Skip to main content
NCBI home page
Nursing Open logoLink to Nursing Open
. 2026 Jul 14;13(7):e70699. doi: 10.1002/nop2.70699

Association Between the Red Blood Cell Distribution Width‐to‐Albumin Ratio, Nursing Notes, and the Length of Hospital Stay in Iron Deficiency Anaemia Complicated With Congestive Heart Failure Patients

Xiao‐fang Fang 1,2, Ya‐ping Xie 1, Yu‐qin Teng 1, Fang Zheng 1, Pei‐yu Lin 1, Yuan‐yuan Hu 3,✉
PMCID: PMC13367939  PMID: 42447273

ABSTRACT

Aims

To evaluate the influence of the red blood cell distribution width‐to‐albumin ratio (RAR) and sentiment scores from nursing notes on the length of hospital stay (LOS) and mortality risk among iron deficiency anaemia (IDA) patients with congestive heart failure (CHF).

Design

A retrospective study.

Methods

Data for this retrospective research were obtained from the Medical Information Mart for Intensive Care IV (MIMIC‐IV) database. Multivariable logistic regression analysis was conducted to determine the individual associations of RAR and sentiment subjectivity scores with 7‐day hospital LOS. Additional multivariable logistic analyses and subgroup analyses were performed to explore the joint effect of RAR and subjectivity combined on 7‐day hospital LOS.

Result

655 IDA patients with CHF were included. Among them, 384 patients (58.63%) had a hospital LOS exceeding 7 days. Multivariable logistic regression revealed that the RAR and the scaled subjectivity were linked to the 7‐day hospital LOS. The combination of RAR and subjectivity was significantly associated with the 7‐day hospital LOS. Compared to the low subjectivity and low RAR group, the group with high subjectivity and high RAR was identified as a risk factor for a 7‐day hospital LOS.

Conclusion

High RAR and high subjectivity were associated with a higher risk of 7‐day hospital LOS in IDA patients with CHF, but the two predictors did not interact. Prospective validation of RAR‐based risk stratification and of joint RAR–subjectivity classification as an exploratory descriptive tool is warranted before clinical implementation.

Implications for Patient Care

Nurses may use RAR and nursing note subjectivity as complementary information to identify patients at higher risk of prolonged hospitalisation and prioritise nursing assessment and care planning accordingly.

Reporting Method

Adhered to the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines.

Patient or Public Contribution

No patient or public contribution.

Keywords: IDA, nursing, RARCHF

1. Introduction

Anaemia, as a major global public health issue, is primarily caused by inadequate dietary iron intake, impaired haemoglobin synthesis, and hemolytic disorders (Safiri et al. 2021). Iron plays a pivotal role in oxygen transport via haemoglobin within red blood cells (RBC). An imbalance in iron metabolism can deplete iron reserves. It leads to decreased haemoglobin production, limited RBC formation, and eventually microcytic hypoproliferative anaemia, commonly known as iron deficiency anaemia (IDA) (Kumar et al. 2022). As the predominant form of anaemia worldwide, IDA can lead to multisystem dysfunction impacting cognitive development, immune response, and physical performance (McCann et al. 2020). Epidemiological data indicated that the global prevalence of anaemia reached 24.3% in 2021, affecting approximately 1.920 billion individuals, with 66.2% of cases attributed to dietary iron deficiency (ID) (Collaborators 2023).

Anaemia is frequently complicated by congestive heart failure (CHF) (Ohno et al. 2015). CHF, which affects over 64 million individuals worldwide, is a complex and life‐threatening condition characterised by high morbidity and mortality, poor functional capacity and quality of life, and exerts significant economic impact (Savarese et al. 2023). Anaemia is linked to increased hospitalisation, morbidity, and mortality. The causes of anaemia in CHF are multifaceted, involving deficiencies in hematinics, specifically IDA (Siddiqui et al. 2022). Notably, IDA may induce a hyperdynamic circulatory state, thereby promoting the development of high‐output HF (Yousaf et al. 2023). A retrospective observational study using the National Inpatient Sample found that in 2019, about 7% of 112,864 CHF hospitalisations were complicated by IDA. IDA in CHF patients leads to prolonged hospitalisation, increased costs, and a higher risk of specific complications (Alharbi et al. 2024).

Red cell distribution width (RDW), an indicator of RBC heterogeneity, is not only used for anaemia differential diagnosis but also relates to systemic inflammatory response, oxidative stress, and prognosis of various diseases (Fava et al. 2019; Salvagno et al. 2015). Albumin, a widely used inflammatory index, is a known prognostic factor for cardiac diseases, including CHF (El Iskandarani et al. 2021). However, its prognostic value may be limited by factors such as chronic illness, malnutrition, and inflammation (Eckart et al. 2020). The RDW‐to‐albumin ratio (RAR) combines these two indices and may provide more comprehensive information than either parameter alone. RAR has been employed to assess adverse outcomes in stroke (Zhao et al. 2021), cancer (Lu et al. 2022), and depression (Shangguan et al. 2025). Further studies have demonstrated that elevated RAR is associated with increased all‐cause and cause‐specific mortality in the general population, as well as higher mortality in CHF patients (Hao et al. 2024; Ni et al. 2022). Clinical narrative content can be extracted from physician notes and quantified using natural language processing techniques (Liu et al. 2022). Nursing notes, including patient symptom descriptions and clinical observations, may capture early signs of clinical deterioration or psychosocial risk factors not reflected by traditional objective measures. Sentiment analysis can evaluate emotional valence by parsing textual lexical symbols (Gohil et al. 2018). However, no studies have yet investigated the combined prognostic value of RAR and nursing note scores‐derived sentiment analysis in IDA patients with CHF. This real‐world data study aims to explore the influence of this combined index (integrating RAR with sentiment scores of nursing notes) on hospital length of stay (LOS) and mortality in IDA patients with CHF, potentially providing new evidence for early identification and personalised intervention in high‐risk populations.

2. Methods

2.1. Data Sources and Study Population

This study was a database‐based retrospective observational study and was reported in accordance with the STROBE statement. This retrospective study utilised data from Medical Information Mart for Intensive Care IV (MIMIC‐IV), which was the first publicly available critical care database. MIMIC‐IV contains comprehensive clinical data including demographic information, laboratory tests, medications, vital signs, procedures, diagnoses, drug administration records, and follow‐up survival status. All patient identifiers were anonymised before public release; thus, ethical review and patient‐informed consent were not required (Johnson et al. 2023).

From the initial cohort of 50,907 adult patients (≥ 18 years) with anaemia in MIMIC‐IV, the study population was identified using predefined inclusion/exclusion criteria: Inclusion criteria: (1) IDA diagnosis according to ICD‐9 (280.0/280.8/280.9) or ICD‐10 (D50.0/D50.8/D50.9) codes (n = 6586); (2) with CHF diagnosis (n = 1400). Exclusion criteria: (1) Hospital LOS < 24 h (n = 72); (2) Missing RDW values (n = 21); (3) Missing albumin values (n = 652). After the screening, 655 IDA patients with CHF were included in the final analyses. Of the 655 included patients, 308 (47.023%) had an Intensive Care Unit (ICU) admission record, while the remaining 347 (52.977%) were admitted to general wards.

2.2. Study Outcome

The primary outcome was hospital LOS. Based on the literature consensus from previous studies on HF or critical care hospitalisation (Aydin and Abdi 2025; Marcus et al. 2026; Stricker et al. 2003), patients were stratified into two groups: < 7 days and ≥ 7 days. Secondary outcomes included 1‐year, 3‐year, and 5‐year survival rates.

2.3. Sentiment Analysis

Data extraction was conducted using structured query language (SQL) via PostgreSQL (version 15.2). For this study, sentiment analysis was conducted on all nursing notes for each patient during their hospitalisation using the Python programming language with the TextBlob natural language processing library. According to the official TextBlob documentation, the underlying Pattern module includes a heuristic negation detection mechanism (e.g., reversing polarity for ‘not good’). The primary evaluation indexes included: sentiment polarity score (range: [−1, 1], with higher values indicating more positive sentiment) and sentiment subjectivity score (range: [0, 1], with higher values reflecting greater subjectivity). The mean values of all nursing note scores for each patient were used as their sentiment characteristic indicators. Detailed analytical methods can be found in the referenced literature (Waudby‐Smith et al. 2018).

