Introduction

Coronary artery disease (CAD) is a leading cause of death worldwide, significantly impacting both developed and developing nations1,2,3,4,5. Despite advancements in treatments such as aspirin, statins, and percutaneous coronary intervention, mortality is particularly high among critically ill CAD patients in the ICU, who often present with poor baseline health and rapid disease progression, leading to a 1-year mortality rate of up to 16.1%6,7,8. While CAD ranks as the second leading cause of death worldwide, after malignant tumors9,10research specifically focused on optimizing prognostic management for this critically ill subpopulation remains limited.

Blood pressure variability (BPV), the fluctuation of blood pressure(BP) over time, is influenced by hemodynamics, neurohumoral regulation, behavior, and environment11,12,13,14,15,16. These fluctuations, beyond absolute BP values, can damage coronary arteries and increase cardiac load and oxygen demand, negatively impacting cardiac prognosis17,18. Studies have shown a strong link between elevated BPV and cardiovascular events19,20,21,22,23,24.Studies in ICU patients link increased BPV to higher mortality, prolonged ICU stays, and organ dysfunction. However, existing ICU BPV research has limitations: small sample sizes, specific subpopulations, and inconsistent BPV metrics23,25,26. This heterogeneity hinders firm conclusions about optimal BPV measurement and its prognostic value across the broader ICU population, especially concerning whether increased BPV directly leads to poor outcomes23,25.

The impact of BPV on outcomes in critically ill patients with coronary artery disease is particularly understudied. Existing CAD research often focuses on ambulatory monitoring in non-critically ill individuals16,17neglecting the unique physiological challenges and interventions in the ICU (e.g., vasopressors, mechanical ventilation), which can significantly influence BPV. Therefore, findings from ambulatory settings may not be directly applicable to critically ill patients. While traditional metrics for assessing BPV, such as standard deviation (SD) and coefficient of variation (CV), primarily focus on the overall spread of BP values, they do not capture the sequential nature of BP fluctuations. To address this, newer metrics like average real variability (ARV) have emerged. ARV offers a more reliable assessment by considering the temporal dynamics of BP changes and is less affected by outliers, making it particularly suitable for ICU settings27,28. This study seeks to refine the assessment of BPV’s prognostic significance in critically ill patients with CAD by examining the association between 24-hour ARV and mortality.

Methods

Database

This study is based on the Medical Information Mart for Intensive Care-IV (MIMIC-IV) version 2.2 database, which supports a retrospective cohort analysis. The MIMIC-IV database contains clinical data from over 50,000 patients admitted to Beth Israel Deaconess Medical Center from 2008 to 2019. It provides extensive, real-world clinical data encompassing vital signs, laboratory test results, medication records, diagnosis codes, imaging reports, prognosis, and other pertinent information relevant to ICU patients in a tertiary care setting. Data extraction and processing were conducted using Structured Query Language to facilitate subsequent analysis. All researchers involved in this study completed relevant training courses provided by the National Institutes of Health and successfully passed necessary assessments to gain access to the MIMIC-IV database (certificate number: 52898099). In this research, we utilized version 2.2 of the database and employed PostgreSQL v13.0 for data retrieval (http://www.postgresql.org/). As all personal data in the database has been deidentified before analysis, the requirement for institutional review board approval was waived, and patient consent was not required29,30.

Study populations

This study extracted patient data from the MIMIC-IV version 2.2 database. The inclusion and exclusion criteria were as follows:

Inclusion Criteria:

  1. (1)

    Patients were included if they had the history of CAD. The diagnosis of CAD was based on ICD-9 and ICD-10(Supplementary Table S1).

  2. (2)

    Age ≥ 18 years.

  3. (3)

    First admission to the ICU.

Exclusion Criteria:

  1. (1)

    ICU stay of less than 24 h.

  2. (2)

    Continuous BP measurements with intervals exceeding 1 h during the ICU stay, or fewer than 24 recorded BP measurements.

  3. (3)

    SBP < 0 mmHg or > 400 mmHg, or DBP < 0 mmHg or > 300 mmHg.

