Association of Mean Perfusion Pressure Variability and Short-Term Mortality in Critically Ill Patients

Abnormal blood pressure variability (BPV) is associated with various organ injuries, high risk of cardio-cerebrovascular events and mortality. To investigate the association between mean perfusion pressure variability (MPPV) and mortality in critically ill patients admitted to the Intensive Care Unit (ICU), we analyzed data stored in the eICU-CRD database. MPPV was measured as the standard deviation (SD), coecient of variation (CV), average real variability (ARV) and variation independent of the mean (VIM) of the rst 24 hours MPP data within the rst 72 hours in the rst ICU stay. A total of 6049 patients were included. Apart from SD, survivors had signicantly lower MPPV in ARV, CV but higher VIM compared with non-survivors. After accounting for confounders, highest MPPV in decile using four measurements were associated with increased risk of hospital mortality compared with those in the fth and sixth decile. In addition, lowest MPPV with dimension (mmHg) in decile also correlated to an increase in the risk of hospital mortality compared with the fth and sixth decile. These relationships remained remarkable in the sensitive analyses. Increased MPPV and decreased MPPV with dimension were associated with short-term mortality in critically ill patients.


Introduction
During hospitalization in the intensive care unit (ICU), blood pressure (BP) is regularly monitored to observe organ perfusion in a safe, non-invasive or invasive manner. BP is a substitute index for blood perfusion of terminal organs such as brain, heart and kidney which are prone to ischemia 1 . And variability is a normal physiological property of BP, which may contribute to morbidity and mortality through unbalanced load 2,3 . For a long time, studies on long-term blood pressure variability (BPV) have con rmed that abnormal BPV is associated with various organ injuries, high risk of cardio-cerebrovascular events and mortality [4][5][6][7][8] .
Short term high BPV in critically ill patients is also found to be linked to mortality and organ injuries like acute kidney injury (AKI) 9 . In contrast, a recent study showed that the decrease (between − 5% and 5%) in mean arterial pressure (MAP) uctuation calculated by (nighttime MAP -daytime MAP) / 24-hour MAP may be related to adverse outcomes in critically ill patients 10 . However, the threshold at which variability is too high or too low to be clinically signi cant remains unclear. And previous studies about the impact of short-term BPV on critically ill patients are limited by sample and evidence is not strong enough.
Moreover, traditional index like MAP has some physiological de ciencies, especially the failure to consider venous out ow pressure. Obtained by the difference between MAP and central venous pressure (CVP), mean perfusion pressure (MPP) was recently proposed to personalized management tissue perfusion pressure instead of MAP 11,12 .
Therefore, we sought to describe the relationship between MPP variability (MPPV) and hospital mortality among critically ill patients. We hypothesized that an optimal range of MPPV can be determined to reduce hospital mortality in critically ill patients.

Study population
This study utilized data stored in the eICU Collaborative Research Database (eICU-CRD) v2.0 13 , which is a unique and publicly accessible multicenter database covering more than 200,000 ICU admissions 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 rst 72 hours in the rst ICU stay and (3) at least 20 MPP readings in the daytime and at least seven in the nighttime 15 . Daytime is de ned as 7 am to 11 pm, otherwise as nighttime. Those who received dialysis, died during the rst 24 hours, complicated with chronic kidney disease stage 5, and with incomplete data or extreme MPP data were excluded.

Data extraction
We extracted MPP data, demographic data, baseline ICU characteristics, Charlson comorbidity index 16 , and admission illness severity scores [including the Sequential Organ Failure Assessment (SOFA) 17 and Oxford Acute Severity of Illness Score (OASIS) 18 ]. Criteria for sepsis were de ned based on those described earlier by Angus et al 19 instead of sepsis 3.0 because most microbiology data was unavailable in eICU-CRD. Additionally, the need for mechanical ventilation, incidence of AKI, use of vasopressor, antihypertensive drugs and sedatives were also collected.