To address concerns regarding the validity of TextBlob on clinical narrative, an independent rater‐consistency validation was performed: 50 nursing notes were stratified‐sampled across polarity × subjectivity quartiles and independently rescored by a large language model (LLM) used as an independent rater (LLM‐as‐rater) under a prespecified rubric matched to the TextBlob polarity ([−1, +1]) and subjectivity ([0, +1]) scales. LLM‐as‐rater was selected over human raters because it ensures consistent rubric application, can rapidly and systematically extract relevant content from unstructured text to support clinical workflows, and has demonstrated high inter‐rater reliability with human experts (intraclass correlation coefficient, ICC = 0.818), while reducing evaluation time from approximately 600 to 22 s (Croxford et al. 2025; Seidl et al. 2026). Spearman ρ, Bland–Altman bias, and quadratic‐weighted Cohen's κ were calculated between TextBlob and LLM ratings.

2.4. Collected Variables

The variables collected included: (1) Demographics: age (years), gender (male and female), smoking (yes and no), insurance (medicare, medicaid and other), body mass index (BMI, kg/m2); (2) Comorbidities: hypertension, diabetes, coronary heart disease (CHD), sepsis, obesity, chronic obstructive pulmonary disease (COPD), myocardial infarct (MI), peripheral vascular disease (PVD), and cerebrovascular disease (CVD). Charlson comorbidity index (CCI) was also collected; (3) Laboratory parameters (first 24 h of admission): partial thromboplastin time (PTT, sec), prothrombin time (PT, sec), white blood cell (WBC, K/μL), RBC (m/μL), RDW (%), albumin (g/dL), anion gap (AG, mEq/L), bicarbonate (mEq/L), blood urea nitrogen (BUN, mg/dL), calcium (mg/dL), chloride (mEq/L), creatinine (mg/dL), haematocrit (HCT, %), haemoglobin (g/dL), international normalised ratio (INR), platelets (K/μL), sodium (mEq/L), potassium (mEq/L), total bilirubin (TBil, mg/dL), mean corpuscular haemoglobin concentration (MCHC, g/dL), total cholesterol (TC, mg/dL), and triglyceride (TG, mg/dL). The albumin‐corrected anion gap (ACAG) was calculated according to the previously established formula (Hu et al. 2023): ACAG = AG + [4.4‐albumin (g/dL)] × 2.5. The RAR was derived as RAR = RDW (%)/albumin (g/dL) (Chen et al. 2024). Additionally, clinical treatment data such as mechanical ventilation (MV), diuretics (furosemide and bumetanide), and blood transfusion were extracted. Additional clinical variables were collected to address residual confounding: ICU admission status, ICU length of stay, and three established critical‐care severity scores: Simplified Acute Physiology Score II (SAPS II), Oxford Acute Severity of Illness Score (OASIS), and Systemic Inflammatory Response Syndrome (SIRS) criteria for patients admitted to the ICU. For biochemical confirmation of IDA, serum ferritin, transferrin saturation (TSAT), serum iron, and total iron‐binding capacity (TIBC) were extracted, with biochemically confirmed IDA defined as ferritin < 30 ng/mL or TSAT < 20%. For CHF verification, N‐terminal pro‐B‐type natriuretic peptide (NT‐proBNP) was extracted where available. Estimated glomerular filtration rate (eGFR) was calculated.

2.5. Statistical Analysis

To minimise bias, variables with > 20% missing data (e.g., BMI, TC, and TG) were excluded from the analysis. Multiple imputation was performed for variables with ≤ 20% missing data using the ‘mice’ package in R. Categorical variables in the imputed dataset were described using frequencies and percentages, with comparisons between two groups made via chi‐square tests. The Shapiro–Wilk test revealed that the continuous variables did not conform to a normal distribution. Thus, the continuous variables were presented as median [P25, P75] and compared using the Mann–Whitney U test. Collinearity was evaluated using multicollinearity diagnostic analysis, with variance inflation factors (VIF) > 10 indicating significant collinearity.

Tree‐based ranking by random forest and LightGBM was employed to evaluate the relative importance of baseline variables, particularly RAR and nursing note subjectivity scores, in relation to 7‐day hospital LOS. Logistic regression models were constructed to examine the combined effect of RAR and subjectivity on hospital LOS by comparing model R 2 values: Model 1 incorporated RAR and baseline difference variables; Model 2 included subjectivity score and baseline difference variables; Model 3 combined both RAR and subjectivity with baseline difference variables. In multivariable regression, the full range of the subjectivity score (0–1) represented a 1‐unit change, leading to artificially small adjusted odds ratios (OR) that were not clinically interpretable (Waudby‐Smith et al. 2018). To address this, we applied a 10‐fold scaling transformation to subjectivity scores. Discrimination of Model 1, Model 2, and Model 3 was quantified by the area under the receiver‐operating‐characteristic curve (ROC‐AUC). Clinical utility was assessed by decision curve analysis (DCA). Incremental value of subjectivity over RAR was tested by integrated discrimination improvement (IDI) and continuous net reclassification improvement (NRI). Log‐linear regression was used for the continuous LOS sensitivity analysis because hospital LOS is right‐skewed.

Shapley Additive exPlanations (SHAP) values were then used to quantify the independent prognostic contributions of RAR and subjectivity. Restricted cubic spline (RCS) analyses were conducted to explore non‐linear relationships between RAR/subjectivity and 7‐day hospital LOS, followed by trend regression analyses after tertiles grouping. The stability of the RCS‐derived RAR cut‐off was assessed by 1000 bootstrap percentile confidence intervals. Multiplicative interaction between RAR and subjectivity was assessed within logistic regression, and additive interaction was quantified by the relative excess risk due to interaction (RERI) and the synergy index (SI) with the delta‐method 95% CI.

To assess the robustness of the primary findings, several sensitivity analyses were performed. First, to verify the accuracy of ICD‐based diagnoses, analyses were repeated in biochemically confirmed IDA patients and in the NT‐proBNP‐available subset. Second, to evaluate selection bias, a sensitivity comparison of baseline characteristics between the 655 included patients and the 652 patients excluded for missing albumin was performed. Third, to assess residual confounding by illness severity and ICU admission, ICU admission status was added to Model 3 in the full cohort; analyses were then restricted to ICU‐admitted patients with further adjustment for SAPS II, OASIS, or SIRS. The incremental value of RAR and subjectivity over existing ICU severity scores (SAPS II, OASIS, SIRS) was assessed by comparing AUCs in the ICU subset. Fourth, to address residual confounding by renal function, eGFR was added as an additional covariate. Fifth, to explore potential mediation by loop diuretics in heart failure, we refitted the model after removing diuretics alone or all treatment covariates (diuretics, transfusion, MV).

Patients were then divided into four combined groups based on RCS‐derived optimal cutoff points for RAR and subjectivity: combination 1 (low RAR and low subjectivity), combination 2 (low RAR and high subjectivity), combination 3 (high RAR and high subjectivity), and combination 4 (high RAR and low subjectivity). The association between these combined groups and 7‐day hospital LOS was assessed using chi‐square tests, multivariable logistic regression, and subgroup analyses.

Kaplan–Meier (K–M) curves were used to assess the long‐term prognostic value of individual and combined RAR‐subjectivity in IDA patients with CHF. Long‐term mortality (1‐, 3‐, 5‐, and 10‐year) was additionally analysed with multivariable Cox proportional‐hazards models; the proportional‐hazards assumption was tested by Schoenfeld residuals. All analyses were performed using R 4.3 (rms, pROC, ResourceSelection, dcurves, survival, mice, rmda, car, randomForest, lightgbm, PredictABEL, shapviz, boot, epiR) and Python 3.11 (statsmodels 0.14, scikit‐learn 1.4, lifelines 0.30, scipy 1.13). All hypothesis tests were two‐sided with statistical significance set at p < 0.05.

