Severe fluctuation in mean perfusion pressure is associated with increased risk of in-hospital mortality in critically ill patients with central venous pressure monitoring: A retrospective observational study

Background The mean perfusion pressure (MPP) was recently proposed to personalize tissue perfusion pressure management in critically ill patients. Severe fluctuation in MPP may be associated with adverse outcomes. We sought to determine if higher MPP variability was correlated with increased mortality in critically ill patients with CVP monitoring. Methods We designed a retrospective observational study and analyzed data stored in the eICU Collaborative Research Database. Validation test was conducted in MIMIC-III database. The exposure was the coefficient of variation (CV) of MPP in the primary analyses, using the first 24 hours MPP data recorded within 72 hours in the first ICU stay. Primary endpoint was in-hospital mortality. Results A total of 6,111 patients were included. The in-hospital mortality of 17.6% and the median MPP-CV was 12.3%. Non-survivors had significantly higher MPP-CV than survivors (13.0% vs 12.2%, p<0.001). After accounting for confounders, the highest MPP-CV in decile (CV > 19.2%) were associated with increased risk of hospital mortality compared with those in the fifth and sixth decile (adjusted OR: 1.38, 95% Cl: 1.07–1.78). These relationships remained remarkable in the multiple sensitivity analyses. The validation test with 4,153 individuals also confirmed the results when MPP-CV > 21.3% (adjusted OR: 1.46, 95% Cl: 1.05–2.03). Conclusions Severe fluctuation in MPP was associated with increased short-term mortality in critically ill patients with CVP monitoring.


Study population
This study utilized data stored in the eICU Collaborative Research Database (eICU-CRD) v2.0 [13], a unique and publicly accessible multicenter database covering more than 200,000 ICU admissions. The data stored in the database was collected through the Philips eICU program, a critical care telehealth program that delivers information to caregivers at the bedside. Vital signs were generally interfaced as 1-minute averages, and archived into the database as 5-minute median values [14]. The inclusion criteria were (1) age 16 years or more; (2) at least 24 hours of continuous MAP and CVP invasive monitoring within the first 72 hours in the first ICU stay and (3) at least 20 MPP readings in the daytime and at least seven in the nighttime [15]. Daytime is defined as 7 am to 11 pm, otherwise as nighttime. Those who received dialysis, died during the first 24 hours, were complicated with chronic kidney disease stage 5, intracranial hypertension, abdominal compartment syndrome and with incomplete data or extreme MPP data were excluded. Extreme MPP refers to the values of MAP not between 0 mmHg to 150 mmHg, and the values of CVP not between -10 mmHg to 50 mmHg. Patients with CKD stage 5 were excluded because they may undergo dialysis, which will significantly affect MPP and increase variability.

Data extraction
We extracted MPP data, demographic data, baseline ICU characteristics, Charlson comorbidity index [16], and admission illness severity score (the Sequential Organ Failure Assessment (SOFA) [17]). Criteria for sepsis were defined based on those described earlier by Angus et al [18] instead of sepsis 3.0 because most microbiology data was unavailable in eICU-CRD. Additionally, the need for mechanical ventilation, the incidence of AKI, use of vasopressor, antihypertensive drugs, and sedatives were also collected. As MPP is a dynamic process, timeweighted average MPP (TWA-MPP) during the first 24 hours of ICU stay was calculated as the area under the MPP-versus-time plot as follows to truly reflect the average level of MPP.
where X n is the value of the variable of interest at the timepoint t n .

Data cleaning
We chose the values of MAP between 0 mmHg and 150 mmHg, and the values of CVP between -10 mmHg and 50 mmHg.

Exposure
Short-term MPPV was measured as the coefficient of variation (CV) of 24-hour MPP data (MPP-CV), defined as the standard deviation (SD) divided by the mean MPP value.

Outcomes
The primary outcome was in-hospital mortality.

Statistical analysis
This is a post hoc analysis. Statistical analyses were performed using R version 3.63 (R Foundation for Statistical Computing, Vienna, Austria; www.r-project.org). Firstly, the baseline characteristics were compared between survivors and non-survivors. Categorical variables were presented as percentages and compared using a chi-square test. Continuous variables were expressed as median (25th, 75th percentile) and compared using Wilcoxon rank-sum test. To get a better understanding of the relationship between MPPV and TWA-MPP as well as other blood pressure variability (BPV), we used correlation matrices to show the correlation coefficient and then analyzed the association between MAP variability (MAPV) and prognosis.
Secondly, generalized additive models with a logit link function were built to plot associations between MPP-CV and in-hospital mortality, adjusted by age, gender, BMI, ethnicity, Charlson comorbidity index, SOFA score, admission type (elective surgery, emergency surgery or medicine), cardiovascular surgery, history of tachyarrhythmia, sepsis, incidence of AKI in the first day of ICU admission, the need for mechanical ventilation, the use of vasopressor, antihypertensive drug, sedatives, and TWA-MPP.
Thirdly, taking MPP-CV as classification variables, we used multivariable logistic regression models to assess the relationship between the hospital mortality and deciles of each parameter in which the median two deciles, the fifth decile together with the sixth decile, were chosen as reference. The multivariable logistic regression models were adjusted by the same variables mentioned above.

