Searching Optimal Target of Perfusion Pressure for Managing Acute Kidney Injury in Critically Ill Patients: An Analysis of Three Large Open Databases


 Background: The optimal perfusion pressure target for acute kidney injury (AKI) in critically ill patients remains uncertain. We investigated the association between mean perfusion pressure (MPP) and AKI among critically ill patients and estimated its optimal range.Methods: We analyzed data stored in the Medical Information Mart for Intensive Care (MIMIC) -III, eICU Collaborative Research Database (eICU-CRD), and MIMIC-IV databases. Critically ill patients receiving invasive measurements of MPP for at least 12 hours within the first 24 hours of ICU stay were included. The exposure of interest was the time-weighted average MPP (TWA-MPP) in the first 24 hours. The primary outcome was the incidence of AKI in the next 48 hours. Results: We enrolled 7,992, 8,604, and 6,730 patients from the MIMIC-III, eICU-CRD, and MIMIC-IV databases, respectively. TWA-MPP had higher areas under the curve than mean arterial pressure in predicting AKI in the next 48 hours (0.63 vs 0.57, 0.62 vs 0.58, and 0.64 vs 0.58 in three databases, all p < 0.001). We observed the lowest adjusted risk of AKI when TWA-MPP above 72, 65, and 69 mmHg in the MIMIC-III, eICU-CRD, and MIMIC-IV databases, respectively. Pooled analyses indicated that per 10% increase of proportion of MPP above 65 mmHg was associated with decreased incidence of AKI (adjusted odds ratio = 0.93, 95% confidence interval = 0.92–0.94, p < 0.001). Furthermore, pooled analyses showed that the lowest risk of new-onset, persistence, and progression of AKI was estimated when TWA-MPP above 74, 70 and 65 mmHg, respectively. Conclusions: MPP outperformed mean arterial pressure as a perfusion predictor of AKI. MPP of 65 mmHg or higher may be the optimal target for managing AKI in critically ill patients. The target rises to higher when reversing or preventing AKI.


Introduction
Acute kidney injury (AKI) is common in critically ill patients and is associated with poor short-term and long-term outcomes [1,2]. It manifests in clinical and biochemical derangements characterized by a signi cant reduction in glomerular ltration rate (GFR) [3]. Sepsis, nephrotoxins, and hypoperfusion could contribute to the onset and progression of AKI. The mainstay of management is early intervention to address the inciting cause, with supportive care to restore kidney perfusion and oxygenation [4].
The kidney can maintain constant renal blood ow and GFR over a wide range of arterial pressures (80-180 mm Hg) with changes in vascular tone of the afferent and efferent arterioles of the glomeruli [5]. This phenomenon named autoregulation is believed to be mainly mediated by tubuloglomerular feedback and the renal myogenic response [6]. However, the autoregulation could fail in critically ill patients whose kidney is susceptible to comprised blood pressure, sepsis, and drugs interfering with neurohormonal regulation [5]. Until now, the threshold of perfusion pressure to manage AKI is not clear in critically ill patients who usually have unstable hemodynamics.
Previous studies have focused on maintaining an adequate mean arterial pressure (MAP) to manage AKI [7][8][9][10][11][12]. However, MAP is arterial in ow pressure, which fails to take venous out ow pressure into account. Central venous pressure (CVP), an indicator of the out ow pressure, is closely related to AKI in critically ill patients [13,14]. Therefore, the mean perfusion pressure (MPP), obtained by the difference between MAP and CVP, was recently proposed to be used instead of MAP to personalize the management of tissue perfusion pressure [15,16]. Earlier studies indicated that lower MPP or higher MPP de cits are correlated with the occurrence or progression of AKI [17][18][19][20][21][22]. Still, these studies were limited by the sample size, patients with a speci c disease, and composition of different outcomes of AKI, which limited the generalization to all critically ill patients. The optimal range of MPP is still uncertain for the management of AKI, which contains preventing progression, persistence and new-onset of AKI.
Given the sparsity of evidence to support speci c perfusion pressure targets in critically ill patients, we sought to describe the relationship between perfusion pressure and incidence and evolution of AKI in critically ill patients. We hypothesized that (1) MPP performs better than MAP in predicting AKI; (2) a threshold MPP may identify an optimal MPP range for the lowest risk of AKI; (3) different outcomes may have some differences in the optimal target.