3. Results

3.1. Baseline Information of Participants

A total of 655 IDA patients with CHF were included in this study, of which 384 (58.626%) had a hospital LOS ≥ 7 days. A baseline comparison of data from patients between the < 7‐day hospital LOS group and ≥ 7‐day hospital LOS group revealed that there were differences in comorbidities (sepsis, CCI), treatment (MV, diuretics, and blood transfusion), nursing note score (subjective), and some laboratory measures (WBC, albumin, ACAG, bicarbonate, calcium) (all p < 0.05) (Tables S1 and S2). In the ≥ 7‐day hospital LOS group, sepsis prevalence (17.188% vs. 6.642%), CCI index (8.000 vs. 7.000), MV (51.302% vs. 27.306%), diuretics (88.218% vs. 71.587%), and blood transfusion (20.312% vs. 5.535%) were higher compared to those in the < 7‐day hospital LOS group. Additionally, subjectivity (0.369 vs. 0.355), WBC (8.600 vs. 7.900), and ACAG (17.500 vs. 16.750) were elevated. Conversely, levels of albumin (3.300 vs. 3.500), bicarbonate (24.000 vs. 26.000), and calcium (8.600 vs. 8.700) were lower. Furthermore, patients with hospital LOS ≥ 7 days had a higher RAR than those with hospital LOS < 7 days (5.086 vs. 4.649) (p < 0.05) (Table S2).

3.2. Ranking of Importance of RAR and Subjectivity

Considering the association between albumin, bicarbonate, and ACAG, and given that RAR includes albumin components, albumin and bicarbonate indices were excluded from further analysis to avoid multicollinearity problems. Through multicollinearity diagnostic analysis, the results revealed that the remaining 10 difference variables had no collinearity problems (all VIF < 10) (Table S3). To further assess the importance of RAR and subjectivity, LightGBM and random forest were employed. The results showed that among the 10 variables, RAR and nursing note subjectivity scores ranked in the top three (Table S4). This finding confirmed the critical role of RAR and subjectivity scores in influencing the hospital LOS in IDA patients with CHF.

3.3. Association Between RAR, Subjectivity, and 7‐Day Hospital LOS

3.3.1. Logistic Regression Analysis for Binary LOS Outcome (≥ 7 Days)

Different logistic regression models were constructed to determine the association between RAR, nursing note subjectivity, and the 7‐day hospital LOS (Table 1). In model 1, RAR was a risk factor for hospital LOS ≥ 7 days (OR = 1.222, 95% CI: 1.055–1.417), with a Nagelkerke R 2 of 0.199. In model 2, scaled subjectivity was a risk factor for the hospital LOS ≥ 7 days (OR = 1.450, 95% CI: 1.068–1.970), with a Nagelkerke R 2 of 0.196. In Model 3, which included both RAR and scaled subjectivity, the relationship of RAR and scaled subjectivity with the hospital LOS ≥ 7 days still existed (ORRAR = 1.211, 95% CI: 1.045–1.404; ORScaled subjectivity = 1.428, 95% CI: 1.045–1.951), and the Nagelkerke R 2 was 0.208. Discrimination was acceptable for all three models (Model 1: AUC = 0.723, 95% CI: 0.684–0.762; Model 2: AUC = 0.725, 95% CI: 0.686–0.764; Model 3: AUC = 0.730, 95% CI: 0.692–0.769). DCA showed positive net benefit across thresholds of 0.30–0.85 (Figure 1A,B). Adding subjectivity to RAR yielded a small but notable improvement in risk reclassification (IDI = 0.006; continuous NRI = 0.111), suggesting that the combined use of both indicators provides better risk stratification than RAR alone.

TABLE 1.

The association of RAR and subjectivity with 7‐day hospital LOS in IDA complicated with CHF patients by logistic regression.

Model OR 95% CI p Nagelkerke R 2
Model 1 0.199
RAR 1.222 1.055–1.417 0.008
Model 2 0.196
Scaled subjectivity 1.450 1.068–1.970 0.017
Model 3 0.208
Scaled subjectivity 1.428 1.045–1.951 0.025
RAR 1.211 1.045–1.404 0.011
Open in a new tab

Note: In the three models, ACAG, calcium, CCI, WBC, diuretics, blood transfusion, MV, and sepsis were adjusted.

Abbreviations: ACAG, albumin‐corrected anion gap; CCI, Charlson comorbidity index; CHF, congestive heart failure; CI, confidence interval; IDA, iron deficiency anaemia; LOS, length of stay; MV, mechanical ventilation; OR, odds ratio; RAR, red blood cell distribution width‐to‐albumin ratio; WBC, white blood cell.

FIGURE 1.

FIGURE 1

Open in a new tab

ROC curves and DCA for the association of RAR and subjectivity with 7‐day hospital LOS. (A) ROC curves for all three models. (B) Decision curve analysis for all three models. DCA, decision curve analysis; LOS, length of stay; RAR, red blood cell distribution width‐to‐albumin ratio; ROC, receiver operating characteristic.

3.3.2. SHAP and RCS Analyses

Model 3 was interpreted through the SHAP method, which assessed the contribution of each feature to the model output to identify the most significant variables. A higher SHAP value indicated a greater probability of hospital LOS ≥ 7 days. The findings revealed that RAR was the most influential feature, followed by CCI and scaled subjectivity (Figure 2A). The summary plot illustrated that the SHAP values of RAR and scaled subjectivity had a wide range and exhibited a monotonically increasing trend (Figure 2B). The scatter plot results indicated that when the RAR was around 5 and the scaled subjectivity was around 3.5, the SHAP value was positive (namely, hospital LOS ≥ 7 days), and their linear relationship was presented (Figure 2C,D). After adjusting for other difference variables, trend regression results demonstrated that the risk of a ≥ 7‐day hospital LOS rose with increasing RAR or scaled subjectivity (all p for trend < 0.05) (Table 2), and the trend seemed to be linear (OR values were gradually increased for RAR or scaled subjectivity).

FIGURE 2.

FIGURE 2

Open in a new tab

Interpretation of model 3 using the SHAP method. (A) The SHAP bar graph; (B) The SHAP summary plot; (C) Scatter plot of RAR and SHAP values; (D) Scatter plot of scaled subjectivity and SHAP values. In the scatter plot, a positive SHAP value indicated that a ≥ 7‐day hospital LOS was more likely. ACAG, albumin‐corrected anion gap; CCI, Charlson comorbidity index; LOS, length of stay; MV, mechanical ventilation; RAR, red blood cell distribution width‐to‐albumin ratio; SHAP, Shapley Additive exPlanations; WBC, white blood cell.

TABLE 2.

Trend regression analysis of RAR or scaled subjectivity as a categorical variable on 7‐day hospital LOS.

Variable OR 95% CI p
RAR
2.767–4.447 Reference
4.452–5.375 1.005 [0.668, 1.511] 0.982
5.385–14.636 1.661 [1.056, 2.615] 0.028
p for trend 1.275 [1.018, 1.596] 0.034
Scaled subjectivity
0.000–3.470 Reference
3.472–3.784 1.474 [0.972, 2.234] 0.068
3.786–5.054 1.663 [1.076, 2.570] 0.022
p for trend 1.293 [1.040, 1.608] 0.021
Open in a new tab

Abbreviations: CI, confidence interval; LOS, length of stay; OR, odds ratio.

Based on the scatter plot of SHAP and the trend regression results, RCS analysis was further employed to visualise the relationship between RAR or subjectivity and the risk of a ≥ 7‐day hospital LOS. We found that the relationship between RAR levels and the risk of ≥ 7‐day hospital LOS was linear (p for non‐linear = 0.524 and adjusted P for non‐linear = 0.906) (Figure 3A,B). The optimal cut‐off value for RAR was approximately 4.90 (bootstrap 95% CI: 4.19–5.47). In addition, the relationship between the subjectivity level and ≥ 7‐day hospital LOS was also linear after adjustment (P for non‐linear = 0.037 and adjusted p for non‐linear = 0.514) (Figure 3C,D). Before and after adjustment, the optimal cut‐off value of subjectivity was 0.364 (bootstrap 95% CI: 0.345–0.382). All the analyses suggested a significant linear relationship between RAR or subjectivity and the risk of hospital LOS ≥ 7 days.

