Abstract
This study aims to investigate the potential of the red cell distribution width to total calcium ratio (RCR) as a biomarker for in-hospital mortality in patients with non-idiopathic pulmonary fibrosis (IPF) interstitial lung diseases. A retrospective cohort analysis was carried out utilizing the Medical Information Mart for Intensive Care database, including 1138 patients with non-idiopathic pulmonary fibrosis interstitial lung diseases. Patients were divided into a survivor group (n = 1023) and a non-survivor group (n = 115) based on in-hospital mortality. The Boruta algorithm combined with a machine learning-based random forest algorithm was employed to calculate Shapley Additive Explanations (SHAP) values to identify clinical indicators significantly contributing to in-hospital mortality. A nomogram model based on logistic regression was constructed to assess the relationship between RCR and in-hospital mortality. Compared to the survivor group, the non-survivor group’s average age was significantly older (73.00 ± 10.67 years vs 69.83 ± 13.24 years, P = .013), and RCR was significantly elevated in the non-survivor group (1.83 ± 0.30 vs 1.73 ± 0.27, P <.001). After adjusting for white blood cell count, blood urea nitrogen (BUN), sodium levels, and pneumonia in the model, the odds ratio for RCR was 2.283 (95% CI: 1.108–4.649, P = .024). BUN was identified as a mediator, accounting for approximately 14.6% of the indirect effect. Subgroup analyses revealed a stronger association of RCR with in-hospital mortality in female patients, those aged ≤65 years, and patients with hypertension. The nomogram model’s C-index was 0.771 for the training set and 0.764 for the validation set. The training set’s area under the curve was 0.771 (95% CI: 0.712–0.829), while the validation set’s was 0.764 (95% CI: 0.706–0.821). RCR serves as a simple and effective biomarker for predicting in-hospital mortality risk in patients with non-idiopathic pulmonary fibrosis, with BUN playing a mediating role in this association.
Keywords: mediating effect, nomogram model, non-idiopathic pulmonary fibrosis, red cell distribution width, serum calcium
1. Introduction
Non-idiopathic pulmonary fibrosis (IPF) refers to all fibrotic interstitial lung diseases (F-ILDs) excluding IPF, encompassing a variety of etiologies and clinical phenotypes, with the core features being chronic inflammation of the lung interstitium and progressive fibrosis.[1] Non-IPF interstitial lung diseases (non-IPF ILDs) encompass a heterogeneous group of fibrotic lung disorders with diverse etiologies, including connective tissue disease-associated ILD, hypersensitivity pneumonitis, and unclassifiable forms.[2,3] Notably, some non-IPF ILDs can develop into a progressive fibrotic phenotype (PF-ILD), which is frequently linked to a poor prognosis. Clinical characteristics of PF-ILD include worsening lung function, exacerbation of symptoms, and CT evidence of fibrotic progression. For patients with non-IPF ILD, assessing the potential risk of in-hospital mortality holds significant clinical implications.
Red cell distribution width (RDW) is a conventional parameter reflecting the heterogeneity of peripheral blood erythrocyte volume and has demonstrated significant value in the clinical assessment of various diseases in recent years.[4] RDW and inflammatory reactions are tightly related, as the release of inflammatory factors can lead to an imbalance between erythropoiesis and erythrocyte destruction.[5] Total serum calcium also plays a significant role in inflammatory responses.[6] Hypocalcemia may exacerbate inflammatory responses and tissue damage by affecting the function of immune cells and modulating inflammatory signaling pathways.[7] The red cell distribution width to total calcium ratio (RCR) is a composite reflection of RDW and total serum calcium. In view of the fact that RDW and serum calcium respectively reflect 2 complementary pathophysiological processes of inflammation and metabolic homeostasis, RCR may better reflect the comprehensive severity of the disease than either parameter alone.[8] In patients with pulmonary diseases, inflammatory responses can activate immune cells, leading to the release of various cytokines and inflammatory mediators. This process simultaneously affects protein metabolism and renal function, resulting in alterations in blood urea nitrogen (BUN) levels.[9]
Although previous studies on RCR have primarily focused on conditions such as pancreatitis and cardiovascular diseases, it is important to consider that the progression of non-IPF ILD is closely linked to inflammation. Additionally, RCR does not require complex calculations or multiple tests, making it clinically practical for assessing prognosis in non-IPF ILD patients. In this study, we hypothesize that an elevated RCR is independently associated with higher in-hospital mortality in patients with non-IPF interstitial lung disease, and that this association is partially mediated by elevated BUN levels.
2. Materials and methods
2.1. Study population
This retrospective cohort study utilized the Medical Information Mart for Intensive Care IV database (MIMIC-IV). The MIMIC-IV database contains clinical data from over 380,000 patients admitted to the Beth Israel Deaconess Medical Center in Boston, Massachusetts, between 2008 and 2019. The participants were included based on the following criteria: diagnosed with non-IPF based on ICD coding (ICD 10: J8410; ICD 9: 515); aged over 18 years and under 90 years. Exclusion criteria included: patient who were not being hospitalized for the first time; patients lacking corresponding clinical data (see Fig. 1). Patients were categorized into 2 groups based on in-hospital mortality: the survivor group and the non-survivor group.
Figure 1.
Flow chart of patient inclusion.
