Association between early post-insertion arterial oxygen tension and documented acute upper-body deep or central venous thrombosis in critically ill adults: A retrospective MIMIC-IV cohort study

Qing Cui; Anshuai Fang; Haiyan Zhang; Qingtong Meng; Sijia Zhou; Yueming Peng; Weixiang Luo

doi:10.1371/journal.pone.0359719

. 2026 Oct 5;21(10):e0359719. doi: 10.1371/journal.pone.0359719

Association between early post-insertion arterial oxygen tension and documented acute upper-body deep or central venous thrombosis in critically ill adults: A retrospective MIMIC-IV cohort study

Qing Cui ^1,^#, Anshuai Fang ^1,^#, Haiyan Zhang ^2,^‡, Qingtong Meng ^3,^‡, Sijia Zhou ^2,^‡, Yueming Peng ^4,^*, Weixiang Luo ^4,^*

Editor: Giovanni Giordano⁵

PMCID: PMC13637912 PMID: 42832483

Abstract

Abnormal oxygenation may influence thrombosis, but evidence in catheterized critically ill patients remains limited. We examined whether the first arterial partial pressure of oxygen (PaO₂) measured within 24 hours after qualifying invasive venous catheter insertion was associated with documented acute upper-body deep or central venous thrombosis during hospitalization. This retrospective observational cohort study used the Medical Information Mart for Intensive Care IV database, version 2.2 (MIMIC-IV v2.2) (2008–2019). Adult intensive care unit admissions were included when a qualifying invasive venous catheter was present and an arterial PaO₂ measurement was obtained after insertion, within 24 hours, and while the catheter remained in situ. PaO₂ was analyzed continuously and categorically (≤80, 81–100, and >100 mmHg). Logistic regression models sequentially adjusted for patient characteristics, laboratory measurements, catheter characteristics, measurement timing, inspired oxygen concentration, respiratory support, and non-respiratory organ dysfunction. Multiple imputation addressed missing covariate data, and patient-clustered robust standard errors accounted for repeated admissions. The cohort included 11,277 hospital admissions from 10,789 patients; 195 admissions (1.73%) had the primary outcome. In the unadjusted model, higher PaO₂ was associated with lower odds of the outcome (odds ratio [OR] per 10-mmHg increase, 0.970; 95% confidence interval [CI], 0.952–0.987), but this association attenuated after adjustment. In the fully adjusted model, PaO₂ was not associated with the outcome (OR, 1.008; 95% CI, 0.989–1.027; P = 0.430). Most secondary analyses supported the primary result, although the 60-minute sensitivity analysis yielded adjustment-dependent estimates. In the complete cohort, a single early post-insertion PaO₂ measurement was not independently associated with documented acute upper-body deep or central venous thrombosis. These findings do not support PaO₂-based thrombosis risk stratification or modification of oxygen therapy for thrombosis prevention. Future studies should incorporate repeated oxygenation measurements and accurately timed, imaging-confirmed thrombotic outcomes.

Introduction

Central venous access devices are widely used in critically ill patients for the administration of vasoactive medications, fluids, parenteral nutrition, renal replacement therapy, and other supportive treatments. However, upper-extremity deep or central venous thrombosis is an important complication of vascular access and may lead to catheter dysfunction, treatment interruption, recurrent venous thromboembolism, pulmonary embolism, and post-thrombotic morbidity [1–4]. Reported thrombosis rates vary substantially according to the population studied, catheter type, outcome definition, follow-up duration, and whether systematic imaging surveillance is performed. Peripherally inserted central catheters (PICCs) have been associated with a higher occurrence of venous thromboembolism than conventional central venous catheters in several observational syntheses, particularly in critically ill populations, although findings have not been uniform across study designs [2,3]. Thrombotic susceptibility in the ICU is also influenced by patient and treatment characteristics, including malignancy, sepsis, invasive mechanical ventilation, vasoactive medication use, immobility, and the presence of central venous catheters [5]. In addition, catheter diameter, the catheter-to-vein relationship, tip position, insertion technique, and other device-related factors may modify thrombotic risk [6].

Abnormal oxygenation may also be relevant to thrombosis, but the direction and magnitude of the relationship remain uncertain. Recurrent hypoxia–reoxygenation, as observed in obstructive sleep apnea, has been linked to oxidative stress, systemic inflammation, endothelial dysfunction, altered blood viscosity, and increased expression of procoagulant mediators, providing a potential connection between oxygen disturbance and the components of Virchow’s triad [7]. However, a systematic review of environmental hypoxia in healthy adults found only small and heterogeneous changes in coagulation, platelet activity, and fibrinolysis and did not establish hypoxia as an independent cause of hypercoagulability [8]. Disease-specific clinical studies have reported lower arterial partial pressure of oxygen (PaO₂) among patients with venous thromboembolism or pulmonary embolism in chronic obstructive pulmonary disease [9,10]. These findings may nevertheless reflect respiratory failure, inflammation, immobility, invasive ventilation, or the effects of pulmonary embolism on gas exchange rather than an isolated effect of low PaO₂. At the opposite end of the oxygen spectrum, exposure to higher inspired oxygen concentrations may increase oxidative stress and reduce antioxidant activity [11]. Trials of higher versus lower oxygenation targets in mechanically ventilated ICU patients have also used heterogeneous and overlapping target ranges, leaving uncertainty regarding the clinical effects of different oxygen exposures [12]. Consistent with this uncertainty, large randomized trials comparing lower or conservative oxygenation strategies with higher or usual oxygenation strategies have not demonstrated a consistent clinical advantage of one strategy across critically ill populations [13–15]. Thus, physiological or clinical observations arising from intermittent hypoxia, environmental hypoxia, chronic lung disease, or administered oxygen strategies cannot be directly extrapolated to a single arterial PaO₂ measurement obtained during routine critical care.

In critically ill patients, PaO₂ is influenced by inspired oxygen concentration, ventilatory support, pulmonary gas exchange, perfusion, disease severity, and the timing of arterial blood-gas sampling. Many of these factors are also related to the indication for vascular access, catheter selection, and thrombotic risk. Consequently, an apparent association between PaO₂ and thrombosis may be confounded by the clinical and device-related circumstances surrounding oxygen measurement and catheter placement. Direct evidence examining whether an early arterial PaO₂ measurement is associated with upper-body venous thrombosis among catheterized ICU patients, after accounting for these factors, remains limited. We therefore conducted a retrospective observational cohort study using the Medical Information Mart for Intensive Care IV database to examine the association between the first arterial PaO₂ measured within 24 hours after qualifying catheter insertion and documented acute upper-body deep or central venous thrombosis during hospitalization. PaO₂ was evaluated as both a continuous and categorical exposure, potential nonlinear associations were assessed, and the robustness of the findings was examined after adjustment for catheter characteristics, measurement timing, respiratory support, inspired oxygen concentration, and illness severity.

Materials and methods

Study design and objective

This retrospective observational cohort study used deidentified electronic health record data from the Medical Information Mart for Intensive Care IV database and was reported in accordance with the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement. The study aimed to evaluate whether arterial partial pressure of oxygen (PaO₂) measured during the first 24 hours after qualifying invasive venous catheter insertion was associated with documented acute upper-body deep or central venous thrombosis during the corresponding hospitalization in critically ill adults. We further examined whether this association remained after adjustment for patient characteristics, catheter-related factors, the timing of PaO₂ measurement, inspired oxygen concentration, respiratory support, and illness severity.

The hospital admission was used as the primary unit of analysis. When more than one ICU stay within the same hospital admission met the eligibility criteria, only the earliest eligible ICU stay was retained. A patient could contribute more than one eligible hospital admission. This study evaluated admission-level associations and was not designed as a time-to-event analysis. Because thrombosis diagnoses lacked reliable onset timestamps, temporal precedence of the selected PaO₂ measurement could not be confirmed for every admission.

Data source

Data were obtained from the Medical Information Mart for Intensive Care IV database, version 2.2 (MIMIC-IV v2.2), a deidentified electronic health record database containing clinical information from patients treated at Beth Israel Deaconess Medical Center in Boston, Massachusetts, between 2008 and 2019 [16]. The database includes hospital- and ICU-level information on demographics, hospital and ICU encounters, laboratory measurements, bedside observations, procedures, respiratory support, medications, and diagnosis codes. Access to MIMIC-IV is restricted to credentialed researchers who complete the required human-subjects research training and sign the PhysioNet data use agreement. The authors accessed the database through the standard credentialing process and had no special access privileges.

Ethics statement

The collection and sharing of the MIMIC-IV research resource were reviewed by the Institutional Review Board of Beth Israel Deaconess Medical Center, which approved the data-sharing initiative and granted a waiver of individual informed consent [16]. The present study was a secondary analysis of deidentified data accessed under the PhysioNet Credentialed Health Data Use Agreement. The investigators had no direct contact with participants and no access to identifiable information. No additional institutional ethics approval was obtained for this analysis because the study used only an existing deidentified database and involved no participant recruitment, intervention, or collection of identifiable information. No additional individual informed consent was obtained.

Study population

We screened invasive venous catheter records linked to ICU stays in MIMIC-IV. ICU stays were eligible if patients were aged 18 years or older at hospital admission, remained in the ICU for at least 24 hours, had at least one qualifying central venous catheter episode overlapping the ICU stay, and had at least one eligible arterial PaO₂ measurement obtained after catheter insertion and within 24 hours of catheter placement while at least one qualifying catheter remained in situ.

Catheters documented in lower-body locations were excluded because the study outcomes were restricted to upper-body deep or central venous thrombosis. Catheters documented in upper-body locations were retained. Otherwise eligible catheter records were not excluded solely because their anatomical location was incompletely documented.

The initial extraction included 30,003 invasive venous catheter records corresponding to 18,056 ICU stays among 15,916 patients. After application of the eligibility criteria, 11,474 ICU stays remained. A further 197 ICU stays were excluded because an earlier eligible ICU stay occurred during the same hospital admission. The final cohort comprised 11,277 hospital admissions from 10,789 unique patients. The cohort-selection process is shown in Fig 1.

Fig 1 — ICU, intensive care unit; PaO₂, arterial partial pressure of oxygen. The hospital admission was used as the unit of analysis. When more than one eligible ICU stay occurred during the same hospital admission, only the earliest eligible ICU stay was retained. A patient could contribute more than one eligible hospital admission.

Exposure assessment

The primary exposure was arterial PaO₂ measured after placement of a qualifying central venous catheter. For each eligible ICU stay, catheter episodes were reconstructed using the documented insertion and removal times. The beginning of the qualifying catheter episode was defined as the index time.

An arterial PaO₂ measurement was eligible if it was obtained:

At or after the index catheter insertion time;
Within 24 hours after catheter insertion;
During the corresponding ICU stay; and
While at least one qualifying catheter remained active.

The first arterial PaO₂ measurement satisfying all four criteria was selected as the exposure value. The interval between catheter insertion and the selected PaO₂ measurement was calculated in minutes. For an additional post hoc descriptive analysis, the interval from ICU admission to the selected PaO₂ measurement was also calculated in minutes.

PaO₂ was analyzed in two prespecified forms:

As a continuous variable, with odds ratios reported per 10-mmHg increase; and
As a categorical variable divided into ≤80 mmHg, 81–100 mmHg, and >100 mmHg, with ≤80 mmHg used as the reference group.

In a secondary sensitivity analysis, the ratio of PaO₂ to the fraction of inspired oxygen was analyzed as an alternative oxygenation exposure.

Outcome definitions

The primary outcome was documented acute upper-body deep or central venous thrombosis during the hospital admission. The outcome was identified using a prespecified list of International Classification of Diseases, Ninth and Tenth Revision diagnosis codes recorded in the MIMIC-IV diagnoses_icd table.

The primary outcome definition included acute thrombosis involving deep veins of the upper extremity, axillary veins, subclavian veins, or internal jugular veins. Superficial venous thrombosis and thrombosis identified only by anatomically unspecified upper-extremity codes were not included in the primary outcome.

Two additional outcome definitions were evaluated in sensitivity analyses:

A broader secondary outcome that included other documented upper-body venous thrombosis diagnoses, including superficial or less anatomically specific venous thrombosis; and
A strict outcome restricted to primary-outcome diagnoses with laterality potentially compatible with the qualifying catheter: a right-sided diagnosis with a right-sided catheter, a left-sided diagnosis with a left-sided catheter, or a bilateral diagnosis. Diagnoses with unspecified laterality did not satisfy the strict definition.

The complete ICD-9-CM and ICD-10-CM code list and the rules used to construct the primary, broader, and strict outcome definitions are provided in Supplementary Table S10 in S1 File. No chart-adjudicated or imaging-adjudicated reference standard was available; therefore, the sensitivity and specificity of the coding algorithm could not be estimated.

Because the diagnosis codes were assigned at the hospitalization level and lacked reliable timestamps, the exact onset time of thrombosis could not be determined. Accordingly, it could not be confirmed that the selected PaO₂ measurement preceded thrombus formation in every admission. The available data also could not establish that the coded thrombosis was caused by the qualifying catheter. Therefore, the outcome was described as a documented hospitalization-level thrombotic outcome rather than confirmed catheter-related thrombosis.

Covariates

We collected demographic characteristics, comorbidity burden, laboratory measurements, catheter-related characteristics, measurement timing, respiratory variables, and indicators of illness severity as potential confounding factors. These variables included age at hospital admission, sex, race or ethnicity, Charlson Comorbidity Index, hemoglobin, platelet count, baseline lactate, catheter type, catheter laterality, the number of active catheters, the interval between catheter insertion and PaO₂ measurement, FiO₂ at the time of PaO₂ measurement, respiratory-support status, and the non-respiratory component of the first-day SOFA score. Race or ethnicity was used in descriptive analyses and an additional sensitivity analysis but was not included in the primary fully adjusted model.

Hemoglobin and platelet count were obtained from the most recent complete blood count recorded within the 24 hours preceding the selected PaO₂ measurement. Lactate was preferentially obtained from the same arterial blood gas record as PaO₂; when unavailable in the same record, the most recent lactate measurement within the preceding 24 hours was used. Hemoglobin, platelet count, and lactate were analyzed as continuous variables.

Statistical analysis

Continuous variables were summarized as medians with interquartile ranges, and categorical variables as frequencies and percentages. Standardized mean differences were used to describe between-group differences.

The association between PaO₂ and the primary outcome was evaluated using logistic regression. PaO₂ was analyzed continuously, with odds ratios reported per 10-mmHg increase, and categorically as ≤80, 81–100, and >100 mmHg, with ≤80 mmHg as the reference group.

Four models were constructed. Model 0 was unadjusted. Model 1 included age, sex, and Charlson Comorbidity Index. Model 2 additionally included hemoglobin, platelet count, lactate, catheter type, catheter laterality, multiple active catheters, and the log-transformed interval from catheter insertion to PaO₂ measurement. Model 3 further included FiO₂, respiratory-support status, and the non-respiratory component of the first-day SOFA score.

Missing values in the prespecified model covariates were handled using multiple imputation by chained equations, with 20 imputed datasets [17]. Each dataset was analyzed separately using patient-clustered robust standard errors, and estimates were combined using Rubin’s rules.

Sensitivity analyses included complete-case analysis; exclusion of PaO₂ values greater than 400 mmHg; exclusion of values above the 99th percentile; restriction to each patient’s first eligible admission; restriction to a high-reliability FiO₂ subset, defined as FiO₂ obtained from the same blood-gas record or from a ventilator setting recorded within 60 minutes of the PaO₂ measurement; additional adjustment for race or ethnicity; alternative outcome definitions; and substitution of the PaO₂/FiO₂ ratio for PaO₂. An additional narrow-window sensitivity analysis restricted the cohort to admissions in which the selected PaO₂ measurement was obtained within 60 minutes after catheter insertion. Models 2 and 3 were refitted in this restricted cohort using the same multiple-imputation framework, patient-clustered HC0 robust standard errors, and Rubin’s-rules pooling as in the primary analysis. Results of this analysis are presented in Supplementary Table S12 in S1 File. Because the strict outcome included only 69 events, Firth penalized logistic regression was used as an additional sensitivity analysis. Restricted cubic spline functions were used to examine potential nonlinear associations between PaO₂ and the primary outcome [18]. Exploratory interaction analyses were conducted according to measurement-time window, invasive mechanical ventilation status, and PICC status. A post hoc module-specific analysis examined changes in the PaO₂ estimate after adjustment for different groups of covariates. Model performance and stability were assessed using the apparent area under the receiver operating characteristic curve, Brier score, generalized variance inflation factors, leverage, Cook’s distance, and DFBETA statistics. No adjustment for multiple comparisons was applied to secondary and exploratory analyses; these findings were interpreted as hypothesis-generating. Further details are provided in the Supplementary Methods.

No a priori sample-size calculation was performed because this was a retrospective database study. All hospital admissions meeting the prespecified eligibility criteria were included. All analyses were conducted using R version 4.3.3. Tests were two-sided, and P < 0.05 was considered statistically significant.

Results

Study population and baseline characteristics

A total of 11,277 hospital admissions from 10,789 unique patients were included in the analytic cohort. The median age was 67.7 years (interquartile range [IQR], 58.2–76.4 years), 64.0% of admissions involved male patients, and the median Charlson Comorbidity Index was 5 (IQR, 4–7). Based on the first qualifying PaO₂ measurement obtained within 24 hours after catheter insertion, 1,143 admissions (10.1%) had a PaO₂ of ≤80 mmHg, 1,348 (12.0%) had a PaO₂ of 81–100 mmHg, and 8,786 (77.9%) had a PaO₂ of >100 mmHg. The primary outcome occurred in 32 admissions (2.8%) in the PaO₂ ≤ 80-mmHg group, 29 admissions (2.2%) in the 81–100-mmHg group, and 134 admissions (1.5%) in the > 100-mmHg group, for a total of 195 events (1.73%). In an additional post hoc descriptive analysis, the median interval from ICU admission to the selected PaO₂ measurement was 299 minutes (IQR, 176–525) in admissions without the primary outcome and 494 minutes (IQR, 232–1,160) in admissions with the primary outcome. Baseline characteristics according to PaO₂ category and primary-outcome status are presented in Tables 1 and 2, respectively.

Table 1. Baseline characteristics according to PaO₂ category.