FIGURE 3.

FIGURE 3

Open in a new tab

RCS model for 7‐day hospital LOS.(A) Unadjusted RCS model for RAR in 7‐day hospital LOS; (B) Fully adjusted restricted cubic spline model for RAR in 7‐day hospital LOS. (C) Unadjusted restricted cubic spline model for subjectivity in 7‐day hospital LOS. (D) Fully adjusted restricted cubic spline model for subjectivity in 7‐day hospital LOS. CI, Confidence interval; LOS, length of stay; OR, odds ratio; RCS, restricted cubic spline.

3.3.3. Sensitivity Analysis for Continuous LOS Outcome

To further validate the robustness of the primary findings, we performed a sensitivity analysis treating hospital LOS as a continuous outcome using log‐linear OLS regression after natural log transformation. Consistent with the primary results, both RAR (+6.1% per unit, 95% CI: +1.9% to +10.4%, p = 0.004) and scaled subjectivity (+9.8% per unit, 95% CI: +0.6% to +19.7%, p = 0.036) were significantly associated with prolonged hospital stay (Table S5).

3.4. The Influence of RAR Combined With Subjectivity on 7‐Day Hospital LOS in IDA Patients With CHF

Both RAR and scaled subjectivity were independently associated with the risk of 7‐day hospital LOS. The multiplicative interaction was non‐significant (p = 0.459), and the additive interaction was also null (RERI = −0.080, 95% CI −0.426–0.195; SI = 0.797, 95% CI 0.192–4.370). Given the absence of multiplicative or additive interaction, joint risk stratification by RAR and subjectivity—rather than their synergistic or causal interaction—was further examined. Findings indicated that individuals with high subjectivity and high RAR experienced the highest rate of 7‐day hospital LOS at 35.20% (χ 2 = 24.065, p < 0.001) (Table 3). The results of the multivariable logistic regression model revealed that high subjectivity and high RAR were significant risk factors for 7‐day hospital LOS compared to low subjectivity and low RAR (OR = 2.063, 95% CI: 1.290–3.298). Additionally, CCI, diuretics, blood transfusion, and MV were identified as risk factors for 7‐day hospital LOS (all p < 0.05) (Table 4). Further subgroup analyses stratified by these four variables demonstrated that the association of RAR combined subjectivity with 7‐day hospital LOS persisted in patients with diuretics, high CCI, no MV, or no blood transfusion (Table S6).

TABLE 3.

Relationship of combined RAR and sentiment subjectivity to 7‐day hospital LOS in IDA patients with CHF.

Total (n = 655) LOS χ 2 p
< 7 days (n = 271) ≥ 7 days (n = 384)
Combination, n (%) 24.065 < 0.001
Low subjectivity and low RAR 198 (30.20) 107 (39.50) 91 (23.70)
Low RAR and high subjectivity 134 (20.50) 55 (20.30) 79 (20.60)
High subjectivity and high RAR 192 (29.30) 57 (21.00) 135 (35.20)
High RAR and low subjectivity 131 (20.00) 52 (19.20) 79 (20.60)
Open in a new tab

Abbreviations: CHF, congestive heart failure; IDA, iron deficiency anaemia; LOS, length of stay; RAR, red blood cell distribution width‐to‐albumin ratio.

TABLE 4.

The influence of combined RAR and sentiment subjectivity on 7‐day hospital LOS in IDA patients with CHF.

OR 95% CI p
ACAG 1.014 0.964–1.067 0.589
Calcium 0.853 0.679–1.070 0.173
CCI 1.111 1.033–1.194 0.005
WBC 0.989 0.967–1.012 0.354
Diuretics 2.983 1.927–4.617 < 0.001
Blood transfusion 2.908 1.509–5.604 0.001
MV 1.704 1.147–2.532 0.008
Sepsis 1.621 0.871–3.016 0.127
Combination 0.022
Low subjectivity and low RAR Reference
Low RAR and high subjectivity 1.600 0.988–2.590 0.056
High subjectivity and high RAR 2.063 1.290–3.298 0.002
High RAR and low subjectivity 1.497 0.912–2.456 0.110
Open in a new tab

Abbreviations: ACAG, albumin‐corrected anion gap; CCI, Charlson comorbidity index; CHF, congestive heart failure; CI, Confidence interval; IDA, iron deficiency anaemia; LOS, length of stay; MV, mechanical ventilation; OR, odds ratio; RAR, red blood cell distribution width‐to‐albumin ratio; WBC, white blood cell.

3.5. Sensitivity Analyses

Several sensitivity analyses were performed to assess the robustness of the primary findings. TextBlob showed near‐perfect agreement with an LLM‐as‐rater for subjectivity (κ = 0.829) (Figure 4A,B). Compared with excluded patients, included patients were systematically more critically ill (SMD range 0.280–0.820) (Table S7). Among the 367 patients with available iron‐related data, 316 (86.104%) met biochemical criteria for IDA (ferritin < 30 ng/mL or TSAT < 20%). NT‐proBNP was available in 203 patients (median 2849 pg/mL, IQR 1306–9280), well above the heart‐failure rule‐in threshold, supporting the ICD‐based CHF diagnosis. The RAR effect remained robust after adjustment for ICU admission, SAPS II, and eGFR, as well as in biochemically confirmed IDA patients. Removal of diuretics or all treatment covariates did not materially alter the results (Table S8). In the ICU subset, RAR + subjectivity (AUC = 0.631) outperformed SAPS II (0.577), OASIS (0.586), and SIRS (0.520) (Table S9). Detailed results of all sensitivity analyses are presented in Tables S7–, S9.

FIGURE 4.

FIGURE 4

Open in a new tab

Agreement between TextBlob and an independent LLM‐as‐rater for sentiment scores. (A) Bland–Altman plot for polarity. (B) Bland–Altman plot for subjectivity. LLM, Large Language Model.

3.6. Prognostic Value of RAR, Subjectivity, and Their Combination for Long‐Term Survival in IDA Patients With CHF

In addition to the association of RAR, subjectivity, or a combination of both on 7‐day hospital LOS risk, we also explored their relationship with long‐term outcomes. Multivariable Cox proportional‐hazards models adjusted for the same eight covariates as the LOS model showed that each one‐unit increment of RAR was independently associated with 1‐year (HR = 1.219, 95% CI 1.098–1.354, p < 0.001), 3‐year (HR = 1.208, 95% CI 1.095–1.332, p < 0.001), 5‐year (HR = 1.217, 95% CI 1.107–1.339, p < 0.001) and 10‐year mortality (HR = 1.223, 95% CI 1.114–1.343, p < 0.001); the Schoenfeld residual test was non‐significant for RAR (p = 0.95), supporting the proportional‐hazards assumption. Scaled subjectivity additionally and independently predicted long‐term mortality, reaching significance at 5‐ and 10‐year horizons (5‐year HR = 1.396, 95% CI 1.020–1.910, p = 0.037; 10‐year HR = 1.394, 95% CI 1.025–1.896, p = 0.034; Schoenfeld p = 0.83) (Table 5). Binary high/low stratification of subjectivity and RAR at the RCS‐derived cut‐off in the Cox framework yielded similar effect directions. High RAR was associated with significantly lower survival rates at 1, 3, and 5 years (adjusted HR = 1.340–1.430, all p < 0.05) (Figure 5A–C). High subjectivity also predicted lower survival, with borderline significance at 1 year (adjusted HR = 1.315, p = 0.054) and significance at 3 and 5 years (adjusted HR = 1.363–1.406, both p < 0.05) (Figure 5D–F). In joint stratification (reference: C1, low RAR + low subjectivity), the high RAR + high subjectivity group (C3) had the poorest prognosis, with significantly elevated mortality risk across all time points (adjusted HR = 1.713–1.811, all p < 0.05) (Figure 5G–I).