The MIMIC – IV database used in this study has been approved by the institutional review board of Beth Israel Deaconess Medical Center (2001 P – 001699/14). Both ethics review committees have approved the use of this database for research purposes. The personal information of patients in the database has been anonymized, and ethical review and informed consent have been waived.
2.2. Study variables
Baseline characteristics included the age, sex, marital status, hemoglobin, platelet count, RDW, white blood cell count (WBC), red blood cell count (RBC), anion gap, total calcium, chloride, glucose, potassium, sodium, international normalized ratio, creatinine, BUN, body mass index, and RCR of the enrolled patients. RCR was calculated as the ratio of RDW to total calcium ion levels.[10] Comorbidities included the presence of hypertension, pneumonia, stroke, chronic kidney disease, malignancy, type 2 diabetes mellitus (T2DM), chronic bronchitis, coronary heart disease, and chronic obstructive pulmonary disease. All laboratory parameters analyzed were based on the results from the first measurements taken upon hospital admission.
2.3. Statistical analysis
Data analysis was performed using R software (version 4.1.0). In order to evaluate the normality of continuous variables, the Kolmogorov-Smirnov test was implemented. The results of normally distributed data are presented as means ± standard deviations, and statistical comparisons are conducted using independent t-tests. For continuous variables with a non-normal distribution, data are described using medians and interquartile ranges, and statistical analysis was conducted using the Mann–Whitney U test. Categorical variables are presented as frequencies (percentages) and analyzed using chi-square tests. First, the Boruta algorithm was used to identify clinical indicators that have a significant impact on inpatient mortality within the overall study population. The Shapley additive explanations (SHAP) values were calculated using a machine learning-based random forest algorithm to visualize the importance of feature variables, allowing for the identification of indicators significantly associated with in-hospital mortality. Three logistic regression models were constructed: Model 1 included only the RCR; Model 2 included gender and age as covariates; and Model 3 built upon Model 2 by incorporating feature variables selected based on SHAP values. Multicollinearity was assessed using the variance inflation factor, with a threshold set at 5. The nonlinear relationship between in-hospital mortality and RCR was assessed using a restricted cubic spline with 4 knots. To develop a nomogram model, the study population was randomly divided into training (60%) and testing (40%) sets. The nomogram was constructed using the training set and validated in the testing set. The model’s discriminative ability and accuracy were evaluated using the area under the receiver operating characteristic curve (AUC) and the concordance index (C-index). Clinical decision curves were plotted to evaluate the potential benefits of the model for patients. Mediation analysis was conducted to assess whether BUN mediated the impact of RCR on mortality in the hospital. Prior to fitting the mediation models, we evaluated the linearity assumptions. Finally, subgroup analyses were performed to evaluate the significance of RCR effects across different subpopulations. For the clinical indicators included in the study, missing values were deleted. A 2-tailed P-value of <.05 was considered statistically significant.
3. Results
3.1. Patient characteristics
This study included 1138 patients with non-IPF ILD, comparing baseline characteristics between the survival group (n = 1023) and the in-hospital mortality group (n = 115). The results indicated that the non-survivor group had a substantially older age than the survivor group (73.00 ± 10.67 years vs 69.83 ± 13.24 years, P = .013). The RCR was significantly higher in the non-survivorgroup (1.83 ± 0.3 vs 1.73 ± 0.27, P <.001). Regarding comorbidities, the proportion of patients with pneumonia was significantly higher in the non-survivor group compared to the survival group (57.39% vs 25.12%, P <.001) (see Table 1).
Table 1.
General information of included patients.
| Variable Names | Overall (n = 1138) | Survivor (n = 1023) | Non-survivor (n = 115) | P-value |
|---|---|---|---|---|
| Age (year) | 70.15 ± 13.03 | 69.83 ± 13.24 | 73.00 ± 10.67 | .013 |
| Gender (n, %) | ||||
| Female | 543 (47.72) | 488 (47.70) | 55 (47.83) | 1.000 |
| Male | 595 (52.28) | 535 (52.30) | 60 (52.17) | |
| Marriage (n, %) | ||||
| Married | 570 (50.09) | 518 (50.64) | 52 (45.22) | .316 |
| Other | 568 (49.91) | 505 (49.36) | 63 (54.78) | |
| Hemoglobin (g/dL) | 11.18 ± 2.04 | 11.20 ± 2.03 | 10.95 ± 2.07 | .205 |
| Platelet (109/L) | 228.10 ± 98.87 | 229.00 ± 97.77 | 220.11 ± 108.26 | .361 |
| RDW (%) | 15.01 ± 1.92 | 14.96 ± 1.9 | 15.45 ± 2.01 | .009 |
| RBC (1012/L) | 3.75 ± 0.67 | 3.76 ± 0.67 | 3.62 ± 0.7 | .029 |
