Red blood cell distribution width/albumin ratio is a novel risk factor of incidence and long-term mortality in chronic heart failure patients: three large cohorts from China and America

2.1 Study participants

Three cohorts were absorbed in this study. The NHANES and MIMIC were prospective cohort studies, while JHDH was cross-sectional. The following analysis is shown in Figure 1.

• Download figure
• Open in new tab

Figure 1.

The flow chart of this study

NHANES, a periodic health-related survey in America, was processed by the Centers for disease control and Prevention’s National Center for Health Statistics. This dataset currently contained 111,797 patients and was open to the public free of charge. The dataset contained numerous survey examination data, including measurements, precisely designed health questionnaires and laboratory tests. The participants from JAN 01, 1999, to DEC 31, 2020 were included in this work.

MIMIC database (MIMIC III v1.4), a publicly single-centre critical care database, was developed by the Institutional Review Boards of Beth Israel Deaconess Medical Center and the Massachusetts Institute of Technology. The database contains charted events, laboratory events, and documents from the International Classification of Diseases (ICD-9) codes. The dataset is also free of charge, similarily to NHANES. However, users must pass a test to be eligible to register for the database and be approved by the MIMIC-III database manager. After passing the training course, the first author (Dr. Lin Zhang) was supported for extracting data from this database for the study (certification number record ID: 55394138).

In JHDH, the patients diagnosed with non-CHF and CHF in the second cardiac ward of Jinghai Hospital, according to the 2022 AHA/ACC/HFSA HF Guideline[17, 18]. This cohort aims to compare the difference in the distribution of RAR between patients in CHF and non-CHF. The hospital’s ethics committee passed the approval (JHYYLL-2022-0307).

2.2 The definition of heart failure

In NHANES, the diagnosis of CHF relied on a Medical Conditions Questionnaire. Only individuals who answered the questionnaire were included in the study. According to “yes” and “no”, the crowd was divided into the non-CHF or CHF group. With the CHF guideline[17, 18], CHF-related risk factors were determined, including hypertension, atherosclerotic cardiovascular disease (ASCVD), angina, and heart attack.

In MIMIC, the diagnosis of CHF relied on the ICD9 code. According to the CHF guidelines[17, 18], if patients had risk factors (e.g., hypertension, stroke, and diabetes) for CHF while they did not develop as CHF, we classified them into the pre-CHF group. We organized them into the CHF group if they had developed CHF. The diagnosis material of 131 related ICD9 codes was shown in Table S1.

In JHDH, the diagnosis of CHF relied on CHF guidelines[17, 18], especially on clinical symptoms, B-type natriuretic peptide (BNP), and left ventricular ejection fraction (LVEF). With clinical symptoms or signs (e.g., breathlessness and fatigue), LVEF < 50%, and BNP > 100 pg/mL, the hospitalized patients were included in the CHF or non-CHF groups.

2.3 CHF mortality

For NHANES, National Death Index was linked with NHANES and directly provided the mortality-related data. As previously reported[19], CHF all-cause mortality information was derived and matched with unique individuals’ IDs. Followup time was defined as the period of blood draws time to death or end of follow-up (i.e., 31 December 2019).

For MIMIC, the mortality data were directly derived from the CSV file called PATIENTS.csv, which maintained the birth information, and date of death derived from the local social security office. Surviving volunteers had not recorded specific follow-up times, and patients who died with follow-up time were included. According to whether the CHF patients stayed for more than three years, the dead CHF patients were divided into busy within 3 years or more than three years.

2.4 Statistical analyses

All analysis processes were finished with R software (version 4.2.2). In cleaning clinical data, the R packages nhanesR (version 0.9.4.1) were served for NHANES, and pgAdmin PostgreSQL (version 14.1)[20–22] was applied in MIMIC.

In the analysis of variance, if the continuous variables were distributed with Gaussian, the Student’s t-test was applied, and else Mann–Whitney U test. A chi-square test was performed to factor data (e.g., sex and ethnicity). The continuous variables were presented as means ± SDs and proportions for factor variables.

Logistic analysis was developed to explore the odds ratio (OR) in the no-CHF and CHF groups. Cox analysis was used to calculate hazard ratios (HRs) and 95% confidence intervals (CIs) of RAR for CHF all-cause mortality. Time-to-death among groups is presented with Kaplan–Meier curves. Importantly, to optimize the cut-off for RAR in CHF patient mortality, restricted cubic spline (RCS)[23] was applied in this work. The RCS model (with 3-8 knots) examined the optimal RAR value association with CHF mortality. Only a two-tailed P-value of <0.05 was considered statistically significant.