Based on these criteria, a total of 4588 CAD patients who were initially admitted to the ICU between 2008 and 2019 were included in this study.

Data extractions

This study utilized PostgreSQL and Navicat Premium 16 software for data extraction, acquiring the necessary information through Structured Query Language.

BP measurements were a crucial component of this study. All BP measurements were obtained via continuous intra-arterial monitoring recorded in the MIMIC-IV database. SBP values were identified by itemid 220,050, and DBP values by itemid 220,051.

Drawing from previous literature19,20,23,24,25,27,31,32,33 and clinical significance, the study incorporated the following potential confounding variables:

  • Baseline demographic information: age and gender.

  • Mean vital signs within the first 24 h after ICU admission: heart rate, mean arterial pressure, and respiratory rate.

  • Laboratory parameters measured included white blood cells (WBC), hemoglobin (HGB), platelets (PLT), and blood creatinine.

  • Disease states: myocardial infarction, congestive heart failure, cerebrovascular disease, chronic lung disease, hypertension, diabetes, atrial fibrillation, sepsis, peptic ulcer disease.

  • Severity of illness scoring: Simplified Acute Physiology Score II (SAPS II).

  • Treatment measures: including the use of invasive mechanical ventilation within the first 24 h after ICU admission34administration of vasopressors and positive inotropic agents (dopamine, epinephrine, norepinephrine, dobutamine, vasopressin, dobutamine, and milrinone), use of antihypertensive medications (including angiotensin-converting enzyme inhibitors, angiotensin receptor blockers, α receptor blockers, β receptor blockers, calcium channel blockers, and diuretics), implementation of antiplatelet therapy, and administration of statins.

Definition of BPV and ARV

BPV refers to the fluctuations in BP over time, reflecting the complex interplay of various physiological regulatory mechanisms. While traditionally assessed using metrics like SD and CV, these measures have limitations. SD quantifies the dispersion of BP values around the mean but ignores the temporal sequence of these values. CV, calculated as the SD divided by the mean, provides a relative measure of variability, but it also fails to account for the order of BP readings. To evaluate the dynamic changes in clinical signs and to provide a more reliable measure of BPV over time, this study employs average real variability to assess BPV during the first 24 h after patients are admitted to the ICU. The calculation formula is as follows:

$$\:\text{A}\text{R}\text{V}=\frac{1}{\text{N}-1}{\sum\:}_{\text{k}=1}^{\text{N}-1}|{\text{B}\text{P}}_{\text{k}+1}-{\text{B}\text{P}}_{\text{k}}|$$

where N represents the number of valid BP measurements and k represents the order of measurements27,28.ARV, a newer BPV metric, overcomes these limitations. ARV is calculated as the average absolute difference between successive BP measurements. This crucial distinction allows ARV to capture the time-dependent nature of BP fluctuations, differentiating between BP sequences with identical SD or CV values but different patterns of oscillation. For instance, two patients could have the same SD of BP, but one might have frequent, rapid fluctuations while the other has slow, gradual changes; ARV would be higher in the former case. Furthermore, ARV is less sensitive to outliers and variations in sampling frequency compared to SD and CV, making it a more robust and reliable indicator of BPV.Based on the clinical data from all patients involved in this research, the ARV of SBP (ARV-SBP) and the ARV of DBP (ARV-DBP) were calculated.

To further provide a comprehensive account of our investigations into different blood pressure variability parameters, Average Real Variability of Mean Blood Pressure (ARV-MBP) was also calculated. The detailed analysis and results for ARV-MBP are provided in the Supplementary Table S2.

Study endpoint

The primary endpoint for this study was one-year mortality following ICU admission. The date of death, used to determine one-year mortality, was obtained from the “dod” (date of death) column in the “patients” table of the MIMIC-IV database, which is based on hospital and state records. All patients were followed up for one year to ascertain this primary outcome.In-hospital mortality was evaluated as a secondary endpoint. Information regarding in-hospital mortality was derived from the “hospital expire flag” column in the “admissions” table of the MIMIC-IV database.