| WBC (1012/L) | 8.98 (6.40;12.16) | 8.70 (6.40;11.70) | 11.10 (8.60;15.55) | <.001* |
| Anion gap (mEq/L) | 13.56 ± 3.02 | 13.50 ± 2.95 | 14.03 ± 3.56 | .073 |
| Calcium total (mmol/L) | 8.70 ± 0.65 | 8.72 ± 0.64 | 8.47 ± 0.73 | <.001 |
| Chloride (mmol/L) | 101.87 ± 5.29 | 101.96 ± 5.18 | 101.05 ± 6.16 | .078 |
| Glucose (mg/dL) | 110.00 (94.00;138.00) | 109.00 (93.00;137.00) | 118.00 (105.00;156.00) | <.001* |
| Potassium (mmol/L) | 4.14 ± 0.53 | 4.14 ± 0.52 | 4.15 ± 0.61 | .832 |
| Sodium (mmol/L) | 138.28 ± 4.12 | 138.4 ± 4.01 | 137.23 ± 4.83 | .004 |
| INR | 1.20 (1.10;1.40) | 1.20 (1.10;1.40) | 1.30 (1.20;1.70) | <.001* |
| Creatinine (mg/dL) | 0.90 (0.70;1.30) | 0.90 (0.70;1.30) | 0.90 (0.70;1.45) | .191* |
| BUN (mg/dL) | 19.00 (13.00;28.75) | 18.00 (13.00;27.00) | 23.00 (15.00;34.00) | <.001* |
| BMI (kg/m2) | 27.48 ± 5.13 | 27.40 ± 5.13 | 28.16 ± 5.14 | .134 |
| RCR | 1.74 ± 0.28 | 1.73 ± 0.27 | 1.83 ± 0.30 | <.001 |
| Hypertension (n, %) | ||||
| No | 698 (61.34) | 628 (61.39) | 70 (60.87) | .994 |
| Yes | 440 (38.66) | 395 (38.61) | 45 (39.13) | |
| Pneumonia (n, %) | ||||
| No | 815 (71.62) | 766 (74.88) | 49 (42.61) | <.001 |
| Yes | 323 (28.38) | 257 (25.12) | 66 (57.39) | |
| Stroke (n, %) | ||||
| No | 1046 (91.92) | 943 (92.18) | 103 (89.57) | .427 |
| Yes | 92 (8.08) | 80 (7.82) | 12 (10.43) | |
| CKD (n, %) | ||||
| No | 879 (77.24) | 787 (76.93) | 92 (80.00) | .531 |
| Yes | 259 (22.76) | 236 (23.07) | 23 (20.00) | |
| Malignant tumor (n, %) | ||||
| No | 929 (81.63) | 832 (81.33) | 97 (84.35) | .506 |
| Yes | 209 (18.37) | 191 (18.67) | 18 (15.65) | |
| T2DM (n, %) | ||||
| No | 842 (73.99) | 749 (73.22) | 93 (80.87) | .097 |
| Yes | 296 (26.01) | 274 (26.78) | 22 (19.13) | |
| Chronic bronchitis (n, %) | ||||
| No | 1037 (91.12) | 933 (91.20) | 104 (90.43) | .919 |
| Yes | 101 (8.88) | 90 (8.80) | 11 (9.57) | |
| CHD (n, %) | ||||
| No | 758 (66.61) | 683 (66.76) | 75 (65.22) | .819 |
| Yes | 380 (33.39) | 340 (33.24) | 40 (34.78) | |
| COPD (n, %) | ||||
| No | 787 (69.16) | 706 (69.01) | 81 (70.43) | .836 |
| Yes | 351 (30.84) | 317 (30.99) | 34 (29.57) | |
BMI = body mass index, BUN = blood urea nitrogen, CHD = coronary heart disease, CKD = chronic kidney disease, COPD = chronic obstructive pulmonary disease, INR = international normalized ratio, RBC = red blood cell count, RCR = the red cell distribution width to total calcium ratio, RDW = red cell distribution width, T2DM = type 2 diabetes mellitus, WBC = white blood cell count.
, compared by Mann–Whitney U test.
3.2. Results of the Boruta algorithm
The Boruta feature importance box plot (Fig. 2A) illustrates the importance of each variable, with RCR, RDW, WBC, BUN, and pneumonia identified as significant predictors. The SHAP bees plot (Fig. 2B) indicates that pneumonia has a relatively large impact on mortality. WBC and BUN levels also make notable positive contributions to the risk of mortality.
Figure 2.
Results of feature selection (A) Feature selection based on the Boruta algorithm. The Boruta algorithm was employed to analyze the relationship between various clinical indicators and in-hospital mortality. The Z-score associated with a shadow feature for each variable is represented on the vertical axis. The boxplot illustrates the Z-scores of each variable as calculated by the model; (B) SHAP values calculated from random forest This panel presents the distribution of each feature’s impact on the model output. Each point represents an individual patient in the dataset. The color of the points indicates the feature value: purple denotes larger values, while yellow indicates smaller values.
3.3. Logistic regression analysis
In model 1, the odds ratio (OR) for the RCR index was 3.282 (95% CI: 1.704–6.258), P <.001. In model 2, the OR was 3.462 (95% CI: 1.779–6.678), P <.001. Model 3 yielding an OR of 2.283 (95% CI: 1.108–4.649), P = .024, which remained statistically significant. Model 3 included adjustments for age, gender, white blood cell count, BUN, sodium, and pneumonia. The variables in Model 3 were tested and found to be free of multicollinearity. The risk of in-hospital mortality was substantially higher in the Q4 group than in the Q1 group, as indicated by the stratified analysis. Notably, this association attenuated after full adjustment for potential confounders including white blood cell count, BUN, serum sodium, and pneumonia status. Specifically, the OR for the highest RCR quartile (Q4) versus the lowest (Q1) decreased from 2.397 (95% CI: 1.37–4.338; P = .003) in Model 2 to 1.662 (95% CI: 0.911–3.111; P = .103) in Model 3, and the P for trend weakened to 0.045. These findings suggest that part of the observed association may be mediated or confounded by systemic inflammation, renal function, electrolyte imbalance, or acute comorbidities such as pneumonia (Table 2). The restricted cubic spline curve suggests that there is no nonlinear relationship between RCR and in-hospital mortality (Fig. 3).
