Individualized Dynamic Assessment of Mean Arterial Pressure in Hospitalized Older Patients with Acute Kidney Injury: A Retrospective Cohort Study Authors: Qiangguo Ao, Yabin Zhang, Yang Liu, Rui Zhu, Zhen Wu, Jiayu Guo, Jie Zhang, Ying Ding, Qingli Cheng Published: 2026 ABSTRACT PurposeAcute kidney injury (AKI) carries a high burden in hospitalized older adults. While mean arterial pressure (MAP) is a key modifiable factor, the prognostic value of its dynamic, individualized assessment beyond static thresholds remains poorly defined in this population.Patients and MethodsIn this retrospective cohort of 1248 hospitalized patients aged ≥65 years with AKI, we proposed and validated a three-dimensional hemodynamic assessment framework, evaluating: (1) the absolute MAP value at AKI onset (MAPonset); (2) the relative change from patient-specific baseline (MAP change rate); and (3) the time of recovery to baseline. We defined distinct hemodynamic phenotypes and assessed their association with major adverse kidney events within 90 days (MAKE-90) using Cox regression. The incremental predictive value was evaluated with random forest models.ResultsBoth MAPonset ≤65 mmHg (HR 1.33, 95% CI 1.05–1.69) and a decline in MAP of ≥10% from baseline (HR 1.26, 95% CI 1.03–1.54) independently predicted MAKE-90. Crucially, failure to recover within 72 hours was associated with a higher risk (HR 1.71, 95% CI 1.37–2.15). The integrated phenotype of severe initial hypotension with delayed recovery identified patients with the poorest prognosis (HR 1.72, 95% CI 1.36–2.17). Incorporating these dynamic parameters significantly improved the prediction of MAKE-90 (AUC increased from 0.71 to 0.76).ConclusionThis study suggests that a dynamic, individualized MAP assessment might refine risk stratification beyond static thresholds in hospitalized older adults with AKI. Internal evaluation of the integrated dynamic phenotype showed improved predictive performance for adverse outcomes. However, given the retrospective design and substantial potential for confounding, these findings are exploratory. They require rigorous external validation in prospective cohorts before any consideration for clinical application. Introduction Acute kidney injury (AKI) is a common and serious complication among hospitalized older patients. The incidence increases with age, at 11.55% for patients aged 65–79 years and 15.64% for those over 80 years.1,2 The older population is more susceptible to AKI due to multiple comorbidities, polypharmacy, and diminished renal reserve.1 Older patients with AKI have poorer outcomes, including increased mortality, prolonged hospital stays, and a markedly elevated risk of progression to chronic kidney disease.2–4 Therefore, it is important to figure out modifiable risk factors influencing the progression of AKI to improve the prognosis of this vulnerable cohort. Among the complex mechanisms involved in the onset and progression of AKI, arterial blood pressure is one of the few central and directly modifiable factors in clinical practice.5 Mean arterial pressure (MAP) represents the average arterial pressure over a cardiac cycle, and it is the major determinant of renal perfusion.6,7 Given the increased sensitivity to perfusion change in aging kidneys, the fluctuations in MAP are of great clinical significance for older patients. Specific MAP thresholds are commonly used for risk stratification and management in perioperative and critically ill patients. The MAP below 65 mmHg was found to be the most commonly cited threshold consistently associated with postoperative AKI.8 Investigations have also explored whether targeting higher MAP levels (≥80 mmHg) might confer additional renal protection in certain populations.9 However, the applicability of these largely static thresholds to the hospitalized older adults with AKI remains uncertain. This