3.1 General characteristics

60,754 participants were included in the NHANES cohort, including 29,220 (48.1%) males, 31,534 (51.9%) females, with an average age of 49.8, and 19,480 (32.1%) were over 60 years old. The NHANES cohort included non-CHF (n=58,687) and CHF (n=2,067) individuals, which could represent 218,996,994 Americans (Table 1) according to the weights. The RAR level in CHF groups (3.49±0.02) is higher than the non-CHF group (3.10±0.00). Among the 42 characteristics, only four indicators had no significance, including basophils percentage (BaP), alanine transaminase (ALT), aspartate transaminase (AST), and sodium (Na). The results are shown in Table 1.

22,672 participants were included in the MIMIC cohort, including 12,883 (56.8%) males, and 9,789 (43.2%) females, with an average age of 68.3, and 16,123 (71.1%) were over 60 years old. The MIMIC cohort included pre-CHF (n=13,281) and CHF (n=9,391) individuals (Table 2). The RAR level in CHF groups (5.08 ± 1.52) is higher than the pre-CHF group (4.74 ± 1.59). According to the results shown in Table 2, only four characteristics had shown no significant difference, including diabetes, platelet count (Pla), white blood cells (WBC), and anion gap (AG).

15,102 participants were included in the JHDH cohort, including 7,271 (48.1%) males, and 7,831 (51.9%) females, with an average age of 66.3, and 10,725 (71.0%) were over 60 years old. The JHDH cohort included non-CHF (n=12,760) and CHF (n=2,342) individuals (Table S2). The RAR level in CHF groups (4.07 ± 1.08) is higher than the non-CHF group (3.44 ± 0.81). Among the total 38 characteristics, only four had no significant difference, including monocyte percentage (MonP), monocyte (Mon), basophil count (Ba), and total bile acid (TBA).

Finally, the percentage of CHF in four quartiles of RAR was summarised in Table 3. In NHANES, the prevalence of CHF in Q1, Q2, Q3, and Q4 were 8.88%, 16.86%, 27.65%, and 46.61%. In MIMIC, these were 27.57%, 43.62%, 47.05%, and 47.49%. In JHDH, these were 4.75%, 9.04%, 17.42%, and 30.68%. All three cohorts indicated a higher RAR value means a significantly higher prevalence of CHF.

3.2 Logistic regression

Univariate and adjusted Logistic congression were processed among the above differential clinical features in the three cohorts (Figure 2). The univariate Logistic model indicated the OR in MIMIC, JHDH, and NHANES were 1.15; 95% CI 1.13-1.17, 1.87; 95% CI 1.79-1.96, and 95% CI 2.65; 2.42-2.91, respectively.

• Download figure
• Open in new tab

Figure 2.

The univariate and adjusted Logistic regression. (A) The MIMIC cohort in non-CHF and CHF groups. (B) The JHDH cohort in non-CHF and CHF groups. (C) The NHANES cohort in pre-CHF and CHF groups.

The adjusted model was processed. In MIMIC, the RAR was adjusted by age, Pho, Ca, RBC, PTT, PT, Pla, MCHC, MCH, INR, Hg, HCT, BUN, Na, K, Glu, SCR, Cl, HCO₃, and AG. Because RAR was calculated by ALB and RDW, both variables were not included to adjust RAR to eliminate overfitting. In JHDH, the RAR was adjusted by age, ALT, AST, ALP, EoP, Glu, GP, HDL, Hg, LDL, Lym, LymP, MPV, NEUT, MCH, MCHC, PDW, RBC, TC, TG, and TP. In NHANES, the RAR was adjusted by age, LymP, Mon, SegneP, EoP, Lym, MonP, SeneP, Eo, Hg, Hem, Ca, TC, HCO₃, Glu, Fe, TP, UA, Cl, and GLB.

Among the three cohorts with adjusted, RAR was the highest risk factor associated with non-CHF and CHF patients, JHDH (OR:1.45; 95% CI 1.33-1.58) and NHANES (OR:1.93; 95% CI 1.41-2.65). And in pre-CHF and CHF groups, RAR was still a risk factor, MIMIC (OR:1.05; 95% CI 1.02-1.07). More importantly, in the included element, the adjusted RAR was the highest risk factor in two cohorts, JHDH and NHANES.

The Logistic regression for OR values at the three centres according to the quartiles of RAR are shown in Table 4. The Q1 was applied as a reference. In NHANES, the OR value of Q2, Q3 and Q4 were 2.02(1.19,3.43), 3.24(1.95,5.39), and 4.95(2.44,10.02), respectively. In MIMIC, the OR value of adjusted Q2, Q3, and Q4 were 1.62(1.49,1.77), 1.68(1.54,1.84), and 1.59(1.44,1.75), respectively. In JHDH, the adjusted OR value of Q2, Q3, and Q4 were 1.7(1.39,2.07), 2.99(2.44,3.65), and 5.15(4.04,6.57), respectively. The three cohorts indicated that a higher level of RAR shows a higher OR value.