Table 2.
Analysis of the association between RCR and in-hospital mortality.
| Model 1 | P-value | Model 2 | P-value | Model 3 | P-value | |
|---|---|---|---|---|---|---|
| RCR index | 3.282 (1.704–6.258) | <0.001 | 3.462 (1.779–6.678) | <.001 | 2.283 (1.108~4.649) | .024 |
| Q1 | Ref | – | Ref | – | Ref | – |
| Q2 | 1.171 (0.619–2.234) | 0.627 | 1.175 (0.621–2.244) | .621 | 0.961 (0.494~1.883) | .908 |
| Q3 | 1.848 (1.034–3.391) | 0.041 | 1.807 (1.010–3.321) | .050 | 1.504 (0.818~2.832) | .195 |
| Q4 | 2.352 (1.347–4.247) | 0.003 | 2.397 (1.37–4.338) | .003 | 1.662 (0.911~3.111) | .103 |
| P for trend | – | <.001 | – | .001 | – | .045 |
Model 1: no covariates were adjusted. Model 2: age and gender were adjusted. Model 3: age, gender, white blood cells BUN = sodium and pneumonia were adjusted.
BUN = blood urea nitrogen, RCR = the red cell distribution width to total calcium ratio.
Figure 3.
The relationship between RCR with in-hospital mortality in non-IPF ILD. ILD = interstitial lung disease, IPF = idiopathic pulmonary fibrosis, RCR = the red cell distribution width to total calcium ratio.
3.4. Nomogram model
A nomogram model was constructed (Fig. 4). The predictive model’s C-index in the training set was 0.771, and the Hosmer and Lemeshow goodness-of-fit test P-value was 0.977; AUC: 0.771, 95% CI: 0.712–0.829. In the validation set, the C-index of the predictive model was 0.764, with a Hosmer and Lemeshow goodness-of-fit test P-value of 0.052; AUC: 0.764, 95% CI: 0.706–0.821. The calibration curves and decision curve analysis curves of both the training set and the validation set indicate that the prediction model has good accuracy and clinical benefit ability (Fig. 5).
Figure 4.
The nomogram model for predicting in – hospital death constructed based on logistic regression.
Figure 5.
Assessment of the nomogram model; (A–C) ROC curve, calibration curve, and DCA curve for the training set; (D–F) ROC curve, calibration curve, and DCA curve for the validation set. DCA = decision curve analysis, ROC = receiver operating characteristic.
3.5. Results of mediation effect analysis
The mediation effect analysis revealed that the total effect of RCR on in-hospital mortality during patient hospitalization was 0.030 (95% CI: 0.020–0.040, P <.001). The indirect effect (ACME) mediated by BUN was 0.005 (95% CI: 0.002–0.010, P <.001), accounting for 14.6% of the total effect (95% CI: 7.6%–44.0%, P <.001). See Table 3 and Figure 6.
Table 3.
BUN as a mediator variable between RCR and the correlation of in-hospital death.
| Mediation effect | Estimate | 95% CI Lower | 95% CI Upper | P-value |
|---|---|---|---|---|
| Total effect | 0.030 | 0.020 | 0.040 | <.001 |
| ACME | 0.005 | 0.002 | 0.010 | <.001 |
| ADE | 0.025 | 0.016 | 0.030 | <.001 |
| Proportion mediated | 0.146 | 0.076 | 0.440 | <.001 |
ACME = average causal mediation effect, ADE = average direct effect, BUN = blood urea nitrogen, CI = confidence interval, RCR = the red cell distribution width to total calcium ratio.
Figure 6.
Mediation analysis evaluating the role of BUN as an intermediary in the relationship between in-hospital mortality and RCR. BUN = blood urea nitrogen, RCR = the red cell distribution width to total calcium ratio.
3.6. Subgroup analysis
After stratifying by gender, the association strength was higher in female patients (OR = 4.54, 95% CI: 1.78–11.58, P = .002), while the association in male patients was relatively weaker (OR = 2.44, 95% CI: 0.99–6.03, P = .053). In patients aged ≤65 years, the correlation between in-hospital mortality and RCR was larger (OR = 3.86, 95% CI: 1.76–8.46, P <.001). Among patients with hypertension, the association was stronger (OR = 4.45, 95% CI: 1.43–13.84, P = .010), whereas in patients without hypertension, the association was somewhat weaker (OR = 2.96, 95% CI: 1.32–6.62, P = .008). See Table 4.
Table 4.