Study population
The inclusion criteria were (1) age 16 years or more and (2) at least 12 hours of continuous MAP and CVP invasive monitoring within the rst 24 hours in the rst ICU stay. Those who received dialysis, died during the rst 24 hours, complicated with chronic kidney disease stage 5, or with incomplete data or low-quality MPP data were excluded.

Data extraction
We extracted all variables through Structured Query Language, including time-weighted average MPP (TWA-MPP), demographic data, baseline ICU characteristics, Charlson comorbidity index [27], admission illness severity scores [including the Sequential Organ Failure Assessment (SOFA) [28] and Oxford Acute Severity of Illness Score (OASIS) [29]], and the incidence of AKI.

De nitions
The sepsis was de ned using the third sepsis de nition [30,31]. The AKI was de ned using serum creatinine and urine volume according to the AKI guidelines [3], in which the baseline creatinine was estimated using a creatinine determined by back-calculation using simpli ed Modi ed Diet in Renal Disease formula assuming an estimated GFR of 75 ml/min per 1.73 m 2 [32]. Exposure to nephrotoxins were de ned as use of vancomycin, non-steroidal anti-in ammatory drugs, amphotericin, aminoglycosides, angiotensin inhibitors or angiotensin receptor blockers and calcineurin inhibitors within 48 hours before and after ICU admission. Vasopressor dose was expressed as norepinephrine equivalents based on a recently published review of vasoactive-intropic score [33]. More detail on the study data and de nition of data was presented in additional le 1.

Exposure
Since MPP is a dynamic process, TWA-MPP during the rst 24 hours of ICU stay was calculated as the area under the MPP-versus-time plot as follows. TWA = [(t 2 -t 1 )(X 1 + X 2 )/2+(t 3 -t 2 )(X 2 + X 3 )/2+…+(t n -t n−1 )(X n−1 +X n )/2]/(t n -t 1 ) where X n is the value of the variable of interest at the timepoint t n .

Outcomes
The primary outcome was the incidence of AKI recognized between 24 and 72 hours following the ICU admission. Secondary outcomes included the severe AKI (de ned as stage 2 or 3) and the evolution of AKI, categorized as new-onset, persistence, and progression of AKI in the same time period.
New-onset of AKI was de ned as AKI identi ed between 24-72 hours of ICU stay in those patients without AKI recognized within the rst 24 hours. In those patients with AKI recognized within the rst 24 hours, the persistence of AKI was de ned as the absence of reversal of AKI which meant alive and as the absence of AKI in the next 48 hours [34]; the progression of AKI was de ned as dead or as the stage of AKI increase in the next 48 hours.