TABLE 5.

Multivariable Cox proportional‐hazards models for long‐term mortality.

Outcome RAR Scaled subjectivity
HR (95% CI) p HR (95% CI) p
1‐year 1.219 (1.098–1.354) < 0.001 1.438 (0.992–2.083) 0.055
3‐year 1.208 (1.095–1.332) < 0.001 1.378 (0.995–1.910) 0.054
5‐year 1.217 (1.107–1.339) < 0.001 1.396 (1.021–1.910) 0.037
10‐year 1.223 (1.114–1.343) < 0.001 1.394 (1.025–1.897) 0.034
Open in a new tab

Note: Schoenfeld residual tests were non‐significant for RAR (p = 0.95) and scaled subjectivity (p = 0.83), supporting the proportional‐hazards assumption. ACAG, calcium, CCI, WBC, diuretics, blood transfusion, MV, and sepsis were adjusted.

Abbreviations: ACAG, albumin‐corrected anion gap; CCI, Charlson comorbidity index; CI, confidence interval; HR, hazard ratio; MV, mechanical ventilation; RAR, red blood cell distribution width‐to‐albumin ratio; WBC, white blood cell.

FIGURE 5.

FIGURE 5

Open in a new tab

Survival curves for IDA patients with CHF. (A) 1‐year survival between the high and low RAR groups. (B) 3‐year survival between the high and low RAR groups. (C) 5‐year survival between the high and low RAR groups. (D) 1‐year survival between the high and low subjectivity groups. (E) 3‐year survival between the high and low subjectivity groups. (F) 5‐year survival between the high and low subjectivity groups. (G) 1‐year survival among the combination groups of RAR and subjectivity. (H) 3‐year survival among the combination groups of RAR and subjectivity. (I) 5‐year survival among the combination groups of RAR and subjectivity. In the combination group, combination 1 (low subjectivity and low RAR) was a reference. Adjusted HR and 95% CI in each panel are derived from a multivariable Cox model with the binary high/low stratification (RAR cut‐off ≈ 4.90; subjectivity cut‐off ≈ 0.364) as the independent variable and the eight‐covariate adjustment set as in Model 3 (Adjusted for ACAG, calcium, CCI, WBC, diuretics, blood transfusion, MV, and sepsis). ACAG, albumin‐corrected anion gap; CI, confidence interval; CCI, Charlson Comorbidity Index; CHF, congestive heart failure; HR, hazard ratio; IDA, iron deficiency anaemia; MV, mechanical ventilation; RAR, red blood cell distribution width‐to‐albumin ratio; WBC, white blood cell count.

4. Discussion

Hospital LOS is a key indicator of resource consumption, with extended stays driving up healthcare costs and increasing the financial strain on patients (Messahel 1995). CHF is the leading cause of morbidity and mortality in older adults worldwide. IDA aggravates CHF symptoms by reducing oxygen‐carrying capacity and impairing myocardial metabolism, significantly increasing the risk of complications and hospital LOS (Alharbi et al. 2024; Mohan et al. 2023; Thammitage et al. 2025). Early identification of high‐risk patients with longer hospital LOS or poor long‐term prognosis is essential to improve prognosis and reduce healthcare burden. From a nursing perspective, timely recognition of at‐risk patients is fundamental to prioritising nursing assessment, care planning, and resource allocation.

In this study, 655 IDA patients with CHF were analysed, and it was found that the subjective laboratory indicators RAR and subjectivity scores of nursing notes were related to the 7‐day hospital LOS. Patients with high RAR (≥ 4.90) were more likely to be hospitalised for 7 days. RAR integrates RDW and albumin, capturing both haematopoietic dysfunction and hypoalbuminemia (Xu et al. 2022). Low albumin, a marker of malnutrition and inflammation, may be associated with volume overload, chronic inflammation, hepatic congestion, malnutrition, and cachexia (Ajoolabady et al. 2022; Soeters et al. 2019), resulting in a prolonged hospital LOS. Additionally, patients with a high subjective score (≥ 0.364) were also more likely to be hospitalised for 7 days. Subjective signals from nursing notes may reflect the extent to which nurses use evaluative language, clinical judgement terms, and affect descriptors in their notes‐linguistic patterns that typically signal patient complexity or perceived deterioration (McCoy et al. 2015). Conceptually, low‐subjectivity notes are concise and fact‐based, while high‐subjectivity notes are characterised by language indicating clinical concern, uncertainty, or active interpretation of patient status (e.g., descriptors of distress, behavioural change, functional decline, or risk perception). These features often emerge when nurses encounter complex care needs or early signs of deterioration that are not yet reflected in routine vital signs or laboratory values. High subjectivity thus serves as a proxy for nursing clinical reasoning and heightened vigilance‐it indicates that the nurse has synthesised multiple observations, exercised professional judgement, and identified the patient as warranting closer monitoring. This interpretation is consistent with prior evidence that nurses' narrative documentation contains meaningful signals about patient deterioration and care complexity beyond structured data (Collins et al. 2013; Keenan et al. 2008).

In our analysis, RAR showed the strongest association with both short‐term LOS and long‐term (1‐ to 10‐year) mortality. Nursing‐note subjectivity contributed an additional, modest independent signal for both LOS and long‐term mortality among patients with documented nursing notes, without interacting multiplicatively or additively with RAR. The fact that subjectivity remained associated with outcomes even after accounting for RAR and other covariates suggests that nursing documentation captures clinically relevant aspects of patient status. Importantly, high‐subjectivity notes may signal that a nurse has detected emerging care needs or instability not yet reflected in objective data. If this clinical judgement remains implicit rather than explicitly communicated, the patient may receive insufficient monitoring. The subjectivity score thus identifies a potential information gap between bedside observation and structured data, serving as a useful descriptive marker to direct nursing attention.

This study also examined joint risk stratification by the objective laboratory measure (RAR) and subjectivity in nursing notes for 7‐day hospital LOS. The formal multiplicative and additive interaction tests were null, indicating that the joint pattern reflects independent additive contributions rather than synergy. Patients with high RAR and high subjectivity were at higher risk of prolonged hospital LOS. Furthermore, the combined signal remained significant across subgroups including those with diuretic use, high CCI, or no transfusion. These subgroups were linked to low albumin, multi‐organ dysfunction, or uncorrected severe anaemia (Charlson et al. 2022; Lee et al. 2021; Theile et al. 2023; Yang et al. 2023), which may be worse off and have longer hospital stays. In addition, patients with high RAR combined with high subjectivity had a poor long‐term prognosis. This finding provided a new risk stratification tool for clinical practice and facilitated the early identification of high‐risk patients in need of intensive intervention.

Practically, for bedside nurses, the ‘double‐high’ flag can be identified automatically via the electronic nursing dashboard or shift handover report—requiring no extra chart review, as both RAR and the subjectivity score are generated from existing EHR data and nursing notes. At the start of each shift, the flag is visible on the dashboard alongside other patient information, making it available for nurses' awareness during routine care review. However, it is important to emphasise that the flag represents an exploratory research finding derived from a retrospective study, rather than a clinically validated decision support tool. While automated generation is technically feasible using existing EHR data, technical feasibility does not equate to clinical readiness. The subjectivity score, in particular, is derived from a general‐purpose NLP tool (TextBlob) not designed or validated for clinical use. Therefore, the joint pattern should be interpreted strictly as an exploratory descriptive tool for research purposes, and should not be used to guide patient care outside of prospective validation studies. Any clinical translation would require rigorous evaluation of its effectiveness, safety, and fairness across diverse populations and settings.

