN10695161069516Sex (male), n (%)677 (63.3)335 (64.9)0.033683.1 (63.8)327.7 (63.7)0.003<0.1Age (years)51.71 (16.05)49.39 (17.82)0.13751.00 (16.05)51.20 (18.05)0.012<0.1BMI (kg/m2)25.11 (3.82)25.23 (4.09)0.03125.14 (3.87)25.10 (4.11)0.011<0.1Smoking, n (%)345 (32.3)177 (34.3)0.043354.3 (33.1)172.0 (33.4)0.007<0.1Drinking, n (%)381 (35.6)197 (38.2)0.053390.7 (36.5)190.0 (36.9)0.009<0.1Hypertension, n (%)293 (27.4)129 (25.0)0.055285.0 (26.6)139.3 (27.1)0.01<0.1COPD, n (%)6 (0.6)5 (1.0)0.0477.6 (0.7)3.6 (0.7)<0.001<0.1Diabetes, n (%)161 (15.1)94 (18.2)0.085171.9 (16.1)82.9 (16.1)0.001<0.1Temperature (℃)37.26 (0.81)37.20 (0.71)0.08437.23 (0.81)37.22 (0.73)0.014<0.1Pulse (bpm)106.83 (21.58)107.94 (22.03)0.051107.30 (21.76)107.41 (22.23)0.005<0.1Respirations (bpm)26.83 (7.72)26.28 (7.68)0.07126.65 (7.67)26.60 (7.67)0.006<0.1Mean arterial pressure (mmHg)99.70 (16.19)101.17 (16.99)0.089100.21 (16.21)100.13 (16.97)0.005<0.1WBC (×109/L)14.43 (6.25)14.21 (6.28)0.03414.34 (6.21)14.26 (6.29)0.013<0.1HCT (%)39.94 (9.25)40.24 (10.07)0.0340.03 (9.41)40.02 (9.79)0.001<0.1ALB (g/L)34.18 (5.94)34.82 (6.43)0.10234.40 (6.14)34.42 (6.24)0.003<0.1TG (mmol/L)6.58 (9.38)7.73 (9.91)0.1197.00 (10.00)7.00 (9.34)0.001<0.1BUN (mmol/L)8.40 (6.53)8.05 (6.38)0.0548.29 (6.41)8.30 (6.51)0.001<0.1Cr (mmol/L)123.85 (121.43)118.75 (103.75)0.045121.99 (117.96)120.94 (104.27)0.009<0.1Ca (mmol/L)1.92 (0.33)1.96 (0.34)0.1171.93 (0.32)1.93 (0.35)0.003<0.1PH7.40 (0.08)7.38 (0.09)0.2157.39 (0.08)7.39 (0.08)0.008<0.1SIRS2.32 (0.91)2.28 (0.91)0.0422.31 (0.91)2.30 (0.92)0.011<0.1APACHE II12.53 (4.94)12.92 (5.20)0.07712.69 (5.06)12.72 (5.15)0.005<0.1Data are presented as mean ± standard deviation (continuous variables) or number (percentage) (categorical variables); patients in the two groups were matched using inverse probability of treatment weighting (IPTW); a standardized mean difference (SMD) < 0.1 indicates good balance between groupsBMI body mass index, COPD chronic obstructive pulmonary disease, Bpm beats per minute, MAP mean arterial pressure, WBC white blood cell count, HCT hematocrit, ALB albumin, TG triglycerides, BUN blood urea nitrogen, Cr creatinine, Ca calcium, SIRS Systemic Inflammatory Response Syndrome, APACHEII Acute Physiology and Chronic Health Evaluation II, IPTW the inverse probability of treatment weighting Baseline characteristics of patients in conservative and liberal PaO2 groups after the IPTW matching Data are presented as mean ± standard deviation (continuous variables) or number (percentage) (categorical variables); patients in the two groups were matched using inverse probability of treatment weighting (IPTW); a standardized mean difference (SMD) < 0.1 indicates good balance between groups BMI body mass index, COPD chronic obstructive pulmonary disease, Bpm beats per minute, MAP mean arterial pressure, WBC white blood cell count, HCT hematocrit, ALB albumin, TG triglycerides, BUN blood urea nitrogen, Cr creatinine, Ca calcium, SIRS Systemic Inflammatory Response Syndrome, APACHEII Acute Physiology and Chronic Health Evaluation II, IPTW the inverse probability of treatment weighting Association between oxygen partial pressure and clinical outcomes On ICU day 1 (Table S2), the mean PaO2 was 119.3 ± 34.6 mmHg in the liberal group and 83.5 ± 28.5 mmHg in the conservative group. Throughout the first 3 days of ICU stay, median PaO2 values were significantly higher in the liberal group than in the