uncertainty stems from two fundamental limitations. First, the majority of studies used single-point measurements, failing to capture the blood pressure variability of the older population.10,11 Second, they ignored individual differences in baseline blood pressure.6,12,13 Since most older patients often present with chronic hypertension or age-related hemodynamic alterations, MAP thresholds for this population may be significantly different.14,15 Therefore, an individualized assessment to use relative changes from baseline is likely superior to absolute cut-off values. This is supported by evidence in surgical settings. A study in older adults undergoing transcatheter aortic valve replacement found that a >20% drop in MAP from baseline was a predictor of postoperative acute kidney injury.16 Similarly, perioperative studies showed that maintaining systolic blood pressure above 80% or 90% of the preoperative baseline is an effective strategy for preventing AKI.17,18 However, critical knowledge gaps persist when translating this individualized approach to non-surgical older inpatients with AKI. Moreover, the prognostic significance of the recovery dynamics of MAP following AKI onset has been entirely overlooked. To address these gaps, we conducted this study to introduce and validate a novel, three-dimensional framework for MAP assessment in older inpatients with AKI. We hypothesized that a composite evaluation encompassing (1) the absolute insult severity (MAP at onset), (2) the relative insult magnitude (MAP change from baseline), and (3) the recovery kinetics (time for MAP to return to baseline) would provide superior prognostic insight compared to conventional static metrics. By defining distinct hemodynamic phenotypes through this integrated lens, we aimed to evaluate whether this dynamic MAP assessment can improve risk stratification and identify high-risk patients who may warrant closer monitoring and timely supportive care. Methods Study Design and Population This retrospective study included patients aged 65 years or older diagnosed with AKI according to the 2012 Kidney Disease: Improving Global Outcomes (KDIGO) guidelines19 at the Geriatric Medicine Research Center of Chinese PLA General Hospital in China between January 2008 and December 2018. Exclusion criteria were: (1) death within 24 hours of admission; (2) baseline eGFR <15 mL/min/1.73 m2 or a history of recent renal replacement therapy or AKI episode (within 3 months before admission); (3) incomplete key clinical data or missing blood pressure measurements; (4) loss to follow-up (Figure 1). Complete data for MAPonset and MAP change rate were available for all 1248 patients. A total of 1,157 patients had complete 72-hour serial measurements for recovery assessment. The study protocol was reviewed and approved by the Ethics Committee of the Chinese PLA General Hospital (Approval No. 2024-477-03), with a waiver of informed consent due to the retrospective nature of the study. The study was conducted in accordance with the Declaration of Helsinki. Figure 1Patient enrollment and grouping flowchart for this study.Diagram showing how an acute kidney injury (AKI) inpatient cohort was narrowed and then stratified by mean arterial pressure (MAP) metrics. Starting set: 2,184 AKI inpatients (Jan 2008 to Dec 2018). Those aged greater than or equal to 65 years: 1,628. Exclusions listed in a side box: death within 24 hours (65); baseline eGFR less than 15 mL/min/1.73 m2 or recent renal replacement therapy/AKI episode (79); incomplete key clinical data or missing blood pressure data (230); loss to follow-up (6). Final analytic sample: 1,248 participants. From the final sample, three parallel classification branches are shown: 1) MAP at AKI onset, which was categorized according to both absolute value and change rate from baseline: For absolute values: less than or equal to 65 mmHg; greater than 65 to less than or equal to 80 mmHg; greater than 80 mmHg. For change-rate bands: less than equal to negative 20 percent, greater than negative 20 percent to less than or equal to negative 10 percent, greater than negative 10 percent to less than or equal to 10 percent, greater than 10 percent to less than or equal to 20 percent, and greater than 20 percent. 