Association between ROX index and in-hospital mortality in critically ill patients with chronic obstructive pulmonary disease: A retrospective cohort study based on two large databases

Abstract

Objective

The ROX index, combining oxygenation and respiratory rate, is widely used to assess respiratory failure. However, its prognostic value in critically ill patients with chronic obstructive pulmonary disease (COPD) remains unclear. This study aimed to explore the association between the ROX index and mortality in ICU patients with COPD.

Methods

This retrospective cohort study included patients with COPD from two large databases: MIMIC-IV (v3.1) and eICU-CRD (v2.0). The ROX index was calculated within the first 24 hours of ICU admission. The primary outcome was in-hospital mortality, with ICU mortality and 28-day mortality as secondary outcomes. Cox regression models, Kaplan–Meier survival curves, restricted cubic spline (RCS) analysis, and subgroup analyses were used to evaluate the association between ROX and mortality.

Results

A total of 1,639 patients from the MIMIC-IV cohort and 2,170 from the eICU-CRD cohort were included. In multivariable Cox regression, a higher ROX index was independently associated with a lower risk of in-hospital mortality (MIMIC-IV: HR = 0.96, 95% CI: 0.93–0.98; eICU-CRD: HR = 0.95, 95% CI: 0.92–0.98). Similar associations were observed for ICU and 28-day mortality. RCS analysis demonstrated a linear negative correlation between ROX and the risk of death. Subgroup analyses showed consistent results across various clinical strata. The E-value analysis suggested that a considerable amount of unaccounted confounding would be necessary to invalidate the observed associations.

Conclusions

The ROX index is inversely associated with mortality in critically ill patients with COPD, with higher values indicating better prognosis. It may serve as a simple, non-invasive, and valuable tool for early risk stratification in the ICU setting.

Graphical Abstract

Keywords

ROX index COPD critical care in-hospital mortality prognostic marker

1. Introduction

Chronic obstructive pulmonary disease (COPD) is a major global health burden, affecting an estimated 391 million people worldwide in 2019, and ranking as the third most common cause of death worldwide.¹ In the United States alone, COPD affects over 16 million adults, with many more remaining undiagnosed.² The disease imposes a substantial economic and healthcare burden due to frequent hospitalizations, extended ICU stays, and elevated rates of readmission and mortality. Patients with COPD admitted to the ICU commonly present with acute exacerbations, respiratory failure, or sepsis, and experience a hospital mortality rate varying from 15% to 40%, depending on comorbidities and disease severity.³ Despite advances in ventilatory support and ICU management, early and accurate risk stratification remains a challenge in this population.

The respiratory rate-oxygenation (ROX) index, calculated as the ratio of SpO₂ to the FiO₂ and respiratory rate, is commonly utilized to predict the need for intubation in patients undergoing high-flow nasal cannula (HFNC) therapy.^4,5 Its predictive value has been validated primarily in acute hypoxemic respiratory failure (AHRF), including COVID-19 pneumonia.^6,7 Physiologically, the ROX index reflects the balance between oxygenation efficiency and respiratory workload, thus providing an integrated, dynamic snapshot of respiratory function. Unlike traditional indices such as the PaO₂/FiO₂ ratio or alveolar–arterial (A–a) oxygen gradient, the ROX index does not require arterial blood sampling, making it a non-invasive and repeatable parameter suitable for bedside monitoring and early prognostication.

However, the clinical utility of the ROX index beyond HFNC contexts remains underexplored. Patients with COPD, in particular, represent a distinct subgroup with chronic respiratory impairment, altered ventilatory mechanics, and heterogeneous responses to oxygen therapy.⁸ These pathophysiological complexities may influence the prognostic performance of the ROX index.

Therefore, the present study aimed to evaluate the association between the ROX index and in-hospital mortality among critically ill COPD patients using two large, publicly available ICU databases. We further examined the relationship between ROX and secondary outcomes, including ICU and 28-day mortality, tested the consistency of associations across clinical subgroups, and assessed robustness using E-value analysis to account for unmeasured confounding.

2. Methods

2.1. Data source

This retrospective cohort study was conducted using two large, publicly available critical care databases: the Medical Information Mart for Intensive Care IV (MIMIC-IV, version 3.1) and the eICU Collaborative Research Database (eICU-CRD, version 2.0). MIMIC-IV is a comprehensive database developed by the MIT Lab for Computational Physiology in collaboration with Beth Israel Deaconess Medical Center, containing de-identified health-related data from ICU patients admitted between 2008 and 2022. It encompasses diverse clinical data, including demographic details, physiological measurements, laboratory findings, diagnostic codes, therapeutic interventions, and patient outcomes.⁹ The eICU-CRD, developed by Philips Healthcare in partnership with the MIT Lab for Computational Physiology, encompasses data from over 200,000 ICU admissions across multiple hospitals in the United States between 2014 and 2015.¹⁰

Access to both databases was granted following the completion of the necessary training and data use agreements through the PhysioNet Credentialed Health Data Access process (certification number: 60106105). All patient data were fully de-identified, and no identifiable patient information was used in this study. This study was conducted in accordance with the principles of the Declaration of Helsinki (1975), as revised in 2024. The reporting of this study conforms to the STROBE guidelines for observational studies.¹¹

2.2. Study population

In this study, we initially identified patients with a diagnosis of COPD who were admitted to the ICU for the first time. COPD was identified using ICD-9 codes (491.20, 491.21, 491.22, and 496) and ICD-10 codes (J44, J44.0, J44.1, and J44.9). The detailed ICD codes are provided in Table S1. Repeated ICU admissions across hospitalizations were identified using the patient identifier (subject_id in MIMIC-IV and patientunitstayid in eICU-CRD) and excluded from the analysis by retaining only the first ICU stay for each patient. The following exclusion criteria were then applied: (1)age<18; (2)ICU stay< 24 hours; (3) a documented diagnosis of interstitial lung disease (ILD) or lung cancer; and (4) missing data required for calculation of the ROX index, including respiratory rate, pulse oximetry (SpO₂), and fraction of inspired oxygen (FiO₂). After applying these criteria, a total of 1,639 patients from the MIMIC-IV cohort and 2,170 patients from the eICU-CRD cohort were included in the final analysis (Figure 1).

Figure 1.