Statistical analysis
The categorical variables were presented as counts and percentages, including 95% con dence intervals (CI) where applicable, and continuous variables as medians and interquartile ranges (IQR). Patients were categorized into groups according to the analyzed cohort, including MIMIC-III, eICU-CRD and MIMIC-IV.
In order to determine the performance of TWA-MPP versus TWA-MAP in predicting the incidence of AKI during 24 to 72 hours of the ICU/hospital stay, we performed receiver operating characteristics areaunder-the-curve (AUC) analyses using the pROC package, with 95% CI calculated by DeLong method.
Then, the net reclassi cation index and the integrated discrimination improvement were calculated to ascertain that TWA-MPP or TWA-MAP improved the discriminatory ability when added to the base multivariate models. C-statistics and differences in the C-statistics for these multivariate models were also calculated using bootstrapping with 1000 replicates.
We built general additive models with a logit link function to plot associations between TWA-MPP and the AKI incidence. As evidence showed nonlinear relationships between TWA-MPP and outcomes, we chose the two-line piecewise linear models with a single change point to estimate change points and associations using adjusted odds ratio (AOR) and 95% con dence intervals (CI). Furthermore, we assessed the association between the AKI incidence and the proportion of MPP within the optimal interval. All the above models accounted for factors that may in uence outcomes, including age, gender, BMI, ethnicity, surgical admission, cardiovascular ICU, history of hypertension, history of diabetes, history of chronic kidney disease, history of chronic heart failure, cumulative vasopressor dose, sedatives, mechanical ventilation, transfusion of RBCs, AKI at the rst 24 hours, sepsis, exposure to ≥ 2 nephrotoxins, modi ed SOFA score (SOFA minus cardiovascular component) and OASIS score.
To assess the robustness of our ndings, we conducted a secondary analysis in the vasopressor-treated population, which is the focus in the management of blood pressure. Second, we conducted a secondary analysis using median MPP as the exposure instead of TWA-MPP. Third, baseline creatinine determined by the nadir creatine during 7 days before and after ICU admission was used to con rm the association between TWA-MPP and different AKI evolutions.
To evaluate the consistency of our ndings in different population that may have different optimal target, the association of TWA-MPP and the incidence of AKI were analyzed across patients who were elderly (age ≥ 65 years) with hypertension or not, with diabetes or not, with chronic kidney disease or not, with sepsis or not, with septic shock or not, with vasopressor usage or not, with higher than median SOFA score or not at the rst day of ICU admission.
All analyses were performed with R version 3.6.3 software, and a two-sided p < 0.05 was considered statistically signi cant.  Table 1. The majority of patients in the three cohorts were male, and most patients were admitted to the cardiovascular ICU with an initial diagnosis of cardiovascular disease.  Figure S1). A total of 73.0%, 52.3%, and 70.3% of the study population had recognized AKI in the 24 to 72 hours in three databases. Compared with those without AKI, the patients with AKI were more likely to be elderly, complicated more comorbidities (hypertension, diabetes, chronic kidney disease, chronic heart failure), having a higher rate of AKI in the rst day, and higher severity of illness (modi ed SOFA score and OASIS score), and a higher probability to receive support therapy (vasopressors, mechanical ventilation, and transfusion) in three databases. The patients with AKI also had signi cantly higher TWA-MPP and TWA-MAP levels as compared to those without AKI. More hospital information, MPP data and outcomes are presented in Additional le 2: Table S1.

Results
Performance of MPP vs MAP for predicting AKI As MAP was used to represented renal perfusion pressure previously, we compared the performance of MPP versus MAP for predicting the AKI incidence in the next 48 hours (Additional le 1: Figure S2). The results consistently showed that TWA-MPP had signi cantly larger areas under the curve (AUCs) than TWA-MAP for predicting the incidence of AKI (0.63 vs 0.57, 0.62 vs 0.58, 0.64 vs 0.58 in three databases, all p values < 0.001). TWA-MPP also performed better than TWA-MAP in predicting incidence of AKI using net reclassi cation index, integrated discrimination improvement, and C statistic (Additional le 1: Table  S2) in the fully adjusted models. These ndings supported that MPP was superior to MAP as a perfusion predictor of the AKI incidence.
Primary outcomes TWA-MPP in the rst 24 hours of ICU stay was associated with AKI in the next 48 hours in three databases ( Fig. 2A and 2C). We observed the lowest adjusted risk of AKI when TWA-MPP was greater than 72, 65, and 69 mmHg in MIMIC-III, eICU-CRD, and MIMIC-IV, respectively ( Table 2). Pooled analyses indicated that per 5 mmHg increase of TWA-MPP below the pooled change point (71 mmHg) was associated with decreased incidence of AKI (pooled AOR = 0.79, 95% CI = 0.76-0.82, p < 0.001). The table shows that the lowest probability of incidence of AKI and severe AKI, new-onset, persistence, and progression of AKI was observed when TWA-MPP below a change point. The lowest of these change points was 65 mmHg. Thus, our ndings support an optimal target of 65 mmHg or higher in MPP for the management of AKI. Adjusted factors included the top 19 factors listed in Table  1.
As eICU-CRD database represented multicenter and recent practice after publishing the 2012 Kidney Disease: Improving Global Outcomes AKI guideline, and 65 mmHg was the lowest change point of MPP for the incidence of AKI in the three database, we chose MPP of 65 mmHg as a conservative perfusion pressure target for the management of AKI. We furtherly found that the increase in the proportion of MPP > 65 mmHg was associated with less incidence of AKI in unadjusted and adjusted analyses in three databases ( Fig. 2B and 2D). Pooled analyses indicated that per 10% increase of proportion of MPP above 65 mmHg was associated with decreased incidence of AKI (pooled AOR = 0.93, 95% CI = 0.92-0.94, p < 0.001).

