Prognostic Value of the ROX Index for Mortality in Chronic Obstructive Pulmonary Disease: A Retrospective Cohort Study Using the MIMIC-IV Database

Introduction

Chronic obstructive pulmonary disease (COPD) is a prevalent respiratory condition characterized by persistent and progressive airflow limitation caused by airway and/or alveolar abnormalities, including chronic bronchitis and emphysema.^1,2 Its typical clinical presentation includes chronic cough and sputum production, which significantly impair patients’ quality of life. As a leading cause of global morbidity, mortality, and healthcare utilization, COPD poses a substantial and growing public health burden.³ Prevalence increases with age, and World Health Organization projections indicate that COPD is set to become the third leading cause of death worldwide by 2030.^4,5 Consequently, the early identification of high-risk patients is critical for improving outcomes. However, existing prognostic tools for COPD, such as the BODE index or DECAF score, often depend on complex composite measures, specialized laboratory tests, or assessments unsuitable for rapid, bedside application during acute exacerbations.^6,7 This highlights a need for simple, readily available indices for risk stratification in acute care settings.

The ROX index, defined as the ratio of pulse oximetric saturation (SpO₂) to the fraction of inspired oxygen (FiO₂) divided by respiratory rate (RR), meets this need by simultaneously quantifying gas exchange efficiency (SpO₂/FiO₂) and respiratory load (RR). It gained widespread adoption during the COVID-19 pandemic for prognosticating outcomes in acute hypoxemic respiratory failure (AHRF)^8,9 and is primarily used to assess the response to high-flow nasal cannula (HFNC) therapy and predict the need for intubation.^10–13 Its physiological rationale is particularly relevant to COPD exacerbations, which are frequently driven by worsening ventilation-perfusion mismatch (affecting SpO₂/FiO₂) and dynamic hyperinflation leading to tachypnea and increased work of breathing (reflected in RR). Due to its ease of use, non-invasive nature, and suitability for repeated assessment, the index represents a practical bedside tool with potential advantages in settings requiring rapid clinical decisions.^14,15

The prognostic role of the ROX index in COPD, however, remains less clear. Although it has been used to predict the treatment outcomes of HFNC in patients with acute exacerbation of COPD,¹⁶ its relationship with mortality has not been established. To address this, we conducted a retrospective cohort study using the MIMIC-IV database to investigate the association between the ROX index and 28-day all-cause mortality in critically ill COPD patients, identify potential thresholds, and assess the consistency of this association across subgroups.

Materials and Methods

The data utilized in this study were obtained from the Medical Information Mart for Intensive Care IV (MIMIC-IV) database. This database encompasses approximately 190,000 hospital admissions of patients who were admitted to the Beth Israel Deaconess Medical Center (BIDMC) over the period from 2008 to 2019. It serves as a publicly accessible repository containing data on critically ill patients admitted to the intensive care unit (ICU) of a tertiary academic medical center. The database contains comprehensive patient information, including demographic characteristics, vital signs, laboratory results, diagnoses, medication records, in-hospital survival status, and follow-up mortality data obtained from the U.S. Social Security Administration Death Master File. All patient identifiers and private information were anonymized to protect patient rights; This study has obtained ethical approval from the Medical Research Ethics Committee of the Affiliated Shunde Hospital of Jinan University (code JDSY-LL-2026002). Author KL completed the required online course and examination and was granted access to the MIMIC-IV database (Certification ID: 54890087).

The patient selection criteria were as follows: Inclusion criteria consisted of: (1) adult patients (aged ≥18 years); (2) a diagnosis of chronic obstructive pulmonary disease (COPD), confirmed by International Classification of Diseases codes recorded during hospitalization (ICD-9: 491.20, 491.21, 491.22, 496; ICD-10: J44.0, J44.1, J44.9). Exclusion criteria were: (1) no ICU admission during hospitalization; (2) missing records of oxygen saturation (SpO₂), respiratory rate (RR), or fraction of inspired oxygen (FiO₂) within the first 24 hours of ICU admission, which are required for calculating the ROX index; and (3) multiple ICU or hospital admissions, for whom only the first admission was retained. Data extraction was performed using Navicat Premium software and Structured Query Language (SQL). This being a retrospective cohort study utilizing an existing database, a priori power calculation was not performed. The sample included all eligible patients, and the large cohort size (N=2,176) is considered adequate for robust multivariable modeling and subgroup analyses.