Flowchart of patient inclusion and exclusion criteria. MIMIC, Medical Information Mart for Intensive Care; eICU-CRD, eICU Collaborative Research Database; COPD, chronic obstructive pulmonary disease; ICU, intensive care unit; ILD, interstitial lung disease.

2.3. Data collection

For each eligible patient, demographic information (age, sex, and race), vital signs (heart rate, mean blood pressure, respiratory rate, oxygen saturation), and clinical data were extracted as the first recorded values within the first 24 hours. Laboratory parameters collected included white blood cell (WBC) count, red blood cell (RBC) count, hemoglobin, eosinophil count, red cell distribution width (RDW), platelet count, blood urea nitrogen (BUN), creatinine, glucose, arterial blood gases (pH, PCO₂, and PO₂), electrolytes (sodium and potassium), and other relevant biomarkers. Comorbidities such as diabetes mellitus, congestive heart failure (CHF), atrial fibrillation, hypertension, malignancy, pneumonia, acute kidney injury (AKI), and sepsis were identified based on ICD diagnosis codes and charted data. Disease severity was assessed using the SOFA score and the SAPS II. Neurological status was evaluated using the Glasgow Coma Scale (GCS). Therapeutic interventions during ICU stay, including the use of vasoactive agents, mechanical ventilation (MV), antibiotics, glucocorticoids, and diuretics, were also recorded. The ROX index was calculated as SpO₂/FiO₂ divided by respiratory rate.^5,12 All variables used to calculate the ROX index were obtained within the first 24 hours after ICU admission to reflect the patients’ initial respiratory status. When multiple measurements were available during this period, the values closest to the time of ICU admission were used. FiO₂ values were obtained from bedside respiratory monitoring systems or ventilator settings. For patients receiving MV, non-invasive ventilation (NIV), or high-flow nasal cannula (HFNC), FiO₂ values were directly recorded by the ventilator or respiratory support system. For patients receiving conventional oxygen therapy (e.g., nasal cannula, simple mask), FiO₂ values were recorded when explicitly documented. FiO₂ values were standardized as fractions (0.21–1.0) before calculating the ROX index.

2.4. Primary outcome and secondary outcomes

The primary outcome of this study was in-hospital mortality. Secondary outcomes included ICU mortality (death occurring during the ICU stay) and 28-day all-cause mortality. Mortality status and timing were obtained from the structured discharge and vital status records in both databases.

2.5. Statistical analysis

Continuous variables were expressed as medians with interquartile ranges (IQRs), and categorical variables were presented as counts with percentages. Differences between groups were assessed using the Kruskal–Wallis test for continuous variables and the chi-square test or Fisher’s exact test for categorical variables, as appropriate.

As shown in Figure S1, missing data were handled as follows: variables with less than 1% missingness were imputed using their median. For variables with relatively higher missingness (MBP, height, and eosinophil count), multiple imputation was performed using the “mice” package.¹³ The reported results were based on pooled estimates from five imputed datasets.

The association between the ROX index and in-hospital mortality was examined using Cox regression models. ROX was analyzed both as a continuous variable and by tertiles. The proportional hazards assumption was evaluated using Schoenfeld residual tests. The proportional hazards assumption was satisfied in both cohorts (MIMIC-IV: P = 0.601; eICU-CRD: P = 0.407) (Figure S2). Covariates included in the multivariable models were selected based on a combination of clinical relevance, prior literature, and statistical considerations rather than relying solely on statistical significance from univariate screening. Collinearity among covariates was evaluated using the generalized variance inflation factor (GVIF), and no significant multicollinearity was identified (GVIF^(1/(2Df)) < 2 for all variables). Three multivariable models were constructed: Model 1 adjusted for age, gender, and race; Model 2 further adjusted for BMI, heart rate, AKI, and sepsis; Model 3 additionally adjusted for SOFA score, SAPS II score, laboratory parameters (WBC, eosinophils, RDW, BUN, creatinine, pH, and PCO₂), and ICU treatments (use of vasoactive agents, antibiotics, diuretics, and MV).

Restricted cubic spline (RCS) regression was used to evaluate potential nonlinear associations between ROX and mortality risk. The spline model was constructed with four knots placed at the 5th, 35th, 65th, and 95th percentiles of the ROX index distribution. Kaplan–Meier survival curves were generated across ROX tertiles and compared using the log-rank test. Subgroup analyses were performed on an exploratory basis and were not pre-specified in the study protocol. These analyses aimed to evaluate whether the association between the ROX index and mortality differed across key clinical subgroups, such as age, illness severity, and comorbidity status. To assess the robustness of the association and the potential impact of unmeasured confounding, E-value analysis was performed based on the point estimate and 95% confidence interval from the fully adjusted model.¹⁴

A two-tailed P value of < 0.05 was deemed statistically significant. Analyses were performed using R version 4.3.2 and Free Statistics software (version 2.0; available at https://www.clinicalscientist.cn/freestatistics).

3. Results

3.1. Baseline characteristics

A total of 1,639 critically ill patients with COPD from the MIMIC-IV database were included in the final analysis. Baseline characteristics of the study population stratified by ROX index tertiles are shown in Table 1. Patients in the highest ROX tertile (T3) were generally older, had lower heart rates and BMI, and exhibited milder illness severity based on lower SOFA and SAPS II scores. The prevalence of pneumonia and sepsis decreased with increasing ROX tertiles, while the use of MV was more common in the higher ROX group. Notably, patients in T3 also had lower levels of inflammatory markers (e.g., WBC), renal function indicators (BUN and creatinine), and higher PO₂ and pH levels, suggesting better respiratory and metabolic status.

Table 1.

Baseline characteristics of COPD patients according to ROX tertiles.