Characteristic	PaO₂ ≤ 80 mmHg(N = 1,143)	PaO₂ 81–100 mmHg(N = 1,348)	PaO₂ > 100 mmHg(N = 8,786)
Age, years	65.7 (55.2, 75.8)	67.4 (57.3, 76.1)	68.0 (58.7, 76.5)
Charlson Comorbidity Index	6.0 (4.0, 8.0)	6.0 (4.0, 8.0)	5.0 (4.0, 7.0)
Hemoglobin, g/dL	10.3 (9.0, 11.8)	10.5 (9.2, 11.9)	10.5 (9.3, 11.7)
Platelet count, × 10⁹/L	186.0 (131.0, 257.0)	177.0 (130.0, 239.0)	156.0 (120.0, 205.0)
Baseline lactate, mmol/L	1.9 (1.3, 3.0)	2.0 (1.3, 3.0)	2.2 (1.6, 3.1)
Nonrespiratory SOFA score	7.0 (4.0, 9.0)	6.0 (4.0, 8.0)	5.0 (3.0, 7.0)
FiO₂, %	60.0 (50.0, 100.0)	50.0 (50.0, 100.0)	80.0 (50.0, 100.0)
Time from catheter insertion to PaO₂ measurement, min	111.0 (40.0, 336.5)	99.0 (36.0, 248.0)	38.0 (20.0, 124.0)
PaO₂, mmHg	70.0 (64.0, 76.0)	91.0 (86.0, 96.0)	190.0 (137.0, 287.0)
Sex
Male	693 (60.6)	861 (63.9)	5,652 (64.3)
Female	450 (39.4)	487 (36.1)	3,134 (35.7)
Race or ethnicity
White	741 (64.8)	934 (69.3)	6,213 (70.7)
Black	103 (9.0)	81 (6.0)	534 (6.1)
Hispanic or Latino	35 (3.1)	28 (2.1)	289 (3.3)
Asian	30 (2.6)	21 (1.6)	218 (2.5)
Other	48 (4.2)	66 (4.9)	369 (4.2)
Unknown or not reported	186 (16.3)	218 (16.2)	1,163 (13.2)
Respiratory support at PaO₂ measurement
Invasive mechanical ventilation	650 (56.9)	846 (62.8)	5,874 (66.9)
Other recorded respiratory support	229 (20.0)	166 (12.3)	321 (3.7)
No contemporaneous respiratory-support interval identified	264 (23.1)	336 (24.9)	2,591 (29.5)
Catheter type
PICC	140 (12.2)	104 (7.7)	298 (3.4)
Multi-lumen catheter	590 (51.6)	646 (47.9)	2,764 (31.5)
Large-bore or specialized catheter	167 (14.6)	301 (22.3)	3,568 (40.6)
Other qualifying catheter type	246 (21.5)	297 (22.0)	2,156 (24.5)
Catheter laterality
Right	871 (76.2)	1,094 (81.2)	7,713 (87.8)
Left	271 (23.7)	251 (18.6)	1,067 (12.1)
Unknown	1 (0.1)	3 (0.2)	6 (0.1)
Number of active qualifying catheters
One active catheter	937 (82.0)	954 (70.8)	4,647 (52.9)
Multiple active catheters	206 (18.0)	394 (29.2)	4,139 (47.1)
Primary outcome
Documented event, n (%)	32 (2.8)	29 (2.2)	134 (1.5)

Open in a new tab

Data are presented as median (interquartile range) for continuous variables and n (%) for categorical variables.

SMD denotes the absolute standardized mean difference. For continuous variables, SMDs were calculated using means and pooled standard deviations; for categorical variables, SMDs were calculated from proportions.

The primary outcome was documented acute upper-body deep or central venous thrombosis during hospitalization. Hospital-level diagnosis codes could not establish that the event was caused by the qualifying catheter or occurred after the selected PaO₂ measurement.

Abbreviations: FiO₂, fraction of inspired oxygen; IQR, interquartile range; PaO₂, arterial partial pressure of oxygen; PICC, peripherally inserted central catheter; SMD, standardized mean difference; SOFA, Sequential Organ Failure Assessment.

Table 2. Baseline characteristics according to primary-outcome status.

Characteristic	No documented primary outcome(N = 11,082)	Documented primary outcome(N = 195)	SMD
Age, years	67.7 (58.2, 76.4)	64.4 (55.6, 75.2)	0.164
Charlson Comorbidity Index	5.0 (4.0, 7.0)	6.0 (4.0, 7.0)	0.191
Hemoglobin, g/dL	10.5 (9.3, 11.7)	10.0 (8.8, 11.2)	0.253
Platelet count, × 10⁹/L	159.0 (121.0, 214.0)	167.0 (108.0, 249.5)	0.142
Baseline lactate, mmol/L	2.2 (1.5, 3.1)	2.1 (1.4, 3.9)	0.212
Nonrespiratory SOFA score	5.0 (3.0, 7.0)	8.0 (6.0, 10.0)	0.779
FiO₂, %	60.0 (50.0, 100.0)	55.0 (50.0, 100.0)	0.316
Time from catheter insertion to PaO₂ measurement, min	45.0 (22.0, 152.0)	134.0 (60.5, 339.0)	0.422
PaO₂, mmHg	158.0 (106.0, 255.0)	123.0 (93.5, 195.0)	0.279
Sex
Male	7,089 (64.0)	117 (60.0)	0.082
Female	3,993 (36.0)	78 (40.0)	0.082
Race or ethnicity
White	7,771 (70.1)	117 (60.0)	0.212
Black	693 (6.3)	25 (12.8)	0.224
Hispanic or Latino	346 (3.1)	6 (3.1)	0.003
Asian	264 (2.4)	5 (2.6)	0.012
Other	471 (4.3)	12 (6.2)	0.086
Unknown or not reported	1,537 (13.9)	30 (15.4)	0.043
Respiratory support at PaO₂ measurement
Invasive mechanical ventilation	7,232 (65.3)	138 (70.8)	0.118
Other recorded respiratory support	701 (6.3)	15 (7.7)	0.054
No contemporaneous respiratory-support interval identified	3,149 (28.4)	42 (21.5)	0.159
Catheter type
PICC	505 (4.6)	37 (19.0)	0.447
Multi-lumen catheter	3,913 (35.3)	87 (44.6)	0.190
Large-bore or specialized catheter	4,013 (36.2)	23 (11.8)	0.572
Other qualifying catheter type	2,651 (23.9)	48 (24.6)	0.016
Catheter laterality
Right	9,541 (86.1)	137 (70.3)	0.383
Left	1,532 (13.8)	57 (29.2)	0.375
Unknown	9 (0.1)	1 (0.5)	0.079
Number of active qualifying catheters
One active catheter	6,390 (57.7)	148 (75.9)	0.387
Multiple active catheters	4,692 (42.3)	47 (24.1)	0.387
PaO₂ category
≤80 mmHg	1,111 (10.0)	32 (16.4)	0.189
81–100 mmHg	1,319 (11.9)	29 (14.9)	0.087
>100 mmHg	8,652 (78.1)	134 (68.7)	0.212
Missing data, n (%)
Age	0 (0.0)	0 (0.0)
Charlson Comorbidity Index	0 (0.0)	0 (0.0)
Hemoglobin	425 (3.8)	8 (4.1)
Platelet count	390 (3.5)	4 (2.1)
Baseline lactate	584 (5.3)	15 (7.7)
Nonrespiratory SOFA score	892 (8.0)	23 (11.8)
FiO₂	1,181 (10.7)	29 (14.9)
Time from catheter insertion to PaO₂ measurement	0 (0.0)	0 (0.0)
PaO₂	0 (0.0)	0 (0.0)

Open in a new tab

Data are presented as median (interquartile range) for continuous variables and n (%) for categorical variables.

Association between PaO₂ and the primary outcome

When modeled as a continuous variable, higher PaO₂ was associated with lower odds of the primary outcome in the unadjusted model, with an OR of 0.970 for each 10-mmHg increase in PaO₂ (95% CI, 0.952–0.987; P < 0.001). The association remained statistically significant after adjustment for age, sex, and Charlson Comorbidity Index (Model 1: OR, 0.975; 95% CI, 0.957–0.993; P = 0.006). After additional adjustment for laboratory measurements, catheter-related characteristics, and the interval between catheter insertion and PaO₂ measurement, the association was substantially attenuated and was no longer statistically significant (Model 2: OR, 1.003; 95% CI, 0.984–1.022; P = 0.77). No association was observed in the fully adjusted model, which additionally included FiO₂, respiratory-support status, and the non-respiratory SOFA score (Model 3: OR, 1.008; 95% CI, 0.989–1.027; P = 0.43; Table 3).

Table 3. Association between continuous PaO₂ and the documented thrombotic outcome.

Model	Exposure	OR (95% CI)	P value	N	Events
Model 0	PaO₂ per 10-mmHg increase	0.970 (0.952–0.987)	<0.001	11,277	195
Model 1	PaO₂ per 10-mmHg increase	0.975 (0.957–0.993)	0.006	11,277	195
Model 2	PaO₂ per 10-mmHg increase	1.003 (0.984–1.022)	0.774	11,277	195
Model 3	PaO₂ per 10-mmHg increase	1.008 (0.989–1.027)	0.430	11,277	195

Open in a new tab

The outcome was documented acute upper-body deep or central venous thrombosis during hospitalization.

Model 0: unadjusted. Model 1: adjusted for age, sex, and Charlson Comorbidity Index. Model 2: additionally adjusted for hemoglobin, platelet count, baseline lactate, catheter type, catheter laterality, multiple active qualifying catheters, and ln(minutes from catheter insertion to PaO₂ measurement + 1). Model 3: additionally adjusted for FiO₂ at the time of PaO₂ measurement, respiratory-support status, and non-respiratory first-day SOFA score.

Statistical analysis: Missing values in prespecified covariates were handled using multiple imputation by chained equations (20 datasets). Logistic regression models used patient-clustered HC0 robust standard errors, and estimates were combined using Rubin’s rules. All P values are two-sided.

Abbreviations: CI, confidence interval; FiO₂, fraction of inspired oxygen; OR, odds ratio; PaO₂, arterial partial pressure of oxygen; SOFA, Sequential Organ Failure Assessment.

In the categorical analysis, compared with admissions with a PaO₂ of ≤80 mmHg, those with a PaO₂ of >100 mmHg had lower odds of the primary outcome in the unadjusted model (OR, 0.538; 95% CI, 0.364–0.795; P = 0.002) and in Model 1 (OR, 0.603; 95% CI, 0.405–0.899; P = 0.013). However, this association was attenuated after further adjustment and was not statistically significant in Model 2 (OR, 0.961; 95% CI, 0.634–1.456; P = 0.85) or Model 3 (OR, 0.989; 95% CI, 0.652–1.502; P = 0.96). No statistically significant association was observed for the 81–100-mmHg group compared with the ≤ 80-mmHg group in any model (Table 4).

Table 4. Association between PaO₂ categories and the documented thrombotic outcome.

Model	PaO₂ category comparison	OR (95% CI)	P value	N	Events
Model 0	81–100 vs ≤ 80 mmHg	0.763 (0.455–1.281)	0.307	11,277	195
Model 0	>100 vs ≤ 80 mmHg	0.538 (0.364–0.795)	0.002	11,277	195
Model 1	81–100 vs ≤ 80 mmHg	0.802 (0.477–1.349)	0.406	11,277	195
Model 1	>100 vs ≤ 80 mmHg	0.603 (0.405–0.899)	0.013	11,277	195
Model 2	81–100 vs ≤ 80 mmHg	0.927 (0.549–1.565)	0.776	11,277	195
Model 2	>100 vs ≤ 80 mmHg	0.961 (0.634–1.456)	0.850	11,277	195
Model 3	81–100 vs ≤ 80 mmHg	0.900 (0.531–1.526)	0.695	11,277	195
Model 3	>100 vs ≤ 80 mmHg	0.989 (0.652–1.502)	0.960	11,277	195

Open in a new tab

Note: The PaO₂ ≤ 80-mmHg category was the reference group. The outcome was documented acute upper-body deep or central venous thrombosis during hospitalization.

Model 0: unadjusted. Model 1: adjusted for age, sex, and Charlson Comorbidity Index. Model 2: additionally adjusted for hemoglobin, platelet count, baseline lactate, catheter type, catheter laterality, multiple active qualifying catheters, and ln(minutes from catheter insertion to PaO₂ measurement + 1). Model 3: additionally adjusted for FiO₂ at the time of PaO₂ measurement, respiratory-support status, and non-respiratory first-day SOFA score.

Statistical analysis: Missing values in prespecified covariates were handled using multiple imputation by chained equations (20 datasets). Logistic regression models used patient-clustered HC0 robust standard errors, and estimates were combined using Rubin’s rules. All P values are two-sided.

Abbreviations: CI, confidence interval; FiO₂, fraction of inspired oxygen; OR, odds ratio; PaO₂, arterial partial pressure of oxygen; SOFA, Sequential Organ Failure Assessment.

Sensitivity, nonlinear, and subgroup analyses

Most prespecified sensitivity analyses were broadly consistent with the primary complete-cohort result. After exclusion of the non-respiratory SOFA score, restriction to complete cases, exclusion of extreme PaO₂ values, restriction to the first eligible admission for each patient, restriction to admissions with high-reliability FiO₂ measurements, and additional adjustment for race or ethnicity, the adjusted ORs for the primary outcome per 10-mmHg increase in PaO₂ ranged from 1.001 to 1.017, with none reaching statistical significance. Analyses using PaO₂ categories yielded similar results. Detailed estimates from the continuous and categorical sensitivity analyses are presented in Supplementary Tables S1 and S2 in S1 File, respectively. In an additional post hoc descriptive summary of the upper tail of the PaO₂ distribution, 647 admissions with 10 outcome events had PaO₂ values greater than 384 mmHg, 490 admissions with 5 events had values greater than 400 mmHg, and 111 admissions with 2 events had values above the 99th percentile (>461 mmHg). These data are presented in Supplementary Table S11 in S1 File.

The findings were also unchanged when alternative outcome and exposure definitions were applied. For the broader secondary outcome, which occurred in 282 admissions, the adjusted OR per 10-mmHg increase in PaO₂ was 1.002 (95% CI, 0.986–1.018; P = 0.81). For the strict outcome definition, which occurred in 69 admissions, the corresponding OR was 1.008 (95% CI, 0.974–1.042; P = 0.66), with a similar estimate obtained using Firth penalized logistic regression. The PaO₂/FiO₂ ratio was likewise not associated with the primary outcome (OR per 50-unit increase, 1.010; 95% CI, 0.957–1.065; P = 0.73). Detailed results of the alternative outcome and exposure analyses are provided in Supplementary Table S3 in S1 File.

In an additional narrow-window sensitivity analysis restricted to admissions with PaO₂ measured within 60 minutes after catheter insertion, 6,301 admissions from 6,185 patients were included, with 49 primary outcome events. The association was not statistically significant in Model 2 (OR per 10-mmHg increase, 1.026; 95% CI, 0.995–1.058; P = 0.102), whereas fully adjusted Model 3 showed a positive association (OR, 1.041; 95% CI, 1.010–1.073; P = 0.010; Supplementary Table S12 in S1 File).

Restricted cubic spline analysis showed no evidence of an overall or nonlinear association between PaO₂ and the primary outcome (P for overall association = 0.186; P for nonlinearity = 0.109; Fig 2).

There was no evidence of effect modification according to the interval from catheter insertion to PaO₂ measurement (P for interaction = 0.223), invasive mechanical ventilation status (P for interaction = 0.522), or PICC status (P for interaction = 0.453). Subgroup-specific estimates were not statistically significant (Supplementary Table S4 and Supplementary Figure S1 in S1 File).

The mean apparent area under the receiver operating characteristic curve across the 20 imputed datasets was 0.7834, and the mean apparent Brier score was 0.0166. No substantial multicollinearity was identified (maximum adjusted generalized variance inflation factor, 1.318). The maximum PaO₂-specific absolute DFBETA was 0.390. Excluding two admissions identified by the overall DFBETA assessment did not materially change the PaO₂ estimate (OR per 10-mmHg increase, 1.006; 95% CI, 0.987–1.025; P = 0.555). Additional apparent discrimination, calibration, multicollinearity, and influence diagnostics are provided in Supplementary Table S5 and Supplementary Figure S2 in S1 File.

The analysis included 11,277 hospital admissions and 195 outcome events. The solid line represents the adjusted odds ratio, and the shaded area represents the 95% confidence interval. Three knots were placed at the 10th, 50th, and 90th percentiles of the PaO₂ distribution (80, 158, and 346 mmHg, respectively), with 80 mmHg as the reference value. Rug marks indicate the distribution of observed PaO₂ values. The model was adjusted for the covariates included in Model 3. Estimates were combined across 20 multiply imputed datasets using Rubin’s rules, with patient-clustered robust standard errors. Overall association, P = 0.186; nonlinearity, P = 0.109.

Secondary associations of model covariates with the documented thrombotic outcome

In an exploratory examination of the fully adjusted model, several covariates were associated with the documented thrombotic outcome. Compared with multi-lumen catheters, PICC use was associated with higher odds of the outcome (OR, 2.71; 95% CI, 1.77–4.17; P < 0.001), whereas large-bore or specialized catheters were associated with lower odds (OR, 0.44; 95% CI, 0.24–0.80; P = 0.007). Left-sided catheter placement was associated with higher odds than right-sided placement (OR, 1.43; 95% CI, 1.01–2.02; P = 0.044). A longer interval between catheter insertion and PaO₂ measurement was also associated with higher odds of the outcome (OR, 1.33 per one-unit increase in ln[minutes + 1]; 95% CI, 1.15–1.53; P < 0.001), as was a higher non-respiratory SOFA score (OR, 1.15 per point; 95% CI, 1.09–1.20; P < 0.001). These findings were considered secondary associations and were not interpreted as causal effects. Complete estimates for all covariates included in the fully adjusted model are provided in Supplementary Table S8 in S1 File.

Post hoc module-specific analysis

In the post hoc module-specific analysis, the inverse association observed in Model 1 (OR per 10-mmHg increase in PaO₂, 0.975; 95% CI, 0.957–0.993; P = 0.006) was most substantially attenuated after the addition of catheter-related characteristics and the interval from catheter insertion to PaO₂ measurement (OR, 1.004; 95% CI, 0.985–1.023; P = 0.687). Adding the laboratory module alone resulted in minimal change in the effect estimate (OR, 0.974; 95% CI, 0.956–0.992; P = 0.006), whereas addition of the oxygenation and illness-severity module shifted the estimate toward the null (OR, 0.992; 95% CI, 0.973–1.011; P = 0.388). After simultaneous adjustment for all covariate modules in Model 3, the association was no longer statistically significant (OR, 1.008; 95% CI, 0.989–1.027; P = 0.430; Fig 3 and Supplementary Table S7 in S1 File). These analyses were exploratory and descriptive; they were not intended to determine whether any individual covariate was responsible for the observed change in the PaO₂ estimate.

Odds ratios are reported for each 10-mmHg increase in PaO₂. Model 1 was adjusted for age, sex, and Charlson Comorbidity Index. Each module-specific model added the indicated module separately to Model 1. The catheter/timing module included catheter type, catheter laterality, multiple active qualifying catheters, and ln(minutes from catheter insertion to PaO₂ measurement + 1). The laboratory module included hemoglobin, platelet count, and baseline lactate. The oxygenation/severity module included FiO₂ at the time of PaO₂ measurement, respiratory-support status, and the non-respiratory first-day SOFA score. Model 3 included all covariates. Estimates were combined across 20 multiply imputed datasets using Rubin’s rules, with patient-clustered robust standard errors. The module-specific analyses were exploratory and descriptive.

Discussion

In this retrospective observational cohort study of 11,277 hospital admissions involving critically ill adults with qualifying invasive venous catheter episodes, arterial PaO₂ measured within 24 hours after catheter insertion was not associated with documented acute upper-body deep or central venous thrombosis after comprehensive adjustment for patient characteristics, laboratory measurements, catheter-related factors, measurement timing, FiO₂, respiratory-support status, and illness severity. Higher PaO₂ was associated with lower odds of the outcome in the unadjusted and minimally adjusted models; however, the estimate moved toward the null after catheter-related characteristics and the interval from catheter insertion to PaO₂ measurement were incorporated into the model. Most secondary and sensitivity analyses in the complete cohort were broadly consistent with the primary null result, although the narrow-window analysis yielded adjustment-set-dependent estimates.

These findings should not be interpreted as evidence that arterial oxygenation has no biological relationship with thrombosis under all circumstances. Rather, they indicate that a single early PaO₂ measurement obtained during routine ICU care was not independently associated with the documented thrombotic outcome after the major clinical and catheter-related covariates available in the database were considered. The attenuation of the initial association highlights the importance of accounting for the clinical context in which PaO₂ is measured, particularly among critically ill patients receiving invasive vascular access and respiratory support.