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. Furthermore, we deleted the outliers which were de ned as larger than the average MPP values plus four standard deviation (SD) in each patient. Considering the in uence of the extreme value on the result, we removed a total of 1% of patients with extreme high (> 99.5th percentile) and low value (< 0.05th percentile) of the coe cient of variation (CV) of MPP in further analysis.

Exposure
Short-term MPPV was measured as the SD, CV, average real variability (ARV) and variation independent of the mean (VIM) 20 of 24-hour MPP data. The SD and ARV were thought as variability with dimension (mmHg). And CV and VIM were thought as variability without dimension. Detailed formulas are displayed in supplementary Table 1.

Outcomes
The primary outcome was in-hospital mortality.

Statistical analysis
Statistical analyses were performed using R version 3.63 (R Foundation for Statistical Computing, Vienna, Austria; www.r-project.org). 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.
MPPV parameters of the rst complete 24 hours of ICU stay were taken as a continuous variable for the primary analysis. Firstly, general additive models with a logit link function were built to plot associations between MPPV and in-hospital mortality, adjusted by age, gender, BMI, ethnicity, Charlson comorbidity index, SOFA score, OASIS, history of tachyarrhythmia, sepsis, incidence of AKI in the rst day of ICU admission, the need for mechanical ventilation, the use of vasopressor, antihypertensive drug and sedatives. Second, we used multivariable logistic regression models to assess the relationship between the outcome and deciles of each parameter adjusted by the same confounders mentioned before.
There were missing values for body mass index (BMI) (3.2 %) and multiple imputation was used to handle the missing with the mice package in R. In order to verify the robustness of the results, we explored the association of MPPV and ICU mortality. And the association of daytime and nighttime MPPV and hospital mortality were also analyzed. Furthermore, subgroup analyses were conducted in patients who were male or female, elderly (age ≥ 65 years) or not, with or without hypertension, sepsis, median SOFA score on the rst day of ICU admission. For all analyses, a two-tailed P value less than 0.05 was considered statistically signi cant.

Ethics approval and consent to participate
The study was conducted entirely on a publicly available, third-party anonymous public database which was released under the Health Insurance Portability and Accountability Act (HIPAA) safe harbor provision. The re-identi cation risk was certi ed as meeting safe harbor standards by Privacert (Cambridge, MA) (HIPAA Certi cation no. 1031219-2). The ethics committee of The First A liated Hospital of Nanjing Medical University 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 rst ICU stays in eICU-CRD, we nally included 6,049 ful lling the inclusion and exclusion criteria (Fig.1). The baseline characteristics between survivors and non-survivors are shown in Table 1. Though there was no signi cant difference in the use of sedatives and the history of tachyarrhythmia, the survivors were, on average, younger, predominantly male and lower in BMI. Nonsurvivors were signi cantly complicated with more comorbidities, more severe in diseases (higher SOFA score and OASIS), needing more support (mechanical ventilation and vasopressors), less use of antihypertensive drug, higher incidence of AKI and sepsis and lower MPP compared with survivors. More non-survivors had tachyarrhythmia history and less use of sedatives, but the difference did not reach statistical signi cance. Other information about hospitals, initial diagnosis, comorbidities and MPP data of the whole cohort were listed in supplementary Table 2.
The median of the four MPPV parameters were 7.8 mmHg (SD), 12 Fig.1). The correlation coe cient for CV and VIM was 0.98, which was the strongest (supplementary Fig.2).
There was also a strong correlation between MPPV and other BPV like MAP, systolic BP (SBP) and diastolic BP (supplementary Fig.3), among which the correlation coe cients of ARV were higher.
Association with hospital mortality After adjusting for age, gender, BMI, ethnicity, Charlson score, SOFA score, OASIS, history of tachyarrhythmia, sepsis, incidence of AKI in the rst day of ICU admission, the need for mechanical ventilation, the use of vasopressor, antihypertensive drug and sedatives, we found a 'U' shaped curve between variability with dimension indicators (SD and ARV) and hospital mortality using general additive models ( Fig. 2A, 2C). However, hospital mortality simply increased with the dimensionless variability (CV and VIM) increasing (Fig. 2B, 2D).
After grouping in deciles (Fig. 3), multiple logistic regression revealed that both higher and lower MPPV with dimension were related to an increase in the risk of hospital mortality compared with the fth and sixth decile (SD: adjusted odds ratio [OR] (Fig. 3B, 3D) were associated with increased risk of hospital mortality compared with the fth and sixth decile (CV: adjusted OR in the tenth decile: 1.50, 95% Cl: 1.16-1.93; VIM: adjusted OR in the tenth decile: 1.51, 95% Cl: 1.17-1.96). These results were consistent with the changing trend of general additive models.