Statistical analysis

The statistical analysis was conducted using R software (version 4.1.3; R Foundation for Statistical Computing, Vienna, Austria; https://www.r-project.org). The normality of continuous variables was assessed using the Shapiro-Wilk test. Results indicated that all continuous variables deviated from a normal distribution(p < 0.05). Therefore, non-parametric tests were employed throughout the analysis. Descriptive statistics included reporting the median and interquartile range (IQR) for all continuous variables. Categorical variables were presented as numbers and percentages. The Kruskal-Wallis H test, a non-parametric equivalent of one-way ANOVA, was used to compare continuous variables across the three groups, while Fisher’s exact test was utilized for comparing categorical variables. Based on the distributions of ARV-SBP and ARV-DBP, low, medium and high tertiles groups were created (SBP-ARV: <9.86, 9.86–12.69, > 12.69;ARV-DBP<5.28, 5.28–7.07, > 7.07); We selected these confounders on the basis of judgment, previous scientific literature19,20,23,24,25,27,31,32,33. Multicollinearity was tested using the variance inflation factor (VIF) method, with a VIF ≥ 5 indicating the presence of multicollinearity. To assess associations of ARV with short-term and long-term mortality respectively, multivariable logistic regression and Cox proportional hazards regression analyses were performed. ARV was entered as a categorical variable (tertiles) and as a continuous variable. Analyses were first performed in a crude mode l (model 1: adjusted for age, sex and vital signs). Further analyses cumulatively included adjustment for laboratory parameters, disease states, disease severity scores (model 2), and treatment measures (model 3).We conducted restricted cubic spline model to develop smooth curves to examine the possible nonlinear dose-response associations between ARV and mortality. In this model, ARV was used as a continuous variable with four knots (5th, 35th, 65th and 95th) suggested by Harrell. Non-linearity tested by using a likelihood ratio test comparing the model with only a linear term against the model with linear and cubic spline terms. If a non-linear correlation was observed, a two-piecewise regression model was performed to calculate the threshold effect of ARV on mortality in terms of the smoothing plot35. Missing data comprised less than 5% of the dataset, and we dealt with missing data by listwise deletion on an analytical basis. We performed a series of model adjustments to assess the robustness of the findings. We report and compare effect sizes and p-values calculated by all these models. Two-sided p values of less than 0.05 were considered statistically significant.

Results

Population characteristics

Figure 1 delineates the process of cohort selection. A total of 9,582 patients with coronary artery disease were screened from the MIMIC-IV 2.2 database and 4,994 patients were subsequently excluded based on the study criteria, resulting in a final study population of 4,588 patients.

Table 2 summarizes the fundamental characteristics of the study population. Among the participants, 82.3% are over 60 years of age, with 1,232 individuals (26.9%) being female. Moreover, 68.7% of patients received treatment with vasopressors and positive inotropic agents on the first day of ICU admission, while 71.5% received antihypertensive medication on that same day, with additional details on in-hospital and 1-year survival provided in Supplementary Tables S3 and S4.

The study categorized patients into three equal groups (low, medium and high) based on the distribution of ARV-SBP and ARV-DBP measured within 24 h of admission (ARV-SBP: <9.86, 9.86–12.69, > 12.69; ARV-DBP: <5.28, 5.28–7.07, > 7.07). Overall, the baseline characteristics across the different groups are comparable; however, significant differences exist between the SBP and DBP groups regarding gender, the presence of myocardial infarction and heart failure, the administration of vasopressors and positive inotropic agents, and the need for invasive mechanical ventilation (Table 1). Additionally, the SBP group shows significant differences in age, as well as the prevalence of diabetes and hypertension, whereas the DBP group exhibits notable differences in the utilization of antiplatelet agents and statins. Detailed baseline characteristics have been moved to Supplementary Table S5. The follow-up of 4,588 critically ill patients with coronary artery disease revealed that 208 individuals (4.5%) died during hospitalization, and 570 individuals (12.4%) succumbed within one year of admission. These findings indicate that the mortality rate among critically ill patients with coronary artery disease remains notably high.