conservative group (all P < 0.001). In terms of clinical outcomes, after conducting a univariate logistic regression analysis (Fig. 2), higher PaO2 in the liberal group was associated with lower mortality in SAP patients. Specifically, the liberal group showed significantly lower risks of persistent circulatory failure (OR 0.50; 95% CI 0.35–0.69; P < 0.001), persistent renal failure (OR 0.69; 95% CI 0.53–0.91; P = 0.007), infected pancreatic necrosis (OR 0.53; 95% CI 0.39–0.71; P < 0.001), and persistent respiratory failure (OR 0.25; 95% CI 0.20–0.32; P < 0.001). Use of mechanical ventilation (OR 0.36; 95% CI 0.27–0.46; P < 0.001) and continuous renal replacement therapy (OR 0.50; 95% CI 0.37–0.67; P < 0.001) was also significantly reduced in the liberal group. Additional outcomes are summarized in Table 2. Compared with the conservative PaO2 group, the liberal PaO2 group had significantly shorter hospital and ICU lengths of stay (all P < 0.05). After inverse probability of treatment weighting (IPTW), the liberal group also exhibited significantly lower 28-day mortality (38 deaths [7.4%] vs. 130.3 deaths [12.2%]), 90-day mortality (52 deaths [10.1%] vs. 186.6 deaths [17.4%]), and overall mortality (53 deaths [10.3%] vs. 191.2 deaths [17.9%]; all P < 0.01), indicating better survival outcomes associated with higher PaO2.Fig. 2Logistic regression analysis. Association between liberal vs. conservative oxygen therapy and clinical outcomes in patients. This forest plot illustrates the results of univariable logistic regression analyses evaluating the relationship between liberal (liberal PaO2) and conservative (conservative PaO2) oxygen therapy strategies and the incidence of key clinical endpoints in patients with acute pancreatitis. Outcomes examined included mechanical ventilation, continuous renal replacement therapy (CRRT), persistent circulatory failure, persistent multiple organ failure (MOF), persistent renal failure, infected pancreatic necrosis, and persistent respiratory failureTable 2Comparison of clinical outcomes between the two groups after IPTWCharacteristicConservative PaO2Liberal PaO2P value1069.4516.0APFC, n (%)291.5 (27.3)128.0 (24.8)0.310PPC, n (%)9.0 (0.8)8.0 (1.6)0.195WON, n (%)361.5 (33.8)145.0 (28.1)0.025ANC, n (%)661.8 (61.9)319.0 (61.8)0.981Persistent respiratory failure, n (%)900.9 (84.2)301.0 (58.3)<0.001Infected pancreatic necrosis, n (%)211.4 (19.8)59.0 (11.4)<0.001Persistent renal failure, n (%)264.0 (24.7)94.0 (18.2)0.005Persistent organ failure, n (%)922.9 (86.3)318.0 (61.6)<0.001Confirmed organ failure, n (%)1037.9 (97.1)423.0 (82.0)<0.001Confirmed circulatory failure, n (%)194.7 (18.2)52.0 (10.1)<0.001Persistent MOF, n (%)315.6 (29.5)102.0 (19.8)<0.001Persistent circulatory failure, n (%)181.8 (17.0)47.0 (9.1)<0.001Total length of hospital stay26.21 (24.35)22.89 (21.28)0.007Length of ICU stay14.40 (15.43)11.91 (12.88)0.00128-day mortality, n (%)130.3 (12.2)38.0 (7.4)0.00490-day mortality, n (%)186.6 (17.4)52.0 (10.1)<0.001Mortality, n (%)191.2 (17.9)53.0 (10.3)<0.001PCD, n (%)245.5 (23.0)49.0 (9.5)<0.001CRRT, n (%)249.0 (23.3)65.0 (12.6)<0.001Mechanical ventilation, n (%)384.3 (35.9)83.0 (16.1)<0.001APFC acute peripancreatic fluid collection, PPC pancreatic pseudocyst, WON walled-off necrosis, ANC acute necrotic collection, MOF multiple organ failure, PCD percutaneous catheter drainage, CRRT continuous renal replacement therapy Logistic regression