A single arterial PaO₂ value obtained during routine ICU care should not be regarded as an isolated physiological exposure. It is jointly influenced by pulmonary gas exchange, the administered FiO₂, respiratory-support settings, and the patient’s evolving clinical condition. The role of FiO₂ adjustment depends on the estimand of interest because FiO₂ is a direct determinant of the measured PaO₂. Importantly, however, the association was already null in Model 2, before further adjustment for FiO₂, respiratory-support status, and non-respiratory SOFA score (OR per 10-mmHg increase, 1.003; 95% CI, 0.984–1.022; P = 0.774). Thus, the null primary finding cannot be attributed solely to adjustment for FiO₂. The PaO₂/FiO₂ ratio was therefore retained as an alternative exposure in sensitivity analysis rather than included as an additional covariate in the primary models. Systematic reviews of oxygenation targets in mechanically ventilated and general ICU populations have demonstrated substantial heterogeneity in oxygen targets and uncertainty regarding the optimal level of arterial oxygenation, further illustrating that an individual PaO₂ value cannot be interpreted independently of oxygen delivery and respiratory management [12,19].

The additional analysis restricted to PaO₂ measurements obtained within 60 minutes after catheter insertion yielded more heterogeneous estimates. The association was not statistically significant in Model 2 but became positive in the fully adjusted Model 3. This finding should be interpreted cautiously because only 49 outcome events remained in the restricted cohort, and the estimate was sensitive to the adjustment set. In addition, restricting the analysis to patients with very early post-insertion blood-gas measurements may have introduced selection related to clinical severity and measurement timing. Accordingly, this result is best viewed as a secondary, hypothesis-generating finding rather than evidence that overrides the primary complete-cohort analysis.

At the same time, venous thromboembolism in critically ill patients is associated with multiple patient- and treatment-related factors. Previous systematic reviews have identified central venous catheterization, invasive mechanical ventilation, sepsis, vasoactive therapy, malignancy, previous venous thromboembolism, and prolonged ICU treatment as relevant prognostic factors [5,20]. Accordingly, the inverse association observed in the less extensively adjusted models may have reflected differences in clinical condition and treatment context rather than an independent association between PaO₂ and the documented thrombotic outcome.

The post hoc module-specific analysis was consistent with this interpretation. The largest change in the PaO₂ estimate occurred after addition of the catheter and measurement-timing module, whereas adjustment for laboratory variables alone produced little change. Previous evidence indicates that PICC-associated thrombosis may vary with device- and insertion-related characteristics and has been associated with venous blood-flow velocity, insertion site, and time since catheter placement [21]. However, the present analysis cannot establish that any single catheter variable accounted for the attenuation, nor can it determine whether the documented thrombotic events were caused by the qualifying catheter. The module-specific findings should therefore be interpreted as descriptive evidence that the estimated association between PaO₂ and the outcome was sensitive to the broader catheter-related and temporal context, rather than as proof of a specific causal pathway.

Biological plausibility exists for an association between abnormal oxygenation and thrombosis, although the pattern and duration of oxygen exposure are likely to be important. In obstructive sleep apnea, recurrent hypoxia–reoxygenation has been proposed to promote oxidative stress and inflammatory signaling, increase the expression of procoagulant mediators such as plasminogen activator inhibitor-1 and tissue factor, alter blood viscosity, and impair endothelial function. Collectively, these changes may affect hypercoagulability, venous stasis, and endothelial injury, the three components of Virchow’s triad [7]. However, these mechanisms arise in the context of chronic, recurrent intermittent hypoxia and cannot be directly extrapolated to a single arterial PaO₂ measurement obtained during routine ICU care.

Evidence regarding the effects of hypoxia on hemostasis is also inconsistent. A systematic review of environmental hypoxia in healthy adults found only small and heterogeneous changes in coagulation at high altitude. Increased thrombin generation appeared to be accompanied by reduced platelet activation, fibrinolysis was largely unchanged, and available viscoelastic measurements did not consistently indicate increased thrombogenicity. The review therefore could not establish environmental hypoxia as an independent cause of hypercoagulability, particularly because exercise, hydration status, stress, and other concurrent exposures could not be fully separated from the effects of hypoxia [8]. These observations illustrate that the relationship between oxygen tension and hemostasis is context dependent and may be substantially influenced by concurrent physiological and clinical factors.

Some disease-specific clinical studies have reported lower PaO₂ among patients with venous thromboembolism. A meta-analysis of patients with COPD found a small pooled difference in PaO₂ between those with and without venous thromboembolism, while also identifying prolonged immobility, invasive mechanical ventilation, neoplasia, older age, and previous venous thrombosis as associated factors [9]. Similarly, a meta-analysis of patients with acute exacerbations of COPD found that those with pulmonary embolism had lower PaO₂ than those without pulmonary embolism [10]. These findings should be interpreted within the broader pathophysiological context of COPD, in which mitochondrial injury and the release of mitochondrial damage-associated molecular patterns may amplify inflammatory signaling [22]. Thus, lower PaO₂ in these populations may coexist with inflammation, respiratory failure, reduced mobility, mechanical ventilation, and other factors related to thrombotic risk rather than representing an isolated exposure.

These disease-specific findings are not directly comparable with the present study. Previous studies mainly evaluated pulmonary embolism or conventional venous thromboembolism in COPD populations and frequently assessed oxygenation during the same clinical episode in which thromboembolism was diagnosed. Because pulmonary embolism itself may impair gas exchange and reduce PaO₂, reverse causation cannot be excluded [10]. By contrast, our exposure was defined using a PaO₂ measurement obtained within 24 hours after catheter insertion, and the outcome was documented upper-body deep or central venous thrombosis. Nevertheless, because the exact onset time of thrombosis was unavailable, some temporal uncertainty also remains in our study.

Potential adverse effects are also biologically plausible at the higher end of oxygen exposure. Hyperoxia may increase reactive oxygen species production, inflammation, and endothelial injury. A systematic review of randomized perioperative studies found that higher FiO₂ was associated with increases in several oxidative-stress biomarkers and reductions in some antioxidant markers, although the evidence was derived from a small number of predominantly single-center surgical studies [11]. In mechanically ventilated ICU patients, trials comparing higher and lower oxygenation targets have used heterogeneous and sometimes overlapping target ranges, and the optimal oxygenation strategy remains uncertain [12]. Neither body of evidence directly evaluated upper-body venous thrombosis, and an administered FiO₂ strategy or cumulative hyperoxic exposure is not equivalent to a single measured PaO₂ value. Therefore, our results should not be interpreted as evidence that higher PaO₂ protects against thrombosis or that oxygen administration should be increased for thrombosis prevention. Future studies should evaluate repeated or time-weighted oxygenation measurements together with accurately timed and imaging-confirmed thrombotic outcomes.

Several catheter- and patient-related variables were associated with the documented thrombotic outcome in the fully adjusted model. PICC use, left-sided catheter placement, a longer interval from catheter insertion to the selected PaO₂ measurement, and a higher non-respiratory SOFA score were associated with higher odds of the outcome. In contrast, the large-bore or specialized catheter category was associated with lower odds. These findings provide additional insight into the clinical and device-related context in which upper-body venous thrombosis was documented.

The association between PICC use and higher odds of thrombosis is consistent with previous evidence. A systematic review and meta-analysis of 75 studies involving 109,292 hospitalized patients found that PICCs were associated with higher odds of venous thromboembolism than conventional CVCs, with the highest pooled incidence reported in critical care populations [2]. Although the difference was not statistically significant when the analysis was restricted to randomized trials, the overall findings support the view that PICC-associated thrombotic risk may be clinically relevant in selected hospitalized populations. A separate meta-analysis of prospective studies showed that the incidence of symptomatic PICC-related deep vein thrombosis was comparatively low when contemporary insertion practices were used, including ultrasound-guided venipuncture, appropriate catheter-size selection, and verification of catheter-tip position [6]. Taken together, these findings suggest that the thrombotic profile of a PICC is likely determined not only by the device type itself but also by catheter size, the catheter-to-vein relationship, tip position, insertion technique, treatment indication, and patient characteristics. Observational evidence has also suggested that a catheter-to-vein ratio greater than approximately 45% may be associated with a higher occurrence of PICC-related thrombosis, although the optimal threshold remains uncertain [23]. Many of these technical factors were unavailable in MIMIC-IV and may partly account for the association observed in our study.

The higher odds observed with left-sided placement may also have a plausible vascular basis. Previous evidence has linked slower venous blood flow and left basilic-vein placement with asymptomatic PICC-related thrombosis, and most detected events occurred within 3–12 days after insertion [21]. A longer and potentially more angulated venous course on the left side may contribute to altered flow, greater catheter–vessel contact, or local venous stasis. Although the populations and outcome definitions in previous studies differ from ours, the consistency in direction suggests that catheter laterality and local venous anatomy may be relevant when evaluating upper-body thrombotic events.

The association between a higher non-respiratory SOFA score and the thrombotic outcome is consistent with the broader thrombotic vulnerability of critically ill patients. A meta-analysis of 39 ICU cohorts identified central venous catheterization, invasive mechanical ventilation, sepsis, vasoactive medication use, and lack of pharmacologic thromboprophylaxis as factors associated with venous thromboembolism [5]. Higher non-respiratory SOFA scores may therefore reflect a greater burden of organ dysfunction, systemic inflammation, hemodynamic instability, immobility, and invasive treatment. In this context, the SOFA association is clinically plausible as a marker of the overall severity and treatment intensity accompanying thrombotic risk.

The association with a longer interval from catheter insertion to the selected PaO₂ measurement may reflect the evolving clinical course after catheter placement. Patients whose blood gas measurement occurred later may have had more prolonged or complex ICU management, greater cumulative exposure to invasive devices, or more opportunities for thrombotic events to be investigated and documented. This variable was not equivalent to total catheter dwell time, but its association suggests that the temporal context surrounding catheter placement and subsequent monitoring may be relevant to thrombotic risk assessment.

The lower odds observed for the large-bore or specialized catheter category were unexpected. Central venous access devices used in adult critical care differ considerably in their indications, duration of use, insertion settings, and complication profiles [1]. Large-bore or specialized catheters may be inserted for short-term, procedure-specific indications and removed earlier than standard multi-lumen catheters, or they may be used in patient groups with different surveillance and treatment pathways. The inverse association may therefore reflect differences in device indication and patterns of use rather than an inherently lower thrombogenic potential. Nevertheless, this finding highlights the importance of considering catheter categories separately rather than treating all central venous access devices as equivalent.

Overall, these secondary findings suggest that catheter type, laterality, temporal context, and illness severity may be more informative correlates of the documented thrombotic outcome than a single early PaO₂ measurement. Because the analyses were observational and not designed to establish causal effects for individual covariates, these associations should be interpreted as hypothesis-generating and warrant confirmation in datasets containing detailed device characteristics and accurately timed thrombotic events.

Our findings have several practical implications. First, a single arterial PaO₂ measurement obtained within 24 hours after catheter insertion should not be used in isolation to stratify the risk of documented upper-body deep or central venous thrombosis. The attenuation of the association after comprehensive adjustment, together with the broadly consistent findings from most sensitivity, nonlinear, and subgroup analyses, does not support modifying oxygen administration solely to reduce thrombotic risk. Oxygen therapy should continue to be guided by established acute-care oxygenation goals and by the patient’s respiratory, hemodynamic, and overall clinical needs [24].

Second, thrombotic risk assessment in patients with invasive venous catheters should place greater emphasis on the clinical and device-related context, including catheter type, the indication for PICC placement, insertion laterality, catheter size and tip position when available, ongoing device necessity, and the patient’s overall severity of illness(1, 2, 5, 6). Existing clinical guidance on PICC-related thrombosis prevention is heterogeneous, and a substantial proportion of recommendations are supported by low-certainty evidence or expert consensus [25]. These considerations favor individualized catheter selection, careful insertion and maintenance practices, and clinical monitoring rather than the use of a PaO₂ threshold as a basis for thrombosis prevention. Because the exact onset of thrombosis and standardized imaging ascertainment were unavailable, our findings do not provide a basis for introducing new PaO₂-directed oxygen targets, routine ultrasound screening, or anticoagulant prophylaxis strategies.

This study has several strengths. It included a large cohort of 11,277 hospital admissions involving 10,789 critically ill patients with qualifying invasive venous catheter episodes. The exposure definition incorporated a clinically relevant temporal framework by requiring the arterial PaO₂ measurement to occur within 24 hours after catheter insertion, during the ICU stay, and while the qualifying catheter remained in situ. The analyses accounted for a broad range of potential confounders, including demographic and comorbidity characteristics, laboratory measurements, catheter type and laterality, multiple active catheters, the interval from catheter insertion to PaO₂ measurement, FiO₂, respiratory-support status, and non-respiratory organ dysfunction. Patient-clustered robust standard errors were used to account for repeated admissions, and multiple imputation reduced the loss of information associated with missing covariate data. The broadly consistent findings across most continuous, categorical, nonlinear, alternative-outcome, alternative-exposure, subgroup, and sensitivity analyses support the overall stability of the primary complete-cohort result.

Several limitations should also be acknowledged. First, the retrospective observational design precludes causal inference and leaves the possibility of residual or unmeasured confounding despite extensive adjustment. More importantly, the database did not provide a sufficiently precise time of thrombus onset. Consequently, it was not possible to confirm that the selected PaO₂ measurement preceded thrombus formation in every admission. It was therefore also not possible to calculate the interval from PaO₂ measurement to thrombus onset or to perform a valid time-to-event or competing-risk analysis. The outcome was identified from hospitalization-level diagnostic records rather than through a standardized prospective imaging protocol. No chart-adjudicated or imaging-adjudicated reference standard was available, and the sensitivity and specificity of the coding algorithm are therefore unknown. As a result, asymptomatic or clinically unsuspected thromboses may have been missed, and the observed event rate may therefore be lower than rates reported in studies using systematic imaging surveillance. Outcome ascertainment may also have been influenced by diagnostic testing and documentation practices, and the recorded events could not be definitively attributed to the qualifying catheter. The outcome should therefore be interpreted as documented acute upper-body deep or central venous thrombosis during hospitalization rather than confirmed catheter-related thrombosis.

Second, the exposure was based on a single early PaO₂ measurement. This value may not adequately represent the duration, variability, or cumulative burden of hypoxemia or hyperoxemia over the course of critical illness or catheter exposure. PaO₂ is also influenced by FiO₂, ventilatory management, pulmonary gas exchange, perfusion, and the timing of arterial blood-gas sampling. Although these factors were addressed where possible, only patients with an eligible arterial blood-gas measurement were included, which may have introduced selection bias. In addition, the timing of blood-gas measurement was clinically determined rather than protocolized, and restricting analyses to very early measurements may therefore select patients with different illness severity, monitoring intensity, or clinical trajectories. In addition, several potentially relevant catheter characteristics were unavailable or could not be measured reliably, including catheter material, exact diameter, catheter-to-vein ratio, precise tip location, insertion technique, operator-related factors, and complete catheter dwell time. Although catheter insertion and removal records were used to determine whether a qualifying catheter was active at the time of PaO₂ measurement, the catheter dwell time before thrombus onset could not be calculated because reliable thrombus-onset timestamps were unavailable. Information on imaging surveillance and time-varying thromboprophylaxis or anticoagulant treatment was also limited, leaving the possibility of residual confounding and differential outcome detection.

Finally, MIMIC-IV reflects practice at a single academic medical center between 2008 and 2019. Changes in oxygen therapy, vascular-access practice, thromboprophylaxis, and diagnostic imaging over time may limit the applicability of the findings to other institutions and contemporary practice. The primary outcome occurred in 195 admissions, the strict outcome occurred in only 69 admissions, and the 60-minute restricted cohort included only 49 primary outcome events; therefore, some subgroup and sensitivity analyses may have had limited statistical precision and increased susceptibility to model instability. The statistically significant associations observed for several model covariates should also be regarded as secondary, hypothesis-generating findings, particularly because multiple exploratory comparisons were performed.

Conclusion

In the complete cohort of critically ill patients with invasive venous catheters, the first arterial PaO₂ measured within 24 hours after catheter insertion was not associated with documented acute upper-body deep or central venous thrombosis after comprehensive adjustment. The apparent inverse association observed in less-adjusted models was largely attenuated after accounting for catheter characteristics, measurement timing, respiratory support, and illness severity, although the additional 60-minute sensitivity analysis yielded adjustment-dependent estimates. In secondary analyses, catheter type, placement laterality, measurement timing, and non-respiratory organ dysfunction were associated with the documented thrombotic outcome, although these exploratory findings require confirmation. Future studies incorporating repeated oxygenation measurements, detailed catheter characteristics, standardized imaging assessment, and accurately timed thrombotic events are needed to clarify whether dynamic oxygen exposure contributes to upper-body venous thrombosis.

Supporting information

S1 File. Supplementary Information.

Supplementary methods, Tables S1–S12, and Figures S1–S4.

(DOCX)

pone.0359719.s001.docx^{(8.9MB, docx)}

Data Availability

The data used in this study were obtained from the Medical Information Mart for Intensive Care IV database, version 2.2 (MIMIC-IV v2.2), hosted by PhysioNet (https://doi.org/10.13026/6mm1-ek67). MIMIC-IV is a third-party, credentialed-access database. Interested researchers can obtain access to the same source data through the standard PhysioNet credentialing process, including becoming credentialed PhysioNet users, completing the required CITI “Data or Specimens Only Research” training, and signing the applicable PhysioNet Credentialed Health Data Use Agreement. The authors are not permitted to publicly redistribute the patient-level MIMIC-IV data or the derived patient-level analytic dataset. Researchers who independently obtain authorized access to MIMIC-IV version 2.2 can reconstruct the analytic dataset and replicate the reported analyses using the eligibility criteria, exposure and outcome definitions, covariate definitions, data-processing procedures, and statistical methods described in the Methods and Supporting Information. The authors had no special access privileges to the data, and no additional proprietary or author-only patient-level data were used in this study.