Variables	Total (n = 1639)	T1 (n = 546)	T2 (n = 546)	T3 (n = 547)	P value
Age	71.13 (62.77, 78.96)	69.21 (61.30, 77.42)	71.54 (63.35, 79.21)	72.05 (63.56, 79.50)	0.01
Gender, n (%)					0.74
Female	735 (44.84)	238 (43.59)	246 (45.05)	251 (45.89)
Male	904 (55.16)	308 (56.41)	300 (54.95)	296 (54.11)
Race, n (%)					0.52
White	1072 (65.41)	357 (65.38)	358 (65.57)	357 (65.27)
Black	151 (9.21)	56 (10.26)	48 (8.79)	47 (8.59)
Other	146 (8.91)	54 (9.89)	41 (7.51)	51 (9.32)
Unknown	270 (16.47)	79 (14.47)	99 (18.13)	92 (16.82)
Smoke, n (%)	446 (27.21)	161 (29.49)	136 (24.91)	149 (27.24)	0.24
Alcohol, n (%)	182 (11.10)	68 (12.45)	52 (9.52)	62 (11.33)	0.30
Vital signs
BMI (Kg/m2)	28.34 (23.69, 34.64)	29.08 (24.53, 34.48)	28.41 (23.37, 35.48)	27.54 (23.01, 33.77)	0.02
Heart rate (beats/min)	89.00 (76.00, 103.00)	95.00 (81.00, 110.00)	88.00 (75.25, 102.75)	84.00 (74.00, 96.50)	< 0.001
MBP (mmHg)	79.00 (68.00, 92.00)	80.00 (69.00, 93.00)	78.00 (68.00, 91.00)	77.00 (67.00, 91.00)	0.03
Comorbidity
Diabetes, n (%)	588 (35.88)	192 (35.16)	201 (36.81)	195 (35.65)	0.84
Malignant cancer, n (%)	228 (13.91)	94 (17.22)	82 (15.02)	52 (9.51)	< 0.001
CHF, n (%)	761 (46.43)	253 (46.34)	258 (47.25)	250 (45.7)	0.88
Hypertension, n (%)	1162 (70.90)	375 (68.68)	379 (69.41)	408 (74.59)	0.06
Atrial fibrillation, n (%)	651 (39.72)	219 (40.11)	210 (38.46)	222 (40.59)	0.75
Pneumonia, n (%)	748 (45.64)	297 (54.4)	261 (47.8)	190 (34.73)	< 0.001
AKI, n (%)	1503 (91.70)	508 (93.04)	501 (91.76)	494 (90.31)	0.26
Sepsis, n (%)	1362 (83.10)	473 (86.63)	463 (84.8)	426 (77.88)	< 0.001
Severity scores
SOFA	6.00 (4.00, 9.00)	7.00 (4.00, 10.00)	6.00 (4.00, 9.00)	6.00 (4.00, 8.00)	< 0.001
SAPSII	44.00 (36.00, 54.00)	46.50 (37.00, 57.00)	42.00 (35.00, 53.00)	43.00 (35.00, 51.00)	< 0.001
GCS	15.00 (13.00, 15.00)	15.00 (13.00, 15.00)	15.00 (14.00, 15.00)	15.00 (13.00, 15.00)	0.31
Laboratory tests
WBC (k/uL)	11.80 (8.35, 16.40)	12.30 (8.72, 17.78)	11.80 (8.20, 16.80)	11.20 (8.10, 15.35)	0.02
RBC (k/uL)	3.49 (3.00, 4.12)	3.62 (3.05, 4.19)	3.55 (3.03, 4.18)	3.36 (2.92, 3.94)	< 0.001
Hemoglobin (g/dL)	10.40 (8.90, 12.00)	10.80 (9.00, 12.20)	10.50 (9.10, 12.10)	10.00 (8.65, 11.70)	< 0.001
Eosinophils (k/uL)	0.03 (0.00, 0.13)	0.02 (0.00, 0.13)	0.03 (0.00, 0.14)	0.04 (0.00, 0.14)	0.01
RDW (%)	15.10 (14.00, 16.60)	15.20 (14.10, 16.70)	15.10 (14.00, 16.88)	14.80 (13.80, 16.40)	0.01
Platelet (k/uL)	191.00 (137.00, 256.00)	198.00 (146.00, 272.00)	190.00 (140.25, 258.00)	183.00 (129.50, 240.00)	<0.001
BUN (mg/dL)	25.00 (17.00, 40.00)	27.00 (18.00, 41.00)	25.00 (18.00, 41.00)	22.00 (16.00, 36.00)	< 0.001
Creatinine (mg/dl)	1.10 (0.80, 1.70)	1.20 (0.80, 1.90)	1.10 (0.80, 1.80)	1.00 (0.70, 1.60)	< 0.001
Potassium (mmol/L)	4.30 (3.90, 4.80)	4.30 (3.80, 4.88)	4.30 (3.90, 4.80)	4.30 (3.90, 4.80)	0.289
Sodium (mmol/L)	139.00 (136.00, 142.00)	139.00 (135.00, 142.00)	139.00 (135.00, 142.00)	139.00 (136.00, 142.00)	0.074
Glucose (mg/dL)	136.00 (110.00, 178.00)	140.00 (112.00, 180.00)	138.00 (112.00, 182.00)	129.00 (107.00, 170.00)	0.01
PH	7.34 (7.26, 7.40)	7.33 (7.23, 7.39)	7.33 (7.25, 7.39)	7.35 (7.28, 7.41)	< 0.001
PCO2 (mmHg)	47.00 (39.00, 58.00)	47.00 (39.00, 59.00)	47.00 (40.00, 58.00)	46.00 (38.00, 56.00)	0.02
PO2 (mmHg)	90.00 (61.00, 171.00)	82.00 (57.00, 164.00)	88.00 (59.25, 137.00)	101.00 (66.00, 219.00)	< 0.001
Treatment
Vasoactive, n (%)	1000 (61.01)	351 (64.29)	328 (60.07)	321 (58.68)	0.14
Antibiotics, n (%)	1483 (90.48)	501 (91.76)	507 (92.86)	475 (86.84)	0.00
Glucocorticoid, n (%)	596 (36.36)	211 (38.64)	208 (38.1)	177 (32.36)	0.06
Diuretic, n (%)	1186 (72.36)	395 (72.34)	392 (71.79)	399 (72.94)	0.91
MV, n (%)	1577 (96.22)	516 (94.51)	534 (97.8)	527 (96.34)	0.02
Outcomes
Hospital stay (day)	11.26 (6.66, 19.24)	11.84 (6.69, 20.01)	11.44 (6.71, 18.68)	10.30 (6.60, 17.86)	0.48
ICU stay (day)	5.56 (3.00, 10.73)	5.94 (3.19, 11.95)	5.62 (3.02, 11.11)	4.93 (2.77, 9.46)	0.01
Hospital mortality, n (%)	428 (26.11)	184 (33.7)	144 (26.37)	100 (18.28)	< 0.001
CU mortality, n (%)	344 (20.99)	155 (28.39)	109 (19.96)	80 (14.63)	< 0.001
28-day mortality, n (%)	474 (28.92)	197 (36.08)	162 (29.67)	115 (21.02)	< 0.001

Abbreviations: BMI, body mass index; MBP, mean blood pressure; CHF, congestive heart failure; AKI, acute kidney injury; SOFA, Sequential Organ Failure Assessment; SAPS II, Simplified Acute Physiology Score II; GCS, Glasgow Coma Scale; WBC, white blood cell count; RBC, red blood cell count; RDW, red cell distribution width; BUN, blood urea nitrogen; PH, potential of hydrogen (arterial blood pH); PCO₂, partial pressure of carbon dioxide; PO₂, partial pressure of oxygen; MV, mechanical ventilation.