analysis. Association between liberal vs. conservative oxygen therapy and clinical outcomes in patients. This forest plot illustrates the results of univariable logistic regression analyses evaluating the relationship between liberal (liberal PaO2) and conservative (conservative PaO2) oxygen therapy strategies and the incidence of key clinical endpoints in patients with acute pancreatitis. Outcomes examined included mechanical ventilation, continuous renal replacement therapy (CRRT), persistent circulatory failure, persistent multiple organ failure (MOF), persistent renal failure, infected pancreatic necrosis, and persistent respiratory failure Table 2Comparison of clinical outcomes between the two groups after IPTWCharacteristicConservative PaO2Liberal PaO2P value1069.4516.0APFC, n (%)291.5 (27.3)128.0 (24.8)0.310PPC, n (%)9.0 (0.8)8.0 (1.6)0.195WON, n (%)361.5 (33.8)145.0 (28.1)0.025ANC, n (%)661.8 (61.9)319.0 (61.8)0.981Persistent respiratory failure, n (%)900.9 (84.2)301.0 (58.3)<0.001Infected pancreatic necrosis, n (%)211.4 (19.8)59.0 (11.4)<0.001Persistent renal failure, n (%)264.0 (24.7)94.0 (18.2)0.005Persistent organ failure, n (%)922.9 (86.3)318.0 (61.6)<0.001Confirmed organ failure, n (%)1037.9 (97.1)423.0 (82.0)<0.001Confirmed circulatory failure, n (%)194.7 (18.2)52.0 (10.1)<0.001Persistent MOF, n (%)315.6 (29.5)102.0 (19.8)<0.001Persistent circulatory failure, n (%)181.8 (17.0)47.0 (9.1)<0.001Total length of hospital stay26.21 (24.35)22.89 (21.28)0.007Length of ICU stay14.40 (15.43)11.91 (12.88)0.00128-day mortality, n (%)130.3 (12.2)38.0 (7.4)0.00490-day mortality, n (%)186.6 (17.4)52.0 (10.1)<0.001Mortality, n (%)191.2 (17.9)53.0 (10.3)<0.001PCD, n (%)245.5 (23.0)49.0 (9.5)<0.001CRRT, n (%)249.0 (23.3)65.0 (12.6)<0.001Mechanical ventilation, n (%)384.3 (35.9)83.0 (16.1)<0.001APFC acute peripancreatic fluid collection, PPC pancreatic pseudocyst, WON walled-off necrosis, ANC acute necrotic collection, MOF multiple organ failure, PCD percutaneous catheter drainage, CRRT continuous renal replacement therapy Comparison of clinical outcomes between the two groups after IPTW APFC acute peripancreatic fluid collection, PPC pancreatic pseudocyst, WON walled-off necrosis, ANC acute necrotic collection, MOF multiple organ failure, PCD percutaneous catheter drainage, CRRT continuous renal replacement therapy Kaplan–Meier survival and cox regression analyses Kaplan–Meier survival curves demonstrated significantly higher survival probabilities in the liberal PaO2 group compared with the conservative group (Figs. 3, 4, 5). Cox proportional hazards regression analysis confirmed that higher PaO2 levels were associated with reduced mortality (Table 3), with hazard ratios of 0.67 (95% CI 0.47–0.96; P = 0.034) at 28 days, 0.70 (95% CI 0.52–0.94; P = 0.024) at 90 days, and 0.70 (95% CI 0.52–0.95; P = 0.025) for overall mortality, corresponding to risk reductions of 33%, 30%, and 30%, respectively. No significant association between PaO2 and clinical outcomes was observed during the first or second day of ICU stay (Figs. 6, 7, 8). However, by the third ICU day, higher PaO2 showed a significant inverse linear relationship with mortality, with the association appearing to stabilize around a PaO2 level of 100 mmHg.Fig. 3Kaplan–Meier estimates of 28-day survival in patients with severe acute pancreatitisFig. 4Kaplan–Meier estimates of 90-day survival in patients with severe acute pancreatitisFig. 