Funding Statement

This work was supported by a Scientific Research Project of the Shenzhen Society of Health Economics (grant number: not applicable), awarded to Weixiang Luo; the Shenzhen Basic Research Program (Natural Science Foundation) (grant number: JCYJ20240813103813018), awarded to Qing Cui, Anshuai Fang, and Yueming Peng; the Shenzhen Clinical Research Center for Respiratory Disease (grant number: LCYSSQ20220823091203007), for which no individual author was named as the award recipient; and the Shenzhen Key Laboratory of Respiratory Diseases (grant number: SYSPG20241211173920041), for which no individual author was named as the award recipient. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

References

1.Takashima M, Schults J, Mihala G, Corley A, Ullman A. Complication and Failures of Central Vascular Access Device in Adult Critical Care Settings. Crit Care Med. 2018;46(12):1998–2009. doi: 10.1097/CCM.0000000000003370 [DOI] [PubMed] [Google Scholar]
2.Puri A, Dai H, Giri M, Wu C, Huang H, Zhao Q. The incidence and risk of venous thromboembolism associated with peripherally inserted central venous catheters in hospitalized patients: A systematic review and meta-analysis. Front Cardiovasc Med. 2022;9:917572. doi: 10.3389/fcvm.2022.917572 [DOI] [PMC free article] [PubMed] [Google Scholar]
3.Mavrovounis G, Mermiri M, Chatzis DG, Pantazopoulos I. Peripherally Inserted Central Catheter lines for Intensive Care Unit and onco-hematologic patients: A systematic review and meta-analysis. Heart Lung. 2020;49(6):922–33. doi: 10.1016/j.hrtlng.2020.07.008 [DOI] [PubMed] [Google Scholar]
4.Thiyagarajah K, Ellingwood L, Endres K, Hegazi A, Radford J, Iansavitchene A, et al. Post-thrombotic syndrome and recurrent thromboembolism in patients with upper extremity deep vein thrombosis: A systematic review and meta-analysis. Thromb Res. 2019;174:34–9. doi: 10.1016/j.thromres.2018.12.012 [DOI] [PubMed] [Google Scholar]
5.Tran A, Fernando SM, Rochwerg B, Cook DJ, Crowther MA, Fowler RA, et al. Prognostic Factors Associated With Development of Venous Thromboembolism in Critically Ill Patients-A Systematic Review and Meta-Analysis. Crit Care Med. 2022;50(4):e370–81. doi: 10.1097/CCM.0000000000005382 [DOI] [PubMed] [Google Scholar]
6.Balsorano P, Virgili G, Villa G, Pittiruti M, Romagnoli S, De Gaudio AR, et al. Peripherally inserted central catheter-related thrombosis rate in modern vascular access era-when insertion technique matters: A systematic review and meta-analysis. J Vasc Access. 2020;21(1):45–54. doi: 10.1177/1129729819852203 [DOI] [PubMed] [Google Scholar]
7.Raj R, Paturi A, Ahmed MA, Thomas SE, Gorantla VR. Obstructive Sleep Apnea as a Risk Factor for Venous Thromboembolism: A Systematic Review. Cureus. 2022;14(2):e22729. doi: 10.7759/cureus.22729 [DOI] [PMC free article] [PubMed] [Google Scholar]
8.Treml B, Wallner B, Blank C, Fries D, Schobersberger W. The Influence of Environmental Hypoxia on Hemostasis-A Systematic Review. Front Cardiovasc Med. 2022;9:813550. doi: 10.3389/fcvm.2022.813550 [DOI] [PMC free article] [PubMed] [Google Scholar]
9.Huang J, Wu B, Zhang Z, Wang L. Risk Factors Associated With Venous Thromboembolism in Individuals With COPD: A Systematic Review and Meta-analysis. Clin Appl Thromb Hemost. 2025;31:10760296251388989. doi: 10.1177/10760296251388989 [DOI] [PMC free article] [PubMed] [Google Scholar]
10.Fu X, Zhong Y, Xu W, Ju J, Yu M, Ge M, et al. The prevalence and clinical features of pulmonary embolism in patients with AE-COPD: A meta-analysis and systematic review. PLoS One. 2021;16(9):e0256480. doi: 10.1371/journal.pone.0256480 [DOI] [PMC free article] [PubMed] [Google Scholar]
11.Oldman AH, Martin DS, Feelisch M, Grocott MPW, Cumpstey AF. Effects of perioperative oxygen concentration on oxidative stress in adult surgical patients: a systematic review. Br J Anaesth. 2021;126(3):622–32. doi: 10.1016/j.bja.2020.09.050 [DOI] [PMC free article] [PubMed] [Google Scholar]
12.Cumpstey AF, Oldman AH, Martin DS, Smith A, Grocott MPW. Oxygen Targets During Mechanical Ventilation in the ICU: A Systematic Review and Meta-Analysis. Crit Care Explor. 2022;4(4):e0652. doi: 10.1097/CCE.0000000000000652 [DOI] [PMC free article] [PubMed] [Google Scholar]
13.ICU-ROX Investigators and the Australian and New Zealand Intensive Care Society Clinical Trials Group, Mackle D, Bellomo R, Bailey M, Beasley R, Deane A, et al. Conservative Oxygen Therapy during Mechanical Ventilation in the ICU. N Engl J Med. 2020;382(11):989–98. doi: 10.1056/NEJMoa1903297 [DOI] [PubMed] [Google Scholar]
14.Schjørring OL, Klitgaard TL, Perner A, Wetterslev J, Lange T, Siegemund M, et al. Lower or Higher Oxygenation Targets for Acute Hypoxemic Respiratory Failure. N Engl J Med. 2021;384(14):1301–11. doi: 10.1056/NEJMoa2032510 [DOI] [PubMed] [Google Scholar]
15.Martin DS, Gould DW, Shahid T, Doidge JC, Cowden A, Sadique Z, et al. Conservative Oxygen Therapy in Mechanically Ventilated Critically Ill Adult Patients: The UK-ROX Randomized Clinical Trial. JAMA. 2025;334(5):398–408. doi: 10.1001/jama.2025.9663 [DOI] [PMC free article] [PubMed] [Google Scholar]
16.Johnson AEW, Bulgarelli L, Shen L, Gayles A, Shammout A, Horng S, et al. MIMIC-IV, a freely accessible electronic health record dataset. Sci Data. 2023;10(1):1. doi: 10.1038/s41597-022-01899-x [DOI] [PMC free article] [PubMed] [Google Scholar]
17.White IR, Royston P, Wood AM. Multiple imputation using chained equations: Issues and guidance for practice. Statistics in Medicine. 2010;30(4):377–99. doi: 10.1002/sim.4067 [DOI] [PubMed] [Google Scholar]
18.Desquilbet L, Mariotti F. Dose-response analyses using restricted cubic spline functions in public health research. Stat Med. 2010;29(9):1037–57. doi: 10.1002/sim.3841 [DOI] [PubMed] [Google Scholar]
19.Klitgaard TL, Schjørring OL, Nielsen FM, Meyhoff CS, Perner A, Wetterslev J, et al. Higher versus lower fractions of inspired oxygen or targets of arterial oxygenation for adults admitted to the intensive care unit. Cochrane Database Syst Rev. 2023;9(9):CD012631. doi: 10.1002/14651858.CD012631.pub3 [DOI] [PMC free article] [PubMed] [Google Scholar]
20.Yang X, Zhou X, Qiu Y. Risk factors of venous thromboembolism in ICU patients: a systematic review and meta-analysis. Ann Med. 2026;58(1):2614222. doi: 10.1080/07853890.2026.2614222 [DOI] [PMC free article] [PubMed] [Google Scholar]
21.Chen P, Zhu B, Wan G, Qin L. The incidence of asymptomatic thrombosis related to peripherally inserted central catheter in adults: A systematic review and meta-analysis People’s. Nurs Open. 2021;8(5):2249–61. doi: 10.1002/nop2.811 [DOI] [PMC free article] [PubMed] [Google Scholar]
22.Shen Y, Chen L, Chen J, Qin J, Wang T, Wen F. Mitochondrial damage-associated molecular patterns in chronic obstructive pulmonary disease: Pathogenetic mechanism and therapeutic target. J Transl Int Med. 2022;11(4):330–40. doi: 10.2478/jtim-2022-0019 [DOI] [PMC free article] [PubMed] [Google Scholar]
23.Sharp R, Carr P, Childs J, Scullion A, Young M, Flynn T, et al. Catheter to vein ratio and risk of peripherally inserted central catheter (PICC)-associated thrombosis according to diagnostic group: a retrospective cohort study. BMJ Open. 2021;11(7):e045895. doi: 10.1136/bmjopen-2020-045895 [DOI] [PMC free article] [PubMed] [Google Scholar]
24.Piraino T, Madden M, Roberts KJ, Lamberti J, Ginier E, Strickland SL. AARC Clinical Practice Guideline: Management of Adult Patients With Oxygen in the Acute Care Setting. Respir Care. 2022;67(1):115–28. doi: 10.4187/respcare.09294 [DOI] [PubMed] [Google Scholar]
25.Zhang N, Xu Y, Zhu LY, Wang Y, Lu Q, An R, et al. Quality appraisal of clinical guidelines for peripherally inserted central catheter-related thrombosis prophylaxis in patients: a systematic review. BMJ Open. 2024;14(11):e084330. doi: 10.1136/bmjopen-2024-084330 [DOI] [PMC free article] [PubMed] [Google Scholar]

PLoS One. doi: 10.1371/journal.pone.0359719.r001

Decision Letter 0

Paavani Atluri

6 Nov 2025

Dear Dr. Cui,

Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process.

Please submit your revised manuscript by Dec 20 2025 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at plosone@plos.org. When you're ready to submit your revision, log on to https://www.editorialmanager.com/pone/ and select the 'Submissions Needing Revision' folder to locate your manuscript file.

A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.
A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.
An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'.

If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter.

If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: https://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory-protocols. Additionally, PLOS ONE offers an option for publishing peer-reviewed Lab Protocol articles, which describe protocols hosted on protocols.io. Read more information on sharing protocols at https://plos.org/protocols?utm_medium=editorial-email&utm_source=authorletters&utm_campaign=protocols.

We look forward to receiving your revised manuscript.

Kind regards,

Paavani Atluri

Academic Editor

PLOS ONE

Journal Requirements:

When submitting your revision, we need you to address these additional requirements.

1. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at

https://journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main_body.pdf and https://journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_title_authors_affiliations.pdf

2. We note that the grant information you provided in the ‘Funding Information’ and ‘Financial Disclosure’ sections do not match.

When you resubmit, please ensure that you provide the correct grant numbers for the awards you received for your study in the ‘Funding Information’ section.

3. Thank you for stating the following financial disclosure:

[Scientific Research Grant from Shenzhen Society of Health Economics].

Please state what role the funders took in the study. If the funders had no role, please state: "The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript."

If this statement is not correct you must amend it as needed.

Please include this amended Role of Funder statement in your cover letter; we will change the online submission form on your behalf.

4. Thank you for stating the following in your Competing Interests section:

[no].

Please complete your Competing Interests on the online submission form to state any Competing Interests. If you have no competing interests, please state "The authors have declared that no competing interests exist.", as detailed online in our guide for authors at http://journals.plos.org/plosone/s/submit-now

This information should be included in your cover letter; we will change the online submission form on your behalf.

5. When completing the data availability statement of the submission form, you indicated that you will make your data available on acceptance. We strongly recommend all authors decide on a data sharing plan before acceptance, as the process can be lengthy and hold up publication timelines. Please note that, though access restrictions are acceptable now, your entire data will need to be made freely accessible if your manuscript is accepted for publication. This policy applies to all data except where public deposition would breach compliance with the protocol approved by your research ethics board. If you are unable to adhere to our open data policy, please kindly revise your statement to explain your reasoning and we will seek the editor's input on an exemption. Please be assured that, once you have provided your new statement, the assessment of your exemption will not hold up the peer review process.

6. Your ethics statement should only appear in the Methods section of your manuscript. If your ethics statement is written in any section besides the Methods, please delete it from any other section.

If the reviewer comments include a recommendation to cite specific previously published works, please review and evaluate these publications to determine whether they are relevant and should be cited. There is no requirement to cite these works unless the editor has indicated otherwise.

[Note: HTML markup is below. Please do not edit.]

Reviewers' comments:

Reviewer's Responses to Questions

Comments to the Author

1. Is the manuscript technically sound, and do the data support the conclusions?

Reviewer #1: Partly

Reviewer #2: No

**********

2. Has the statistical analysis been performed appropriately and rigorously? -->?>

Reviewer #1: Yes

Reviewer #2: Yes

**********

3. Have the authors made all data underlying the findings in their manuscript fully available??>

The PLOS Data policy

Reviewer #1: No

Reviewer #2: Yes

**********

4. Is the manuscript presented in an intelligible fashion and written in standard English??>

Reviewer #1: Yes

Reviewer #2: Yes

**********

Reviewer #1: Dear Author:

Thank you for submitting the manuscript. PLOS ONE. After a thorough review, I regret to inform you that the manuscript in its current form does not meet the standards required for publication. My concerns are as follows:

First, in determining the specific diseaseWhen considering clinical risk factors, the timing of data collection (especially exploratory indicators) is crucial. In this study, the authors failed to clarify...PaO2and CRT The temporal relationship between developments. However, as far as I know,MIMIC dataThe library does indeed allow researchers to determine the temporal sequence between disease events and laboratory indicators. Logically, your conclusion seems to be based on data collected before thrombosis.PaO2——noTherefore, I strongly urge the authors to clearly and specifically detail the inclusion criteria and the timing of data collection in the manuscript.

SecondlyAlthough the manuscript title indicatesPaO2yes CRT A dangerWhile there are risk factors, intraductal thrombosis can be influenced by a variety of factors, not just one. PaO₂I acceptThe author's extensive and exploratory analysis is commendable.;However,There is a clear disconnect between the title and the actual content of the manuscript. I suggest revising the title to more accurately reflect the scope and findings of the research.。

Finally, return in segments.Regression analysis to determine PaO₂ Threshold is 384 mmHg，This is far higher than commonly observed clinical values. Such levels can only be achieved with pure oxygen inhalation or hyperbaric oxygen therapy. Therefore, the clinical relevance and applicability of this threshold are highly questionable.

I sincerely hope the author canThese key issues were taken seriously, and the manuscript was revised accordingly.

Yours sincerely

Reviewpeople

Reviewer #2: Evaluating risk factors for catheter-related thrombosis in critically ill patients is an important issue.

However, the article requires significant revisions.

There are several publications highlighting the harmful effects of hyperoxia in critically ill patients — this topic should be discussed in the manuscript.

Introduction – I recommend revising the first paragraph to make the message clearer.

We know that the prevalence of thrombosis varies depending on the type of catheter used — I suggest that the authors describe the types of catheters (if available) and provide a discussion on this point.

Did the hospitals have protocols for diagnosing thrombosis? How were the thrombosis cases assessed? Complementary examinations are not described.

I suggest revising Table 1 – for instance, regarding vasopressin, it would be more appropriate to present only the proportion of patients who used it, as well as for anticoagulants.

Table 2 is difficult to interpret due to its layout. I recommend the authors reformat it for clarity.

Finally, I suggest the authors revisit the limitations section of the study, which should include the heterogeneity of the participating centers, the presence or absence of diagnostic protocols, and a discussion on the limitations inherent to a retrospective study.

**********

what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy

Reviewer #1: No

Reviewer #2: No

**********

[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.]

To ensure your figures meet our technical requirements, please review our figure guidelines: https://journals.plos.org/plosone/s/figures

You may also use PLOS’s free figure tool, NAAS, to help you prepare publication quality figures: https://journals.plos.org/plosone/s/figures#loc-tools-for-figure-preparation.

NAAS will assess whether your figures meet our technical requirements by comparing each figure against our figure specifications.

PLoS One. 2026 Oct 5;21(10):e0359719. doi: 10.1371/journal.pone.0359719.r002

Author response to Decision Letter 1

9 Mar 2026

Response to Reviewers

Manuscript ID: PONE-D-25-27701

Title: Higher Partial Pressure of Oxygen is Associated with Reduced Risk of Catheter-Related Thrombosis in Critical Care: Analysis of Threshold Effects Using the MIMIC Database

Dear Academic Editor and Reviewers,

Thank you very much for your thoughtful and constructive feedback on our manuscript. We greatly appreciate the time and expertise you have devoted to reviewing our work. We have carefully considered all comments and have revised the manuscript accordingly. Below, we provide a point-by-point response to each concern raised.

________________________________________

Response to Reviewer #1

Comment 1: The authors failed to clarify the temporal relationship between PaO₂ and CRT development… I strongly urge the authors to clearly and specifically detail the inclusion criteria and the timing of data collection.

We sincerely thank the reviewer for this crucial observation and apologize for the lack of clarity in our original manuscript.

We fully understand the reviewer’s concern. Establishing a valid temporal sequence—where exposure (PaO₂) precedes outcome (CRT)—is fundamental for causal inference in observational studies. Without this clarification, the validity of our conclusions could be questioned.

We confirm that all PaO₂ values used in our analysis were extracted from the first arterial blood gas measurement recorded after ICU admission and, critically, before the diagnosis of CRT (or before ICU discharge for patients who did not develop CRT). This ensures that the exposure temporally precedes the outcome.

We have revised the manuscript as follows:

Original: “The primary exposure variable was arterial oxygen partial pressure (PaO2), which was categorized into three groups… These measurements were obtained from arterial blood gas analyses during the ICU stay.”

Revised: “The primary exposure variable was the first recorded arterial oxygen partial pressure (PaO₂) during the ICU stay, which was confirmed to have occurred prior to the diagnosis of catheter-related thrombosis. This PaO₂ value was categorized into three groups: ≤80 mmHg, >80 to ≤100 mmHg, and >100 mmHg.”

Additionally, we added a sentence in the Statistical Analysis section: “All exposure data (PaO₂) were ascertained before the outcome event (CRT) to establish a valid temporal sequence.”

________________________________________

Comment 2: There is a clear disconnect between the title and the actual content… I suggest revising the title.

We thank the reviewer for this helpful suggestion.

We agree that the original title could be misleading, as it labeled PaO₂ a “risk factor,” whereas our findings indicate it is a protective factor (higher PaO₂ associated with lower CRT risk).

We have revised the title to more accurately reflect our findings:

Original: “Partial Pressure of Oxygen as a Risk Factor for Catheter-Related Thrombosis in Critical Care…”

Revised: “Higher Partial Pressure of Oxygen is Associated with Reduced Risk of Catheter-Related Thrombosis in Critical Care: Analysis of Threshold Effects Using the MIMIC Database”.

________________________________________

Comment 3: The threshold of 384 mmHg is far higher than clinical values… clinical relevance is highly questionable.

We sincerely appreciate the reviewer’s insightful clinical perspective and acknowledge this important concern.

We understand that a PaO₂ of 384 mmHg is supraphysiological and rarely a clinical target, as it typically requires high-concentration oxygen therapy, which carries risks of oxygen toxicity.

In our revision, we clarify that this threshold should be interpreted as a statistical inflection point—not a therapeutic goal. The key clinical message is that maintaining PaO₂ above hypoxemic levels (e.g., >80 mmHg) is protective, and no additional CRT benefit is seen beyond extremely high levels. We now explicitly caution against pursuing supraphysiological PaO₂ for CRT prevention alone.

Revised text in the Discussion:

Original: “Our findings provide valuable clinical reference for oxygen therapy management in ICU patients. The identification of a 384 mmHg PO2 threshold enables clinicians to optimize individualized oxygen therapy protocols…”

Revised: “However, our identification of a PO₂ threshold at 384 mmHg should be interpreted with caution. This value far exceeds typical clinical oxygenation targets and is often only achievable with high or pure oxygen supplementation, which carries its own risks of oxygen toxicity and lung injury(11). Therefore, this threshold should be viewed primarily as a statistical inflection point indicating the saturation of the protective effect against CRT, rather than a therapeutic goal. The more clinically relevant and actionable finding is that maintaining PO₂ above the hypoxemic range (e.g., >80 mmHg) is consistently associated with a significantly reduced risk of CRT. Clinicians should aim for adequate oxygenation to avoid hypoxemia, but should not pursue supraphysiological PO₂ levels solely for the purpose of CRT prevention, given the potential for harm from hyperoxia. Several publications have highlighted the harmful effects of hyperoxia in critically ill patients, including increased oxidative stress, ventilator-induced lung injury, and higher mortality in subgroups such as those with acute respiratory distress syndrome (ARDS) or post-cardiac arrest(12-14). For instance, Girardis et al. (12) in a randomized trial showed that conservative oxygen therapy (targeting PaO₂ 70-100 mmHg) reduced mortality compared to liberal strategies (PaO₂ up to 150 mmHg). This underscores the need to avoid hyperoxia while preventing hypoxemia. The observed association between hypoxemia and increased thrombosis risk is biologically plausible and aligns with previous mechanistic studies. Under hypoxic conditions, the activation of Hypoxia-Inducible Factors (HIFs) promotes thrombosis, though current risk models do not include oxygenation status as a specific factor (20)(21). Our study suggests that a patient's oxygenation level could be a relevant variable to consider in future risk stratification models. Further prospective studies are warranted to determine whether optimizing oxygenation, while strictly avoiding both hypoxia and hyperoxia, can be an effective component of a multifaceted strategy to prevent CRT in ICU patients.”