Treatment patterns also varied across tertiles. Antibiotics and vasoactive agents were used more frequently in the lower ROX group. Glucocorticoids were more frequently used in the lower ROX groups, although the difference did not reach statistical significance (P = 0.06). The length of ICU stay decreased significantly from T1 to T3 (P = 0.01), while hospital stay did not differ significantly across tertiles. Mortality outcomes varied markedly: in-hospital mortality was 33.7% in T1, 26.4% in T2, and 18.3% in T3 (P < 0.001).

Table 2 presents a comparison of baseline characteristics between hospital survivors and non-survivors. Non-survivors were significantly older and had lower BMI, higher heart rates, and worse illness severity scores (SOFA and SAPS II). They also had a higher prevalence of sepsis, AKI, pneumonia, and malignancy. Laboratory findings in non-survivors reflected greater physiologic derangement, including elevated WBC, RDW, BUN, creatinine, and lower eosinophil counts, pH, and PO₂ levels. Non-survivors were more likely to receive vasopressors, antibiotics, and glucocorticoids, whereas ventilator use was unexpectedly higher among survivors, potentially reflecting early withdrawal of care or do-not-intubate orders in high-risk patients.

Table 2.

Baseline characteristics of the survivor and non-survivor groups.

Variables	Overall (n = 1639)	Survivor (n = 1211)	Non-survivor (n = 428)	P value
Age	71.13 (62.77, 78.96)	69.89 (61.39, 77.67)	74.18 (66.59, 81.45)	< 0.001
Gender, n (%)				0.90
Female	735 (44.84)	542 (44.76)	193 (45.09)
Male	904 (55.16)	669 (55.24)	235 (54.91)
Race, n (%)				< 0.001
White	1072 (65.41)	820 (67.71)	252 (58.88)
Black	151 (9.21)	114 (9.41)	37 (8.64)
Other	146 (8.91)	107 (8.84)	39 (9.11)
Unknow	270 (16.47)	170 (14.04)	100 (23.36)
Smoke, n (%)	446 (27.21)	357 (29.48)	89 (20.79)	< 0.001
Alcohol, n (%)	182 (11.10)	134 (11.07)	48 (11.21)	0.93
Vital signs
BMI (Kg/m2)	28.34 (23.69, 34.64)	28.70 (24.08, 35.07)	26.92 (22.62, 33.31)	< 0.001
Heart rate (beats/min)	89.00 (76.00, 103.00)	87.00 (75.50, 102.00)	92.00 (79.00, 109.00)	< 0.001
MBP (mmHg)	79.00 (68.00, 92.00)	79.00 (68.00, 92.00)	77.00 (66.00, 90.00)	0.03
ROX	9.33 (6.13, 12.81)	9.77 (6.48, 13.57)	8.27 (5.28, 11.11)	< 0.001
Comorbidity
Diabetes, n (%)	588 (35.88)	440 (36.33)	148 (34.58)	0.52
Malignant cancer, n (%)	228 (13.91)	152 (12.55)	76 (17.76)	0.01
CHF, n (%)	761 (46.43)	558 (46.08)	203 (47.43)	0.63
Hypertension, n (%)	1162 (70.90)	871 (71.92)	291 (67.99)	0.12
Atrial fibrillation, n (%)	651 (39.72)	457 (37.74)	194 (45.33)	0.01
Pneumonia, n (%)	748 (45.64)	529 (43.68)	219 (51.17)	0.01
AKI, n (%)	1503 (91.70)	1087 (89.76)	416 (97.2)	< 0.001
Sepsis, n (%)	1362 (83.10)	962 (79.44)	400 (93.46)	< 0.001
Severity scores
SOFA	6.00 (4.00, 9.00)	6.00 (4.00, 8.00)	8.00 (5.00, 11.00)	< 0.001
SAPSII	44.00 (36.00, 54.00)	41.00 (34.00, 50.00)	51.00 (42.00, 62.00)	< 0.001
GCS	15.00 (13.00, 15.00)	15.00 (13.00, 15.00)	15.00 (13.00, 15.00)	0.34
Laboratory tests
WBC (k/uL)	11.80 (8.35, 16.40)	11.40 (8.20, 15.40)	13.05 (8.97, 19.02)	< 0.001
RBC (k/uL)	3.49 (3.00, 4.12)	3.52 (3.03, 4.11)	3.44 (2.94, 4.12)	0.15
Hemoglobin (g/dL)	10.61 ± 2.33	10.64 ± 2.29	10.53 ± 2.43	0.40
Eosinophils (k/uL)	0.03 (0.00, 0.13)	0.04 (0.00, 0.15)	0.01 (0.00, 0.07)	< 0.001
RDW (%)	15.10 (14.00, 16.60)	14.90 (13.90, 16.50)	15.50 (14.20, 17.00)	< 0.001
Platelet (k/uL)	191.00 (137.00, 256.00)	194.00 (142.00, 256.00)	186.00 (126.75, 255.00)	0.15
BUN (mg/dL)	25.00 (17.00, 40.00)	23.00 (16.00, 36.00)	30.50 (20.00, 47.25)	< 0.001
Creatinine (mg/dl)	1.10 (0.80, 1.70)	1.10 (0.70, 1.60)	1.30 (0.90, 2.10)	< 0.001
Potassium (mmol/L)	4.30 (3.90, 4.80)	4.30 (3.90, 4.80)	4.30 (3.80, 4.90)	0.93
Sodium (mmol/L)	139.00 (136.00, 142.00)	139.00 (136.00, 142.00)	139.00 (135.00, 142.00)	0.41
Glucose (mg/dL)	136.00 (110.00, 178.00)	134.00 (110.00, 177.00)	142.00 (109.00, 181.75)	0.47
PH	7.34 (7.26, 7.40)	7.35 (7.28, 7.41)	7.30 (7.22, 7.39)	< 0.001
PCO2 (mmHg)	47.00 (39.00, 58.00)	46.00 (39.00, 58.00)	47.50 (39.00, 57.25)	0.80
PO2 (mmHg)	90.00 (61.00, 171.00)	94.00 (62.50, 186.00)	78.50 (53.00, 142.50)	< 0.001
Treatment
Vasoactive, n (%)	1000 (61.01)	674 (55.66)	326 (76.17)	< 0.001
Antibiotics, n (%)	1483 (90.48)	1069 (88.27)	414 (96.73)	< 0.001
Glucocorticoid, n (%)	596 (36.36)	421 (34.76)	175 (40.89)	0.02
Diuretic, n (%)	1186 (72.36)	910 (75.14)	276 (64.49)	< 0.001
MV, n (%)	1577 (96.22)	1187 (98.02)	390 (91.12)	< 0.001