5Kaplan–Meier estimates of overall survival in patients with severe acute pancreatitisTable 3Cox analysis of death outcomesSubgroupn. totaln. event_%HR (95% CI)P value28-day mortality0.0341069131 (12.3)1 (Ref)51638 (7.4)0.67 (0.47–0.96)90-day mortality0.0241069188 (17.6)1 (Ref)51652 (10.1)0.7 (0.52–0.94)Total mortality0.0251069192 (18)1 (Ref)51653 (10.3)0.7 (0.52–0.95)Fig. 6Restricted cubic spline-derived dose–response relationship between day 1 PaO2 levels and all-cause mortalityFig. 7Restricted cubic spline-derived dose–response relationship between day 2 PaO2 levels and all-cause mortalityFig. 8Restricted cubic spline-derived dose–response relationship between day 3 PaO2 levels and all-cause mortality Kaplan–Meier estimates of 28-day survival in patients with severe acute pancreatitis Kaplan–Meier estimates of 90-day survival in patients with severe acute pancreatitis Kaplan–Meier estimates of overall survival in patients with severe acute pancreatitis Table 3Cox analysis of death outcomesSubgroupn. totaln. event_%HR (95% CI)P value28-day mortality0.0341069131 (12.3)1 (Ref)51638 (7.4)0.67 (0.47–0.96)90-day mortality0.0241069188 (17.6)1 (Ref)51652 (10.1)0.7 (0.52–0.94)Total mortality0.0251069192 (18)1 (Ref)51653 (10.3)0.7 (0.52–0.95) Cox analysis of death outcomes Restricted cubic spline-derived dose–response relationship between day 1 PaO2 levels and all-cause mortality Restricted cubic spline-derived dose–response relationship between day 2 PaO2 levels and all-cause mortality Restricted cubic spline-derived dose–response relationship between day 3 PaO2 levels and all-cause mortality Subgroup and sensitivity analyses In the subgroup analysis (Table 4), the analysis was conducted based on patients’ smoking history, age, BMI, SIRS score, and APACHE II score. There were significant differences in treatment effects among different clinically relevant subgroups in the liberal group, highlighting the key interaction between oxygen therapy effects and patient-specific pathophysiology. The results showed that patients without a smoking history (HR 0.66, 95% CI 0.45–0.95), aged < 60 years (HR 0.54, 95% CI 0.35–0.83), with BMI < 24 (HR 0.56, 95% CI 0.33–0.92), SIRS score ≥ 2 points (HR 0.69, 95% CI 0.50–0.95), and APACHE II ≥ 8 points (HR 0.65, 95% CI 0.48–0.89) were associated with a lower 28-day mortality risk. Table 4Subgroup analysisSubgroupVariablen. total28-day mortality (%)HR (95% CI)P valueSmoking statusNoConservative PaO2724134 (18.5)1 (Ref)Liberal PaO233933 (9.7)0.66 (0.45–0.95)0.033YesConservative PaO234558 (16.8)1 (Ref)Liberal PaO217720 (11.3)0.8 (0.48–1.31)0.381Age<60Conservative PaO2720118 (16.4)1 (Ref)Liberal PaO237125 (6.7)0.54 (0.35–0.83)0.006≥60Conservative PaO234974 (21.2)1 (Ref)Liberal PaO214528 (19.3)0.85 (0.56–1.31)0.478BMI<24Conservative PaO241377 (18.6)1 (Ref)Liberal PaO219717 (8.6)0.56 (0.33–0.92)0.029≥24Conservative PaO2656115 (17.5)1 (Ref)Liberal PaO231936 (11.3)0.8 (0.55–1.16)0.255SIRS<2Conservative PaO218324 (13.1)1 (Ref)Liberal PaO2986 (6.1)0.8 (0.35–1.85)0.631≥2Conservative PaO2886168 (19)1 (Ref)Liberal PaO241847 (11.2)0.69 (0.5–0.95)0.025APACHEII<8Conservative PaO215812 (7.6)1 (Ref)Liberal PaO2705 (7.1)1.41 (0.53–3.7)0.526≥8Conservative PaO2911180 (19.8)1 (Ref)Liberal PaO244648 (10.8)0.65 (0.48–0.89)0.009 Table 4Subgroup analysisSubgroupVariablen. total28-day mortality (%)HR (95% CI)P valueSmoking statusNoConservative PaO2724134 (18.5)1 (Ref)Liberal