Revised text in the conclusion

Original: Higher PO2 levels are associated with reduced CRT risk in ICU patients, with an optimal threshold at 384 mmHg. These findings suggest the importance of maintaining adequate oxygenation for CRT prevention while avoiding excessive oxygen supplementation.

Revised: Our findings suggest that avoiding hypoxemia—not inducing hyperoxia—is the clinically relevant strategy for CRT prevention

________________________________________

Response to Reviewer #2

Comment 1: Discuss the harmful effects of hyperoxia in critically ill patients.

We sincerely appreciate the reviewer’s suggestion to expand the discussion on the harmful effects of hyperoxia in critically ill patients, as this ensures a more balanced and clinically relevant presentation.

In response, we have expanded the Discussion section to include a dedicated paragraph on hyperoxia's adverse effects, incorporating additional references to key publications for a comprehensive view.

Revised text in the introduction

Original: However, excessive oxygen levels may also have detrimental effects, highlighting the complexity of this relationship.

Revised: However, excessive oxygen levels may also have detrimental effects, highlighting the complexity of this relationship(11). Nevertheless, hyperoxia can lead to adverse outcomes such as oxidative stress and lung injury, necessitating a balanced approach(12-14).

Revised text in the discussion

This is now integrated into our interpretation of the 384 mmHg threshold (see response to Reviewer #1, Comment 3).________________________________________

Comment 2: Revise the first paragraph of the Introduction for clarity.

We sincerely appreciate the reviewer's suggestion to revise the first paragraph of the Introduction for greater clarity, as this helps ensure the manuscript is more accessible to readers.

In response, we have revised the first paragraph to make the message clearer by reorganizing the content with better flow, shorter sentences, and explicit transitions.

Original: Catheter-related thrombosis (CRT) represents a significant complication in intensive care unit (ICU) patients(1). Recent multicenter studies have reported that central venous catheters are utilized in over 85% of ICU patients, with CRT incidence rates varying considerably from 3.6% to 59%(2). This complication not only prolongs hospital stays and increases healthcare costs but also poses substantial risks for life-threatening complications such as pulmonary embolism(3). Furthermore, CRT has been associated with increased mortality rates and adverse clinical outcomes in critically ill patients(4)

Revised: Catheter-related thrombosis (CRT) is a major complication in intensive care unit (ICU) patients, affecting over 85% of those with central venous catheters(1,2). Incidence rates range from 3.6% to 59%, leading to prolonged hospital stays, increased healthcare costs, and severe risks such as pulmonary embolism and higher mortality(3,4). These impacts highlight the urgent need for better prevention strategies

________________________________________

Comment 3: Describe catheter types if available.

We acknowledge this limitation. The MIMIC-IV database identifies CRT via diagnosis codes but does not specify catheter types (e.g., PICC vs. CVC).

We have added this as a limitation: “Fourth, the MIMIC-IV database does not provide granular details on the catheters used, including material type (e.g., polyurethane vs. silicone). Given that these factors are well-established determinants of thrombosis risk, their absence constitutes a significant limitation and may introduce unmeasured heterogeneity into our findings.”

________________________________________

Comment 4: Clarify how thrombosis was diagnosed.

We appreciate the reviewer’s attention to the details of the diagnostic methods and thank you for emphasizing their importance to the reliability of our study.

We agree with your suggestion and have added further details in the Methods section under “Outcome Measurement,” describing the diagnostic assessment approach based on the MIMIC database. We have expanded the Outcome Measurement section:

Original: The primary outcome was the development of catheter-related thrombosis during the ICU stay. Thrombosis was identified through documented diagnosis codes and clinical notes.

Revised: The primary outcome was the development of catheter-related thrombosis during the ICU stay. Thrombosis was identified through documented ICD-9/10 diagnosis codes (e.g., 453.81 for acute venous embolism and thrombosis) and clinical notes in the MIMIC-IV database. Diagnosis typically involved complementary examinations such as duplex ultrasound or computed tomography angiography, as per standard protocols at Beth Israel Deaconess Medical Center, though specific per-case details are not uniformly recorded.

________________________________________

Comment 5: Improve Tables 1 and 2.

we sincerely thank the reviewer for their valuable suggestions regarding the readability and presentation of our tables. we fully understand the reviewer’s concerns: in the original Table 1, the reporting of both "yes" and "no" percentages for binary variables such as vasopressin and anticoagulant use was redundant; and the original layout of Table 2 was overly compact, making it difficult to interpret the association estimates across different adjustment models. In response, we have carefully revised both tables in accordance with the reviewer’s recommendations to enhance clarity and data accessibility.

The specific revisions are as follows:

Table 1: For vasopressin and anticoagulant use, we now report only the proportion of patients who received these interventions (i.e., the “Yes” percentage), removing the redundant “No” category.

Table 2: We have completely reformatted the table to clearly separate the results for continuous and categorical PO₂ analyses, and to align the odds ratios (ORs), 95% confidence intervals (CIs), and P-values for the non-adjusted, Model I, and Model II regression models in a vertically structured and reader-friendly format.

________________________________________

Comment 6: Expand the limitations section.

we sincerely thank the reviewer for their insightful comments regarding the limitations of our study. we fully understand the reviewer’s concerns. Although the MIMIC-IV database is derived from a single-center ICU (Beth Israel Deaconess Medical Center), the retrospective nature of our analysis, the absence of standardized diagnostic protocols for catheter-related thrombosis, and the lack of detailed information on catheter characteristics (such as material type) represent important methodological constraints that warrant explicit acknowledgment.

In response, we sincerely thank the reviewer for their insightful comments regarding the limitations of our study.

Original: “…unmeasured confounders such as genetic background and specific medication regimens.

Revised: This study has several limitations that warrant consideration. First, our analysis is based on the MIMIC-IV database, which is a single-center retrospective cohort. While this provides a large and detailed dataset, the findings may not be fully generalizable to other healthcare settings or populations with different baseline characteristics. Second, as a retrospective study, our analysis is inherently susceptible to residual confounding and selection bias. Although we adjusted for a comprehensive set of clinically relevant covariates, unmeasured or unrecorded confounders (e.g., specific genetic predispositions, detailed medication regimens, or fluid balance) may still influence the observed associations. Third, the diagnosis of catheter-related thrombosis (CRT) relied on ICD coding and supporting clinical documentation, without a standardized, protocol-driven diagnostic approach across the ICU. This may introduce outcome misclassification, potentially leading to either under- or over-estimation of the true CRT incidence. Fourth, the MIMIC-IV database does not provide granular details on the catheters used, including material type (e.g., polyurethane vs. silicone). Given that these factors are well-established determinants of thrombosis risk, their absence constitutes a significant limitation and may introduce unmeasured heterogeneity into our findings. Finally, our data spans from 2008 to 2019, and clinical practices regarding both oxygen therapy and thromboprophylaxis may have evolved since then, which could affect the contemporary applicability of our results.

________________________________________

We believe these revisions have significantly strengthened the manuscript. Thank you again for your valuable feedback.

Sincerely,

Qing Cui, Anshuai Fang, Haiyan Zhang, Qingtong Meng, Sijia Zhou, Yueming Peng, Weixiang Luo

Attachment

Submitted filename: Response to Reviewers.docx

pone.0359719.s004.docx^{(37.2KB, docx)}

PLoS One. doi: 10.1371/journal.pone.0359719.r003

Decision Letter 1

Giovanni Giordano

15 Jun 2026

Dear Dr. Cui,

Please submit your revised manuscript by Jul 30 2026 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at plosone@plos.org. When you're ready to submit your revision, log on to https://www.editorialmanager.com/pone/ and select the 'Submissions Needing Revision' folder to locate your manuscript file.

A letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.
A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.
An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'.

As the corresponding author, your ORCID iD is verified in the submission system and will appear in the published article. PLOS supports the use of ORCID, and we encourage all coauthors to register for an ORCID iD and use it as well. Please encourage your coauthors to verify their ORCID iD within the submission system before final acceptance, as unverified ORCID iDs will not appear in the published article. Only the individual author can complete the verification step; PLOS staff cannot verify ORCID iDs on behalf of authors.

We look forward to receiving your revised manuscript.

Kind regards,

Giovanni Giordano

Academic Editor

PLOS One

Journal Requirements:

[Note: HTML markup is below. Please do not edit.]

Reviewers' comments:

Reviewer's Responses to Questions

Comments to the Author

Reviewer #1: All comments have been addressed

Reviewer #3: (No Response)

Reviewer #4: (No Response)

**********

2. Is the manuscript technically sound, and do the data support the conclusions??>

Reviewer #1: No

Reviewer #3: Partly

Reviewer #4: Partly

**********

3. Has the statistical analysis been performed appropriately and rigorously? -->?>

Reviewer #1: No

Reviewer #3: No

Reviewer #4: No

**********

4. Have the authors made all data underlying the findings in their manuscript fully available??>

The PLOS Data policy

Reviewer #1: Yes

Reviewer #3: Yes

Reviewer #4: Yes

**********

5. Is the manuscript presented in an intelligible fashion and written in standard English??>

Reviewer #1: Yes

Reviewer #3: (No Response)

Reviewer #4: Yes

**********

Reviewer #1: The authors have substantially revised the manuscript, and their effort is appreciated. However, the revised manuscript still does not resolve the fundamental methodological concerns raised in the previous review.

First, the temporal relationship between PaO₂ measurement and CRT development remains insufficiently clarified. If an arterial blood gas value obtained after ICU admission is used as the exposure for catheter-related thrombosis, this value must have a clear and clinically meaningful temporal relationship with catheter insertion, catheter dwell time, and thrombosis development. For example, if a patient with shock undergoes arterial blood gas testing immediately after ICU admission and has a CVC inserted at that time, but develops deep venous thrombosis on ICU day 10, it remains questionable whether the initial PaO₂ value can reasonably represent the relevant exposure for that catheter-related thrombotic event. In the revised manuscript, the authors only state that the first recorded PaO₂ during ICU stay occurred before CRT diagnosis. This statement does not adequately address the concern. Demonstrating that PaO₂ preceded the diagnostic code for CRT is not sufficient to establish a valid exposure window. The authors still do not report the interval from ICU admission to PaO₂ measurement, from catheter insertion to PaO₂ measurement, or from PaO₂ measurement to CRT diagnosis. Nor do they show whether the timing of blood gas sampling was comparable between patients with and without CRT.

Second, the reported threshold of 384 mmHg remains clinically uninterpretable. Although the title has been revised, the authors continue to retain this threshold as an important finding. They acknowledge that this value far exceeds usual clinical oxygenation targets and should be interpreted as a statistical inflection point rather than a therapeutic goal. However, this does not resolve the problem. In real ICU practice, a PaO₂ of 384 mmHg is uncommon and usually reflects transient exposure to very high FiO₂, peri-intubation or perioperative oxygen administration, resuscitation, or other specific oxygenation contexts rather than a stable and clinically generalizable oxygenation target. Therefore, this threshold has very limited clinical relevance.

More importantly, the 384 mmHg cutoff may be an artifact arising from the construction of the MIMIC-based dataset and the modeling strategy. Potential sources of bias include indication bias in arterial blood gas sampling, exposure misclassification due to the absence of FiO₂, PEEP, oxygen delivery method, mechanical ventilation status, and PaO₂/FiO₂ ratio, sparse data in the high-PaO₂ tail, extreme-value effects, case-mix heterogeneity, unmeasured catheter dwell time and catheter type, and outcome misclassification due to reliance on ICD codes and clinical notes. The authors do not report the distribution of PaO₂ values, the number of patients with PaO₂ >384 mmHg, the number of CRT events above this threshold, the handling of outliers, or sensitivity analyses excluding extreme PaO₂ values. Therefore, the threshold should not be presented as a clinically meaningful finding.

Overall, the authors have not truly addressed the major concern raised in the previous review. At minimum, they should provide detailed timing information linking PaO₂ measurement to ICU admission, catheter insertion, catheter dwell time, and CRT diagnosis. If the MIMIC database cannot provide the variables required to establish a valid exposure window and clinically interpretable oxygenation context, then this database may not be suitable for answering the present research question. The authors should consider validating or re-conducting the study using original clinical data from their own center, where catheter timing, catheter characteristics, oxygenation parameters, and CRT diagnosis can be reliably verified.

For these reasons, I do not recommend acceptance of the manuscript in its current form.

Reviewer #3: Manuscript ID: PONE-D-25-27701R1

Dear Editor,

Thank you very much for the invitation to review this manuscript titled “Higher Partial Pressure of Oxygen is Associated with Reduced Risk of Catheter-Related Thrombosis in Critical Care: Analysis of Threshold Effects Using the MIMIC Database” and for the opportunity to contribute to the evaluation process for PLOS ONE. I would also like to thank the authors for their work on a clinically relevant question and for the careful, point-by-point responses provided in this revised version, which have clearly improved the clarity of the manuscript.

In brief, the authors use the MIMIC-IV database (2008–2019) to investigate the association between arterial partial pressure of oxygen (PaO₂) and catheter-related thrombosis (CRT) in 13,328 adult ICU patients with upper-extremity catheters. PaO₂ was analysed both as a continuous variable and in three categories (≤80, >80–100, >100 mmHg), with progressively adjusted logistic regression models, smooth-curve fitting, segmented (threshold) regression, and subgroup analyses by anticoagulation status, SOFA score and age. Compared with PaO₂ ≤80 mmHg, higher PaO₂ categories were associated with a lower adjusted risk of CRT (adjusted OR 0.49 and 0.47), and the authors report a threshold at 384 mmHg beyond which no further benefit was seen. The revised manuscript reframes this finding as avoidance of hypoxemia rather than induction of hyperoxia, which is appropriate.

My recommendation is MAJOR REVISION. The dataset and question are of interest and the writing has improved, but several methodological and interpretive issues need to be resolved before the conclusions can be considered robust. My specific concerns, in order of importance, are as follows:

1. In Table 3, the effect above the 384 mmHg breakpoint is reported as OR = 0.00 (95% CI 0.00, Inf), p = 0.9795. An odds ratio of zero with an infinite upper bound indicates complete separation / absence of events (or vanishingly few patients) above the breakpoint, i.e. a non-interpretable estimate rather than evidence of a true plateau. Because this threshold now features prominently in both the title and the abstract, it must be substantiated: please report the number of patients and the number of CRT events above 384 mmHg, and reconsider whether the segmented model is adequately powered in that range. If the estimate is driven by sparse data, the threshold should be presented with strong caveats or removed from the title/abstract.

2. When PaO₂ is modelled continuously, the adjusted OR is 1.00 (95% CI 0.99–1.00) per 1 mmHg, with significance driven almost entirely by the large sample size. Describing this as a “significant inverse association” risks overstating a near-null per-unit effect. Please report the OR per a clinically meaningful increment (e.g. per 10 mmHg) and temper the wording accordingly.

3. Patients with PaO₂ >100 mmHg constitute the large majority of the cohort (9,242/13,328) and very likely reflect patients receiving supplemental oxygen or mechanical ventilation, whereas low PaO₂ marks respiratory failure and greater illness. Although APSIII and SOFA were adjusted for, key drivers such as FiO₂, mechanical ventilation, and the PaO₂/FiO₂ ratio are not included. A PaO₂ of 384 mmHg is essentially attainable on high FiO₂, so the exposure is partly a proxy for the treatment received. Please adjust for FiO₂ / ventilation where available, or discuss this as a major source of residual confounding and reverse causation.

4. CRT was identified from ICD-9/10 codes and clinical notes. By the authors’ own cited literature, a large fraction of CRT is asymptomatic (incidence reported up to 59%), yet the observed incidence here is only ~2.2% in the reference group. This strongly suggests that only symptomatic / imaged thrombosis was captured, and detection may be differential (sicker or more frequently imaged patients are more likely to be diagnosed). Please report the total number of CRT events, and discuss the direction of this detection bias; it could plausibly generate the observed association if low-PaO₂ patients are imaged more often.

5. I appreciate the clarification that the first PaO₂ before CRT was used. However, a single baseline value cannot capture the dynamic oxygenation of an ICU stay, and the differing exposure windows introduce an immortal-time-like imbalance. A time-to-event model with PaO₂ as a time-varying exposure would be more appropriate; at minimum, this limitation should be stated explicitly.

6. Reporting completeness (STROBE). Please add the number of patients excluded for missing key variables and a comparison of their characteristics; the number of CRT events per PaO₂ group; and some assessment of model performance (calibration/discrimination). This information is needed to judge the stability of the adjusted estimates given the low event rate.

7. The biological rationale leans heavily on COVID-19 cohorts (e.g. Helms et al., Tang et al.). Coagulopathy in severe SARS-CoV-2 infection may not generalise to unselected ICU patients with catheter-related thrombosis; please acknowledge this and, where possible, cite mechanistic evidence from a broader critical-care population.

A few minor points would also strengthen the manuscript:

8. Table 1 still contains misaligned and internally inconsistent cells; please verify all values, units, and decimal places.

9. The notation alternates between “PO2” and “PaO₂” throughout the text, tables and figure legends; please standardise to PaO₂.

10. Figure 2 should state the number of observations and ideally show the distribution of PaO₂ values, given the wide confidence bands at the extremes.

In summary, this is a potentially useful real-world analysis, but the threshold claim that anchors the title and abstract currently rests on a non-estimable result, and the association is vulnerable to confounding by ventilation and to outcome misclassification.

Addressing points these points is essential before the conclusions can be supported. I believe these issues are addressable, and I would be glad to review a revised version.

I thank the Editor again for the invitation and the authors for their efforts on this work.

Sincerely,

The Reviewer

Reviewer #4: This manuscript investigates the association between arterial oxygen partial pressure (PaO₂) and catheter-related thrombosis (CRT) among ICU patients using the MIMIC-IV database. The topic is clinically relevant because CRT is a common ICU complication and oxygenation status is routinely measured. The large sample size is a major strength, and the authors attempt to explore nonlinear relationships and threshold effects.

However, substantial methodological and interpretative concerns currently limit confidence in the conclusions. The central finding—that higher PaO₂ protects against CRT—may be heavily influenced by confounding, selection bias, exposure misclassification, outcome ascertainment issues, and possible immortal-time bias. Several analytical choices require clarification or re-analysis before publication.

The manuscript has merit but requires major revision.

A. Major Comments

1. Fundamental Concern Regarding CRT Ascertainment

The manuscript states:

"CRT was identified through ICD-9/10 diagnosis codes and clinical notes."

This approach raises major concerns.

CRT is notoriously undercoded in administrative databases. Many upper-extremity catheter thromboses are diagnosed radiologically but never assigned specific ICD codes. Conversely, some ICD thrombotic diagnoses may represent unrelated venous thromboses.

The authors should provide:

Exact ICD-9 and ICD-10 codes used.

Validation strategy.

Sensitivity and specificity of their CRT definition.

Whether thromboses were restricted to upper-extremity catheter-associated events.

How non-catheter-related DVTs were excluded.

Without a validated CRT definition, outcome misclassification could substantially bias results.

2. Exposure Definition is Potentially Problematic

The study uses:

"First recorded arterial PaO₂ during ICU stay."

This single measurement is unlikely to represent oxygen exposure throughout the period during which CRT develops.

Important questions:

How soon after ICU admission was PaO₂ measured?

What was the median time from PaO₂ measurement to CRT diagnosis?

Were patients already thrombosed when the first blood gas was obtained?