Collected variables comprised demographic information (gender, age, height, weight, race [White, Black, or Other], and admission type [Emergency/Elective]); disease severity scores (APS III, SAPS II, SOFA score, and Charlson Comorbidity Index); comorbidities (including severe liver disease, malignant cancer, renal disease, myocardial infarction, congestive heart failure, cerebrovascular disease, and diabetes); vital signs (calculated as mean values from the first 24 hours after ICU admission, including temperature [°C], heart rate [beats/min], respiratory rate [breaths/min], mean blood pressure [mmHg], oxygen saturation [%], and blood glucose [mg/dL]); and laboratory parameters (also averaged over the first 24 hours in the ICU, including WBC [K/μL], hemoglobin [g/dL], platelets [K/μL], bicarbonate [mEq/L], ALT [U/L], AST [U/L], albumin [g/dL], blood urea nitrogen [mg/dL], sodium [mEq/L], calcium [mEq/L], potassium [mEq/L], chloride [mEq/L], anion gap [mmol/L], INR, PT [seconds], APTT [seconds], PH, PaO₂ [mmHg], PaCO₂ [mmHg], oxygenation index [mmHg], creatinine [mg/dL], and lactate [mmol/L]). Furthermore, the presence of an acute exacerbation of chronic obstructive pulmonary disease (AECOPD) during hospitalization was documented.

The ROX index was calculated for each patient using data from the first 24 hours after ICU admission. Specifically, all recorded measurements of SpO₂, FiO₂, and respiratory rate (RR) within this 24-hour window were extracted. In the MIMIC-IV database, these vital signs are typically charted at variable intervals, often hourly or more frequently during active clinical care. For each patient and for each of the three parameters (SpO₂, FiO₂, RR), a mean value was calculated from all available data points within the first 24 hours. The final ROX index was then computed according to the formula: ROX = (SpO₂ / FiO₂) / RR. Table S1 presents baseline characteristics and outcomes of patients with and without complete ROX data, allowing assessment of selection bias.

Patients were divided into four equal groups based on their ROX values: Q1 (ROX≤7.79), Q2 (7.79<ROX≤9.72), Q3 (9.72<ROX≤12.26), and Q4 (ROX>12.26). The primary outcome was 28-day all-cause mortality, and secondary outcomes included 3-month mortality, 1-year mortality, in-ICU mortality, and in-hospital mortality.

Categorical variables are presented as frequencies and percentages, with group comparisons performed using the χ²-test or Fisher’s exact test, as appropriate. Continuous variables were first assessed for normality using the Shapiro–Wilk test. Variables following a normal distribution are reported as mean ± standard deviation (SD) and were compared using the one-way ANOVA or Student’s t-test; non-normally distributed variables are presented as median with interquartile range (IQR) and were compared using the Kruskal–Wallis test or Mann–Whitney U-test.

The nonlinear relationship between ROX and 28-day mortality risk was assessed using restricted cubic splines to identify potential inflection points. Survival curves were plotted using Kaplan–Meier curves, and the Log rank test was applied for comparison. Hazard ratios (HR) and 95% confidence intervals (CI) were calculated using Cox proportional hazards regression, with the proportional hazards assumption verified using Schoenfeld residuals (global test P>0.05). Variables with a P-value < 0.1 in univariate Cox regression analysis were initially considered for inclusion in the multivariate model (Table S2). Based on clinical relevance and prior literature, the final multivariate Cox model was further adjusted for age, race, gender, BMI, admission type, MBP, glucose, SAPSII, Charlson Comorbidity Index, severe liver disease, renal disease, congestive heart failure, sepsis, AECOPD, WBC, hemoglobin, creatinine, sodium, potassium, PT, lactate, PH, PaO2, and PaCO2.

To examine the heterogeneity and robustness of the association between the ROX index and outcomes in patients with COPD, subgroup analyses were conducted based on gender, age (≤65 />65 years), race (White/non-White), BMI (≤30 />30), SAPS II score (≤35 />35), sepsis (yes /no) and AECOPD (yes /no).