Subgroup analysis of relation between RCR and in-hospital mortality.
| Variables | n (%) | OR (95% CI) | P-value | P for interaction |
|---|---|---|---|---|
| All patients | 1138 (100.00) | 3.28 (1.71–6.28) | <.001 | – |
| Gender | ||||
| Female | 543 (47.72) | 4.54 (1.78–11.58) | .002 | .350 |
| Male | 595 (52.28) | 2.44 (0.99–6.03) | .053 | |
| Age | ||||
| >65 | 359 (31.55) | 2.55 (0.77–8.48) | .126 | .572 |
| ≤65 | 779 (68.45) | 3.86 (1.76–8.46) | <.001 | |
| Hypertension | ||||
| No | 698 (61.34) | 2.96 (1.32–6.62) | .008 | .567 |
| Yes | 440 (38.66) | 4.45 (1.43–13.84) | .010 | |
| Stroke | ||||
| No | 1046 (91.92) | 3.11 (1.56–6.21) | .001 | .717 |
| Yes | 92 (8.08) | 4.54 (0.67–31.01) | .123 | |
| CKD | ||||
| No | 879 (77.24) | 4.08 (2.01–8.27) | <.001 | .161 |
| Yes | 259 (22.76) | 1.16 (0.23–5.96) | .855 | |
| Malignant tumor | ||||
| No | 929 (81.63) | 3.22 (1.58–6.56) | .001 | .763 |
| Yes | 209 (18.37) | 4.21 (0.85–20.76) | .078 | |
| T2DM | ||||
| No | 842 (73.99) | 3.51 (1.72–7.16) | <.001 | .730 |
| Yes | 296 (26.01) | 2.60 (0.55–12.31) | .228 | |
| Chronic bronchitis | ||||
| No | 1037 (91.12) | 3.52 (1.78–6.94) | <.001 | .502 |
| Yes | 101 (8.88) | 1.60 (0.17–14.75) | .677 | |
| CHD | ||||
| No | 758 (66.61) | 4.22 (1.94–9.21) | <.001 | .260 |
| Yes | 380 (33.39) | 1.89 (0.59–6.06) | .285 | |
| COPD | ||||
| No | 787 (69.16) | 4.22 (1.96–9.08) | <.001 | .240 |
| Yes | 351 (30.84) | 1.77 (0.51–6.14) | .367 | |
| Marriage | ||||
| No | 570 (50.09) | 2.52 (0.94–6.72) | .065 | .487 |
| Yes | 568 (49.91) | 4.01 (1.68–9.56) | .002 | |
CHD = coronary heart disease, CI = confidence interval, CKD = chronic kidney disease, COPD = chronic obstructive pulmonary disease, OR = odds ratio, RCR = the red cell distribution width to total calcium ratio, T2DM = type 2 diabetes mellitus.
4. Discussion
This study explored the potential of RCR as a biomarker for in-hospital mortality risk in non-IPF ILD patients. The results indicated that RCR not only exhibited significant differences in distinguishing between surviving and deceased patients during hospitalization but also contributed to the development of a nomogram model with strong predictive capability through multivariable adjustment. This provides a robust reference for early detection of high-risk patients in medical care.
Non-IPF ILD refers to interstitial lung diseases characterized by known causes or associations with other conditions, which represent an important component of ILD.[11] Non-IPF ILD encompasses various types of pulmonary fibrosis with known causes or associations with other diseases, primarily including autoimmune-related, exposure-related, and granulomatous diseases.[12] In terms of treatment, antifibrotic agents such as nintedanib and pirfenidone have been expanded from IPF to non-IPF types, while also considering etiology-specific therapies.[13] The prognosis of non-IPF ILD varies by type; non-fibrotic forms or those with early intervention tend to have better outcomes, while those progressing to progressive fibrosis have outcomes similar to IPF. In terms of pathogenesis, non-IPF ILD exhibits complex characteristics across multiple dimensions and layers, featuring common pathways similar to IPF as well as specific mechanisms due to differences in etiology. The immune inflammatory response is a core initiating factor.[14] Previous survival analyses have shown that the rate of 5-year survival for non-IPF patients is 72.8%. (95% CI: 66.0%–78.5%), significantly higher than that of IPF patients, which is 53.7% (95% CI: 46.6%–58.5%).[15] Identifying biological markers associated with mortality in non-IPF patients holds significant clinical value for facilitating early intervention.
The results of present study underscore the clinical utility of RCR as a readily available, cost-effective biomarker that provides prognostic information beyond established risk factors such as age, renal function (as reflected by BUN), and acute infection. Notably, even after rigorous adjustment for confounders, elevated RCR remained independently associated with in-hospital death. This suggests that RCR captures a distinct pathophysiological dimension, likely reflecting the interplay between systemic inflammation and metabolic dysregulation or immune dysfunction. RCR combines 2 routine laboratory indicators, RDW and serum calcium levels, to create a low-cost and easily accessible composite biomarker. Previous research has demonstrated that RCR is capable of estimating the extent and mortality of acute pancreatitis.[16] Similarly, previous studies have evaluated the prognostic value of the neutrophil-to-lymphocyte ratio, platelet-to-lymphocyte ratio, and monocyte-to-high-density lipoprotein ratio (MHR) in patients with pulmonary fibrosis. The AUROC for neutrophil-to-lymphocyte ratio was 0.776, which was superior to that of platelet-to-lymphocyte ratio (0.676) and MHR (0.430).[17] This study is the first to introduce RCR in the context of non-IPF ILD to explore its association with in-hospital mortality. In clinical practice, RCR can be calculated from routine admission blood tests without additional cost or delay. Given the heterogeneity and diagnostic complexity of non-IPF ILDs, early identification of high-risk patients is crucial for timely escalation of care, such as closer monitoring, early antimicrobial therapy, or consideration of antifibrotic treatment.