Fig. 1
figure 1

Flowchart showing the process of cohort selection.

Full size image
Table 1 Baseline characteristics of the population.
Full size table

Relationship between ARV and short-term mortality in patients with CAD

The overall in-hospital mortality rate among patients was 4.5%. In the ARV-DBP tertile groups, in-hospital mortality increased with higher tertiles, rising from 3.6% in the low tertile to 6.1% in the high tertile. In the multivariable logistic regression analysis of ARV-DBP tertiles, compared with the low tertile, the in-hospital mortality risk increased by 80% in the high tertile (OR 1.80, 95% CI 1.28–2.54, P = 0.001), but after adjusting for confounding factors, this association was no longer significant. In the multivariable logistic regression analysis with ARV-DBP as a continuous variable, an increase in ARV-DBP was positively associated with in-hospital mortality. However, after further adjustment, this association disappeared. In contrast, no significant association was observed between ARV-SBP and in-hospital mortality. After adjusting for potential confounders, no significant differences were observed when ARV-SBP was analyzed as either a continuous or categorical variable (Table 3).

Relationship between ARV and long-term mortality in patients with CAD

A long-term mortality assessment indicated an overall mortality rate of 12.4% within one year. The univariate analysis demonstrated a significant association between ARV-DBP and long-term prognosis. In the Cox regression analysis categorized by tertiles, patients in the high ARV-DBP group exhibited a 1.97-fold increased risk of death within one year compared to those in the low ARV-DBP group. This association remained significant after adjusting for potential confounding factors (HR 1.26, 95% CI 1.01–1.56, P = 0.037, Table 2, Model 3). Treating ARV-DBP as a continuous variable in the Cox regression analysis revealed that each unit increase in ARV-DBP was associated with a 14% rise in the risk of death. This relationship persisted as significant after controlling for confounding factors. The multivariable-adjusted restricted cubic spline curve illustrating the relationship between ARV-DBP and one-year mortality is presented in Fig. 2(A).The curves in particular underscore the relationship between ARV-DBP and risk ratios showed a generally positive correlation (P for non-linearity = 0.85), with an increasing trend in 1-year mortality with increasing DBP fluctuations.

In the analysis of tertiles of ARV-SBP, the reduction in 1-year mortality rate for the medium variability group compared to the low variability group is statistically significant (Table 3). Although this reduction did not show a significant association with the high variability group, the notable difference in the medium variability group suggests a potential nonlinear relationship between SBP variability and 1-year mortality rate. To further investigate this hypothesis, we utilized a restricted cubic spline nonlinear model to more accurately capture the possible nonlinear trend between SBP variability and mortality risk. Figure 2(B) illustrates that, after adjusting for potential confounding factors, there is a U-shaped nonlinear relationship between ARV-SBP and 1-year mortality rate, indicating that both excessively low and high SBP variability can increase mortality risk. The inflection point of this relationship is at an ARV-SBP value of 16.912. When ARV-SBP falls below 16.912, each 1-unit increase in ARV-SBP correlates negatively with the one-year mortality risk, suggesting that increased BP variability at lower levels may confer some protective benefits. Conversely, when ARV-SBP exceeds 16.912, the one-year mortality risk rises significantly with increased SBP variability (Table 4).

Table 2 Multivariable analysis of the association between ARV-DBP and mortality.
Full size table
Table 3 Multivariable analysis of the association between ARV-SBP and mortality.
Full size table
Fig. 2
figure 2

Restricted cubic spline (RCS) curves of 1-year mortality risk according to (A) ARV-DBP values and (B) ARV-SBP values.

Full size image

Data were fit by a Cox proportional hazard regression model based on restricted cubic splines. ARV was entered as continuous variable. Data were adjusted for gender, age, vital signs, Laboratory measurements, comorbidities, severity of illness scoring, and treatment (Model 3).The median ARV was defined as the reference standard. The red area represents the 95% CI.