Why was only the first PaO₂ used instead of:

Time-weighted average PaO₂

Maximum PaO₂

Minimum PaO₂

Area-under-the-curve exposure

Given that oxygenation changes dramatically during ICU admission, reliance on a single value may create substantial exposure misclassification.

3. Temporal Relationship Remains Unclear

The manuscript repeatedly claims:

"PaO₂ occurred before CRT diagnosis."

However, the methodology does not demonstrate this rigorously.

The authors should explicitly report:

Time of catheter insertion.

Time of first PaO₂ measurement.

Time of CRT diagnosis.

CRT is a time-dependent event. Standard logistic regression ignores timing and may introduce bias.

A more appropriate approach would be:

Cox proportional hazards model.

Time-to-event analysis.

Competing-risk framework (death as competing event).

Current analyses do not adequately address temporality.

4. Severe Risk of Confounding by Illness Severity

One striking observation:

Patients with PaO₂ >100 mmHg appear less ill:

Variable ≤80 mmHg >100 mmHg

APSIII 4.04 3.76

SOFA 7.64 6.37

Vasopressor use 18.0% 10.1%

ICU LOS 1.70 1.20

These differences are substantial.

Higher PaO₂ may simply identify patients who:

Are physiologically more stable.

Have better pulmonary function.

Have shorter ICU stays.

Have lower inflammatory burden.

Thus PaO₂ may function as a marker of illness severity rather than an independent causal factor.

Although adjustment was performed, residual confounding remains highly likely.

I strongly recommend:

Propensity score matching.

IPTW analysis.

Sensitivity analyses using doubly robust models.

5. Potential Reverse Causality

Patients with severe illness often develop:

Hypoxemia

Longer catheter dwell times

More thrombosis

Thus:

Hypoxemia → severity ← CRT

rather than:

Hypoxemia → CRT

The authors should discuss this possibility more thoroughly.

Current wording occasionally implies causality that cannot be supported by observational data.

6. Threshold Analysis Appears Statistically Fragile

The reported threshold:

384 mmHg

is difficult to interpret biologically.

Several concerns arise:

A. Sparse Data

Very few ICU patients likely have PaO₂ >384 mmHg.

The manuscript does not provide:

Number of patients above threshold.

Distribution of PaO₂ values.

Histogram or density plot.

B. Implausible Estimate

Table 3 reports:

OR = 0.00 (0.00–Inf)

This indicates model instability.

Such estimates typically occur when:

Data are sparse.

Separation occurs.

Regression becomes unreliable.

Therefore the claimed threshold may be an artifact of modeling.

The threshold analysis should be interpreted cautiously or potentially removed.

7. Inconsistency Between Continuous and Categorical Analyses

The manuscript reports:

Continuous model:

OR ≈ 1.00

Categorical model:

OR ≈ 0.47–0.49

These findings appear discordant.

If a clinically meaningful effect exists, the continuous estimate should better reflect that relationship.

This discrepancy suggests:

Strong nonlinearity,

Modeling issues,

Or influence of categorization.

The authors should formally compare models and present spline plots with confidence intervals.

8. Anticoagulation Variable May Introduce Bias

The study adjusts for anticoagulant administration.

However:

Anticoagulation may lie on the causal pathway.

Patients perceived as thrombosis-prone are more likely to receive anticoagulation.

Adjusting for this variable may create:

Collider bias

Overadjustment

The authors should justify its inclusion and perform sensitivity analyses excluding anticoagulation.

9. Missing Important CRT Risk Factors

Several important determinants of CRT are absent:

Catheter-related

Catheter type

Number of lumens

Catheter size

PICC vs CVC

Insertion technique

Catheter dwell time

Tip position

Clinical

Mechanical ventilation

Cancer status

Sepsis

Platelet count

D-dimer

Coagulation profile

These omissions substantially limit causal inference.

10. Use of the Term "Protective Effect"

The manuscript repeatedly uses phrases such as:

"protective effect"

"risk reduction"

These imply causality.

Given the observational design, wording should be changed to:

"associated with lower odds"

"inverse association"

"correlated with reduced CRT occurrence"

throughout the manuscript.

B. Minor Comments

1. Study Design Misclassification

The manuscript alternates between:

Retrospective cohort

Cross-sectional study

These are not equivalent.

The study should be consistently described as a retrospective observational cohort analysis.

2. PO₂ versus PaO₂

Terminology switches repeatedly between:

PO₂

PaO₂

Use PaO₂ consistently because arterial oxygen tension is specifically being measured.

3. APACHE III / APSIII Confusion

The manuscript uses:

APSIII

APACHE III

interchangeably.

Please clarify which score was included in regression models.

4. Reporting of Continuous ORs

Reporting:

OR=1.00

is uninformative.

Authors should present:

OR per 10 mmHg increase

OR per SD increase

to improve interpretability.

5. Missing Handling of Missing Data

The manuscript states:

Missing key variables were excluded.

The following should be reported:

Number excluded.

Variables missing.

Missingness percentages.

Whether multiple imputation was considered.

6. MIMIC-IV Reporting Standards

Authors should provide:

Exact SQL extraction strategy.

MIMIC-IV version used.

Reproducible code repository.

This aligns with current expectations for MIMIC-based publications.

**********

what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy

Reviewer #1: No

Reviewer #3: No

Reviewer #4: No

**********

To ensure your figures meet our technical requirements, please review our figure guidelines: https://journals.plos.org/plosone/s/figures

You may also use PLOS’s free figure tool, NAAS, to help you prepare publication quality figures: https://journals.plos.org/plosone/s/figures#loc-tools-for-figure-preparation.

NAAS will assess whether your figures meet our technical requirements by comparing each figure against our figure specifications.

Attachment

Submitted filename: PONE-D-25-27701R1_Reviewer_Report.docx

pone.0359719.s003.docx^{(28KB, docx)}

PLoS One. 2026 Oct 5;21(10):e0359719. doi: 10.1371/journal.pone.0359719.r004

Author response to Decision Letter 2

13 Aug 2026

General response

Dear Academic Editor and Reviewers,

We sincerely thank you for the careful and constructive evaluation of our manuscript. We agree that the previous analytical framework did not adequately establish a clinically meaningful temporal relationship between oxygen measurement, catheter placement, and the thrombotic outcome, and that the reported 384-mmHg breakpoint was statistically unstable and clinically difficult to interpret.

Rather than retaining the previous framework, we substantially reconstructed the cohort and repeated the analyses. The revised exposure is the first arterial PaO₂ measurement obtained at or after insertion of a qualifying invasive venous catheter, within 24 hours after insertion, during the ICU stay, and while at least one qualifying catheter remained in situ. Because hospitalization-level diagnosis codes lack reliable thrombus-onset timestamps and cannot establish catheter causation, the outcome is no longer described as confirmed catheter-related thrombosis. It is now described as documented acute upper-body deep or central venous thrombosis during hospitalization.

We completely removed the segmented regression analysis, the 384-mmHg threshold, and all claims that higher PaO₂ has a protective effect. The reconstructed cohort included 11,277 hospital admissions from 10,789 patients, with 195 primary outcome events. In the fully adjusted model, PaO₂ was not associated with the outcome (OR per 10-mmHg increase, 1.008; 95% CI, 0.989–1.027; P = 0.430). The revised model includes catheter characteristics, the interval from catheter insertion to PaO₂ measurement, FiO₂, respiratory-support status, and non-respiratory organ dysfunction. Categorical, restricted cubic spline, alternative-outcome, PaO₂/FiO₂, subgroup, complete-case, and extreme-value sensitivity analyses were consistent with the primary result.

We also added the complete ICD code dictionary and outcome-classification rules in Supplementary Table S10, additional timing and upper-tail PaO₂ distribution data in Supplementary Table S11, and a completed STROBE checklist. The title, abstract, Methods, Results, Discussion, Conclusion, tables, figure captions, and Supporting Information have been extensively revised.

For clarity, the reviewer comments are summarized below in bold headings, followed by our responses and the corresponding locations of the revisions.

Reviewer #1

Comment 1. Temporal relationship between PaO₂ measurement, catheter placement, catheter dwell time, and thrombosis development

Reviewer comment: The temporal relationship remained insufficiently clarified. The reviewer requested detailed timing from ICU admission and catheter insertion to PaO₂ measurement, as well as from PaO₂ measurement to thrombosis diagnosis, and questioned whether a single early value could represent the relevant exposure for a later thrombotic event.

Response: We agree that the previous exposure definition was insufficient. We therefore reconstructed the cohort using qualifying catheter insertion as the index time. An eligible PaO₂ measurement was required to occur at or after catheter insertion, within 24 hours after insertion, during the corresponding ICU stay, and while at least one qualifying catheter remained active. The first measurement meeting all four criteria was selected.

We also characterized the timing of arterial blood-gas sampling. In an additional post hoc descriptive analysis, the median interval from ICU admission to the selected PaO₂ measurement was 299 minutes (IQR, 176–525) in admissions without the primary outcome and 494 minutes (IQR, 232–1,160) in admissions with the primary outcome. The corresponding intervals from catheter insertion to PaO₂ measurement were 45 minutes (IQR, 22–152) and 134 minutes (IQR, 60.5–339), respectively. The catheter-insertion-to-PaO₂ interval was included in the adjusted models after log transformation, and subgroup analyses across ≤1 hour, >1–6 hours, and >6–24 hours showed no evidence of interaction (P for interaction = 0.223).

However, the hospitalization-level diagnosis records do not provide a sufficiently reliable time of thrombus onset. We therefore cannot calculate a valid interval from PaO₂ measurement to thrombus formation, confirm that PaO₂ preceded thrombus formation in every admission, or calculate catheter dwell time before thrombus onset. For the same reason, a valid time-to-event or competing-risk analysis could not be performed. We have made these limitations explicit and no longer describe the outcome as confirmed catheter-related thrombosis.

Changes in the revised submission: Methods—Exposure assessment and Outcome definitions; Results—Study population and baseline characteristics; Table 1B; Supplementary Table S11; Discussion—limitations.

Comment 2. The 384-mmHg threshold was clinically uninterpretable and potentially driven by sparse data and model instability

Reviewer comment: The reviewer questioned the clinical relevance and statistical stability of the reported 384-mmHg breakpoint and requested information on the upper tail of the PaO₂ distribution and sensitivity analyses excluding extreme values.

For transparency, an additional post hoc descriptive summary showed that the reconstructed cohort contained 647 admissions and 10 primary-outcome events with PaO₂ >384 mmHg, 490 admissions and 5 events with PaO₂ >400 mmHg, and 111 admissions and 2 events above the 99th percentile (>461 mmHg). These counts confirm that the upper tail was sparse. Potential nonlinearity is now evaluated using restricted cubic splines, which showed no evidence of an overall association (P = 0.186) or nonlinearity (P = 0.109). Sensitivity analyses excluding PaO₂ >400 mmHg and values above the 99th percentile also yielded null estimates.

Changes in the revised submission: Threshold analysis removed throughout; Results—Sensitivity, nonlinear, and subgroup analyses; revised Fig 2; Supplementary Tables S1, S2, and S11.

Comment 3. Suitability of MIMIC-IV for the research question and need for validation using original clinical data

Reviewer comment: The reviewer suggested that, if the database cannot establish a valid exposure window and oxygenation context, the study question may require validation or re-conduction using original clinical data.

Response: We agree that MIMIC-IV cannot support a causal or time-to-event analysis of confirmed catheter-related thrombosis because reliable thrombus-onset timestamps, standardized imaging ascertainment, and several detailed catheter variables are unavailable. We therefore narrowed the research question to the association between a clearly defined early post-insertion PaO₂ measurement and a hospitalization-level documented upper-body thrombotic outcome.

Within this narrower scope, MIMIC-IV provides catheter insertion and removal records, arterial blood-gas timing, FiO₂, respiratory-support information, laboratory data, and illness-severity variables that permit a transparent observational analysis. We explicitly acknowledge the remaining temporal, exposure, and outcome-ascertainment limitations and recommend prospective validation using repeated oxygenation measurements, detailed catheter data, and accurately timed, imaging-confirmed thrombotic outcomes.

Changes in the revised submission: Title, objective, exposure definition, outcome terminology, Discussion, limitations, and Conclusion.

Reviewer #3

Comment 1. The estimate above 384 mmHg was non-interpretable because of complete separation or sparse events

Reviewer comment: The reviewer noted that OR = 0.00 with an infinite upper confidence limit indicated a non-estimable result rather than a true plateau.

Response: We agree. The previous estimate reflected sparse-data separation and should not have been interpreted as a plateau. The breakpoint model and every reference to 384 mmHg as a threshold have been removed. No threshold result remains in the revised manuscript. Upper-tail counts are reported only as an additional post hoc descriptive summary to demonstrate the sparsity that affected the previous model.

Changes in the revised submission: Threshold analysis removed from all manuscript sections, tables, and figures; Supplementary Table S11.

Comment 2. Report the continuous association per a clinically meaningful increment and temper the interpretation

Reviewer comment: The reviewer recommended reporting the OR per 10 mmHg rather than per 1 mmHg and avoiding overstatement of a near-null effect.

Response: PaO₂ is now reported per 10-mmHg increase throughout the manuscript. The unadjusted estimate was OR 0.970 (95% CI, 0.952–0.987), but the association attenuated after adjustment. The fully adjusted estimate was OR 1.008 (95% CI, 0.989–1.027; P = 0.430). We therefore no longer describe the result as a significant inverse association, risk reduction, or protective effect.

Changes in the revised submission: Abstract; Methods—Statistical analysis; Results; Tables 2 and 3; Discussion; Conclusion.

Comment 3. Account for FiO₂, ventilation, and the PaO₂/FiO₂ ratio

Reviewer comment: The reviewer noted that PaO₂ partly reflects oxygen treatment and recommended adjustment for FiO₂ and ventilation and analysis of the PaO₂/FiO₂ ratio.

Response: We agree that PaO₂ cannot be interpreted independently of oxygen delivery and respiratory support. The fully adjusted model now includes FiO₂ at the time of PaO₂ measurement, respiratory-support status, and the non-respiratory component of the first-day SOFA score. Respiratory support was classified as invasive mechanical ventilation, other recorded respiratory support, or no contemporaneous respiratory-support interval identified.

We also analyzed the PaO₂/FiO₂ ratio as an alternative exposure. It was not associated with the outcome (OR per 50-unit increase, 1.010; 95% CI, 0.957–1.065; P = 0.730). A high-reliability FiO₂ sensitivity analysis restricted FiO₂ to the same blood-gas record or a ventilator setting recorded within 60 minutes of PaO₂ measurement and also yielded a null result.

Changes in the revised submission: Methods—Covariates and Statistical analysis; Results—Sensitivity analyses; Supplementary Tables S1–S3.

Comment 4. Outcome underascertainment and differential detection bias

Reviewer comment: The reviewer requested the total number of thrombotic events and discussion of possible underascertainment and differential imaging or documentation.

Response: We agree. The revised primary outcome occurred in 195 of 11,277 admissions (1.73%): 32 events in the PaO₂ ≤80-mmHg group, 29 in the 81–100-mmHg group, and 134 in the >100-mmHg group. We removed the term confirmed catheter-related thrombosis. The primary outcome is now restricted to documented acute upper-body deep or central venous thrombosis, whereas superficial and anatomically unspecified diagnoses are excluded.

A broader outcome included 282 events, and a strict laterality-compatible deep or central venous outcome included 69 events; both yielded null adjusted PaO₂ estimates. For the strict definition, a right-sided diagnosis required a right-sided qualifying catheter, a left-sided diagnosis required a left-sided qualifying catheter, and a bilateral diagnosis was considered compatible; diagnoses with unspecified laterality were excluded. We now explicitly acknowledge that asymptomatic or clinically unsuspected events may have been missed and that outcome detection may have varied according to diagnostic testing and documentation. If patients with greater illness severity or lower PaO₂ were imaged more frequently, differential ascertainment could have contributed to the apparent inverse association in the less-adjusted models.

Changes in the revised submission: Methods—Outcome definitions; Results—event counts and alternative outcomes; Table 1A; Supplementary Tables S3 and S10; Discussion—limitations.

Comment 5. A single PaO₂ value cannot capture dynamic exposure; a time-varying model would be preferable

Reviewer comment: The reviewer emphasized that one baseline value does not represent oxygenation over the ICU stay and raised concern about differing exposure windows and time-related bias.

Response: We agree that a single early PaO₂ measurement cannot represent cumulative or time-varying oxygen exposure. The revised research question is explicitly limited to the association of one early post-insertion measurement with a hospitalization-level documented outcome. We do not interpret the selected value as a measure of cumulative hypoxemia or hyperoxemia.

A time-varying Cox or competing-risk model could not be implemented validly because reliable thrombus-onset timestamps were unavailable. The revised Discussion explicitly acknowledges the single-measurement limitation, temporal uncertainty, selection related to requiring an eligible arterial blood-gas measurement, and the inability to perform valid time-to-event analyses. Future studies should use repeated or time-weighted oxygenation measurements and accurately timed, imaging-confirmed outcomes.

Changes in the revised submission: Methods—Exposure assessment; Discussion—limitations; Conclusion.

Comment 6. Improve STROBE reporting, including missing data, event counts, and model performance

Reviewer comment: The reviewer requested the number excluded for missing variables, comparison of missing cases, event counts by PaO₂ group, and calibration/discrimination information.

Response: Covariate missingness is now handled using multiple imputation rather than excluding admissions solely for missing prespecified covariates. Supplementary Table S6 reports observed and missing counts, missingness percentages, imputation methods, and imputation diagnostics. A complete-case sensitivity analysis included 8,906 admissions and 139 events.

Event counts by PaO₂ category are reported in the Results and Table 1A. Supplementary Table S5 reports the mean apparent AUC (0.7834), mean apparent Brier score (0.0166), generalized variance inflation factors, leverage, Cook’s distance, DFBETA, and an influence analysis. Supplementary Figure S2 presents apparent calibration and is explicitly described as uncorrected for optimism and not as internal or external validation. A completed STROBE checklist is submitted separately.

Changes in the revised submission: Fig 1; Results; Tables 1A–1B; Supplementary Tables S5–S6; Supplementary Figure S2; S1 Checklist.

Comment 7. The biological rationale relied too heavily on COVID-19 cohorts

Reviewer comment: The reviewer requested acknowledgment that COVID-19-associated coagulopathy may not generalize to an unselected ICU population and recommended broader evidence.

Response: We agree. The biological rationale and Discussion were rewritten and no longer rely on COVID-19 cohorts. The revised text considers intermittent hypoxia in obstructive sleep apnea, environmental hypoxia, COPD-associated thromboembolism, perioperative hyperoxia, and randomized ICU oxygen-target trials. We also emphasize that these settings cannot be directly extrapolated to a single routine ICU PaO₂ measurement.

Changes in the revised submission: Introduction and Discussion.

Comment 8. Table 1 contained misaligned or internally inconsistent cells

Reviewer comment: The reviewer requested verification of values, units, and decimal places and improvement of table alignment.

Response: Table 1 was reformatted, and the values, denominators, units, and decimal places were checked. To improve readability, the three pairwise standardized mean-difference columns were removed from the main Table 1A and transferred to Supplementary Table S9.

Changes in the revised submission: Reformatted Tables 1A–1B; new Supplementary Table S9.

Comment 9. Standardize PO₂ and PaO₂ notation

Reviewer comment: The reviewer requested consistent use of PaO₂ for arterial oxygen tension.

Response: The terminology has been standardized to PaO₂ throughout the title, abstract, main text, tables, figure captions, and supporting information.

Changes in the revised submission: All manuscript and supporting-information files.