Subgroup and sensitivity analyses
Subgroup analyses of increased MPP-CV were conducted in patients who were male or female, elderly (age � 65 years) or not, with or without hypertension, sepsis, higher than median SOFA score or not on the first day of ICU admission, admission type (surgical or medical), cardiovascular surgery or not.
Variation independent of the mean (VIM) [19] of 24-hour MPP data (MPP-VIM) was also analyzed in the sensitivity analyses. Both of the two indicators (CV and VIM) are considered to be relatively independent of the mean value [19]. Detailed formulas are displayed in S1 Table. Furthermore, as circadian rhythm exists in blood pressure, the association between daytime or nighttime MPP-CV and hospital mortality were also analyzed to observe whether the association between MPPV and prognosis is solely contributed by daytime or nighttime MPPV. Finally, based on the median MPP-CV (12%) during the first 12 hours of the 24 hours, we categorized the patients into initial low variability (MPP-CV < 12% in the first 12 hours of the 24 hours) groups and initial high variability (MPP-CV > 12% in the first 12 hours of the 24 hours) groups. Based on the MPP-CV in the second 12 hours of the 24 hours, the initial low variability group was further categorized into persistently low (MPP-CV < 12% in the second 12 hours of the 24 hours) group (group 1) and increasing (MPP-CV > 12% in the second 12 hours of the 24 hours) group (group 2); the initial high variability group was further categorized into decreasing (MPP-CV < 12% in the second 12 hours of the 24 hours) group (group 3) and persistently high (MPP-CV > 12% in the second 12 hours of the 24 hours) group (group 4). By grouping, we tried to observe the difference of hospital mortality under different change modes.
There were missing values for body mass index (BMI) (3.2%) and multiple imputation was used to handle the missing values with the mice package in R. For all analyses, a two-tailed P value less than 0.05 was considered statistically significant.

Validation test
The Medical Information Mart for Intensive Care (MIMIC)-III [20,21]

Ethics approval and consent to participate
The study was conducted entirely on the publicly available, third-party anonymous public databases. The ethics committee of our hospital waived the requirement for approval of this study (2021-QT-08). To apply for access to the database, we completed the National Institutes of Health's web-based course and passed the Protecting Human Research Participants exam (record ID. 32559175, ID. 38120064). All methods were performed in accordance with the relevant guidelines and regulations.

Patient characteristics
After reviewing 166,355 first ICU stays in eICU-CRD, we finally included 6,111 fulfilling the inclusion and exclusion criteria (Fig 1). The baseline characteristics between survivors and non-survivors are shown in Table 1. The survivors were, on average, younger, predominantly male and higher in BMI. Non-survivors were significantly complicated with more comorbidities, more severe in SOFA score, needing more support (mechanical ventilation and vasopressor), less antihypertensive drug use, higher incidence of AKI and sepsis, and lower MPP compared with survivors. More non-survivors had a tachyarrhythmia history and less use of sedatives, but the difference did not reach statistical significance. Other information about hospitals, initial diagnosis, comorbidities, and MPP data of the whole cohort was listed in S2 Table. The median of the MPP-CV was 12.3% in the whole cohort. The 10th and 90th percentile for MPP-CV were 8.1% and 19.2%, respectively. The non-survivors had higher MPP-CV (13.0% vs 12.2%, p<0.001) as compared with survivors.

Association with TWA-MPP and other BPV
The correlation matrix showed us that the correlation coefficient between MPP-CV and MPP-VIM was 0.98, which was very strong. There was no correlation between the two MPPV parameters and the TWA-MPP (S1 Fig). We also explored the correlation coefficients between MPPV and other BPV. Among them, MAP-CV had the highest correlation coefficient (r = 0.77, r 2 = 0.60) with MPP-CV. Although CVP is also a part of MPP in calculation, the correlation between CVP-CV and MPP-CV was weak (r = 0.08, r 2 = 0.006).