3.3 The Cox regression

With the above significant factor of Logistic regression, the univariate and adjusted Cox regression (Figure 3) in MIMIC and NHANES was developed. In NHANES, all 1,742 CHF alive (n=784) and dead (n=958) in the follow-up survey (Table S3) had an average age of 68 years, with 965 males (55.4%). And in the MIMIC, the 6,016 CHF patients alive in 3 years (n=1,507) or not (n=4,509) had an average of 74 years (Table S4), with 3,208 males (53.3%). More importantly, the adjusted hazard ratio (HR) (adjusted by the same factors in the logistic analysis) in MIMIC and NHANES (Figure 3) was 1.12 (95% CI 1.1-1.14) and 2.26 (95% CI 1.52-3.36).

• Download figure
• Open in new tab

Figure 3.

The univariate and adjusted Cox regression. (A) The CHF dead or not within 3 years in the MIMIC cohort. (B) The CHF is dead or not in the follow-up survey in the NHANES cohort.

3.4 RCS analysis for the optimal cut-off value

In RCS analysis, MIMIC adjusted the RAR with age, ALB, RDW, MCH, and MCHC (Figure S1). In NHANES, the RAR was adjusted by age, Hem, RDW, and Glu (Figure S2). According to the minimum fluctuation HR value range of 95% CI, the optimal value of RAR, ALB, and RDW was determined (Figure 4). Though the optimal cut-off value in MIMIC and NHANES differed, the trend is similar. For example, though the cut-off values of RAR are different, 5.06 in MIMIC or 3.4 in NHANES, the increasing RAR indicated a gradually rising HR. And in the ALB, both the MIMIC and NHANES showed a similarly fluctuant trend of HR.

• Download figure
• Open in new tab

Figure 4.

The RCS analysis for RAR, ALB, and RDW in MIMIC and NHANES cohorts.

The three factors were levels according to the optimal value of RAR, ALB, and RDW (Figure 4). For example, an individual in MIMIC with a RAR >5.06 was divided into the high RAR group, else the low RAR group. The RDW and ALB classified methods were similar to RAR.

3.5 KM survival curve

With the dichotomous RAR, ALB, and RDW levels, the KM curve was used to observe the CHF-related follow-up population. More importantly, CHF patients with RAR > 3.4 in NHANES or 5.06 in MIMIC showed a poor survival rate than the others (Figure 5). And the CHF patients with ALB < 4.09 in NHANES or 3.20 in MIMIC exhibited a poor survival rate than the others (Figure 5). RDW had a similar phenomenon to RAR (Figure S3). The median follow-up time of MIMIC and NHANES was 364 days and 62 months.

• Download figure
• Open in new tab

Figure 5.

The KM curve of RAR and ALB in NHANES and MIMIC. (A) The KM curve of RAR. (B) The KM curve of ALB. (C) The mortality rate of 19 RAR, ALB, and RDW quantiles in MIMIC. (D) The nomogram plot in MIMIC.

MIMIC had more follow-up CHF patients than NHANES, which is suitable for the mortality statistics in multiple quantiles. The various quantile mortality rates in MIMIC were processed. When quantiles were set to more than 19, the 6th quantiles in ALB had a zero death population. Therefore, 19 quantiles of RAR, ALB, and RDW mortality rates were processed (Table S5). In the mortality of the 19 quantiles, higher RAR and RDW indicated higher mortality (Figure 5. C). And lower ALB demonstrated higher mortality. However, the mortality rate of RAR is more stable than RDW (Figure 5. C). And the mortality rate of RDW drastically fluctuated with a gradually increased. Regarding ALB, the fluctuating wave was even more drastic than RDW.

In the 1st quantile (Table S5), the mortality rate of RAR was lower than RDW (0.62 vs 0.67). And in the 19th quartile, the mortality rate of RAR is higher than RDW (0.87 vs 0.82). RAR’s 19th percentile mortality rate is 1.4 times higher than the 1st percentile. At the same time, it is 1.22 times for RDW. RAR was superior to RDW and ALB in predicting CHF death due to its stable predictive value at different quantile values (Figure 5. C). Based on the Cox analysis of CHF in MIMIC, nomogram plots were processed to predict the three-year survival of patients at RAR, RDW, ALB, and age (Figure 5. D).