Comment 10. Figure 2 should report the sample size and show the PaO₂ distribution

Reviewer comment: The reviewer requested the number of observations and a visual indication of the exposure distribution because of wide confidence intervals at the extremes.

Response: The revised restricted cubic spline figure includes rug marks showing the observed PaO₂ distribution and displays the 95% confidence interval across the plotted range. The caption now states that the analysis included 11,277 hospital admissions and 195 outcome events. Supplementary Table S11 also reports the overall PaO₂ distribution and upper-tail counts.

Changes in the revised submission: Revised Fig 2 and caption; Supplementary Table S11.

Reviewer #4

Major Comment 1. Fundamental concern regarding outcome ascertainment

Reviewer comment: The reviewer requested the exact ICD codes, validation strategy, sensitivity and specificity, anatomical restriction, and an explanation of how unrelated DVTs were excluded.

Response: We agree that the previous terminology overstated what could be established from the available data. The revised outcome is not described as confirmed catheter-related thrombosis. It is now defined as documented acute upper-body deep or central venous thrombosis during hospitalization. The revised algorithm uses the MIMIC-IV diagnoses_icd table and no longer relies on clinical notes.

The primary definition includes acute ICD-9-CM and ICD-10-CM codes for deep upper-extremity, axillary, subclavian, and internal jugular venous thrombosis. Superficial, chronic, and anatomically unspecified diagnoses are excluded from the primary definition. A broader acute outcome and a strict laterality-compatible deep or central venous outcome were evaluated as sensitivity analyses. For the strict definition, right-sided diagnoses were required to match a right-sided qualifying catheter, left-sided diagnoses were required to match a left-sided qualifying catheter, bilateral diagnoses were considered compatible, and diagnoses with unspecified laterality were excluded. The complete code list and classification rules are provided in Supplementary Table S10.

No chart-adjudicated or imaging-adjudicated reference standard was available; therefore, we cannot estimate or claim the sensitivity or specificity of the coding algorithm. We also cannot exclude every non-catheter-related upper-body thrombosis or establish catheter causation. These limitations are now stated explicitly.

Changes in the revised submission: Methods—Outcome definitions; Supplementary Tables S3 and S10; Discussion—limitations.

Major Comment 2. The exposure definition based on a single first PaO₂ measurement was problematic

Reviewer comment: The reviewer asked how soon PaO₂ was measured, whether thrombosis may already have been present, and why time-weighted, maximum, minimum, or cumulative measures were not used.

Response: The exposure is no longer the first PaO₂ recorded during the ICU stay. It is the first PaO₂ obtained at or after qualifying catheter insertion, within 24 hours, during the corresponding ICU stay, and while the catheter remained active. This definition provides a reproducible early post-insertion measurement and avoids using later measurements that may be increasingly affected by the subsequent ICU course.

We agree that the selected value remains a single measurement and cannot represent time-weighted, maximum, minimum, or cumulative oxygen exposure. Reliable thrombus-onset timestamps were unavailable, so repeated oxygen measurements could not be aligned with event onset, and pre-existing thrombosis at the time of the selected measurement cannot be excluded in every admission. These limitations are now stated explicitly. The timing from ICU admission and catheter insertion to the selected PaO₂ is reported in the Results and Supplementary Table S11.

Changes in the revised submission: Methods—Exposure assessment and Outcome definitions; Results—timing description; Supplementary Table S11; Discussion—limitations.

Major Comment 3. Temporal relationship remained unclear; Cox or competing-risk analysis was suggested

Reviewer comment: The reviewer requested catheter insertion time, PaO₂ time, and thrombosis time and suggested time-to-event methods.

Response: Exact catheter insertion and selected PaO₂ times were available and are now linked explicitly. The interval from catheter insertion to PaO₂ was incorporated into the adjusted models, and timing by outcome status is reported. However, thrombosis was available only as a hospitalization-level diagnosis without a sufficiently reliable onset time. Using a diagnosis-code date as biological onset would create false precision.

We therefore retained hospitalization-level logistic regression, clearly limited the estimand to an association with a documented hospitalization-level outcome, and removed claims that PaO₂ preceded thrombus formation. We now state explicitly that valid Cox, time-to-event, and competing-risk analyses could not be performed.

Changes in the revised submission: Methods—Exposure assessment and Outcome definitions; Results; Supplementary Table S11; Discussion—limitations.

Major Comment 4. Severe confounding by illness severity; propensity-score or weighting methods were recommended

Reviewer comment: The reviewer noted major baseline differences and recommended propensity-score matching, IPTW, or doubly robust analyses.

Response: We agree that illness severity and treatment context are important sources of confounding. The revised fully adjusted model includes age, sex, Charlson Comorbidity Index, hemoglobin, platelet count, lactate, catheter type, catheter laterality, multiple active catheters, the interval from catheter insertion to PaO₂ measurement, FiO₂, respiratory-support status, and the non-respiratory component of the first-day SOFA score. The attenuation of the inverse estimate after catheter, timing, oxygenation, and severity variables were added is reported transparently.

We considered the suggested propensity-score methods. However, PaO₂ is a continuous physiologic measurement rather than an assigned binary treatment. Conventional matching or binary-treatment IPTW would require an arbitrary PaO₂ cutoff, discard exposure information, and change the estimand. These approaches would also not resolve unavailable thrombus-onset times or unmeasured confounding. We therefore retained prespecified continuous and categorical analyses with sequential multivariable adjustment, multiple imputation, patient-clustered robust standard errors, nonlinear analysis, and extensive sensitivity analyses. The manuscript consistently interprets all estimates as associations rather than causal effects.

Changes in the revised submission: Methods—Covariates and Statistical analysis; Results—sequential and module-specific models; Discussion—interpretation and limitations.

Major Comment 5. Potential reverse causality

Reviewer comment: The reviewer suggested that illness severity could cause both hypoxemia and thrombosis rather than PaO₂ causing thrombosis and requested more cautious interpretation.

Response: We agree. The revised models account for available measures of illness severity, respiratory treatment, laboratory status, catheter characteristics, and measurement timing. The inverse estimate observed in the unadjusted and minimally adjusted models moved to the null after these variables were included. We now interpret this attenuation as evidence that the crude association was sensitive to clinical and treatment context, not as evidence of an independent or causal effect of PaO₂.

Because thrombus-onset time was unavailable, reverse temporal ordering cannot be excluded. This limitation is stated explicitly, and all causal terminology has been removed.

Changes in the revised submission: Results—sequential and module-specific analyses; Discussion—interpretation and limitations; Conclusion.

Major Comment 6. The threshold analysis was statistically fragile

Reviewer comment: The reviewer questioned the biological plausibility and statistical stability of the 384-mmHg threshold, noting the likely sparsity of observations in the upper tail and the non-estimable estimate above the breakpoint.

Response: We agree. The segmented regression analysis, the 384-mmHg breakpoint, and all threshold-based claims have been removed from the title, abstract, Methods, Results, tables, figures, Discussion, and Conclusion. In an additional post hoc descriptive summary, 647 admissions with 10 primary-outcome events had PaO₂ values greater than 384 mmHg, 490 admissions with 5 events had values greater than 400 mmHg, and 111 admissions with 2 events had values above the 99th percentile (>461 mmHg). These counts confirmed the sparsity of the upper tail. Potential nonlinearity is now evaluated using restricted cubic splines, which showed no evidence of an overall association (P = 0.186) or nonlinearity (P = 0.109). Sensitivity analyses excluding PaO₂ values greater than 400 mmHg and values above the 99th percentile also yielded null estimates.

Changes in the revised submission: Threshold analysis removed throughout; Results—Sensitivity, nonlinear, and subgroup analyses; Fig 2; Supplementary Tables S1, S2, and S11.

Major Comment 7. Inconsistency between continuous and categorical analyses

Reviewer comment: The reviewer noted discordance between a near-null continuous estimate and substantially lower categorical odds.

Response: The discrepancy is no longer present after cohort reconstruction and comprehensive adjustment. In the fully adjusted model, the continuous estimate was OR 1.008 (95% CI, 0.989–1.027) per 10-mmHg increase. The categorical estimates were OR 0.900 (95% CI, 0.531–1.526) for 81–100 versus ≤80 mmHg and OR 0.989 (95% CI, 0.652–1.502) for >100 versus ≤80 mmHg. Restricted cubic spline analysis also showed no overall or nonlinear association.

Changes in the revised submission: Results; Tables 2–3; Fig 2.

Major Comment 8. Anticoagulation may introduce overadjustment or collider bias

Reviewer comment: The reviewer questioned adjustment for anticoagulant administration because its timing and indication may be related to perceived thrombotic risk.

Response: We agree that time-varying anticoagulation is difficult to position analytically when indication and timing relative to exposure and outcome are uncertain. Anticoagulant administration is not included in the revised primary adjustment model. The Discussion now acknowledges that detailed time-varying information on thromboprophylaxis and therapeutic anticoagulation was limited, leaving potential residual confounding. We did not make causal claims about the role of anticoagulation.

Changes in the revised submission: Methods—Covariate specification; Discussion—limitations.

Major Comment 9. Important catheter and clinical risk factors were missing

Reviewer comment: The reviewer listed catheter type, lumens, size, dwell time, tip position, ventilation, cancer, sepsis, platelet count, D-dimer, and coagulation variables.

Response: The revised analysis incorporates catheter type, catheter laterality, multiple active qualifying catheters, platelet count, hemoglobin, lactate, FiO₂, respiratory-support status, Charlson Comorbidity Index, and non-respiratory SOFA. PICC status was also evaluated in subgroup analysis.

Nevertheless, catheter material, exact diameter, catheter-to-vein ratio, exact tip position, insertion technique, operator-related factors, complete dwell time before thrombus onset, D-dimer, detailed coagulation measures, and detailed time-varying anticoagulation were unavailable or could not be measured reliably. These limitations are now stated explicitly. The associations of individual covariates are presented as secondary, hypothesis-generating findings and are not described as independent risk factors.

Changes in the revised submission: Methods—Covariates; Results—secondary covariate associations; Supplementary Table S8; Discussion—limitations.

Major Comment 10. Causal wording such as “protective effect” and “risk reduction” should be removed

Reviewer comment: The reviewer requested consistent noncausal terminology.

Response: We agree. Terms including protective effect, risk reduction, and higher PaO₂ reduces thrombosis have been removed. The revised manuscript consistently uses association, odds, and documented outcome. The Conclusion states that the findings do not support PaO₂-based thrombosis risk stratification or modification of oxygen therapy for thrombosis prevention.

Changes in the revised submission: Title; Abstract; Results; Discussion; Conclusion.

Minor Comment 1. Use one study-design term consistently

Reviewer comment: The reviewer noted inconsistent use of retrospective cohort and cross-sectional terminology.

Response: The study is now consistently described as a retrospective observational cohort study. Cross-sectional terminology has been removed.

Changes in the revised submission: Title, Abstract, Methods, and Discussion.

Minor Comment 2. Standardize PO₂ and PaO₂

Reviewer comment: The reviewer requested consistent arterial oxygen terminology.

Response: PaO₂ is now used consistently throughout the manuscript and supporting information.

Changes in the revised submission: All files.

Minor Comment 3. Clarify APSIII versus APACHE III

Reviewer comment: The reviewer requested clarification regarding the severity score used in the models.

Response: Neither APSIII nor APACHE III is included in the revised primary models. Illness severity is represented by the non-respiratory component of the first-day SOFA score, which is consistently defined and reported.

Changes in the revised submission: Methods—Covariates and Statistical analysis; tables and supporting information.

Minor Comment 4. Report continuous ORs per a meaningful increment

Reviewer comment: The reviewer recommended reporting the continuous association per 10 mmHg or per standard deviation.

Response: All continuous PaO₂ associations are now reported per 10-mmHg increase.

Changes in the revised submission: Abstract; Methods; Results; tables; figures; supporting information.

Minor Comment 5. Report handling of missing data

Reviewer comment: The reviewer requested the number and percentages of missing variables and consideration of multiple imputation.

Response: Missing values in prespecified model covariates were handled using multiple imputation by chained equations with 20 imputed datasets. Supplementary Table S6 reports observed and missing counts, missingness percentages, imputation methods, and diagnostics. Supplementary Figures S3 and S4 provide graphical imputation diagnostics. A complete-case sensitivity analysis was also performed.

Changes in the revised submission: Methods—Statistical analysis; Supplementary Table S6; Supplementary Figures S3–S4.

Minor Comment 6. Provide the MIMIC-IV version and reproducible extraction and analysis code

Reviewer comment: The reviewer requested the exact database version, SQL extraction strategy, and a reproducible code repository or package.

Response: Thank you for this important suggestion. The database version is now specified as MIMIC-IV v2.2, and the Methods and Supporting Information have been expanded to describe the cohort-construction logic, eligibility criteria, exposure window, outcome definitions, ICD code list, covariate definitions, missing-data procedures, model specifications, and sensitivity analyses. We agree that sharing well-documented analysis code would further improve reproducibility. The reported results were generated using the final analytic workflow; however, the relevant code currently remains distributed across several working scripts and has not yet been consolidated and documented as a single publication-ready repository. To avoid providing an incomplete or potentially misleading package, we have not included executable code with this revision. If the Editor considers code submission essential for further evaluation, we would be grateful for the opportunity to prepare and provide a fully consolidated and documented package. Patient-level MIMIC-IV data are not redistributed because they are subject to the PhysioNet Credentialed Health Data Use Agreement; qualified researchers may obtain the data through the same credentialing process, and the authors had no special access privileges.

Changes in the revised submission: Methods—Data source and Statistical analysis; Supporting Information; Data Availability Statement in the submission system.

Closing statement

We again thank the Academic Editor and Reviewers for their detailed and constructive comments. The review prompted a fundamental reconstruction of the study, removal of the unstable threshold claim, more complete adjustment for catheter and oxygenation context, and a substantially more cautious interpretation. We believe that the revised manuscript now addresses the concerns transparently and reports the null fully adjusted result without causal or threshold-based claims.

Sincerely,

Qing Cui, on behalf of all auth

Attachment

Submitted filename: renamed_f23a0.docx

pone.0359719.s006.docx^{(41.2KB, docx)}

PLoS One. doi: 10.1371/journal.pone.0359719.r005

Decision Letter 2

Giovanni Giordano

10 Sep 2026

Dear Dr. Cui,

Thank you for submitting your manuscript to PLOS One. After careful consideration, we feel that it has merit but does not fully meet PLOS One’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process.

Please submit your revised manuscript by Oct 25 2026 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at plosone@plos.org. When you're ready to submit your revision, log on to https://www.editorialmanager.com/pone/ and select the 'Submissions Needing Revision' folder to locate your manuscript file.

A letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.
A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.
An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'.

If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: https://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory-protocols. Additionally, PLOS One offers an option for publishing peer-reviewed Lab Protocol articles, which describe protocols hosted on protocols.io. Read more information on sharing protocols at https://plos.org/protocols?utm_medium=editorial-email&utm_source=authorletters&utm_campaign=protocols.

We look forward to receiving your revised manuscript.

Kind regards,

Giovanni Giordano

Academic Editor

PLOS One

Journal Requirements:

1. If the reviewer comments include a recommendation to cite specific previously published works, please review and evaluate these publications to determine whether they are relevant and should be cited. There is no requirement to cite these works unless the editor has indicated otherwise.

2. Please review your reference list to ensure that it is complete and correct. If you have cited papers that have been retracted, please include the rationale for doing so in the manuscript text, or remove these references and replace them with relevant current references. Any changes to the reference list should be mentioned in the rebuttal letter that accompanies your revised manuscript. If you need to cite a retracted article, indicate the article’s retracted status in the References list and also include a citation and full reference for the retraction notice.

Additional Editor Comments:

Dear Dr. Cui and colleagues,

Thank you for submitting the revised version of your manuscript, “Association between early post-insertion arterial oxygen tension and documented acute upper-body deep or central venous thrombosis in critically ill adults: a retrospective MIMIC-IV cohort study” (PONE-D-25-27701R2).

The manuscript has undergone substantial improvement. In particular, the reconstructed cohort, removal of the previously proposed PaO₂ threshold and protective-effect interpretation, revised outcome definition, expanded adjustment strategy, and extensive sensitivity analyses have addressed the major methodological concerns raised during the previous rounds. The revised analysis now reports a well-estimated null association between early post-insertion PaO₂ and the documented thrombotic outcome.

A small number of issues should nevertheless be addressed before the manuscript can be considered for acceptance.

1. Timing between catheter insertion and PaO₂ measurement

The interval between catheter insertion and the selected PaO₂ measurement differs substantially according to outcome status (median 45 vs 134 minutes), and the timing variable itself is associated with the documented thrombotic outcome. Although this interval has appropriately been included in the adjusted models and subgroup analyses according to timing showed no significant interaction, these analyses do not fully address the potential selection and temporal ambiguity introduced by the exposure window.

Please therefore provide an additional sensitivity analysis restricting the cohort to a narrow early post-insertion measurement window, preferably PaO₂ measurements obtained within 1 hour after catheter insertion. Please report the number of included admissions and events and the corresponding adjusted PaO₂ estimate. If another narrow window is considered more appropriate based on the data distribution, please justify the choice.

The related limitation should also be described in terms of temporal uncertainty and possible selection related to measurement timing. Because the exact thrombus-onset time is unavailable, the manuscript should avoid implying that a conventional immortal-time bias can be definitively established.

2. Role of FiO₂ in the adjustment strategy

The Reviewer raised the possibility that adjustment for FiO₂ could contribute to over-adjustment because FiO₂ is a determinant of the measured PaO₂. However, the manuscript already presents Model 2, which does not include FiO₂, respiratory-support status, or non-respiratory SOFA and still shows no association between PaO₂ and the outcome (OR 1.003, 95% CI 0.984–1.022; P = 0.774).

No additional model therefore appears necessary. Please make this point explicit in the response letter and briefly acknowledge in the manuscript that the interpretation of adjustment for FiO₂ depends on the estimand and that the null finding was already present before inclusion of the oxygenation/severity variables. The PaO₂/FiO₂ analysis should continue to be described as an alternative-exposure analysis rather than as an adjustment variable in the primary model.

3. Study-design terminology and temporal ordering

The manuscript is currently described as a retrospective observational cohort study. Given that the exact onset of thrombosis cannot be established and that it cannot be confirmed that the selected PaO₂ measurement preceded thrombus formation in every admission, please further clarify this temporal limitation in the Methods and/or Limitations.

The current study-design terminology may be retained if adequately qualified, although “retrospective observational study” would also be acceptable if the Authors prefer a more conservative formulation.

4. Outcome ascertainment

The manuscript already appropriately acknowledges the limitations of ICD-based outcome ascertainment, including the lack of standardized prospective imaging, unknown sensitivity and specificity of the coding algorithm, possible under-detection of asymptomatic events, and differential ascertainment.

Please add a brief explicit statement noting that the observed event rate may therefore be lower than rates reported in studies using systematic imaging surveillance.

Overall, the revised manuscript is now close to being suitable for publication. Please submit a final revised version addressing the points above together with a concise point-by-point response.

Kind regards,

Giovanni Giordano

[Note: HTML markup is below. Please do not edit.]