PaO233933 (9.7)0.66 (0.45–0.95)0.033YesConservative PaO234558 (16.8)1 (Ref)Liberal PaO217720 (11.3)0.8 (0.48–1.31)0.381Age<60Conservative PaO2720118 (16.4)1 (Ref)Liberal PaO237125 (6.7)0.54 (0.35–0.83)0.006≥60Conservative PaO234974 (21.2)1 (Ref)Liberal PaO214528 (19.3)0.85 (0.56–1.31)0.478BMI<24Conservative PaO241377 (18.6)1 (Ref)Liberal PaO219717 (8.6)0.56 (0.33–0.92)0.029≥24Conservative PaO2656115 (17.5)1 (Ref)Liberal PaO231936 (11.3)0.8 (0.55–1.16)0.255SIRS<2Conservative PaO218324 (13.1)1 (Ref)Liberal PaO2986 (6.1)0.8 (0.35–1.85)0.631≥2Conservative PaO2886168 (19)1 (Ref)Liberal PaO241847 (11.2)0.69 (0.5–0.95)0.025APACHEII<8Conservative PaO215812 (7.6)1 (Ref)Liberal PaO2705 (7.1)1.41 (0.53–3.7)0.526≥8Conservative PaO2911180 (19.8)1 (Ref)Liberal PaO244648 (10.8)0.65 (0.48–0.89)0.009 Subgroup analysis Sensitivity analysis (Table 5), different hypotheses were established by methods such as excluding specific populations to explore the stability of the results. It was found that the research results remained consistent and stable after establishing hypotheses by excluding patients with recurrent acute pancreatitis (HR 0.66, 95% CI 0.48–0.92, P = 0.019), excluding people with chronic obstructive pulmonary disease (COPD) (HR 0.64, 95% CI 0.44–0.94, P = 0.024), including patients admitted within 3 days of onset (HR 0.71, 95% CI 0.53–0.96, P = 0.031). In addition, sensitivity analysis using different weighting methods was performed on the data after propensity score matching, and after adjusting for the covariate FiO2 with multiple factors (Table S5). The results also showed stability. Table 5Sensitivity analysisSubgroupn. total28-day mortality (%)HR (95% CI)P valueSensitivity analysis 10.0341069131 (12.3)1 (Ref)51638 (7.4)0.67 (0.47–0.96)Sensitivity analysis 20.019912167 (18.3)1 (Ref)43342 (9.7)0.66 (0.48–0.92)Sensitivity analysis 30.024679118 (17.4)1 (Ref)34034 (10)0.64 (0.44–0.94)Sensitivity analysis 40.0311063191 (18)1 (Ref)51153 (10.4)0.71 (0.53–0.96)Sensitivity analysis 1: All patientsSensitivity analysis 2: Excluding patients with recurrent acute pancreatitisSensitivity analysis 3: Excluding people with chronic obstructive pulmonary disease (COPD)Sensitivity analysis 4: Including patients admitted to hospital within 3 days of onset Table 5Sensitivity analysisSubgroupn. total28-day mortality (%)HR (95% CI)P valueSensitivity analysis 10.0341069131 (12.3)1 (Ref)51638 (7.4)0.67 (0.47–0.96)Sensitivity analysis 20.019912167 (18.3)1 (Ref)43342 (9.7)0.66 (0.48–0.92)Sensitivity analysis 30.024679118 (17.4)1 (Ref)34034 (10)0.64 (0.44–0.94)Sensitivity analysis 40.0311063191 (18)1 (Ref)51153 (10.4)0.71 (0.53–0.96)Sensitivity analysis 1: All patientsSensitivity analysis 2: Excluding patients with recurrent acute pancreatitisSensitivity analysis 3: Excluding people with chronic obstructive pulmonary disease (COPD)Sensitivity analysis 4: Including patients admitted to hospital within 3 days of onset Sensitivity analysis Sensitivity analysis 1: All patients Sensitivity analysis 2: Excluding patients with recurrent acute pancreatitis Sensitivity analysis 3: Excluding people with chronic obstructive pulmonary disease (COPD) Sensitivity analysis 4: Including patients admitted to hospital within 3 days of onset Discussion This retrospective cohort shows that higher PaO2 is associated with lower mortality in patients with SAP. Specifically, maintaining a higher PaO2 (≥80 mmHg) in the first 3 days