Abbreviations: BMI, body mass index; MBP, mean blood pressure; ROX, respiratory rate-oxygenation index; CHF, congestive heart failure; AKI, acute kidney injury; SOFA, Sequential Organ Failure Assessment; SAPS II, Simplified Acute Physiology Score II; GCS, Glasgow Coma Scale; WBC, white blood cell count; RBC, red blood cell count; RDW, red cell distribution width; BUN, blood urea nitrogen; PH, potential of hydrogen (arterial blood pH); PCO₂, partial pressure of carbon dioxide; PO₂, partial pressure of oxygen; MV, mechanical ventilation.

3.2. Relationship between ROX and mortality

To identify covariates for multivariable modeling, univariate Cox regression analysis was first conducted (Table S2). Collinearity was assessed using the GVIF, and no significant multicollinearity was detected (Table S3), supporting the simultaneous inclusion of these variables in the multivariable models. Cox regression analysis revealed a significant inverse association between the ROX index and in-hospital mortality in both the MIMIC-IV and eICU-CRD cohorts. When treated as a continuous variable, each one-unit increase in ROX was linked to a lower hazard of in-hospital death (HR = 0.96, 95% CI: 0.93–0.98 in MIMIC-IV; HR = 0.95, 95% CI: 0.92–0.98 in eICU-CRD; both P < 0.01) after adjustment for all covariates in Model 3 (Table 3). When analyzed categorically by tertiles, individuals in the highest ROX group (T3) exhibited a markedly reduced risk of in-hospital mortality compared to those in the lowest tertile (T1). In fully adjusted models, the hazard ratio for T3 was 0.62 (95% CI: 0.48–0.79, P < 0.001) in the MIMIC-IV cohort and 0.52 (95% CI: 0.36–0.76, P = 0.001) in the eICU-CRD cohort, with a statistically significant trend observed across increasing tertiles (P for trend < 0.001 in both cohorts). Sensitivity analyses were performed to assess the robustness of the association between ROX and in-hospital mortality. As shown in Table S4, the association remained consistent after excluding eosinophil count, SAPS II, or SOFA score from the multivariable models in both the MIMIC-IV and eICU-CRD cohorts, indicating that the primary findings were not materially affected by missing eosinophil data or potential redundancy between severity scores.

Table 3.

Analysis for association between ROX and in-hospital mortality.

Variables	Model1		Model2		Model3
Variables	HR (95%CI)	P	HR (95%CI)	P	HR (95%CI)	P
MIMIC IV 3.1 cohort
ROX (Continuous)	0.94 (0.92, 0.96)	<0.001	0.94 (0.92, 0.96)	<0.001	0.96 (0.93, 0.98)	<0.001
ROX (Tertiles)
T1	1(Ref)		1(Ref)		1(Ref)
T2	0.72 (0.57, 0.89)	0.003	0.74 (0.59, 0.92)	0.007	0.81 (0.64, 1.02)	0.073
T3	0.5 (0.39, 0.64)	<0.001	0.53 (0.41, 0.68)	<0.001	0.62 (0.48, 0.79)	<0.001
P for trend		<0.001		<0.001		<0.001
eICU-CRD cohort
ROX (Continuous)	0.95 (0.93, 0.98)	<0.001	0.96 (0.93, 0.99)	0.004	0.95 (0.92, 0.98)	0.001
ROX (Tertiles)
T1	1(Ref)		1(Ref)		1(Ref)
T2	0.76 (0.59, 0.98)	0.036	0.79 (0.61, 1.02)	0.075	0.81 (0.62, 1.06)	0.124
T3	0.52 (0.37, 0.73)	<0.001	0.57 (0.40, 0.81)	0.002	0.52 (0.36, 0.76)	0.001
P for trend		<0.001		0.001		0.001

Model 1: Adjusted for age, gender, and race.

Model 2: Adjusted for age, gender, race, body mass index (BMI), heart rate, acute kidney injury (AKI), and sepsis.

Model 3: Adjusted for age, gender, race, BMI, heart rate, AKI, sepsis, Sequential Organ Failure Assessment (SOFA) score, Simplified Acute Physiology Score II (SAPS II), white blood cell count (WBC), eosinophil count, red cell distribution width (RDW), blood urea nitrogen (BUN), creatinine, arterial pH, partial pressure of carbon dioxide (PCO₂), and ICU interventions including vasoactive agents, antibiotics, diuretics, and mechanical ventilation (MV).

Similar associations were observed for secondary outcomes. As shown in Table S5, higher ROX values were independently associated with lower ICU mortality. In the fully adjusted model, the hazard ratio comparing the highest to the lowest tertile (T3 vs. T1) was 0.64 (95% CI: 0.48–0.85, P = 0.002) in the MIMIC-IV cohort and 0.45 (95% CI: 0.28–0.75, P = 0.002) in the eICU-CRD cohort. Similarly, for 28-day mortality (Table S6), the inverse association remained significant, with hazard ratios of 0.58 (95% CI: 0.45–0.73, P < 0.001) in MIMIC-IV and 0.29 (95% CI: 0.18–0.47, P < 0.001) in eICU-CRD for patients in the highest ROX tertile.