Sensitivity and subgroup analyses
For the sensitivity analyses, we analyzed the association between MPPV and ICU mortality. We observed similar trends in each variability index (supplementary Fig.5). Multiple logistic regression also con rmed our ndings (supplementary Fig.6 Since there was time interval in the calculation of daytime or nighttime ARV, we chose SD and CV representatively to analyze the association between day and night MPPV and hospital mortality. The results still showed good consistency (supplementary Fig.7, supplementary Fig.8).
The association of high MPPV and in-hospital mortality was analyzed across patients who were male or female, elderly (age ≥ 65 years) or not, with or without hypertension, sepsis, median SOFA score on the rst day of ICU admission (Fig. 4). In patients who had SOFA score of ≥ eight or had hypertension history, higher variability is associated with higher risk of death in hospital. Surprisingly, sepsis patients with high variability did not increase hospital death risk.

Discussion
In this multicenter, retrospective, cohort study among critically ill patients, we clari ed the clinically signi cant range of MPP abnormalities for the rst time, and found that (1) increased MPPV (SD > 10.4 mmHg, CV > 19%, ARV > 5.6 mmHg, VIM > 0.62 units) was associated with the risk of hospital mortality which was not in uenced by confounding factors including absolute blood pressure levels.
(2) decreased MPPV with dimension instead of those without dimension (SD < 5.9 mmHg, ARV < 2.2 mmHg) was also associated with the risk of hospital mortality.
Increased BPV is related to the prognosis of critically ill patients. A prospective, observational study conducted by Xie et al. reported a signi cant relationship between systolic BPV and the occurrence of AKI as well as a weak link with hospital mortality after adjusting potential confounding factors 9 . Two studies about intraoperative BPV also con rmed that a higher BPV is linked with postoperative AKI and postoperative mortality in noncardiac surgery 20,21 . A post hoc analysis of the HeadPoST study has reported that increased BPV was associated with poor stroke outcome 22 . In endovascular therapy treated acute stroke patients, higher BPV in the rst 24 hours was associated with poor 90-day neurological outcome 23 . In our study, all variability indicators con rmed the link between high variability and increased risk of hospital mortality as well as ICU mortality. In addition, same conclusion could also be drawn when we analyzed daytime and nighttime MPPV separately.
Why is increased MPPV related to prognosis? As we all know, human body is always in dynamic balance to regulate a variety of external stimuli to maintain homeostasis. Critically ill patients are known for high incidence of anxiety 24 , delirium 25 , sleep loss 26 , abnormal central and autonomic nervous regulation 27 , which are all related to increased morbidity and mortality [27][28][29] , and the patients in ICU are constantly affected by the environment day and night. Among previous randomized controlled trial studies in noncritically ill patients, it has been con rmed that generalized anxiety disorder was associated with a signi cant increase in systolic BPV, which may increase the variability by increasing sympathetic dominance 30 . The addition of an anxiolytic to the pharmacotherapy regimens could reduce and stabilize the circadian rhythm of blood pressure 31 . Moreover, partial sleep deprivation 32 , fragmented sleep 33 and cold weather 34 could also lead to an increased BPV. In our study, we chose MPP as the target of variability study to explain the effect of organ perfusion variation on prognosis more pertinently. In addition to the above possible speculation, we supposed that this may be due to the ischemia-reperfusion injury. Prolonged ischemia can cause cell dysfunction and death 35 . Increased BPV can be observed in the animal model of ischemia-reperfusion and lead to various organ injuries 36,37 . Whether improving MPPV can affect the prognosis of critically ill patients still needs further study.