2) MAP recovery pattern: no decline at onset; recovery within 24 h; within 48 h; within 72 h; or no recovery by 72 h. 3) Combined onset-and-recovery categories: no decline at onset; mild decline with recovery; mild decline without recovery; severe decline with recovery; severe decline without recovery.AKI cohort selection and MAP-based subgroup definitions with counts and exclusion reasons.Abbreviations: AKI, acute kidney injury; MAP, mean arterial pressure; eGFR, estimated glomerular filtration rate; MAPonset: MAP at the onset of AKI. Patient enrollment and grouping flowchart for this study. AKI Stage and Baseline Serum Creatinine Definition AKI was staged according to the 2012 KDIGO guidelines.19 Baseline serum creatinine (Scr) was defined as the value within 7 days before AKI; if unavailable, the most recent stable Scr (7–365 days prior) was used, with the average taken for multiple readings. eGFR was calculated using the CKD-EPI 2009 formula.20 Blood Pressure Measurement and Definitions All blood pressure data were obtained non-invasively using calibrated oscillometric monitors (Omron HEM-7121, manufactured by Omron Dalian Co., Ltd., China). Measurements were performed by trained nursing staff following a standardized protocol: after the patient had rested quietly for at least 5 minutes, with an appropriately sized cuff placed on the upper arm at heart level. To align with standard care practices in different clinical settings, the measurement position was as follows: baseline (outpatient) blood pressure was measured in the seated position, while all measurements during hospitalization were performed in the supine position, according to routine inpatient nursing standards at our institution. MAP was calculated as (1/3×Systolic blood pressure) + (2/3× diastolic blood pressure). Core hemodynamic variables were defined as follows (detailed in Table 1):Table 1Incidence of MAKE-90 According to MAP GroupsCharacteristicSample sizeDefinitionNon-MAKE90MAKE90P Value1248 (1157)/617631MAPonset (%)<0.001≤65mmHg147MAPonset≤65mmHg56 (38.10)91 (61.90)65–80mmHg group4756580 mmHg group626MAPonset>80 mmHg352 (56.23)274 (43.78)MAP change rate (%)0.005≤-20%267Change rate≤-20%111 (41.57)156 (58.43)−20% to −10%306−20%20%64Change rate>20%35 (54.69)29 (45.31)MAP recovery pattern (%)0.005No-decline615MAPonset > 90% of MAPbaseline330 (53.66)285 (46.34)Recovery within 24 h237(MAPonset≤90% of MAPbaseline) and (MAP at 24h> 90% of MAPbaseline)116 (48.95)121 (51.05)Recovery within 48 h74(MAPonset≤90% of MAPbaseline) and (MAP at 48h> 90% of MAPbaseline)31 (41.89)43 (58.11)Recovery within 72 h63(MAPonset≤90% of MAPbaseline) and (MAP at 72h> 90% of MAPbaseline)29 (46.03)34 (53.97)No-recovery by 72 h168(MAPonset≤90% of MAPbaseline) and (MAP at 72h≤90% of MAPbaseline)64 (38.10)104 (61.90)Integrated phenotype (%)No-decline615MAPonset > 90% of MAPbaseline330 (53.66)285 (46.34)0.007Mild decline with recovery148(−20% < MAP change rate≤ −10%) and (MAP at 72h > 90% of MAPbaseline)73 (49.32)75 (50.68)Mild decline without recovery139(−20% < MAP change rate≤ −10%) and (MAP at 72h ≤ 90% of MAPbaseline)58 (41.73)81 (58.27)Severe decline with recovery98(MAP change rate ≤ −20%) and (MAP at 72h > 90% of MAPbaseline)47 (47.96)51 (52.04)Severe decline without recovery157(MAP change rate ≤ −20%) and (MAP at 72h ≤ 90% of MAPbaseline)62 (39.49)95 (60.51)Notes: The analysis included 1248 patients for MAPonset and the MAP change rate. For MAP recovery pattern and integrated phenotype, a