Association with hospital mortality
Before and after adjusting for all the confounders, we found that hospital mortality increased when the MPP-CV increased (Fig 2A). After grouping in deciles (Fig 2B), univariate logistic regression revealed that higher MPP-CV (CV > 19.2%) were related to an increase in the risk of hospital mortality compared with the fifth and sixth decile (adjusted odds ratio [OR] in the tenth decile: 1.91, 95% confidence interval [Cl]:1.51-2.41). Multivariable logistic regression also revealed an increase in the risk of hospital mortality when MPP-CV > 19.2% (adjusted OR in the tenth decile: 1.38, 95% Cl:1.07-1.78).
Considering the high correlation between MPPV and MAPV, we also analyzed the relationship between MAP-CV and prognosis in two databases. In eICU-CRD database, MAP-CV and mortality showed a U-shaped curve as compared to the median two deciles. In the MIMIC database, however, there was no significant correlation between increased MAP-CV and prognosis (S2 Fig). In terms of predicting hospital mortality, MPPV has a slightly advantage than MAPV (S3 Table).  In the subgroup analyses (Fig 3), higher MPP-CV is associated with higher risk of in-hospital mortality in the patients with a SOFA score � 8. In contrast, high MPP-CV did not increase the risk of in-hospital mortality in sepsis patients. The results drawn in MPP-VIM are consistent (S5 Fig).

MPP-CV change modes
According to the MPP-CV in the two periods of the first 24 hours with MPP data recorded (0-12 hours and 12-24hours), the patients were divided into four groups (Fig 4). Most patients belonged to the persistently low variability group (N = 2302). The decreasing group had the smallest number of patients (N = 791). Patients with persistently high variability had the highest hospital mortality (21.3%), and patients with persistently low variability had the lowest hospital mortality (15.7%).

Validation test
In the cohort of MIMIC-III database, 4,153 patients were enrolled with the same inclusion and exclusion criteria (Fig 5A). Although the overall MPP-CV value of MIMIC-III cohort is slightly higher in distribution, it can still be observed that hospital mortality increased when the MPP-CV increased (Fig 5B), and the highest decile of MPP-CV (CV > 21.3%) were related to an increase in the risk of hospital mortality compared with the fifth and sixth decile (adjusted OR in the tenth decile: 1.46, 95% Cl:1.05-2.03) (Fig 5C).

Main findings
The MPP was recently proposed to personalized management tissue perfusion pressure instead of MAP in critically ill patients. However, we knew little about the relationship between MPPV and mortality. In this multicenter, retrospective cohort study among critically ill patients with CVP monitoring, we aimed to clarify the clinically significant range of MPPV abnormalities for the first time. We found that the median MPP-CV was 13.2% in critically ill patients during the first 24 hours of CVP monitoring. And severely high MPP-CV that reached around 20% or more in two cohorts, occurring in about 10% of the study participants, was associated with the increased risk of in-hospital mortality.

Fig 3. Adjusted odds ratios and 95% CIs for hospital mortality associated with the increased MPP-CV in different
subgroups. Subgroup analyses of increased MPP-CV were conducted in patients who were male or female, elderly (age � 65 years) or not, with or without hypertension, sepsis, higher than median SOFA score or not on the first day of ICU admission, admission type (surgical or medical), cardiovascular surgery or not. The above associations were adjusted by age, gender, BMI, ethnicity, Charlson comorbidity index, SOFA score, admission type (elective surgery, emergency surgery or medicine), cardiovascular surgery, history of tachyarrhythmia, sepsis, incidence of AKI in the first day of ICU admission, the need for mechanical ventilation, the use of vasopressor, antihypertensive drug, sedatives and time-weighted average MPP. https://doi.org/10.1371/journal.pone.0287046.g003