3.3. Kaplan-Meier curves and RCS

Kaplan–Meier analysis showed significantly better 28-day survival in patients with higher ROX index tertiles in both cohorts (Figure 2, log-rank P < 0.001), indicating that higher ROX values on ICU admission are associated with lower short-term mortality. RCS regression was applied to explore potential nonlinear associations between the ROX index and mortality outcomes, including in-hospital, ICU, and 28-day mortality (Figure 3). In both cohorts, the RCS curves revealed an approximately linear inverse relationship between ROX values and mortality risk, with no indication of significant non-linearity. The mortality risk decreased steadily with increasing ROX values, reinforcing the utility of ROX as a continuous prognostic marker rather than requiring arbitrary cutoffs.

Figure 2.

Kaplan–Meier survival curves for 28-day mortality stratified by ROX index tertiles in the MIMIC-IV and eICU-CRD cohorts. (A) MIMIC IV 3.1 database (B) eICU-CRD database.

Figure 3.

Restricted cubic spline analyses depicting the nonlinear association between ROX index and mortality outcomes in critically ill patients with COPD. Panels A–C represent the MIMIC-IV cohort, showing associations with in-hospital mortality (A), ICU mortality (B), and 28-day mortality (C). Panels D–F correspond to the eICU-CRD cohort, showing the same outcomes respectively. ROX index was modeled as a continuous variable using RCS with full covariate adjustment. Solid lines represent hazard ratios (HRs) and shaded areas indicate 95% confidence intervals.

3.4. Subgroup analysis

Subgroup analyses were performed to evaluate the consistency of the association between the ROX index and 28-day mortality across clinically relevant subpopulations (Figure 4). In both cohorts, the inverse association between ROX and mortality remained generally stable across groups stratified by age, sex, BMI, atrial fibrillation, pneumonia, vasoactive medication use, and SOFA score. To assess the potential influence of hypercapnia, patients were additionally stratified by baseline PCO₂ into normocapnic (≤45 mmHg) and hypercapnic (>45 mmHg) groups. No significant interaction between ROX and PCO₂ status was observed (P for interaction > 0.05), suggesting that the prognostic value of ROX was not significantly modified by hypercapnia.

Figure 4.

Subgroup analyses of the association between ROX index and in-hospital mortality in critically ill patients with COPD. Forest plots display the adjusted hazard ratios (HRs) and 95% confidence intervals (CIs) for in-hospital mortality per unit increase in ROX index across clinically relevant subgroups in the MIMIC-IV (A) and eICU-CRD (B) cohorts.

3.5. Sensitivity analysis for unmeasured confounding

To test the robustness of the association between the ROX index and in-hospital mortality, E-value analysis was conducted using fully adjusted Cox models. As shown in Figure S3, the E-values for the point estimates were 2.27 (MIMIC-IV) and 2.52 (eICU-CRD), with lower 95% CI bounds of 1.79 and 1.71, respectively. These findings suggest that substantial unmeasured confounding would be required to negate the observed associations, supporting their internal validity.

4. Discussion

In this large, multicenter retrospective study of critically ill COPD patients, a higher ROX index within 24 hours of ICU admission was independently associated with reduced in-hospital, ICU, and 28-day mortality. This relationship remained stable after comprehensive adjustment for confounders and was consistent across subgroups in both the MIMIC-IV and eICU-CRD cohorts. RCS analysis demonstrated a linear inverse association between ROX and mortality risk, while E-value analysis indicated that substantial unmeasured confounding would be necessary to explain away the observed findings, supporting their robustness.

The ROX index represents the interplay between oxygenation efficiency and respiratory workload.¹⁵ A lower ROX index implies either worse oxygenation (lower SpO₂/FiO₂) or an increased respiratory drive (higher respiratory rate), both of which are hallmarks of respiratory decompensation.¹⁶ Previous studies have predominantly focused on the ROX index as a predictor of intubation necessity in patients undergoing HFNC therapy. Roca et al. first introduced the ROX index in patients with AHRF, and subsequent studies confirmed its value in identifying HFNC failure, especially in COVID-19-associated pneumonia.^17–19 However, fewer studies have examined its role as a broader prognostic marker. Although the ROX index was originally developed for predicting high-flow nasal cannula failure in non-intubated patients, its components (SpO₂, FiO₂, and respiratory rate) represent fundamental indicators of respiratory physiology. In the present study, ROX was used as a marker of respiratory status rather than as a predictor of HFNC failure. Nevertheless, the applicability of the ROX index in mechanically ventilated populations warrants further validation in future studies. Recently, Liu et al. reported a nonlinear association between the ROX index and 28-day mortality in patients with AHRF, identifying a markedly increased risk when ROX fell below 8.28.⁷ In contrast, our RCS analysis in critically ill COPD patients suggested a monotonic inverse relationship, with mortality risk gradually decreasing as ROX increased and without an obvious nonlinear threshold. This discrepancy may be related to differences in study populations and clinical contexts, as our study focused specifically on a high-risk ICU population with COPD rather than a broader AHRF population receiving HFNC. Taken together, our findings expand the existing literature by supporting the prognostic value of the ROX index in critically ill COPD patients across two large independent cohorts.

There are several plausible mechanisms by which a low ROX index may reflect poor prognosis in COPD. First, a low SpO₂/FiO₂ ratio suggests impaired oxygenation, which may result from ventilation–perfusion mismatch, alveolar hypoventilation, or advanced emphysematous changes.^20,21 Second, an elevated respiratory rate indicates increased work of breathing and impending respiratory muscle fatigue, especially in COPD patients with pre-existing dynamic hyperinflation and reduced diaphragmatic efficiency.^22,23 Together, these factors may signal a failing respiratory system and the need for more intensive support. Moreover, persistent tachypnea has been associated with systemic inflammation, autonomic dysregulation, and adverse cardiovascular events—all of which may contribute to higher mortality.