Decrease MPPV can also associated with an increased risk of death in hospital. This kind of patients may lose the function of physiological regulation and circadian rhythm of blood pressure. The blood pressure of normal people shows a dipper pattern. Abnormal diurnal variation of blood pressure was a risk factor for cardiovascular disease 38 . Previous studies have investigated that reduced day/night uctuations were associated with ICU and hospital mortality 10 and nocturnal mean arterial pressure rising may be an important risk factor for high short-term and long-term mortality in critically ill patients 39 . In our study, decreased MPPV with dimension was linked to hospital mortality and the risks were even higher as had been expected in ARV, an indicator closely related to time. However, this connection was affected by absolute blood pressure levels, which made the dimensionless MPPV without statistical signi cance. Besides, for the time series nature of ARV, it showed better sensitivity to identify MPP which changed little over time and to predict poor prognosis. Early studies have proposed that ARV was a more reliable representation of time series variability and added more prognostic value than SD 40,41 . Our subgroup analyses for SD and ARV also displayed a better stability of ARV.
Our subgroup analyses showed that higher MPPV did not associate with in-hospital mortality in patients with sepsis. Though a prospective study with a small sample size observed a possible association between early SBP complexity and 28-day mortality in patients with severe sepsis 42 , they only analyzed SBP variability on the rst ve-minute window. Patients with sepsis are often characterized with increased MPPV during uid resuscitation, but would not develop adverse outcomes. Given patients with hypertension who seem to be more susceptible to higher MPPV, more attention may be paid to the ICU MPP stability management of patients with hypertension in the future. This is the rst clinical investigation to explore the association between the variability of MPP and hospital mortality in critically ill patients. The advantage of this post hoc analysis was that the eICU-CRD database contained comprehensive and high-quality data and had an average ve-minute of measurement interval of MAP and CVP 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 combined four variability indicators, and conducted sensitivity and subgroup analyses to make the results more robust.
Our study has some limitations. First, although the data stored in eICU-CRD database was multicenter, effective external veri cation was not been carried out due to the lack of su cient monitoring frequency in other databases. Second, due to the heterogeneity of monitoring methods and frequency, it is hard to determine the reference values of the MPPV results of each parameter. Third, the post hoc analysis has its inherent defects and unavoidable bias.

Conclusion
Increased MPPV and decreased MPPV with dimension were associated with short-term mortality in critically ill patients. Authors' contributions YDP and BYW designed the study, conducted the data collection, data analysis, data interpretation, and wrote the manuscript. CYX designed the study and reviewed the manuscript. HJM designed the study, conducted the data interpretation, and reviewed the manuscript. All authors reviewed and approved the version submitted for publication.   The associations between adjusted in-hospital mortality risk and MPPV tted by general additive models and the histograms of MPPV. The gure shows that in all MPPV parameters, in-hospital mortality increases with variability. In two absolute MPPV parameters, SD and ARV, decreased variability also associated with high in-hospital mortality. The above associations were adjusted by age, gender, BMI, ethnicity, Charlson comorbidity index, SOFA score, OASIS score, history of tachyarrhythmia, sepsis, incidence of AKI in the rst day of ICU admission, the need for mechanical ventilation, the use of vasopressor, antihypertensive drug and sedatives.

Figure 3
The deciles of MPP variability and adjusted odds ratio of in-hospital mortality using the fth and sixth deciles as a reference. Multiple logistic regression reveals that for absolute variability indicators, both increased and decreased MPPV would lead to an increase in the risk of hospital mortality compared with the fth and sixth decile. But in relative variability indicators, only higher MPPV were associated with increased risk of hospital mortality compared with the fth and sixth decile, which were consistent with the change trend of general additive models.