Table 4 The non-linearity relationship between ARV-SBP and 1-year mortality.
Full size table

Discussion

In 4,588 critically ill CAD patients, this study found that 24-hour ARV-DBP positively correlated with 1-year mortality, independently of confounders. ARV-SBP showed a U-shaped relationship with 1-year mortality (nadir at 16.912 mmHg). Neither ARV-DBP nor ARV-SBP was significantly linked to in-hospital mortality after adjustment.

While prior BPV research often focused on hypertensive populations, recent studies link BPV to adverse outcomes in non-hypertensive individuals36,37,38,39,40including increased myocardial infarction and mortality risk in the elderly41. However, Harefa et al.42 found no significant ARV-BPV link to in-hospital events in acute myocardial infarction patients, potentially due to smaller samples or different BPV metrics. Critically ill CAD patients present unique challenges due to dynamic physiological changes and intensive interventions impacting BPV, differentiating them from general or hypertensive populations and warranting specific investigation.

Comparing our findings with studies also using the MIMIC-IV database, He et al.25 found higher systolic BPV increased mortality in critically ill CAD patients. Our study, employing ARV to better capture temporal BP dynamics, aligns on the importance of BPV for long-term risk but differs in specific associations and metrics. Liu et al.23 reported no significant ARV-DBP link to short or long-term mortality in ICU patients with acute MI, contrasting with our finding of a positive correlation between ARV-DBP and 1-year mortality in a broader critically ill CAD cohort. These discrepancies may stem from our larger sample, broader patient population, the use of ARV, and more extensive adjustments for ICU-specific treatments (e.g., vasoactive drugs, antihypertensives, statins), which significantly influence BPV and outcomes.

The seemingly paradoxical finding of a significant association between 24-hour BPV and 1-year mortality, but not in-hospital mortality, warrants careful consideration. Several potential, non-mutually exclusive explanations may account for this apparent discrepancy. Firstly, the intensive in-hospital management of critically ill patients within the ICU environment, characterized by continuous monitoring and aggressive interventions such as fluid resuscitation, vasopressor administration, mechanical ventilation, and rapid antihypertensive titration, can substantially alter BP dynamics. These interventions, while crucial for immediate stabilization, may effectively mask the immediate influence of baseline BPV on short-term outcomes26. Secondly, it is plausible that the initial 24-hour BPV serves as an indicator of underlying vascular and autonomic dysregulation, reflecting a pre-existing vulnerability that manifests its prognostic significance over a longer timeframe43,44. The mechanisms through which BPV contributes to cardiovascular damage, including endothelial dysfunction and the progression of atherosclerosis, are likely to exert their influence more prominently in the long term. Thirdly, the acute and dominant drivers of in-hospital mortality in critically ill CAD patients, such as sepsis, heart failure, or cerebrovascular disease, could effectively overshadow the more subtle contribution of BPV to short-term mortality. In essence, the signal of BPV’s impact on in-hospital mortality may be obscured by the noise of these more immediate and potent factors. Furthermore, the inherent heterogeneity of our critically ill CAD patient population, despite adjustments for confounders, may introduce variability that obscures a clear short-term association. Residual confounding cannot be entirely excluded. Finally, while our sample size is considerable, it is possible that the statistical power was insufficient to detect a potentially smaller effect of BPV on the less frequent event of in-hospital mortality, compared to one-year mortality. These factors collectively suggest that while intensive care may mitigate the immediate mortality risk associated with BPV, its prognostic significance for longer-term outcomes, reflecting deeper vascular pathophysiology, remains evident. Further research specifically designed to investigate the nuanced relationship between BPV and short-term mortality in critically ill patients is warranted.

Adjusted analyses revealed a U-shaped nonlinear relationship between ARV-SBP and 1-year mortality. Risk decreased with ARV-SBP up to 16.912 mmHg, then significantly increased above this threshold, suggesting dual mechanisms impacting long-term prognosis. Below this inflection point, moderate ARV-SBP—potentially reflecting physiological regulation, stable organ perfusion, or even beneficial effects of timely vasoactive therapy—was associated with lower mortality. Conversely, very low variability might indicate autonomic dysfunction43,44. Above this threshold, excessive ARV-SBP likely signifies disrupted cardiovascular autonomic regulation, hemodynamic instability, increased cardiac workload, and underlying vascular impairment, leading to higher mortality.