after admission to the ICU is closely related to a reduced 28-day mortality rate and improved clinical outcomes. Subgroup analysis further revealed that patients aged < 60 years, with BMI < 24, SIRS score ≥ 2, APACHE II ≥ 8, and no smoking history had particularly significant survival benefits; sensitivity analysis confirmed the stability of the results. Our research findings differ from the conventional concept of conservative oxygen therapy in SAP patients and clarify the unique pathophysiological characteristics of SAP. Therefore, this study indicates that higher PaO2 is associated with a lower incidence of organ failure (especially AHRF) and lower mortality. The optimal oxygen targets in critically ill patients remain controversial, with considerable variation across studies. For instance, a post-hoc analysis of the ALBIOS trial, which defined hyperoxia as PaO2 > 100 mmHg and normoxia as PaO2 ≤ 100 mmHg, found no significant difference in 28-day mortality or long-term survival among patients with sepsis [13]. In contrast, another study reported significantly lower 28-day mortality and 90-day mortality among patients with hyperoxia (PaO2 > 80 mmHg) during the first 3 days of ICU stay [11]. Although hypoxemia has traditionally been defined as PaO2 ≤ 60 mmHg, and lung-protective ventilation strategies often target PaO2 between 60 and 80 mmHg [14]. Although this target appears reasonable for patients with preserved oxygen delivery and utilization in tissues [3]. However, recent studies suggest that higher PaO2 may improve survival in conditions, such as sepsis, traumatic brain injury, and acute respiratory distress syndrome (ARDS), with ideal oxygenation targets varying by disease [11, 15–17]. This non-random therapeutic effect that occurs due to differences in treatment subjects is called treatment heterogeneity, the need to tailor oxygen therapy based on patient-specific and disease-specific characteristics [18, 19]. Based on the above theory, our research results are also consistent. The survival benefit observed with higher PaO2 in SAP may be attributed to mitigation of early inflammatory cascades, microcirculatory dysfunction, and mitochondrial impairment—hallmarks of SAP pathophysiology [20]. The initial pancreatic injury triggers the release of proteases (e.g., trypsin and elastase) and lipases (e.g., phospholipase A2), which damage pulmonary endothelium, increase alveolar–capillary permeability, and promote intrapulmonary shunting, ultimately impairing oxygen diffusion [21–24]. SIRS often presents early in SAP, disrupts cellular oxygen utilization, and contributes to AHRF. Animal studies have implicated calcium overload and oxidative stress in amplifying inflammation [25]. Hypoxia stimulates reactive oxygen species (ROS) production and calcium release, activating inflammatory pathways (e.g., AP-1, STAT3, and MAPK) and increasing cytokines, such as IL-6, TNF-α, and IL-1 [26–28], thereby exacerbating airway inflammation and lung ischemia–reperfusion injury. Concurrently, mitochondrial apoptosis pathways are activated through cytochrome c release and opening of the mitochondrial permeability transition pore (MPTP), leading to loss of membrane potential, uncoupled respiration, and impaired ATP synthesis [29–32]. Further pancreatic necrosis releases damage-associated molecular patterns (DAMPs), which activate immune cells and amplify the inflammatory cascade [33, 34]. Necrotic