Implications of study findings
In our study, two variability indicators confirmed the link between high variability and increased risk of hospital mortality. In addition, the same conclusion could also be drawn when analyzing daytime and nighttime MPP-CV separately. The mortality under different variability modes also confirmed the correlation between increased MPPV and prognosis. The exposure was mainly focused on MPP-CV, an indicator of the relative scatter of the values, which is easy to calculate and understand. The risk of short-term mortality increased when MPP-CV (SD/mean) reached around 20% or more. According to our results, if a patient has an average MPP of 60 mmHg and MPP-CV < 20%, then SD should be less than 12 mmHg. That is to say, 95% of the MPP readings should be within the range of 36.5 to 83.5 mmHg (mean ± 1.96 SD) based on a hypothesis of normal distribution of the MPP. Obviously, MPP-CV of over 20% represented a severe fluctuation in MPP, but it indeed occurred in about 10% of the study population in both cohorts. Therefore, it is physiologically acceptable that avoiding severe fluctuation in MPP (MPP-CV < 20%) may be a potential target for better hemodynamic management enhancing the outcomes of these patients.
One previous study has shown that intraoperative systolic BPV was associated with shortterm mortality in patients undergoing aortocoronary bypass surgery [22]. Our study mainly focused on the MPP variability in critically ill patients. Why is severe fluctuation in MPP related to adverse outcomes? In effect, MPP is determined approximately by the product of cardiac output and systemic vascular resistance [23]. Any condition of circulation that affects either of these two factors also affects the fluctuation of MPP. The change of blood volume, electrolytes, acute deterioration of cardiac function could affect the cardiac output, and the vasoactive agents, sedatives and pain could impact the SVR. Therefore, severe fluctuation of MPP indirectly represents a significant variation of management in blood volume, cardiac function, electrolytes control, use of vasoactive agents, sedatives and pain management, contributing to increased mortality risk. A global perspective pointed out that rapid changes in the infusion rate of vasoactive drugs or clinicians who desired to maintain higher blood pressure levels than expected without proper de-escalation are likely to cause serious adverse complications [24]. The relationship between positive fluid balance and the development or worsening of organ dysfunction as well as excess mortality has also been confirmed [25]. Undoubtedly, severe fluctuation of MPP represents a marker of illness severity to some extent, although the illness severity was accounted for in the multivariable models. Further randomized controlled trials are required to confirm the potential causal relationship between increased MPPV and mortality.
Interestingly, the patients with higher SOFA scores seem to be more susceptible to higher MPPV according to the subgroup analyses. Critically ill patients with SOFA � 8 may be presented with a higher rate of microcirculation dysfunction. In this case, the tissue oxygen extraction capacity is lost, and a more severe or prolonged duration of hypotension will aggravate tissue hypoxia. Therefore, MPP stability management should be strengthened when treating patients with more severe multiple organ dysfunction. Our subgroup analyses also showed that higher MPPV did not associate with in-hospital mortality in patients with sepsis. Our

PLOS ONE
result disagreed with a previous prospective study, which showed a correlation between early higher SBP complexity and increased risk of 28-day mortality in 51 patients with severe sepsis [26]. But their study only analyzed SBP variability on the first five-minute window. Patients with sepsis are often characterized by increased MPP and MPPV during fluid resuscitation but would not necessarily develop adverse outcomes. Various studies have shown that high shortterm to long-term BPV is associated with adverse outcome in patients with hypertension [27][28][29]. Our study further clarified the relationship between short-term blood pressure variability and poor prognosis in hypertensive population which may be related to arteriosclerosis and decreased ability to regulate blood pressure and made them more vulnerable to ischemiareperfusion injury.
Most of the variability in MPP can be explained by MAPV. However, the correlation between MAP-CV and prognosis showed no significant difference after multiple adjustments, suggesting that MAPV was of less robustness than MPPV to predict prognosis. Although there is still controversy about fluid resuscitation under the guidance of CVP [30][31][32], we argue that more focus should be paid to MPPV in critically ill patients with CVP monitoring, as MPP comprehensively reflects the overall perfusion [4].

Strengths and limitations
This is the first clinical investigation to explore the association between the MPPV and hospital mortality in critically ill patients. The advantage of this post hoc analysis was that both eICU-CRD and MIMIC-III databases contained comprehensive and high-quality data, which guaranteed the reliability of variability calculation. Moreover, the inclusion of the 24-hour measurement ensured that all patients were exposed to a complete diurnal cycle. Finally, we conducted sensitivity analyses and validation test to make the results robust.
Our study has some limitations. First, the post hoc analysis has its inherent defects and unavoidable bias. Second, our study population is limited to the patients with central venous pressure monitoring, who are more severely ill and cannot be extended to the whole population of critically ill patients. Third, it was hard to prove the causal relationship between MPPV and the primary endpoint as the study was observational, despite using two databases to confirm the association. The question of whether MPPV was a marker of severity of illness or a potential target to improve prognosis required randomized trials to answer. Fourth, our study did not account for advanced hemodynamic data such as cardiac index, peripheral vascular resistance and mechanical ventilation parameters like positive end-expiratory pressure.

Conclusion
Severe MPP fluctuation was associated with short-term mortality in critically ill patients with CVP monitoring. Therefore, it may need to be avoided in the management of critically ill patients.  Table. Calculation formula of variability parameters. Note: n is the number of MPP readings, �

Supporting information
x is the mean value and w refers to the time of each interval. For VIM, linear regression fitting log (SD) with log (x) was performed. The "k" was the exponential of β0 and the "b" was the β1 of the linear regression model.