RDW serves as an indicator of the heterogeneity of red blood cell volume and is typically associated with inflammatory responses, oxygenation status, and nutritional status.[18] Elevated RDW levels are commonly observed in patients with chronic diseases and may reflect an exacerbation of systemic inflammatory status.[19] Research related to acute ischemic stroke suggests that the release of inflammatory factors can disrupt the normal erythropoiesis process in the bone marrow, leading to varying sizes of newly formed red blood cells and subsequently causing an increase in RDW.[20] This study found that patients in the non-survivor group had significantly elevated RCR levels. Furthermore, there was an upward trend in the likelihood of in-hospital mortality with rising RCR quartiles (all P for trend <.05). These results indicate that RCR also holds reference value for patients with non-IPF ILD.
The calcium levels participate in a variety of crucial cellular processes, including cell signaling, apoptosis, and muscle contraction.[21,22] Hypocalcemia may lead to impaired respiratory muscle function, affecting lung ventilation and oxygenation, thereby exacerbating the severity of pneumonia.[23] A study focusing on pediatric populations with pulmonary inflammatory diseases found that calcium and phosphorus imbalance may be involved in the inflammatory processes of febrile respiratory tract infections.[24] An analysis of COVID-19 patients reported that calcium ion imbalance is more prevalent among severe or critically ill COVID-19 patients, suggesting that the occurrence of hypocalcemia may be associated with impaired immune function.[25] Additionally, calcium ions play a crucial role in the cardiovascular system. Low calcium levels may lead to decreased cardiac contractility, affecting cardiac output.[26] RCR may provide a more thorough representation of the patient’s general condition. An elevated RCR may indicate that an enhanced inflammatory response, coupled with calcium homeostasis imbalance, collectively drives the deterioration of the patient’s condition.
This study also identified a mediating effect of BUN in the relationship between in-hospital mortality and RCR. BUN was found to be an important variable mediating the relationship between RCR and in-hospital mortality, accounting for approximately 14.6% of the total effect. This suggests that RCR may indirectly influence patient survival by affecting renal function. This finding supports the critical role of renal function in the deterioration of respiratory system diseases. In patients with pulmonary diseases, changes in BUN levels not only reflect renal function status but are also closely related to nutritional status, inflammatory response, and overall metabolic state.[27,28] The glomerular filtration rate is reduced as a result of impaired renal function, resulting in reduced BUN excretion and consequently elevated BUN levels. In patients with pulmonary diseases, particularly those with severe pulmonary conditions, a systemic inflammatory response may occur, resulting in an increase in protein catabolism.[29] Mediation analysis results suggest that for patients with high RCR values, monitoring and managing renal function can provide a more comprehensive assessment of disease progression. Meanwhile, it should be noted that the mediation analysis suggests a partial indirect association mediated by BUN, but due to the observational nature of the study, causal inference cannot be drawn. The observed pathway may be influenced by residual confounding or reverse causality.
Currently, there is a lack of clinical prediction models specifically for non-IPF ILD. The Boruta algorithm was used for feature selection. The Boruta algorithm’s superiority is its capacity to produce “shadow features” that can be compared to the actual features, allowing for a systematic assessment of the importance of each variable, significantly enhancing the model’s accuracy, stability, and interpretability.[30] Building on this, the study further utilized SHAP values to visualize the importance of the selected features. The constructed nomogram model achieved AUCs of 0.771 in the training set and 0.764 in the validation set. This suggests that the model not only effectively identifies patients at high risk of mortality during hospitalization but also demonstrates good consistency and stability. The nomogram simplifies the interpretation of complex data, providing individualized risk assessments for patients in clinical applications. Notably, although the strength of the correlation between mortality in hospitals and RCR varied across different subgroup analyses, a consistent trend was maintained overall. This was particularly evident in female patients, those aged under 65 years, and patients with comorbid hypertension. These variations may be related to gender differences, underlying disease states, and individual sensitivities to calcium metabolism regulation.
Additionally, this investigation is subject to certain constraints. First, as a retrospective cohort study, it is unable to entirely eradicate the impact of confounding factors. Second, this research focused on predictive performance and did not delve into the model’s performance in real-world applications, such as monitoring disease progression, treatment response, and prognostic outcomes, as well as comparative analyses with other noninvasive indicators. The lack of external validation also limits the generalizability of the results. Third, the current phase is a clinical observational study, lacking foundational experiments to validate the research hypothesis. Future studies should involve long-term, multicenter follow-up research and the application of repeated measures data to verify the model’s practicality.
5. Conclusion
RCR is associated with in-hospital mortality in patients with non-IPF ILD. BUN may partially account for the observed association between RCR and in-hospital mortality, implying a potential indirect pathway. This study provides a predictive tool for disease progression in non-IPF ILD, contributing to the improvement of patient management strategies. While our findings highlight its potential for early risk stratification, prospective validation in multicenter cohorts is warranted before clinical application.