total of 1157 patients’ data were used for analysis.Abbreviations: AKI, acute kidney injury; MAKE-90, major adverse kidney events at 90 days; MAP, mean arterial pressure; MAPbaseline, baseline mean arterial pressure; MAPonset, mean arterial pressure at the onset of AKI. Table 1Incidence of MAKE-90 According to MAP GroupsCharacteristicSample sizeDefinitionNon-MAKE90MAKE90P Value1248 (1157)/617631MAPonset (%)<0.001≤65mmHg147MAPonset≤65mmHg56 (38.10)91 (61.90)65–80mmHg group4756580 mmHg group626MAPonset>80 mmHg352 (56.23)274 (43.78)MAP change rate (%)0.005≤-20%267Change rate≤-20%111 (41.57)156 (58.43)−20% to −10%306−20%20%64Change rate>20%35 (54.69)29 (45.31)MAP recovery pattern (%)0.005No-decline615MAPonset > 90% of MAPbaseline330 (53.66)285 (46.34)Recovery within 24 h237(MAPonset≤90% of MAPbaseline) and (MAP at 24h> 90% of MAPbaseline)116 (48.95)121 (51.05)Recovery within 48 h74(MAPonset≤90% of MAPbaseline) and (MAP at 48h> 90% of MAPbaseline)31 (41.89)43 (58.11)Recovery within 72 h63(MAPonset≤90% of MAPbaseline) and (MAP at 72h> 90% of MAPbaseline)29 (46.03)34 (53.97)No-recovery by 72 h168(MAPonset≤90% of MAPbaseline) and (MAP at 72h≤90% of MAPbaseline)64 (38.10)104 (61.90)Integrated phenotype (%)No-decline615MAPonset > 90% of MAPbaseline330 (53.66)285 (46.34)0.007Mild decline with recovery148(−20% < MAP change rate≤ −10%) and (MAP at 72h > 90% of MAPbaseline)73 (49.32)75 (50.68)Mild decline without recovery139(−20% < MAP change rate≤ −10%) and (MAP at 72h ≤ 90% of MAPbaseline)58 (41.73)81 (58.27)Severe decline with recovery98(MAP change rate ≤ −20%) and (MAP at 72h > 90% of MAPbaseline)47 (47.96)51 (52.04)Severe decline without recovery157(MAP change rate ≤ −20%) and (MAP at 72h ≤ 90% of MAPbaseline)62 (39.49)95 (60.51)Notes: The analysis included 1248 patients for MAPonset and the MAP change rate. For MAP recovery pattern and integrated phenotype, a total of 1157 patients’ data were used for analysis.Abbreviations: AKI, acute kidney injury; MAKE-90, major adverse kidney events at 90 days; MAP, mean arterial pressure; MAPbaseline, baseline mean arterial pressure; MAPonset, mean arterial pressure at the onset of AKI. Incidence of MAKE-90 According to MAP Groups Notes: The analysis included 1248 patients for MAPonset and the MAP change rate. For MAP recovery pattern and integrated phenotype, a total of 1157 patients’ data were used for analysis. Abbreviations: AKI, acute kidney injury; MAKE-90, major adverse kidney events at 90 days; MAP, mean arterial pressure; MAPbaseline, baseline mean arterial pressure; MAPonset, mean arterial pressure at the onset of AKI. (1)Baseline MAP (MAPbaseline): Baseline MAP was defined as the mean of all available outpatient seated blood pressure measurements recorded within the 12 months before admission. All patients had at least one such record during this period.(2)MAP at AKI onset (MAPonset): The first MAP value recorded on the calendar day of AKI diagnosis. It was categorized into three groups: ≤65, 65–80, and >80 mmHg, thresholds informed by prior literature.8,9(3)MAP change rate: Calculated as (MAPonset-MAPbaseline)/MAPbaseline×100%. It was grouped into five groups: ≤−20%, −20% to −10%, −10% to 10%, 10% to 20%, and >20%.(4)MAP recovery pattern: MAP was tracked daily for 72 hours post-AKI (daily value = average of morning fasting and pre-dinner readings). For patients with an initial insult (MAP change rate ≤ −10%), recovery was defined as the first day within 72 hours when their MAP returned to within ±10% of MAPbaseline. Patients were thus classified into five groups: “no decline”, “recovery within 24h”, “recovery within 48h”, “recovery within 72h”, and “no recovery by 72h”.