Sensitivity and subgroup analyses
First, the change points of MPP were in the range of > 65 mmHg in critically ill patients who received vasopressors during the rst 24 hours (Additional le 2: Figure S3, Additional le 2: Table S4). Second, the optimal target of MPP did not shift when median MPP was included into the analyses instead of TWA-MPP (Additional le 2: Figure S4, Additional le 2: Table S5). Third, the different de nitions of baseline creatinine determined by the nadir creatine during 7 days before and after ICU admission did not totally affect the association between TWA-MPP and the AKI incidence (Additional le 2: Figure S5, Additional le 2: Table S6).
The association of TWA-MPP and the AKI incidence was analyzed across patients who with or without elderly (≥ 65) and hypertension, diabetes, chronic kidney disease, sepsis, vasopressor-treated, higher above median SOFA score at the rst day of ICU admission (Fig. 4). The subgroup analyses consistently support the MPP of 65 mmHg or higher in the management of AKI.

Discussion
The ndings of this investigation can be summarized as follows: (1)  Early in the 1950s, Semple et al. demonstrated that an increase in MAP was required to maintain the amount of blood ow through the kidneys when the renal venous pressure increased [35]. Thus, MPP may be better than MAP at re ecting renal perfusion, as the former takes the venous into account, especially in those with venous congestion named 'renal congestion'. However, comparative studies of the predictive ability of MPP versus MAP for clinical outcomes are scarce. Our data consistently supported that MPP outperformed MAP as a predictor of AKI. Our study's C-statistic was not as high as that reported by Gul et al. [20]. They reported a C-statistic of 0.81 using MPP during transcatheter aortic valve implantation to predict AKI. The low predicting ability of MPP for AKI incidence may be caused by the study population's heterogeneity and many other causes of AKI. Our results supported the hypothesis that the MPP can replace the MAP as a clinical target in selected patients to personalize tissue perfusion pressure management.
Is there an optimal target of MPP for AKI in critically ill patients? Previous studies have focused on the MAP, which is a critical component of MPP. Heretofore, an MAP of at least 65 mmHg has been assumed to be within the zone of autoregulation for multiple organs and thus protects from organ failure in various types of shock except hemorrhagic shock [5,36,37]. In a multicenter randomized controlled trial on MAP targets in patients with septic shock, targeting an MAP of 80-85 mmHg did not result in signi cant differences in mortality compared with that with an MAP of 70-75 mmHg, but result in a lower rate of doubling of serum creatinine or receiving dialysis [7]. In addition, several recently published observational studies also favored an MAP of normal or near normal levels (75-85 mmHg) in septic shock to decrease the risk of AKI [8][9][10]12]. Our study was consistent with the above studies in that the lowest risk of AKI was observed when MPP was greater than 65 mmHg, which is usually equivalent to an MAP of > 75 mmHg when CVP is approximately 10 mmHg. As compared with previous studies, one strength of our research was that the optimal target using MPP as measurement should be more precise because of MPP considering venous return and performing better than MAP as a predictor of AKI. Another advantage was that we estimated the optimal target using the predicted incidence rather than ROC analyses or groups utilized in the past [8,10,12], which make the optimal target of MPP clearer. The third advantage was that, with a large sample size and heterogeneity, our study had better generalization to critically ill patients. More randomized trials are necessary to prove our estimated optimal target of MPP.