Missing data were handled by multiple imputation (5 imputations) using the “mice” package in R. Variables with a missing rate exceeding 20% were excluded. Multicollinearity was assessed using the variance inflation factor (VIF), with VIF<5 indicating no significant collinearity (Table S3). All analyses were two-sided, and a P<0.05 was considered statistically significant. Analyses were performed using R version 4.2.3.

Results

The MIMIC-IV database (version 2.0) included 14,279 adult patients diagnosed with COPD. After applying the exclusion criteria, the final study cohort comprised 2,176 patients, who were stratified into four groups based on ROX index quartiles. The detailed patient selection process is presented in Figure 1.

Figure 1 Flowchart of patient selection. Quartile 1: ROX index≤7.79; Quartile 2: 7.79<ROX index≤9.72; Quartile 3: 9.72<ROX index≤12.26; Quartile 4: ROX index>12.26.

The study ultimately included 2,176 patients with a median age of 71.55 (63.58, 78.91) years, including 1,250 males (57.44%). As the ROX index progressively increased, patients exhibited gradual decreases in BMI, SAPS II score, APS III score, heart rate, respiratory rate, white blood cell count, and lactate levels, while oxygen saturation and the oxygenation index showed a corresponding upward trend. Detailed baseline characteristics are presented in Table 1.

Table 1 Baseline Characteristics of Patients with Chronic Obstructive Pulmonary Disease According to the ROX Index

Figure 2 shows the restricted cubic spline curve depicting the association between the ROX index and all-cause mortality in COPD patients. An L-shaped relationship was observed between the ROX index and mortality, both during hospitalization and throughout the follow-up period. To quantify this association, a simple piecewise linear model was fitted. When the ROX index was below 9.72, each unit increase in the ROX index was associated with a significant reduction in 28-day mortality risk (HR per SD, 0.84 [95% CI 0.78–0.90], P<0.001). Above a ROX index of 9.72, the risk curve plateaued, and no significant association was found between the ROX index and 28-day mortality (HR per SD, 1.01 [95% CI 0.96–1.07], P =0.628). To assess the robustness of the identified inflection point, we performed a bootstrap sensitivity analysis with 1,000 resamples, which yielded a 95% confidence interval of 8.74 to 10.35 for the threshold value (Figure S1 and Figure S2), confirming the stability of the 9.72 cutoff. Furthermore, similar associations were observed in Cox proportional hazards analyses for 3-month mortality, 1-year mortality, in-ICU mortality, and in-hospital mortality.

Figure 2 Restricted cubic splines showing the association between the ROX index and all-cause mortality risk in patients with chronic obstructive pulmonary disease. (A) 28-day mortality. (B) 3-month mortality. (C) 1-year mortality. (D) In-ICU mortality. (E) In-hospital mortality.

As shown in Figure 3, patients were categorized into four groups according to ROX index quartiles. Kaplan-Meier survival analysis revealed a progressive decrease in 28-day mortality with increasing ROX index (35.29% vs. 21.88% vs. 18.93% vs. 17.46%, log-rank P<0.001), although the difference between Quartile 3 and Quartile 4 did not reach statistical significance. Similar trends were observed for 3-month and 1-year mortality.

Figure 3 Kaplan–Meier curves for all-cause mortality stratified by the ROX index levels. (A) 28-day mortality. (B) 3-month mortality. (C) 1-year mortality. Quartile 1: ROX index≤7.79; Quartile 2: 7.79<ROX index≤9.72; Quartile 3: 9.72<ROX index≤12.26; Quartile 4: ROX index>12.26.

As shown in Table 2, multivariable regression analysis of outcomes across the four patient groups revealed that, using Quartile 1 as the reference, Quartile 2 (HR=0.75, 95% CI 0.59–0.96, P=0.021), Quartile 3 (HR=0.60, 95% CI 0.46–0.77, P<0.001), and Quartile 4 (HR=0.58, 95% CI 0.45–0.75, P<0.001) all showed significantly lower 28-day mortality. Similar trends were observed for 3-month and 1-year mortality. Regarding in-ICU and in-hospital mortality, while the difference between Quartile 2 and Quartile 1 was not statistically significant (P>0.05), both Quartile 3 and Quartile 4 demonstrated significantly reduced all-cause mortality compared to Quartile 1. Pairwise comparisons among quartiles were further examined using multivariable Cox regression (Tables S4 and S5).