Acknowledgments
We acknowledge the contributions of the MIMIC-IV program registry for the development and continuous updates of the MIMIC-IV database.
Author contributions
Conceptualization: Zengxia Ma, Mingxia Sun.
Data curation: Aiqing Lin, Mingxia Sun.
Formal analysis: Aiqing Lin.
Funding acquisition: Mingxia Sun.
Investigation: Zengxia Ma, Mingxiang Zhou, Mingxia Sun.
Methodology: Mingxia Sun.
Project administration: Mingxia Sun.
Resources: Aiqing Lin, Mingxiang Zhou, Mingxia Sun.
Software: Aiqing Lin, Mingxia Sun.
Supervision: Aiqing Lin, Mingxia Sun.
Validation: Aiqing Lin, Zengxia Ma, Mingxiang Zhou, Mingxia Sun.
Visualization: Aiqing Lin, Mingxia Sun.
Writing – original draft: Aiqing Lin, Zengxia Ma, Mingxiang Zhou, Mingxia Sun.
Writing – review & editing: Aiqing Lin, Zengxia Ma, Mingxia Sun.
Abbreviations:
- AUC
- area under the receiver operating characteristic curve
- BMI
- body mass index
- BUN
- blood urea nitrogen
- CHD
- coronary heart disease
- CKD
- chronic kidney disease
- COPD
- chronic obstructive pulmonary disease
- CTD-ILD
- connective tissue disease-related ILD
- F-ILDs
- fibrotic interstitial lung diseases
- HP
- hypersensitivity pneumonitis
- INR
- international normalized ratio
- MIMIC-IV
- Medical Information Mart for Intensive Care IV database
- OR
- odds ratio
- RA-ILD
- rheumatoid arthritis-related ILD
- RBC
- red blood cell count
- RCR
- the red cell distribution width to total calcium ratio
- RDW
- red cell distribution width
- SHAP
- Shapley additive explanations
- T2DM
- type 2 diabetes mellitus
- VIF
- variance inflation factor
- WBC
- white blood cell count
The author(s) declare that financial support was received for the research and/or publication of this article. This work was supported by the Medical and Health Science and Technology Project of Shandong Province (No: 202403020519).
This study was conducted in accordance with the principles outlined in the Declaration of Helsinki. The use of the MIMIC-IV database was approved by the Institutional Review Boards of Beth Israel Deaconess Medical Center and the Massachusetts Institute of Technology. As the MIMIC-IV database provides data freely, informed consent and ethical approval statements were not required for this study.
The authors have no conflicts of interest to disclose.
The datasets generated during and/or analyzed during the current study are available from the corresponding author on reasonable request.
How to cite this article: Lin A, Ma Z, Zhou M, Sun M. Correlation between red cell distribution width to total calcium ratio and in-hospital mortality in patients with non-idiopathic pulmonary fibrosis interstitial lung diseases: A retrospective cohort study from the MIMIC-IV database. Medicine 2026;105:8(e47560).
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
Contributor Information
Aiqing Lin, Email: laqsdsgw@126.com.
Zengxia Ma, Email: 443999470@qq.com.
Mingxiang Zhou, Email: 1252588313@qq.com.
References
- [1].Goos T, De Sadeleer LJ, Yserbyt J, et al. Progression in the management of non-idiopathic pulmonary fibrosis interstitial lung diseases, where are we now and where we would like to be. J Clin Med. 2021;10:1330. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [2].Kam MLW, Tiew PY, Chai HZ, Low SY. Cluster phenotypes in a non-idiopathic pulmonary fibrosis fibrotic interstitial lung diseases cohort in Singapore. J Thorac Dis. 2022;14:2481–92. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [3].Erçen Diken O, Güngör O, Akkaya H. Evaluation of progressive pulmonary fibrosis in non-idiopathic pulmonary fibrosis-interstitial lung diseases: a cross-sectional study. BMC Pulm Med. 2024;24:403. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [4].García-Escobar A, Lázaro-García R, Goicolea-Ruigómez J, et al. Red blood cell distribution width is a biomarker of red cell dysfunction associated with high systemic inflammation and a prognostic marker in heart failure and cardiovascular disease: a potential predictor of atrial fibrillation recurrence. High Blood Pressure Cardiovasc Prevent. 2024;31:437–49. [DOI] [PubMed] [Google Scholar]
- [5].Zhang L, Yu CH, Guo KP, Huang C-Z, Mo L-Y. Prognostic role of red blood cell distribution width in patients with sepsis: a systematic review and meta-analysis. BMC Immunol. 2020;21:40. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [6].Zhuang Z, Wang X, Huang M, Luo Y, Yu H. Serum calcium improved systemic inflammation marker for predicting survival outcome in rectal cancer. J Gastrointest Oncol. 2021;12:568–79. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [7].Ghafouri E, Bigdeli M, Khalafiyan A, et al. Unmasking the complex roles of hypocalcemia in cancer, COVID-19, and sepsis: engineered nanodelivery and diagnosis. Environ Res. 2023;238:116979. [DOI] [PubMed] [Google Scholar]