The different goals for managing AKI may require a different target of MPP. In our pooled analyses, the lowest adjusted probability of new-onset, persistence, and progression of AKI was estimated when TWA-MPP above 74, 70, 65 mmHg, respectively. The change points gradually decreased when the requirement of protection for the kidney was also reduced. We proposed that the MPP of 65 mmHg or higher may be an optimal target for AKI in critically ill patients, in which 65 mmHg is also the change point for the progression of AKI, higher than previously Ostermann et al. [19] reported (60 mmHg) probably due to the large sample size. Therefore, maintaining MPP well above 65 mmHg is prudent in critically ill patients. Based on these ndings, we proposed the concept frame of MPP management for AKI (Additional le 2: Figure S6), which need be con rmed in the future.
Personal hemodynamic management is proposed to achieve optimal/adequate blood ow using individual targets and adaptive multiparametric approaches [16]. Previous studies mainly focused on the MPP de cit [17,18,21,22], which is de ned as the difference between the basal MPP and the present MPP, indicating that an MPP goal should be achieved to decrease the MPP de cit. We proposed the optimal target of MPP was 65 mmHg or higher, which is near lower limits of normal range (about 70-95 mmHg). Our study ndings were concordant with and proved those of Panwar et al. [22], who recently reported that the vasopressor-treated patients with shock are often exposed to a signi cant degree and duration of relative hypotension (MPP de cits), which are associated with new-onset adverse kidneyrelated outcomes through a prospective, multicenter observational study. Recently a large database study showed that a strategy aimed at maintaining MAP above 65 mmHg appears to be as good as one based on the percentage reduction from baseline [38]. The question of whether MPP management should be based on an absolute MPP threshold or on a relative decline from the individual patient's baseline MPP should be investigated in the future.
Elderly patients or patients with hypertension, who usually have impaired autoregulation of blood ow to multiple organs, may bene t from a higher MPP. Our studies did not show higher change points for the incidence of AKI in elderly patients with hypertension in three databases, suggesting that the presence or absence of advanced age/hypertension may not signi cantly in uence the target MPP. This result was in line with recent research suggesting that elderly patients do not necessarily need higher blood pressures [39,40]. In contrast, our study only implied a higher threshold (84 mmHg) may be needed in those patients with chronic kidney disease to reduce the incidence of AKI, which suggests that blood ow in the kidney becomes pressure passive due to severe kidney injury [41].
The present analysis had some limitations. First, when considering the ndings, their post hoc nature should be taken into account. Residual confusion may also in uence our ndings, although we tried to explain this point through adjustments of the common risk factors. Second, the data were obtained from the U.S., and therefore, the results may not be fully applied to ICUs with different practices or resources.
Additionally, the current ndings cannot be interpreted regarding patients who died within 24 hours of admission to the ICU or received dialysis on the rst day of the ICU stay. While our ndings support the optimal range of MPP for the incidence and evolution of AKI, stronger evidence, such as large randomized controlled trials, is necessary to establish causality.

Conclusions
Our analyses suggest that MPP outperformed MAP as a perfusion predictor of AKI. The estimated optimal target of MPP during the rst 24 hours was 65 mmHg or higher for AKI in critically ill patients.
The target of MPP may rise when needed to prevent AKI or reverse AKI. The optimal range should be veri ed in future randomized trials.

Competing interests
The authors declare that they have no competing interests.
Funding Figure 1 Patient ow chart.         Subgroup analyses of the estimated optimal target for AKI. The threshold of mean perfusion pressure was expressed as change points (95% CI). Each subgroup had a change point of time-weighted average mean perfusion pressure above 65 mmHg.

Figure 4
Subgroup analyses of the estimated optimal target for AKI. The threshold of mean perfusion pressure was expressed as change points (95% CI