Table 2 Associations Between ROX Index Quartiles and All-Cause Mortality at Different Time Points

As shown in Figure 4, subgroup analyses of COPD patients with ROX index below 9.72 demonstrated significant associations between ROX index and 28-day mortality across all subgroups: male (HR=0.81, 95% CI 0.74–0.89, P<0.001), female (HR=0.89, 95% CI 0.80–0.99, P=0.045), age ≤65 years (HR=0.74, 95% CI 0.63–0.88, P<0.001), age >65 years (HR=0.88, 95% CI 0.81–0.96, P<0.001), White race (HR=0.81, 95% CI 0.75–0.88, P<0.001), non-White race (HR=0.88, 95% CI 0.78–0.99, P=0.044), BMI ≤30 (HR=0.83, 95% CI 0.75–0.92, P<0.001), BMI >30 (HR=0.85, 95% CI 0.77–0.94, P=0.002), SAPS II score ≤35 (HR=0.74, 95% CI 0.58–0.93, P=0.011), SAPS II score >35 (HR=0.84, 95% CI 0.78–0.91, P<0.001), with sepsis (HR=0.84, 95% CI 0.77–0.91, P<0.001), without sepsis (HR=0.87, 95% CI 0.76–0.99, P=0.046), with AECOPD (HR=0.77, 95% CI 0.65–0.92, P=0.003), and without AECOPD (HR=0.85, 95% CI 0.79–0.92, P<0.001). Except for age, no significant interactions were detected between the ROX index and most other variables (all P for interaction > 0.05).

Figure 4 Forest plot showing subgroup analyses of the association between the ROX index and 28-day mortality in patients with a ROX index<9.72.

Discussion

This study revealed a significant nonlinear association between the ROX index and mortality in COPD patients, demonstrating its utility as an effective tool for assessing mortality risk in this population.

This study revealed a significant L-shaped association between the ROX index and mortality in patients with COPD. This relationship may stem from the index’s formula, which integrates two core pathophysiological processes during acute exacerbations of COPD: impaired gas exchange (reflected by SpO₂/FiO₂) and increased respiratory load (reflected by respiratory rate).^3,17 Specifically, when the ROX index falls below the critical threshold of 9.72, a decrease in its value suggests that the patient’s pulmonary function may be in a state of acute decompensation. This state is characterized by a vicious cycle of hypoxemia and tachypnea triggered by worsened ventilation/perfusion mismatch and dynamic pulmonary hyperinflation,¹⁸ indicating rapidly escalating respiratory muscle fatigue and a sharply elevated risk of mortality. Conversely, when the index exceeds this threshold, the respiratory system stabilizes, and mortality risk plateaus. At this stage, patient prognosis is likely influenced more by non-respiratory factors, such as underlying comorbidities and organ functional reserve.

This study is the first to validate a significant L-shaped nonlinear association between the ROX index and 28-day mortality in critically ill COPD patients, establishing a specific prognostic threshold of 9.72. This finding extends and refines our previous work in acute hypoxemic respiratory failure (AHRF) cohort, suggesting that the L-shaped association may represent a general prognostic pattern across diseases.^19,20 The critical distinction, however, lies in the disease-specific threshold. While reported values are 8.28 for AHRF and approximately 10 for sepsis, the threshold for COPD is 9.72. This difference likely originates from the unique pathophysiology of COPD, including chronic airflow limitation, dynamic hyperinflation, and a tendency toward hypercapnia.

The ROX index is an established bedside tool for assessing HFNC outcomes in AHRF, with a value >6.88 also shown to predict treatment success in COPD-related respiratory failure.¹⁶ Additionally, the SpO₂/FiO₂ ratio has been demonstrated to exhibit a certain correlation with the PaO₂/FiO₂ ratio,^21,22 and may serve as a valid substitute in specific clinical scenarios where arterial blood gas monitoring is unavailable.²³ Compared to the PaO₂/FiO₂ ratio, our findings suggest a distinct role for the ROX index in COPD. Given the high median SpO₂ in our cohort, the PaO₂/FiO₂ ratio may have limited sensitivity. The ROX index adds crucial value by incorporating respiratory rate, thereby capturing the increased respiratory drive typical of COPD exacerbations. It is thus more suitable for early risk stratification in this population, while PaO₂/FiO₂ remains key for managing severe hypoxemia.