acinar cells also promote procoagulant activity, contributing to microthrombosis and microcirculatory failure, characterized by reduced functional capillary density, impaired tissue perfusion, and diminished oxygen delivery [35, 36]. Consistent with these mechanisms, our subgroup analyses showed that patients with heightened systemic inflammation (SIRS ≥ 2, APACHE II ≥ 8) benefited most from higher PaO2. Improved oxygenation may help overcome diffusion barriers, restore mitochondrial function, and mitigate pancreatic necrosis and microvascular dysfunction—key factors in preventing AHRF [37]. The protective association persisted across multiple sensitivity analyses, supporting the robustness of our findings. Moreover, the most pronounced benefit was observed on ICU day 3 at a PaO2 level around 100 mmHg, suggesting a critical time window for oxygen therapy. We hypothesize that the initial 2 days represent a phase of escalating inflammation, where compensatory mechanisms remain active, whereas day 3 marks the peak of systemic inflammation and oxidative stress, during which aggressive oxygenation was most strongly associated with reduced organ failure and mortality. Whether maintaining high PaO2 beyond this hyperinflammatory phase remains beneficial warrants further investigation. Limitations Several limitations should be acknowledged. First, despite statistical adjustments and sensitivity analyses, due to the observational nature of the study and the lack of relevant data, such as invasive/non-invasive mechanical ventilation status, duration of ventilation, positive end-expiratory pressure (PEEP), there may be residual confounding factors. Future prospective studies are needed for verification. Second, this was a single-center study, and differences in ICU practices may limit the generalizability of the findings. Third, although traditional subgroup analyses were conducted, more advanced methods such as machine learning may better account for effect heterogeneity and individual treatment responses. Fourth, this study only obtained data from the first 3 days of patients’ ICU admission and did not control for subsequent interventions and outcomes, which limits the conclusions regarding long-term management. Fifth, this study only verified whether higher arterial partial pressure of oxygen is associated with a lower mortality rate in patients with severe acute pancreatitis and failed to elaborate on the mechanism underlying the higher PaO2 in the liberal oxygen therapy group. Finally, observational studies may introduce time-dependent confounding factors, and it cannot be ruled out that unrecognized deterioration of circulatory or respiratory function during this period led to a decrease in PaO2, thereby resulting in patients being assigned to the conservative oxygen therapy group. Although we adjusted for baseline vital signs, these adjustments may not fully reflect the dynamic physiological deterioration that occurred before grouping. This requires more in-depth research to confirm. Conclusion This study suggests that maintaining a higher PaO2 (≥80 mmHg) during the first 3 days of ICU admission was associated with lower mortality in patients with severe acute pancreatitis. Supplementary Information Additional file 1.Additional file 2.Additional file 3.Additional file 4.Additional file 5. Additional file 1. Additional file 2. Additional file 3. Additional file 4. Additional file 5.