Reviewers' comments:

Reviewer's Responses to Questions

Comments to the Author

Reviewer #3: (No Response)

Reviewer #4: All comments have been addressed

**********

2. Is the manuscript technically sound, and do the data support the conclusions??>

Reviewer #3: Partly

Reviewer #4: (No Response)

**********

3. Has the statistical analysis been performed appropriately and rigorously? -->?>

Reviewer #3: Yes

Reviewer #4: (No Response)

**********

4. Have the authors made all data underlying the findings in their manuscript fully available??>

The PLOS Data policy

Reviewer #3: Yes

Reviewer #4: (No Response)

**********

5. Is the manuscript presented in an intelligible fashion and written in standard English??>

Reviewer #3: Yes

Reviewer #4: (No Response)

**********

Reviewer #3: Dear Editor and Authors,

Thank you for the opportunity to review this manuscript entitled "Association between early post-insertion arterial oxygen tension and documented acute upper-body deep or central venous thrombosis in critically ill adults: a retrospective MIMIC-IV cohort study" (Manuscript ID: PONE-D-25-27701). This is the third review of the manuscript. I am grateful to the authors for the substantial work invested in this revision.

In this study the authors use the MIMIC-IV database to examine whether early post-insertion arterial oxygen tension (PaO2) is associated with documented acute upper-body deep or central venous thrombosis in critically ill adults. Compared with the previous versions, the cohort has been entirely reconstructed (11,277 admissions, 10,789 patients, 195 events, 1.73%), the previously reported 384-mmHg threshold and all "protective" claims have been removed, and the analysis now yields a well-estimated null association (fully adjusted OR 1.008 per 10 mmHg, 95% CI 0.989-1.027, P=0.43), supported by categorical models, a PaO2/FiO2 analysis, restricted cubic splines, multiple imputation, and model-performance and influence diagnostics. I commend the authors; this revision resolves the complete-separation problem and the over-interpretation that affected the original submission, and the manuscript is now scientifically sound and, for a null finding, clearly reportable.

A few points should still be addressed to strengthen the validity and transparency of the work before publication.

MAJOR REVISIONS

1. The interval between catheter insertion and the first PaO2 measurement differs systematically between patients with and without the outcome (approximately 134 vs 45 minutes) and is itself a strong predictor in the models (OR 1.33 per unit, P<0.001). This raises a concern of immortal-time-like/selection bias that adjustment alone may not fully remove, because patients who develop the outcome have a different temporal profile. The authors should discuss this explicitly and provide a sensitivity analysis restricting the cohort to a narrow, homogeneous insertion-to-measurement window (e.g., excluding extreme intervals) to show that the null result is robust to this design feature.

2. FiO2 (and the PaO2/FiO2 ratio) is a determinant of the exposure rather than a classical confounder. Adjusting for it may attenuate any true association toward the null through over-adjustment effects. Because the primary conclusion is a null association, this deserves an explicit acknowledgment, and the authors should present a model that excludes FiO2 and the PaO2/FiO2 term, so that readers can judge whether the null persists without conditioning on a determinant of the exposure.

MINOR REVISIONS

3. Given that the exposure is captured at a single time point and that the design cannot establish that exposure precedes the outcome, the label "cohort study" is somewhat generous. The authors should either soften this terminology and add a brief statement in the Methods/Limitations clarifying the temporal relationship between exposure measurement and outcome ascertainment.

4. Outcome ascertainment relies on administrative ICD coding, which is known to under-capture asymptomatic thrombosis. The authors have honestly disclosed this and added broad and strict outcome sensitivity analyses, both null. I would only ask that the Limitations section state plainly that residual outcome misclassification remains possible and that the observed event rate is lower than rates reported with systematic screening.

In summary, the authors have thoroughly and honestly addressed the concerns raised in the previous rounds; the reconstructed analysis is rigorous and the null finding is appropriately presented. The remaining requests concern the discussion of temporal/selection bias, transparency around adjustment for a determinant of the exposure, and minor clarifications of terminology and outcome ascertainment. I believe the manuscript can be accepted after these points are addressed.

Best regards,

The reviewer

Reviewer #4: (No Response)

**********

what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy

Reviewer #3: No

Reviewer #4: No

**********

To ensure your figures meet our technical requirements, please review our figure guidelines: https://journals.plos.org/plosone/s/figures

You may also use PLOS’s free figure tool, NAAS, to help you prepare publication quality figures: https://journals.plos.org/plosone/s/figures#loc-tools-for-figure-preparation.

NAAS will assess whether your figures meet our technical requirements by comparing each figure against our figure specifications.

Attachment

Submitted filename: PONE-D~2.DOC

pone.0359719.s005.DOC^{(76KB, DOC)}

PLoS One. 2026 Oct 5;21(10):e0359719. doi: 10.1371/journal.pone.0359719.r006

Author response to Decision Letter 3

12 Sep 2026

Dear Academic Editor and Reviewers,

Thank you for your careful and constructive comments. We have addressed all remaining points raised by the Academic Editor and Reviewer #3. The revision includes an additional sensitivity analysis restricted to PaO₂ measurements obtained within 60 minutes after catheter insertion, clarification of the role of FiO₂ in the adjustment strategy, further qualification of the temporal relationship between PaO₂ measurement and thrombosis ascertainment, and additional discussion of limitations related to ICD-based outcome ascertainment.

A detailed point-by-point response to each comment is provided in the separately uploaded Response to Reviewers document. All corresponding changes have been incorporated into the revised manuscript and Supporting Information.

We sincerely thank the Academic Editor and Reviewers for their guidance.

Attachment

Submitted filename: Response_to_Reviewers_auresp_3.docx

pone.0359719.s007.docx^{(37.2KB, docx)}

PLoS One. doi: 10.1371/journal.pone.0359719.r007

Decision Letter 3

Giovanni Giordano

17 Sep 2026

Association between early post-insertion arterial oxygen tension and documented acute upper-body deep or central venous thrombosis in critically ill adults: a retrospective MIMIC-IV cohort study

PONE-D-25-27701R3

Dear Dr. Cui,

We’re pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it meets all outstanding technical requirements.

Within one week, you’ll receive an e-mail detailing the required amendments. When these have been addressed, you’ll receive a formal acceptance letter and your manuscript will be scheduled for publication.

An invoice will be generated when your article is formally accepted. Please note, if your institution has a publishing partnership with PLOS and your article meets the relevant criteria, all or part of your publication costs will be covered. Please make sure your user information is up-to-date by logging into Editorial Manager at Editorial Manager® and clicking the ‘Update My Information' link at the top of the page. For questions related to billing, please contact billing support.

If your institution or institutions have a press office, please notify them about your upcoming paper to help maximize its impact. If they’ll be preparing press materials, please inform our press team as soon as possible -- no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact onepress@plos.org.

Kind regards,

Giovanni Giordano

Academic Editor

PLOS One

Additional Editor Comments (optional):

Reviewers' comments:

PLoS One. doi: 10.1371/journal.pone.0359719.r008

Acceptance letter

Giovanni Giordano

PONE-D-25-27701R3

PLOS One

Dear Dr. Cui,

I'm pleased to inform you that your manuscript has been deemed suitable for publication in PLOS One. Congratulations! Your manuscript is now being handed over to our production team.

At this stage, our production department will prepare your paper for publication. This includes ensuring the following:

* All references, tables, and figures are properly cited

* All relevant supporting information is included in the manuscript submission,

* There are no issues that prevent the paper from being properly typeset

You will receive further instructions from the production team, including instructions on how to review your proof when it is ready. Please keep in mind that we are working through a large volume of accepted articles, so please give us a few days to review your paper and let you know the next and final steps.

Lastly, if your institution or institutions have a press office, please let them know about your upcoming paper now to help maximize its impact. If they'll be preparing press materials, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact onepress@plos.org.

You will receive an invoice from PLOS for your publication fee after your manuscript has reached the completed accept phase. If you receive an email requesting payment before acceptance or for any other service, this may be a phishing scheme. Learn how to identify phishing emails and protect your accounts at https://explore.plos.org/phishing.

If we can help with anything else, please email us at customercare@plos.org.

Thank you for submitting your work to PLOS One and supporting open access.

Kind regards,

PLOS One Editorial Office Staff

on behalf of

Dr. Giovanni Giordano

Academic Editor

PLOS One

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

S1 File. Supplementary Information.

Supplementary methods, Tables S1–S12, and Figures S1–S4.

(DOCX)

pone.0359719.s001.docx^{(8.9MB, docx)}

Attachment

Submitted filename: Response to Reviewers.docx

pone.0359719.s004.docx^{(37.2KB, docx)}

Attachment

Submitted filename: PONE-D-25-27701R1_Reviewer_Report.docx

pone.0359719.s003.docx^{(28KB, docx)}

Attachment

Submitted filename: renamed_f23a0.docx

pone.0359719.s006.docx^{(41.2KB, docx)}

Attachment

Submitted filename: PONE-D~2.DOC

pone.0359719.s005.DOC^{(76KB, DOC)}

Attachment

Submitted filename: Response_to_Reviewers_auresp_3.docx

pone.0359719.s007.docx^{(37.2KB, docx)}

Data Availability Statement

[pone.0359719.ref001] 1.Takashima M, Schults J, Mihala G, Corley A, Ullman A. Complication and Failures of Central Vascular Access Device in Adult Critical Care Settings. Crit Care Med. 2018;46(12):1998–2009. doi: 10.1097/CCM.0000000000003370 [DOI] [PubMed] [Google Scholar]

[pone.0359719.ref002] 2.Puri A, Dai H, Giri M, Wu C, Huang H, Zhao Q. The incidence and risk of venous thromboembolism associated with peripherally inserted central venous catheters in hospitalized patients: A systematic review and meta-analysis. Front Cardiovasc Med. 2022;9:917572. doi: 10.3389/fcvm.2022.917572 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pone.0359719.ref003] 3.Mavrovounis G, Mermiri M, Chatzis DG, Pantazopoulos I. Peripherally Inserted Central Catheter lines for Intensive Care Unit and onco-hematologic patients: A systematic review and meta-analysis. Heart Lung. 2020;49(6):922–33. doi: 10.1016/j.hrtlng.2020.07.008 [DOI] [PubMed] [Google Scholar]

[pone.0359719.ref004] 4.Thiyagarajah K, Ellingwood L, Endres K, Hegazi A, Radford J, Iansavitchene A, et al. Post-thrombotic syndrome and recurrent thromboembolism in patients with upper extremity deep vein thrombosis: A systematic review and meta-analysis. Thromb Res. 2019;174:34–9. doi: 10.1016/j.thromres.2018.12.012 [DOI] [PubMed] [Google Scholar]

[pone.0359719.ref005] 5.Tran A, Fernando SM, Rochwerg B, Cook DJ, Crowther MA, Fowler RA, et al. Prognostic Factors Associated With Development of Venous Thromboembolism in Critically Ill Patients-A Systematic Review and Meta-Analysis. Crit Care Med. 2022;50(4):e370–81. doi: 10.1097/CCM.0000000000005382 [DOI] [PubMed] [Google Scholar]

[pone.0359719.ref006] 6.Balsorano P, Virgili G, Villa G, Pittiruti M, Romagnoli S, De Gaudio AR, et al. Peripherally inserted central catheter-related thrombosis rate in modern vascular access era-when insertion technique matters: A systematic review and meta-analysis. J Vasc Access. 2020;21(1):45–54. doi: 10.1177/1129729819852203 [DOI] [PubMed] [Google Scholar]

[pone.0359719.ref007] 7.Raj R, Paturi A, Ahmed MA, Thomas SE, Gorantla VR. Obstructive Sleep Apnea as a Risk Factor for Venous Thromboembolism: A Systematic Review. Cureus. 2022;14(2):e22729. doi: 10.7759/cureus.22729 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pone.0359719.ref008] 8.Treml B, Wallner B, Blank C, Fries D, Schobersberger W. The Influence of Environmental Hypoxia on Hemostasis-A Systematic Review. Front Cardiovasc Med. 2022;9:813550. doi: 10.3389/fcvm.2022.813550 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pone.0359719.ref009] 9.Huang J, Wu B, Zhang Z, Wang L. Risk Factors Associated With Venous Thromboembolism in Individuals With COPD: A Systematic Review and Meta-analysis. Clin Appl Thromb Hemost. 2025;31:10760296251388989. doi: 10.1177/10760296251388989 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pone.0359719.ref010] 10.Fu X, Zhong Y, Xu W, Ju J, Yu M, Ge M, et al. The prevalence and clinical features of pulmonary embolism in patients with AE-COPD: A meta-analysis and systematic review. PLoS One. 2021;16(9):e0256480. doi: 10.1371/journal.pone.0256480 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pone.0359719.ref011] 11.Oldman AH, Martin DS, Feelisch M, Grocott MPW, Cumpstey AF. Effects of perioperative oxygen concentration on oxidative stress in adult surgical patients: a systematic review. Br J Anaesth. 2021;126(3):622–32. doi: 10.1016/j.bja.2020.09.050 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pone.0359719.ref012] 12.Cumpstey AF, Oldman AH, Martin DS, Smith A, Grocott MPW. Oxygen Targets During Mechanical Ventilation in the ICU: A Systematic Review and Meta-Analysis. Crit Care Explor. 2022;4(4):e0652. doi: 10.1097/CCE.0000000000000652 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pone.0359719.ref013] 13.ICU-ROX Investigators and the Australian and New Zealand Intensive Care Society Clinical Trials Group, Mackle D, Bellomo R, Bailey M, Beasley R, Deane A, et al. Conservative Oxygen Therapy during Mechanical Ventilation in the ICU. N Engl J Med. 2020;382(11):989–98. doi: 10.1056/NEJMoa1903297 [DOI] [PubMed] [Google Scholar]

[pone.0359719.ref014] 14.Schjørring OL, Klitgaard TL, Perner A, Wetterslev J, Lange T, Siegemund M, et al. Lower or Higher Oxygenation Targets for Acute Hypoxemic Respiratory Failure. N Engl J Med. 2021;384(14):1301–11. doi: 10.1056/NEJMoa2032510 [DOI] [PubMed] [Google Scholar]

[pone.0359719.ref015] 15.Martin DS, Gould DW, Shahid T, Doidge JC, Cowden A, Sadique Z, et al. Conservative Oxygen Therapy in Mechanically Ventilated Critically Ill Adult Patients: The UK-ROX Randomized Clinical Trial. JAMA. 2025;334(5):398–408. doi: 10.1001/jama.2025.9663 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pone.0359719.ref016] 16.Johnson AEW, Bulgarelli L, Shen L, Gayles A, Shammout A, Horng S, et al. MIMIC-IV, a freely accessible electronic health record dataset. Sci Data. 2023;10(1):1. doi: 10.1038/s41597-022-01899-x [DOI] [PMC free article] [PubMed] [Google Scholar]

[pone.0359719.ref017] 17.White IR, Royston P, Wood AM. Multiple imputation using chained equations: Issues and guidance for practice. Statistics in Medicine. 2010;30(4):377–99. doi: 10.1002/sim.4067 [DOI] [PubMed] [Google Scholar]

[pone.0359719.ref018] 18.Desquilbet L, Mariotti F. Dose-response analyses using restricted cubic spline functions in public health research. Stat Med. 2010;29(9):1037–57. doi: 10.1002/sim.3841 [DOI] [PubMed] [Google Scholar]

[pone.0359719.ref019] 19.Klitgaard TL, Schjørring OL, Nielsen FM, Meyhoff CS, Perner A, Wetterslev J, et al. Higher versus lower fractions of inspired oxygen or targets of arterial oxygenation for adults admitted to the intensive care unit. Cochrane Database Syst Rev. 2023;9(9):CD012631. doi: 10.1002/14651858.CD012631.pub3 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pone.0359719.ref020] 20.Yang X, Zhou X, Qiu Y. Risk factors of venous thromboembolism in ICU patients: a systematic review and meta-analysis. Ann Med. 2026;58(1):2614222. doi: 10.1080/07853890.2026.2614222 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pone.0359719.ref021] 21.Chen P, Zhu B, Wan G, Qin L. The incidence of asymptomatic thrombosis related to peripherally inserted central catheter in adults: A systematic review and meta-analysis People’s. Nurs Open. 2021;8(5):2249–61. doi: 10.1002/nop2.811 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pone.0359719.ref022] 22.Shen Y, Chen L, Chen J, Qin J, Wang T, Wen F. Mitochondrial damage-associated molecular patterns in chronic obstructive pulmonary disease: Pathogenetic mechanism and therapeutic target. J Transl Int Med. 2022;11(4):330–40. doi: 10.2478/jtim-2022-0019 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pone.0359719.ref023] 23.Sharp R, Carr P, Childs J, Scullion A, Young M, Flynn T, et al. Catheter to vein ratio and risk of peripherally inserted central catheter (PICC)-associated thrombosis according to diagnostic group: a retrospective cohort study. BMJ Open. 2021;11(7):e045895. doi: 10.1136/bmjopen-2020-045895 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pone.0359719.ref024] 24.Piraino T, Madden M, Roberts KJ, Lamberti J, Ginier E, Strickland SL. AARC Clinical Practice Guideline: Management of Adult Patients With Oxygen in the Acute Care Setting. Respir Care. 2022;67(1):115–28. doi: 10.4187/respcare.09294 [DOI] [PubMed] [Google Scholar]

[pone.0359719.ref025] 25.Zhang N, Xu Y, Zhu LY, Wang Y, Lu Q, An R, et al. Quality appraisal of clinical guidelines for peripherally inserted central catheter-related thrombosis prophylaxis in patients: a systematic review. BMJ Open. 2024;14(11):e084330. doi: 10.1136/bmjopen-2024-084330 [DOI] [PMC free article] [PubMed] [Google Scholar]

PERMALINK

Association between early post-insertion arterial oxygen tension and documented acute upper-body deep or central venous thrombosis in critically ill adults: A retrospective MIMIC-IV cohort study

Qing Cui

Anshuai Fang

Haiyan Zhang

Qingtong Meng

Sijia Zhou

Yueming Peng

Weixiang Luo

Roles

Abstract

Introduction

Materials and methods

Study design and objective

Data source

Ethics statement

Study population

Fig 1. Flowchart of cohort selection.

Exposure assessment

Outcome definitions

Covariates

Statistical analysis

Results

Study population and baseline characteristics

Table 1. Baseline characteristics according to PaO2 category.

Table 2. Baseline characteristics according to primary-outcome status.

Association between PaO2 and the primary outcome

Table 3. Association between continuous PaO2 and the documented thrombotic outcome.

Table 4. Association between PaO2 categories and the documented thrombotic outcome.

Sensitivity, nonlinear, and subgroup analyses

Fig 2. Restricted cubic spline analysis of the association between arterial PaO2 and the documented thrombotic outcome.

Secondary associations of model covariates with the documented thrombotic outcome

Post hoc module-specific analysis

Fig 3. Changes in the estimated association between arterial PaO2 and the documented thrombotic outcome after adjustment for different covariate modules.

Discussion

Conclusion

Supporting information

Data Availability

Funding Statement

References

Decision Letter 0

Paavani Atluri

Roles

Author response to Decision Letter 1

Decision Letter 1

Giovanni Giordano

Roles

Author response to Decision Letter 2

Decision Letter 2

Giovanni Giordano

Roles

Author response to Decision Letter 3

Decision Letter 3

Giovanni Giordano

Roles

Acceptance letter

Giovanni Giordano

Roles

Associated Data

Supplementary Materials

Data Availability Statement

ACTIONS

PERMALINK

RESOURCES

Similar articles

Cited by other articles

Links to NCBI Databases

Table 1. Baseline characteristics according to PaO₂ category.

Association between PaO₂ and the primary outcome

Table 3. Association between continuous PaO₂ and the documented thrombotic outcome.

Table 4. Association between PaO₂ categories and the documented thrombotic outcome.

Fig 2. Restricted cubic spline analysis of the association between arterial PaO₂ and the documented thrombotic outcome.

Fig 3. Changes in the estimated association between arterial PaO₂ and the documented thrombotic outcome after adjustment for different covariate modules.