The integration of respiratory rate gives the ROX index a particular advantage in COPD, where dynamic hyperinflation leads to increased work of breathing. This allows it to signal deterioration even before oxygenation significantly declines.^3,17 It identifies patients at high risk of mortality, facilitating repeated bedside assessment. A key interpretation from our data is that in patients with preserved SpO₂, the predictive power of the ROX index likely stems predominantly from the respiratory rate component. This highlights respiratory effort as a primary risk driver in such exacerbations. The index is most useful in critically ill COPD patients and has limited utility in stable outpatients or those on palliative care. Additionally, we observed differences in the association between the ROX index and mortality at different time points. One possible explanation is that ICU mortality and in-hospital mortality are more directly influenced by the severity of the patient’s acute illness and the therapeutic interventions administered during the ICU stay. In contrast, follow-up mortality rates are more closely correlated with the underlying severity and natural progression of COPD.

Overall, the ROX index offers the advantages of being non-invasive, straightforward, and suitable for continuous dynamic monitoring. It is applicable for early screening, routine monitoring, and treatment response evaluation in critically ill COPD patients. The index’s simplicity and reliance on routinely available parameters (SpO₂, RR, and FiO₂) give it potential utility in scenarios requiring frequent assessment, though this requires validation in such specific environments. Its primary limitation lies in the reduced sensitivity for detecting worsening oxygenation when SpO₂ is near normal,²⁴ although the incorporation of respiratory rate partially mitigates this shortcoming. In contrast, the oxygenation index, as an invasive gold-standard assessment, provides more precise oxygenation data but is limited in routine dynamic evaluation due to operational constraints and the inability for continuous monitoring. Therefore, these two indices hold complementary value in clinical practice: the ROX index is suitable for daily dynamic monitoring and risk stratification, whereas at critical junctures or when the ROX index contradicts the clinical picture, the oxygenation index should be obtained via arterial blood gas analysis for calibration.

This study has several limitations. First, as a single-center retrospective cohort study utilizing the MIMIC-IV database, the findings are derived from the patient population and clinical practices of a specific tertiary academic medical center, which may limit their generalizability to other healthcare settings. Second, although we adjusted for a range of potential confounders, residual confounding may still exist due to unmeasured or incompletely documented variables inherent to retrospective research, such as FEV₁, smoking status, long-term oxygen therapy, and details of inhaled maintenance therapy, furthermore, excluding 2312 patients with missing ROX component data may have introduced selection bias. Third, the prognostic value of the ROX index observed in this critically ill ICU cohort may not be applicable to non-ICU COPD populations, including stable outpatients or patients managed in general wards. Fourth, due to challenges in reliably determining the primary reason for hospital or ICU admission from ICD codes in the MIMIC-IV database, we were unable to assess whether the optimal ROX index threshold varies according to different admission etiologies. Finally, regarding measurement, the ROX index was calculated using averaged values of its components over the first 24 hours of ICU admission to obtain a stable estimate. While this approach reduces the influence of transient fluctuations, it may obscure dynamic temporal changes that could carry additional prognostic significance. Moreover, the use of retrospectively extracted discrete measurements may not fully capture the continuous physiological status of patients.¹⁴

Conclusion

In conclusion, this study demonstrates that the ROX index is an independent and effective predictor of mortality risk in critically ill COPD patients. Notably, this association is nonlinear and threshold-dependent, with prognostic value most pronounced when the index falls below the identified cutoff of approximately 9.72. The significant association between lower ROX index values and increased mortality persisted after multivariable adjustment for available clinical covariates in this retrospective cohort. Our findings support extending the clinical application of the ROX index as a practical bedside tool, particularly for early risk stratification at ICU admission. As these conclusions are derived from ROX index values averaged over the initial 24 hours of ICU care, future studies are warranted to compare the predictive utility of alternative ROX metrics—such as the initial value, the worst value within 24 hours, or serial trends—to identify the most clinically actionable approach.