From a clinical‐pathway perspective, RAR is calculable from two routine laboratory values (RDW and albumin) on any inpatient ward, while ICU severity scores such as SAPS II require ICU‐level physiologic parameters that are not collected outside the ICU; in our head‐to‐head comparison within the ICU subset, RAR alone achieved a higher AUC for LOS ≥ 7 days than SAPS II, OASIS, or SIRS individually, and adding any of these ICU scores to RAR + subjectivity yielded no meaningful incremental discrimination. The joint RAR–subjectivity pattern should therefore be interpreted as an exploratory descriptive tool for clinical attention rather than as evidence for a specific therapeutic intervention; intervention strategies require prospective interventional evidence.

This study was the first to explore the influence of laboratory indicators (RAR) and medical records (subjectivity) on the prognosis of IDA patients with CHF. High RAR and high nursing note subjectivity were associated with prolonged hospital LOS and poor long‐term prognosis. This finding supported the integration of laboratory data with narrative medical records in clinical practice, highlighting that nursing documentation adds unique information beyond what laboratory measures alone can offer. However, this study had the following limitations. (1) Heart‐failure subtypes (HFrEF vs. HFpEF), left‐ventricular ejection fraction, and NYHA functional class could not be reliably extracted from MIMIC‐IV. EF values reside in unstructured echocardiography and discharge‐summary free text, requiring dedicated structured NLP extraction. HFrEF is characterised by systolic dysfunction and neurohormonal activation, with greater disease fluctuation and higher inflammatory levels; HFpEF is characterised by diastolic dysfunction and microvascular impairment, often accompanied by anaemia of chronic disease. These differences may lead to variations in RAR baseline values, predictive thresholds, and the clinical significance of subjectivity between the two subtypes. Therefore, extrapolation of our findings to HFpEF or mixed populations should be made with caution, and future studies should perform subtype‐stratified validation. (2) The exclusion of 652 patients for missing albumin was not missing‐at‐random: included patients were systematically more critically ill. Our conclusions therefore apply specifically to IDA + CHF inpatients in whom albumin is clinically measured, typically of higher acuity, and absolute RAR thresholds should not be extrapolated to ambulatory or low‐acuity populations. (3) MIMIC‐IV is an ICU‐based, single‐centre dataset; external validation in multi‐centre and non‐ICU populations is required. (4) IDA aetiology (nutritional vs. anaemia of chronic disease) could not be reliably differentiated. (5) TextBlob is a general‐purpose lexicon‐based NLP tool that we showed has near‐perfect agreement with an independent LLM rater for subjectivity (Spearman ρ = 0.961) but only moderate agreement for polarity (Spearman ρ = 0.549); domain‐specific clinical NLP tools (e.g., ClinicalBERT) may further refine the sentiment signal in future work. (6) The retrospective observational design precludes causal inference; the results and clinical applicability require prospective validation.

5. Conclusion

Laboratory parameters (RAR) and medical records (subjectivity) were associated with longer hospital LOS and poor long‐term outcomes in IDA patients with CHF. RAR showed a strong association with both prolonged hospital LOS and long‐term mortality; scaled subjectivity was also independently associated with long‐term mortality (10‐year HR = 1.39, p = 0.034) and contributed an additional, modest signal for LOS, and the two factors did not interact. This study highlights that routine nursing documentation contains clinically meaningful signals that can complement laboratory data in identifying patients at risk. The combined stratification may serve as a descriptive tool to help nurses identify patients warranting closer clinical attention. RAR‐based risk stratification and joint RAR–subjectivity classification as an exploratory descriptive tool warrant prospective external validation before clinical implementation. Future prospective studies could evaluate whether nursing‐led protocols triggered by this risk flag improve patient outcomes or care efficiency.

6. Implication for Nursing Practice

The findings of this study have important implications for nursing practice. First, the combined assessment of the objective indicator RAR and subjective emotions recorded in nursing notes can serve as an effective descriptive tool for identifying patients at risk of the 7‐day LOS in IDA patients with CHF, particularly in the subgroup of patients with ‘high RAR and high subjectivity’ who exhibit a higher risk of prolonged LOS. This suggests that nurses should pay attention to the joint association of laboratory indicators and nursing emotional assessments. For bedside nurses, no extra documentation is required to identify the risk—RAR is automatically calculated from laboratory values, and the subjectivity score is extracted from existing nursing notes via TextBlob; both are displayed on the nursing dashboard. At the start of each shift, nurses can quickly identify ‘double‐high’ patients from the dashboard. This information may serve as a reminder to pay closer attention to fluid balance, respiratory function, and mental status during routine assessments, and to clearly communicate the risk status during handover; however, it represents an exploratory research finding and should not be construed as a validated basis for specific clinical actions. This workflow adds no additional documentation or scoring burden. Secondly, this joint association remained significantly related across multiple clinical subgroups (including patients using diuretics, those with high CCI scores, those not receiving MV, or those not receiving blood transfusions), indicating that its risk identification capability has broad clinical applicability and is not influenced by specific treatment modalities. These findings support exploring the integration of RAR and nursing‐record affective analysis into routine risk‐stratification workflows for hospitalised patients, as an aid for identifying patients who may benefit from closer clinical observation. Whether any specific intervention triggered by this risk flag actually improves outcomes requires prospective interventional evaluation and cannot be inferred from this observational design. The study results provide evidence‐based support for the development of personalised management strategies combining objective indicators with subjective assessments.

Author Contributions

X.F. contributed to the conception and design. X.F., Y.X., and Y.T. contributed to the collection and assembly of data. Y.H., F.Z., and P.L. analysed and interpreted the data. All authors wrote and approved the final manuscript.

Funding

This study was supported by the following grant: The Medical and Health Research Project of Zhejiang Province (No. 2025KY1039).

Ethics Statement

This study used anonymised patient data from the Medical Information Mart for Intensive Care IV (MIMIC‐IV) database. As this was a secondary analysis of pre‐existing anonymised data, additional ethics review was not required.

Consent

The authors have nothing to report.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Table S1: The categorical variables information of participants with categorical 7‐day hospital LOS.

NOP2-13-e70699-s002.docx (19.1KB, docx)

Table S2: The continuous variables' information of participants with categorical 7‐day hospital LOS.

NOP2-13-e70699-s006.docx (16.6KB, docx)

Table S3: The collinearity results of 10 different variables.

NOP2-13-e70699-s003.docx (11.8KB, docx)

Table S4: Hierarchy of the importance of baseline difference variables for 7‐day hospital LOS.

NOP2-13-e70699-s005.docx (14KB, docx)

Table S5: Log‐linear OLS regression for continuous LOS as a sensitivity outcome.

NOP2-13-e70699-s007.docx (12.5KB, docx)

Table S6: The subgroup analysis of combined RAR and sentiment subjectivity with 7‐day hospital LOS in IDA patients with CHF.

NOP2-13-e70699-s004.docx (18.9KB, docx)

Table S7: Comparison of baseline characteristics between included and excluded patients.

NOP2-13-e70699-s001.docx (18.5KB, docx)

Table S8: Summary of sensitivity analyses for the association of RAR and subjectivity with LOS ≥ 7 days.

NOP2-13-e70699-s008.docx (17.2KB, docx)

Table S9: Comparison of AUCs for discriminating LOS ≥ 7 days in the ICU subset.

NOP2-13-e70699-s009.docx (12.4KB, docx)

Acknowledgements

The authors have nothing to report.

Data Availability Statement

Data are from the publicly available MIMIC‐IV database (https://physionet.org/content/mimiciv/3.1/). Access to the database required registration and successful completion of the Collaborative Institutional Training Initiative program.