The simplicity of the ROX index makes it particularly attractive in the ICU setting, where real-time assessment of respiratory status is essential. Beyond reflecting oxygenation efficiency and respiratory effort, the ROX index may also capture important pathophysiological changes specific to COPD. In patients with chronic airflow limitation, destruction of alveolar structures and mismatches in ventilation–perfusion ratios are common, often leading to impaired oxygen diffusion and reduced SpO₂/FiO₂ ratios.^24,25 During acute exacerbations, these disturbances are exacerbated, resulting in a further drop in ROX values. In parallel, increased respiratory rate—another component of the index—may indicate rising respiratory drive, compensatory effort, or early respiratory muscle fatigue, particularly in the context of dynamic hyperinflation and diaphragmatic inefficiency, both of which are hallmark features of advanced COPD.^26,27 Although the ROX index does not directly incorporate ventilatory parameters such as PaCO₂, its prognostic significance may partly stem from an indirect reflection of carbon dioxide retention. Hypercapnia is frequently observed in severe COPD and has been linked to poor outcomes, especially when coupled with acidosis or blunted ventilatory response.²⁸ In our study, the inclusion of PCO₂ as an adjustment variable adds analytical strength and helps account for this confounding effect. Whether ROX interacts with PaCO₂ in modifying risk profiles warrants further investigation, potentially through stratified analysis or interaction testing in future studies.

Our findings have meaningful clinical implications. The ROX index is a non-invasive, easy-to-calculate bedside parameter derived from routinely available vital signs and oxygenation data. Its early use may facilitate risk stratification, guide clinical decision-making, and assist in the triage of critically ill COPD patients at ICU admission. Compared with traditional oxygenation indices such as the PaO₂/FiO₂ ratio or the alveolar–arterial (A–a) gradient, ROX does not require arterial blood sampling and can be readily calculated using non-invasive measurements, making it particularly useful in time-sensitive or resource-limited settings. These advantages support the potential integration of ROX into clinical risk assessment frameworks. Future studies may further explore ROX-based predictive models by combining ROX with established severity scores or other key clinical variables to develop more comprehensive prognostic tools.

Despite its strengths, this study has certain limitations. First, COPD was identified based on ICD codes from administrative databases rather than direct clinical confirmation. As the MIMIC-IV and eICU-CRD datasets are fully de-identified, validation using spirometry data or chart review was not possible. Therefore, some degree of disease misclassification cannot be completely excluded. Second, due to the retrospective design, residual confounding cannot be entirely excluded despite adjustment for multiple covariates. Some potentially important factors, such as ventilator parameters or COPD phenotypes, were not available in the databases. Although E-value analysis suggested that substantial unmeasured confounding would be required to fully explain the observed association, the possibility of residual confounding remains. Third, we relied on single-timepoint ROX values within the first 24 hours, and dynamic changes in ROX over time—which may also carry prognostic value—were not analyzed. Fourth, although we validated our findings in an external cohort, both datasets originate from the United States, which may limit generalizability to other healthcare settings or populations. Lastly, cause-specific mortality data were not available, precluding assessment of whether ROX specifically predicts respiratory vs. non-respiratory deaths.

5. Conclusion

In summary, our study found that the ROX index was significantly associated with mortality in critically ill patients with COPD. Given its physiological relevance and ease of calculation at the bedside, the ROX index may serve as a practical tool for early risk stratification in the ICU setting. However, as this was an observational study, the findings should be interpreted as associations rather than causal relationships. Further prospective studies are warranted to evaluate the role of serial ROX measurements and to determine optimal thresholds for clinical decision-making in this population.

Supplemental material

Supplemental material - Association between ROX index and in-hospital mortality in critically ill patients with chronic obstructive pulmonary disease: a retrospective cohort study based on two large databases

Supplemental material for Association between ROX index and in-hospital mortality in critically ill patients with chronic obstructive pulmonary disease: a retrospective cohort study based on two large databases by Guangdong Wang, Tingting Liu, Wenwen Ji, TingTing Li, Zhuoyang Wang, Tinghua Hu, Xiaojian Wang, and Zhihong Shi in Science Progress.

Supplemental material

Supplemental material - Association between ROX index and in-hospital mortality in critically ill patients with chronic obstructive pulmonary disease: a retrospective cohort study based on two large databases

Footnotes

ORCID iD

Guangdong Wang

Ethical considerations

The MIMIC-IV and eICU-CRD databases contain fully de-identified patient data. Access to both databases was granted after completion of the required training and data use agreements through PhysioNet.

Consent to participate

Because this study used publicly available de-identified data, ethical approval and informed consent were waived in accordance with the relevant institutional policies and database regulations.

Author contributions

GW conceived the study design and drafted the manuscript. GW and TL performed data extraction and statistical analysis. WJ,TL ZW and TH performed the statistical analysis. ZS and XW interpreted the data and revised the manuscript critically for important intellectual content. All authors read and approved the final manuscript.

Funding

The authors received no financial support for the research, authorship, and/or publication of this article.

Declaration of conflicting interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Data Availability Statement

The data used in this study are available from the MIMIC-IV database and the eICU Collaborative Research Database. Due to the data use agreements governing these databases, the raw datasets cannot be redistributed by the authors. Access to these databases requires completion of a data use agreement and training through the PhysioNet Credentialed Health Data Access process ().

Supplemental material

Supplemental material for this article is available online.

References

Global burden of chronic respiratory diseases and risk factors , 1990-2019: an update from the Global Burden of Disease Study 2019 [J]. EClinicalMedicine, 2023, 59: 101936. https://doi.org/10.1016/j.eclinm.2023.101936

Wheaton

Liu

Croft

, et al. Chronic Obstructive Pulmonary Disease and Smoking Status - United States, 2017 [J]. MMWR Morb Mortal Wkly Rep, 2019, 68(24): 533–538. https://doi.org/10.15585/mmwr.mm6824a1

Wang

Chen

, et al. Quality of Care and Outcomes of Inpatients with COPD: A Multi-Center Study in China [J]. Int J Chron Obstruct Pulmon Dis, 2025, 20: 1787–1795. https://doi.org/10.2147/COPD.S510613

Roca

Messika

Caralt

, et al. Predicting success of high-flow nasal cannula in pneumonia patients with hypoxemic respiratory failure: The utility of the ROX index [J]. J Crit Care, 2016, 35: 200–205. https://doi.org/10.1016/j.jcrc.2016.05.022

Yau

Lee

DYX

Vasudevan

, et al. Performance of the ROX index in predicting high flow nasal cannula failure in COVID-19 patients: a systematic review and meta-analysis [J]. Crit Care, 2023, 27(1): 320. https://doi.org/10.1186/s13054-023-04567-7