- [8].Huang S, Zhang H, Zhuang Z, et al. Propensity score analysis of red cell distribution width to serum calcium ratio in acute myocardial infarction as a predictor of in-hospital mortality. Front Cardiovasc Med. 2023;10:1292153. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [9].Du J, Niu J, Ma L, Sui Y, Wang S. Association between blood urea nitrogen levels and length of stay in patients with pneumonic chronic obstructive pulmonary disease exacerbation: a secondary analysis based on a multicentre, retrospective cohort study. Int J Chron Obstruct Pulmon Dis. 2022;17:2847–56. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [10].Gupta V, Narang SS, Gill CS, Selhi PK, Gupta M. Red cell distribution width and ratio of red cell distribution width-to-total serum calcium as predictors of outcome of acute pancreatitis. Int J Appl Basic Med Res. 2023;13:5–9. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [11].Valenzuela C, Cottin V. Epidemiology and real-life experience in progressive pulmonary fibrosis. Curr Opin Pulm Med. 2022;28:407–13. [DOI] [PubMed] [Google Scholar]
- [12].Wijsenbeek M, Suzuki A, Maher TM. Interstitial lung diseases. Lancet. 2022;400:769–86. [DOI] [PubMed] [Google Scholar]
- [13].Lamb YN. Nintedanib: a review in fibrotic interstitial lung diseases. Drugs. 2021;81:575–86. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [14].Kamiya M, Carter H, Espindola MS, et al. Immune mechanisms in fibrotic interstitial lung disease. Cell. 2024;187:3506–30. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [15].Tsubouchi K, Hamada N, Tokunaga S, et al. Survival and acute exacerbation for patients with idiopathic pulmonary fibrosis (IPF) or non-IPF idiopathic interstitial pneumonias: 5-year follow-up analysis of a prospective multi-institutional patient registry. BMJ Open Respir Res. 2023;10:e001864. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [16].Shivakumar C, Kampani G. Evaluation of red cell distribution width and its ratio to total serum calcium as predictors of severity in acute pancreatitis. J Assoc Physicians India. 2024;72:64–7. [DOI] [PubMed] [Google Scholar]
- [17].Chen Y, Cai J, Zhang M, Yan X. Prognostic role of NLR, PLR and MHR in patients with idiopathic pulmonary fibrosis. Front Immunol. 2022;13:882217. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [18].Lichtman MA. Red cell distribution width as a bellwether of prognosis. Blood Cells Mol Dis. 2024;109:102884. [DOI] [PubMed] [Google Scholar]
- [19].Cao W, Shao Y, Wang N, Jiang Z, Yu S, Wang J. Pretreatment red blood cell distribution width may be a potential biomarker of prognosis in urologic cancer: a systematic review and meta-analysis. Biomark Med. 2022;16:1289–300. [DOI] [PubMed] [Google Scholar]
- [20].Zhang XQ, Shen JH, Zhou Q, Duan X-J, Guo Y-F. Red cell distribution width to total serum calcium ratio and in-hospital mortality risk in patients with acute ischemic stroke: a MIMIC-IV retrospective analysis. Medicine (Baltimore). 2024;103:e38306. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [21].Staruschenko A, Alexander RT, Caplan MJ, Ilatovskaya DV. Calcium signalling and transport in the kidney. Nat Rev Nephrol. 2024;20:541–55. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [22].Tonon CR, Silva T, Pereira FWL, et al. A review of current clinical concepts in the pathophysiology, etiology, diagnosis, and management of hypercalcemia. Med Sci Monit. 2022;28:e935821. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [23].Fekete M, Lehoczki A, Csípő T, et al. The role of trace elements in COPD: pathogenetic mechanisms and therapeutic potential of zinc, iron, magnesium, selenium, manganese, copper, and calcium. Nutrients. 2024;16:4118. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [24].Mao S, Wu L, Shi W. Calcium, phosphorus, magnesium levels in frequent respiratory tract infections. Ann Med. 2023;55:2304661. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [25].Ghanem L, Essayli D, Kotaich J, Zein MA, Sahebkar A, Eid AH. Phenotypic switch of vascular smooth muscle cells in COVID-19: role of cholesterol, calcium, and phosphate. J Cell Physiol. 2024;239:e31424. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [26].De Nicolo B, Cataldi-Stagetti E, Diquigiovanni C, Bonora E. Calcium and reactive oxygen species signaling interplays in cardiac physiology and pathologies. Antioxidants (Basel). 2023;12:353. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [27].Ba C, Wang H, Jiang C, Shi X, Jin J, Fang Q. Clinical manifestations and prognostic factors analysis of patients hospitalised with acute exacerbation of idiopathic pulmonary fibrosis and other interstitial lung diseases. BMJ Open Respir Res. 2024;11:e001997. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [28].Zhang S, Gao L, Zhao Z, et al. Blood urea nitrogen to serum albumin ratio as a new indicator of disease severity and prognosis in idiopathic pulmonary artery hypertension. Respir Med. 2024;227:107643. [DOI] [PubMed] [Google Scholar]
- [29].Rajesh R, Atallah R, Bärnthaler T. Dysregulation of metabolic pathways in pulmonary fibrosis. Pharmacol Ther. 2023;246:108436. [DOI] [PubMed] [Google Scholar]
- [30].Degenhardt F, Seifert S, Szymczak S. Evaluation of variable selection methods for random forests and omics data sets. Brief Bioinform. 2019;20:492–503. [DOI] [PMC free article] [PubMed] [Google Scholar]