References

  1. Ajoolabady, A. , Nattel S., Lip G. Y. H., and Ren J.. 2022. “Inflammasome Signaling in Atrial Fibrillation: JACC State‐of‐the‐Art Review.” Journal of the American College of Cardiology 79, no. 23: 2349–2366. 10.1016/j.jacc.2022.03.379. [DOI] [PMC free article] [PubMed] [Google Scholar]
  2. Alharbi, A. , Shah M., Bansal N., et al. 2024. “Outcomes and Complications of Heart Failure With Iron Deficiency Anemia: A Nationwide Analysis.” Hospital Practice (1995) 52, no. 3: 84–90. 10.1080/21548331.2024.2337615. [DOI] [PubMed] [Google Scholar]
  3. Aydin, M. Z. , and Abdi I. A.. 2025. “Clinical Relevance of HALP Score in Predicting Hospital Stay Duration and Outcomes in Acute Heart Failure.” International Journal of General Medicine 18: 7095–7100. 10.2147/IJGM.S554891. [DOI] [PMC free article] [PubMed] [Google Scholar]
  4. Charlson, M. E. , Carrozzino D., Guidi J., and Patierno C.. 2022. “Charlson Comorbidity Index: A Critical Review of Clinimetric Properties.” Psychotherapy and Psychosomatics 91, no. 1: 8–35. 10.1159/000521288. [DOI] [PubMed] [Google Scholar]
  5. Chen, S. , Guan S., Yan Z., et al. 2024. “Prognostic Value of Red Blood Cell Distribution Width‐To‐Albumin Ratio in ICU Patients With Coronary Heart Disease and Diabetes Mellitus.” Frontiers in Endocrinology (Lausanne) 15: 1359345. 10.3389/fendo.2024.1359345. [DOI] [PMC free article] [PubMed] [Google Scholar]
  6. Collaborators, G. B. D. A . 2023. “Prevalence, Years Lived With Disability, and Trends in Anaemia Burden by Severity and Cause, 1990‐2021: Findings From the Global Burden of Disease Study 2021.” Lancet Haematology 10, no. 9: e713–e734. 10.1016/S2352-3026(23)00160-6. [DOI] [PMC free article] [PubMed] [Google Scholar]
  7. Collins, S. A. , Cato K., Albers D., et al. 2013. “Relationship Between Nursing Documentation and Patients' Mortality.” American Journal of Critical Care 22, no. 4: 306–313. 10.4037/ajcc2013426. [DOI] [PMC free article] [PubMed] [Google Scholar]
  8. Croxford, E. , Gao Y., First E., et al. 2025. “Evaluating Clinical AI Summaries With Large Language Models as Judges.” Npj Digital Medicine 8, no. 1: 640. 10.1038/s41746-025-02005-2. [DOI] [PMC free article] [PubMed] [Google Scholar]
  9. Eckart, A. , Struja T., Kutz A., et al. 2020. “Relationship of Nutritional Status, Inflammation, and Serum Albumin Levels During Acute Illness: A Prospective Study.” American Journal of Medicine 133, no. 6: 713–722.e717. 10.1016/j.amjmed.2019.10.031. [DOI] [PubMed] [Google Scholar]
  10. El Iskandarani, M. , El Kurdi B., Murtaza G., Paul T. K., and Refaat M. M.. 2021. “Prognostic Role of Albumin Level in Heart Failure: A Systematic Review and Meta‐Analysis.” Medicine (Baltimore) 100, no. 10: e24785. 10.1097/MD.0000000000024785. [DOI] [PMC free article] [PubMed] [Google Scholar]
  11. Fava, C. , Cattazzo F., Hu Z. D., Lippi G., and Montagnana M.. 2019. “The Role of Red Blood Cell Distribution Width (RDW) in Cardiovascular Risk Assessment: Useful or Hype?” Annals of Translational Medicine 7, no. 20: 581. 10.21037/atm.2019.09.58. [DOI] [PMC free article] [PubMed] [Google Scholar]
  12. Gohil, S. , Vuik S., and Darzi A.. 2018. “Sentiment Analysis of Health Care Tweets: Review of the Methods Used.” JMIR Public Health and Surveillance 4, no. 2: e43. 10.2196/publichealth.5789. [DOI] [PMC free article] [PubMed] [Google Scholar]
  13. Hao, M. , Jiang S., Tang J., et al. 2024. “Ratio of Red Blood Cell Distribution Width to Albumin Level and Risk of Mortality.” JAMA Network Open 7, no. 5: e2413213. 10.1001/jamanetworkopen.2024.13213. [DOI] [PMC free article] [PubMed] [Google Scholar]
  14. Hu, B. , Zhong L., Yuan M., et al. 2023. “Elevated Albumin Corrected Anion Gap Is Associated With Poor In‐Hospital Prognosis in Patients With Cardiac Arrest: A Retrospective Study Based on MIMIC‐IV Database.” Frontiers in Cardiovascular Medicine 10: 1099003. 10.3389/fcvm.2023.1099003. [DOI] [PMC free article] [PubMed] [Google Scholar]
  15. Johnson, A. E. W. , Bulgarelli L., Shen L., et al. 2023. “MIMIC‐IV, a Freely Accessible Electronic Health Record Dataset.” Scientific Data 10, no. 1: 1. 10.1038/s41597-022-01899-x. [DOI] [PMC free article] [PubMed] [Google Scholar]
  16. Keenan, G. M. , Yakel E., Tschannen D., and Mandeville M.. 2008. “Documentation and the Nurse Care Planning Process.” In Patient Safety and Quality: An Evidence‐Based Handbook for Nurses, edited by Hughes R. G., Agency for Healthcare Research and Quality (US). [PubMed] [Google Scholar]
  17. Kumar, S. B. , Arnipalli S. R., Mehta P., Carrau S., and Ziouzenkova O.. 2022. “Iron Deficiency Anemia: Efficacy and Limitations of Nutritional and Comprehensive Mitigation Strategies.” Nutrients 14, no. 14: 2976. 10.3390/nu14142976. [DOI] [PMC free article] [PubMed] [Google Scholar]
  18. Lee, T. H. , Kuo G., Chang C. H., et al. 2021. “Diuretic Effect of Co‐Administration of Furosemide and Albumin in Comparison to Furosemide Therapy Alone: An Updated Systematic Review and Meta‐Analysis.” PLoS One 16, no. 12: e0260312. 10.1371/journal.pone.0260312. [DOI] [PMC free article] [PubMed] [Google Scholar]
  19. Liu, Z. , Yang Y., Song H., and Luo J.. 2022. “A Prediction Model With Measured Sentiment Scores for the Risk of In‐Hospital Mortality in Acute Pancreatitis: A Retrospective Cohort Study.” Annals of Translational Medicine 10, no. 12: 676. 10.21037/atm-22-1613. [DOI] [PMC free article] [PubMed] [Google Scholar]
  20. Lu, C. , Long J., Liu H., et al. 2022. “Red Blood Cell Distribution Width‐To‐Albumin Ratio Is Associated With All‐Cause Mortality in Cancer Patients.” Journal of Clinical Laboratory Analysis 36, no. 5: e24423. 10.1002/jcla.24423. [DOI] [PMC free article] [PubMed] [Google Scholar]
  21. Marcus, G. , Maymon S. L., Kalmanovich E., et al. 2026. “Early Discharge After Clinical Stabilization in Acute Decompensated Heart Failure: Associations With Short‐Term Outcomes.” Clinical Research in Cardiology. 10.1007/s00392-026-02858-x. [DOI] [PubMed] [Google Scholar]
  22. McCann, S. , Perapoch Amado M., and Moore S. E.. 2020. “The Role of Iron in Brain Development: A Systematic Review.” Nutrients 12, no. 7: 2001. 10.3390/nu12072001. [DOI] [PMC free article] [PubMed] [Google Scholar]
  23. McCoy, T. H. , Castro V. M., Cagan A., Roberson A. M., Kohane I. S., and Perlis R. H.. 2015. “Sentiment Measured in Hospital Discharge Notes Is Associated With Readmission and Mortality Risk: An Electronic Health Record Study.” PLoS One 10, no. 8: e0136341. 10.1371/journal.pone.0136341. [DOI] [PMC free article] [PubMed] [Google Scholar]
  24. Messahel, F. M. 1995. “Post Cholecystectomy Admission to the Intensive Care Unit. Comparison Between Open, Mini‐Lap and Laparoscopic Techniques.” Anaesthesia 50, no. 10: 901–904. 10.1111/j.1365-2044.1995.tb05861.x. [DOI] [PubMed] [Google Scholar]
  25. Mohan, I. K. , Baba K., Iyyapu R., Thirumalasetty S., and Satish O. S.. 2023. “Advances in Congestive Heart Failure Biomarkers.” Advances in Clinical Chemistry 112: 205–248. 10.1016/bs.acc.2022.09.005. [DOI] [PubMed] [Google Scholar]
  26. Ni, Q. , Wang X., Wang J., and Chen P.. 2022. “The Red Blood Cell Distribution Width‐Albumin Ratio: A Promising Predictor of Mortality in Heart Failure Patients—A Cohort Study.” Clinica Chimica Acta 527: 38–46. 10.1016/j.cca.2021.12.027. [DOI] [PubMed] [Google Scholar]
  27. Ohno, Y. , Hanawa H., Jiao S., et al. 2015. “Liver Congestion in Heart Failure Contributes to Inappropriately Increased Serum Hepcidin Despite Anemia.” Tohoku Journal of Experimental Medicine 235, no. 1: 69–79. 10.1620/tjem.235.69. [DOI] [PubMed] [Google Scholar]
  28. Safiri, S. , Kolahi A. A., Noori M., et al. 2021. “Burden of Anemia and Its Underlying Causes in 204 Countries and Territories, 1990–2019: Results From the Global Burden of Disease Study 2019.” Journal of Hematology & Oncology 14, no. 1: 185. 10.1186/s13045-021-01202-2. [DOI] [PMC free article] [PubMed] [Google Scholar]
  29. Salvagno, G. L. , Sanchis‐Gomar F., Picanza A., and Lippi G.. 2015. “Red Blood Cell Distribution Width: A Simple Parameter With Multiple Clinical Applications.” Critical Reviews in Clinical Laboratory Sciences 52, no. 2: 86–105. 10.3109/10408363.2014.992064. [DOI] [PubMed] [Google Scholar]
  30. Savarese, G. , Becher P. M., Lund L. H., Seferovic P., Rosano G. M. C., and Coats A. J. S.. 2023. “Global Burden of Heart Failure: A Comprehensive and Updated Review of Epidemiology.” Cardiovascular Research 118, no. 17: 3272–3287. 10.1093/cvr/cvac013. [DOI] [PubMed] [Google Scholar]
  31. Seidl, P. , Szep M., Breden S., et al. 2026. “LLM‐Based Extraction of Clinical Data: Potentials and Challenges.” Orthopadie (Heidelberg) 55, no. 1: 17–23. 10.1007/s00132-025-04754-0. [DOI] [PubMed] [Google Scholar]
  32. Shangguan, T. , Xu J., Weng X., and Lin H.. 2025. “Red Blood Cell Distribution Width to Albumin Ratio Is Associated With Increased Depression: The Mediating Role of Atherogenic Index of Plasma.” Frontiers in Psychiatry 16: 1504123. 10.3389/fpsyt.2025.1504123. [DOI] [PMC free article] [PubMed] [Google Scholar]
  33. Siddiqui, S. W. , Ashok T., Patni N., Fatima M., Lamis A., and Anne K. K.. 2022. “Anemia and Heart Failure: A Narrative Review.” Cureus 14, no. 7: e27167. 10.7759/cureus.27167. [DOI] [PMC free article] [PubMed] [Google Scholar]
  34. Soeters, P. B. , Wolfe R. R., and Shenkin A.. 2019. “Hypoalbuminemia: Pathogenesis and Clinical Significance.” JPEN Journal of Parenteral and Enteral Nutrition 43, no. 2: 181–193. 10.1002/jpen.1451. [DOI] [PMC free article] [PubMed] [Google Scholar]
  35. Stricker, K. , Rothen H. U., and Takala J.. 2003. “Resource Use in the ICU: Short‐ Vs. Long‐Term Patients.” Acta Anaesthesiologica Scandinavica 47, no. 5: 508–515. 10.1034/j.1399-6576.2003.00083.x. [DOI] [PubMed] [Google Scholar]
  36. Thammitage, M. B. , Moussa N., Rezvani A., et al. 2025. “The Intersection of Heart Failure and Iron Deficiency Anemia: Diagnostic and Therapeutic Approaches.” Current Cardiology Reviews 21, no. 3: e1573403X1331380. 10.2174/011573403X331380241111091452. [DOI] [PMC free article] [PubMed] [Google Scholar]
  37. Theile, P. , Muller J., Daniels R., Kluge S., and Roedl K.. 2023. “Association Between Red Cell Distribution Width and Outcomes of Nonagenarians Admitted to the Intensive Care Unit‐A Retrospective Cohort Study.” Diagnostics (Basel) 13, no. 20. 10.3390/diagnostics13203279. [DOI] [PMC free article] [PubMed] [Google Scholar]
  38. Waudby‐Smith, I. E. R. , Tran N., Dubin J. A., and Lee J.. 2018. “Sentiment in Nursing Notes as an Indicator of Out‐of‐Hospital Mortality in Intensive Care Patients.” PLoS One 13, no. 6: e0198687. 10.1371/journal.pone.0198687. [DOI] [PMC free article] [PubMed] [Google Scholar]
  39. Xu, W. , Huo J., Chen G., et al. 2022. “Association Between Red Blood Cell Distribution Width to Albumin Ratio and Prognosis of Patients With Sepsis: A Retrospective Cohort Study.” Frontiers in Nutrition 9: 1019502. 10.3389/fnut.2022.1019502. [DOI] [PMC free article] [PubMed] [Google Scholar]
  40. Yang, Y. , Wang Q., Gao L., et al. 2023. “Promising Applications of Red Cell Distribution Width in Diagnosis and Prognosis of Diseases With or Without Disordered Iron Metabolism.” Cell Biology International 47, no. 7: 1161–1169. 10.1002/cbin.12029. [DOI] [PubMed] [Google Scholar]
  41. Yousaf, M. , iIlahi M., Bibi A., et al. 2023. “Chronic Anemia Complicated by Cardiac Failure, Pulmonary Hypertension, and Pericardial Effusion: A Case Report.” Journal of Medical Case Reports 17, no. 1: 44. 10.1186/s13256-022-03686-z. [DOI] [PMC free article] [PubMed] [Google Scholar]
  42. Zhao, N. , Hu W., Wu Z., et al. 2021. “The Red Blood Cell Distribution Width‐Albumin Ratio: A Promising Predictor of Mortality in Stroke Patients.” International Journal of General Medicine 14: 3737–3747. 10.2147/IJGM.S322441. [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