Prakash

Bhattacharya

Yadav

, et al. ROX index as a good predictor of high flow nasal cannula failure in COVID-19 patients with acute hypoxemic respiratory failure: A systematic review and meta-analysis [J]. J Crit Care, 2021, 66: 102–108. https://doi.org/10.1016/j.jcrc.2021.08.012

Liu

Xiao

, et al. Association between the ROX index and mortality in patients with acute hypoxemic respiratory failure: a retrospective cohort study [J]. Respir Res, 2024, 25(1): 143. https://doi.org/10.1186/s12931-024-02771-9

Wedzicha

Seemungal

. COPD exacerbations: defining their cause and prevention [J]. Lancet, 2007, 370(9589): 786–796. https://doi.org/10.1016/S0140-6736(07)61382-8

Johnson

AEW

Bulgarelli

Shen

, et al. MIMIC-IV, a freely accessible electronic health record dataset [J]. Sci Data, 2023, 10(1): 1. https://doi.org/10.1038/s41597-022-01899-x

10.

Pollard

Johnson

AEW

Raffa

, et al. The eICU Collaborative Research Database, a freely available multi-center database for critical care research [J]. Sci Data, 2018, 5: 180178. https://doi.org/10.1038/sdata.2018.178

11.

Von

ELME

Altman

Egger

, et al. The Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement: guidelines for reporting observational studies [J]. Ann Intern Med, 2007, 147(8): 573–577. https://doi.org/10.7326/0003-4819-147-8-200710160-00010

12.

Goh

Chai

Ong

, et al. Early prediction of high flow nasal cannula therapy outcomes using a modified ROX index incorporating heart rate [J]. J Intensive Care, 2020, 8: 41. https://doi.org/10.1186/s40560-020-00458-z

13.

Shah

Bartlett

Carpenter

, et al. Comparison of random forest and parametric imputation models for imputing missing data using MICE: a CALIBER study [J]. Am J Epidemiol, 2014, 179(6): 764–774. https://doi.org/10.1093/aje/kwt312

14.

Vanderweele

Ding

. Sensitivity Analysis in Observational Research: Introducing the E-Value [J]. Ann Intern Med, 2017, 167(4): 268–274. https://doi.org/10.7326/M16-2607

15.

Junhai

Jing

Beibei

, et al. The value of ROX index in predicting the outcome of high flow nasal cannula: a systematic review and meta-analysis [J]. Respir Res, 2022, 23(1): 33. https://doi.org/10.1186/s12931-022-01951-9

16.

Zhou

Liu

Pan

, et al. The ROX index as a predictor of high-flow nasal cannula outcome in pneumonia patients with acute hypoxemic respiratory failure: a systematic review and meta-analysis [J]. BMC Pulm Med, 2022, 22(1): 121. https://doi.org/10.1186/s12890-022-01914-2

17.

Arunachala

Parthasarathi

Basavaraj

, et al. The Validity of the ROX Index and APACHE II in Predicting Early, Late, and Non-Responses to Non-Invasive Ventilation in Patients with COVID-19 in a Low-Resource Setting [J]. Viruses, 2023, 15, 2231. https://doi.org/10.3390/v15112231(11).

18.

KüçüK

Ergan

Yakar

, et al. The Predictive Values of Respiratory Rate Oxygenation Index and Chest Computed Tomography Severity Score for High-Flow Nasal Oxygen Failure in Critically Ill Patients with Coronavirus Disease-2019 [J]. Balkan Med J, 2022, 39(2): 140–147. https://doi.org/10.4274/balkanmedj.galenos.2021.2021-7-32

19.

Girault

Bubenheim

Boyer

, et al. ROX index performance to predict high-flow nasal oxygen outcome in Covid-19 related hypoxemic acute respiratory failure [J]. Ann Intensive Care, 2024, 14(1): 13. https://doi.org/10.1186/s13613-023-01226-6

20.

Erlebach

Pale

Beck

, et al. Limitations of SpO(2)/FiO(2)-ratio for classification and monitoring of acute respiratory distress syndrome-an observational cohort study [J]. Crit Care, 2025, 29(1): 82. https://doi.org/10.1186/s13054-025-05317-7

21.

Klitgaard

SchjøRRING

Nielsen

, et al. Higher versus lower fractions of inspired oxygen or targets of arterial oxygenation for adults admitted to the intensive care unit [J]. Cochrane Database Syst Rev, 2023, 9(9): Cd012631. https://doi.org/10.1002/14651858.CD012631.pub3

22.

Gagnon

Guenette

Langer

, et al. Pathogenesis of hyperinflation in chronic obstructive pulmonary disease [J]. Int J Chron Obstruct Pulmon Dis, 2014, 9: 187–201. https://doi.org/10.2147/COPD.S38934

23.

Koopman

Posthuma

Vanfleteren

, et al. Lung Hyperinflation as Treatable Trait in Chronic Obstructive Pulmonary Disease: A Narrative Review [J]. Int J Chron Obstruct Pulmon Dis, 2024, 19: 1561–1578. https://doi.org/10.2147/COPD.S458324

24.

Polosukhin

Gutor

, et al. Small airway determinants of airflow limitation in chronic obstructive pulmonary disease [J]. Thorax, 2021, 76(11): 1079–1088. https://doi.org/10.1136/thoraxjnl-2020-216037

25.

Cosio Piqueras

Cosio

. Disease of the airways in chronic obstructive pulmonary disease [J]. Eur Respir J Suppl, 2001, 34: 41s-49s. https://doi.org/10.1183/09031936.01.00234601

26.

O'Donnell

Laveneziana

. The clinical importance of dynamic lung hyperinflation in COPD [J]. Copd, 2006, 3(4): 219–232. https://doi.org/10.1080/15412550600977478

27.

Qian

Cai

Sun

, et al. Analyses of Factors Associated with Acute Exacerbations of Chronic Obstructive Pulmonary Disease: A Review [J]. Int J Chron Obstruct Pulmon Dis, 2023, 18: 2707–2723. https://doi.org/10.2147/COPD.S433183

28.

Csoma

Vulpi

Dragonieri

, et al. Hypercapnia in COPD: Causes, Consequences, and Therapy [J]. J Clin Med, 2022, 11(11), 3180. https://doi.org/10.3390/jcm11113180

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.69 MB

0.17 MB