To provide a more comprehensive assessment, we also investigated the ARV-MBP, with detailed results presented in Supplementary Table S2. ARV-MBP showed no independent association with 1-year mortality after full multivariable adjustment. The absence of this association is notable. We posit this notable discrepancy stems from two factors. From an intrinsic perspective, MBP is a calculated parameter. The process of combining SBP and DBP into MBP might attenuate the specific prognostic information carried by the variability of SBP and DBP components individually. SBP variability primarily relates to arterial stiffness and cardiac output fluctuations17,45,46while DBP variability is more sensitive to changes in vascular tone and diastolic coronary perfusion time17,47,48. Therefore, the combining process risks obscuring the specific insights offered by each of these unique pathophysiological mechanisms. Furthermore, a external confounder is goal-directed therapy in the ICU. Clinicians frequently target a specific MBP level when titrating vasopressors, and this continuous intervention may artificially suppress the natural variability of MBP, thereby confounding its value as an intrinsic physiological marker. As the prognostic signal of MBP may be doubly attenuated by both its mathematical derivation and clinical interventions, we conclude that in our cohort of critically ill patients with CAD, the variabilities of the component pressures (SBP and DBP) are more robust and specific long-term prognostic markers than the composite MBP. However, it is crucial to emphasize that these findings pertain to the prognostic limitations of MBP variability and do not diminish the established clinical importance of monitoring the absolute MBP level to guide hemodynamic management and ensure adequate organ perfusion. Moreover, the definitive prognostic significance of ARV-MBP in this population warrants further investigation through studies with higher levels of evidence.

Determining the optimal BPV range in critically ill CAD patients is challenging. Putative mechanisms for BPV-induced cardiovascular target organ damage include direct endothelial injury from perfusion pressure changes49renin-angiotensin system activation promoting vascular smooth muscle cell proliferation and hypertrophy50worsened myocardial ischemia-reperfusion injury inducing cardiomyocyte apoptosis51,52and systemic inflammatory responses53. Ultimately, these can lead to endothelial dysfunction54atherosclerosis development, cardiac remodeling, and impaired diastolic and systolic functions, thereby heightening adverse cardiovascular event risk55,56.

Our findings underscore that BP stability, particularly minimizing ARV-DBP fluctuations, is critical for the long-term prognosis of critically ill CAD patients. Clinically, while moderate ARV-SBP may be beneficial, excessive fluctuations significantly increase mortality risk. Therefore, maintaining moderate BPV while avoiding excessive fluctuations, guided by close BP monitoring and precise control, is vital for improving long-term survival rates in this vulnerable population.

This study has several limitations. Firstly, reliance on continuous intra-arterial BP monitoring, while precise for capturing BPV in critically ill patients, limits direct generalizability to non-ICU settings where non-invasive cuff-based measurements are standard and carries inherent risks57. Secondly, the potential impact of the data’s age and the evolution of medical practices on our findings, as data were collected over an extended period, which may be influenced by evolving medical practices and device updates. This is compounded by the constraint imposed by data de-identification in MIMIC-IV, which prevents detailed analysis stratified by precise patient inclusion periods. Thirdly, due to incomplete information in the database, some potential confounding factors (e.g., echocardiographic results, troponin levels, myocardial enzyme profiles) could not be analyzed. Finally, the applicability of these findings to general ward patients warrants further investigation.

Conclusion

This study demonstrates that 24-hour BPV, measured by ARV, show significant associations with 1-year mortality in critically ill CAD patients. Elevated ARV-DBP independently predicts increased mortality, while ARV-SBP exhibits a U-shaped relationship, with a nadir at 16.912 mmHg, suggesting an optimal range for systolic variability. These findings highlight the prognostic value of BPV and suggest that minimizing diastolic fluctuations and targeting moderate systolic variability may improve long-term outcomes. Further prospective studies are needed to validate these findings and evaluate BPV-targeted interventions.