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

Supplementary Materials

Table S1: The categorical variables information of participants with categorical 7‐day hospital LOS.

NOP2-13-e70699-s002.docx (19.1KB, docx)

Table S2: The continuous variables' information of participants with categorical 7‐day hospital LOS.

NOP2-13-e70699-s006.docx (16.6KB, docx)

Table S3: The collinearity results of 10 different variables.

NOP2-13-e70699-s003.docx (11.8KB, docx)

Table S4: Hierarchy of the importance of baseline difference variables for 7‐day hospital LOS.

NOP2-13-e70699-s005.docx (14KB, docx)

Table S5: Log‐linear OLS regression for continuous LOS as a sensitivity outcome.

NOP2-13-e70699-s007.docx (12.5KB, docx)

Table S6: The subgroup analysis of combined RAR and sentiment subjectivity with 7‐day hospital LOS in IDA patients with CHF.

NOP2-13-e70699-s004.docx (18.9KB, docx)

Table S7: Comparison of baseline characteristics between included and excluded patients.

NOP2-13-e70699-s001.docx (18.5KB, docx)

Table S8: Summary of sensitivity analyses for the association of RAR and subjectivity with LOS ≥ 7 days.

NOP2-13-e70699-s008.docx (17.2KB, docx)

Table S9: Comparison of AUCs for discriminating LOS ≥ 7 days in the ICU subset.

NOP2-13-e70699-s009.docx (12.4KB, docx)

Data Availability Statement

Data are from the publicly available MIMIC‐IV database (https://physionet.org/content/mimiciv/3.1/). Access to the database required registration and successful completion of the Collaborative Institutional Training Initiative program.

Articles from Nursing Open are provided here courtesy of Wiley

RESOURCES