(5)Integrated phenotype: To identify the patient subgroup with the poorest prognosis, we performed a secondary analysis by creating a composite hemodynamic phenotype. Patients were first classified according to the severity of the initial insult based on the MAP change rate: 1) no decline (MAP change rate> −10%); 2) mild decline (−20% 80 mmHg, thresholds informed by prior literature.8,9 MAP at AKI onset (MAPonset): The first MAP value recorded on the calendar day of AKI diagnosis. It was categorized into three groups: ≤65, 65–80, and >80 mmHg, thresholds informed by prior literature.8,9 (3)MAP change rate: Calculated as (MAPonset-MAPbaseline)/MAPbaseline×100%. It was grouped into five groups: ≤−20%, −20% to −10%, −10% to 10%, 10% to 20%, and >20%. MAP change rate: Calculated as (MAPonset-MAPbaseline)/MAPbaseline×100%. It was grouped into five groups: ≤−20%, −20% to −10%, −10% to 10%, 10% to 20%, and >20%. (4)MAP recovery pattern: MAP was tracked daily for 72 hours post-AKI (daily value = average of morning fasting and pre-dinner readings). For patients with an initial insult (MAP change rate ≤ −10%), recovery was defined as the first day within 72 hours when their MAP returned to within ±10% of MAPbaseline. Patients were thus classified into five groups: “no decline”, “recovery within 24h”, “recovery within 48h”, “recovery within 72h”, and “no recovery by 72h”. MAP recovery pattern: MAP was tracked daily for 72 hours post-AKI (daily value = average of morning fasting and pre-dinner readings). For patients with an initial insult (MAP change rate ≤ −10%), recovery was defined as the first day within 72 hours when their MAP returned to within ±10% of MAPbaseline. Patients were thus classified into five groups: “no decline”, “recovery within 24h”, “recovery within 48h”, “recovery within 72h”, and “no recovery by 72h”. (5)Integrated phenotype: To identify the patient subgroup with the poorest prognosis, we performed a secondary analysis by creating a composite hemodynamic phenotype. Patients were first classified according to the severity of the initial insult based on the MAP change rate: 1) no decline (MAP change rate> −10%); 2) mild decline (−20% −10%); 2) mild decline (−20% 25% from baseline).21 Secondary endpoints were 90-day all-cause mortality and persistent renal dysfunction. Statistical Analysis Data analysis was performed using R software (version 4.5.1). Continuous variables are presented as mean ± standard deviation (SD) or median (interquartile range, IQR) based on their distribution. They were compared using an independent-samples t-test (for normally distributed variables) or the Mann–Whitney U/Kruskal–Wallis tests (for non-normally distributed variables). Categorical variables are presented as number (percentage) and compared using the chi-square or Fisher’s exact test, as appropriate. The associations of the three hemodynamic dimensions (MAPonset, MAP change rate, MAP recovery pattern) with the primary outcome were assessed both individually and in combination (via an integrated phenotype). Survival and Regression Analyses: Kaplan-Meier curves with Log rank tests were used. Associations with MAKE-90 were assessed using multivariable Cox proportional hazards models in a three-step approach: Model 1 containing only the parameter of MAP; Model 2 adjusted Model 1 for proteinuria, age, body mass index, smoking status, hypertension, diabetes, hyperlipidemia, hyperuricemia, and chronic kidney disease; Model 3 further adjusted for the following covariates based on Model 2: AKI stage, coronary heart disease, heart failure, history of stroke, chronic obstructive pulmonary disease, use of mechanical ventilation, malignancy, baseline levels of hemoglobin, platelet count, C-reactive protein, albumin, urea, and hepatic injury. The fully adjusted model included 22 covariates with 586 events, yielding an events-per-variable ratio of approximately 26.6, well above the conventional threshold of 10–20, supporting the stability of the estimates. For MAKE-90, time-to-event was defined as the time from AKI onset to the first occurrence of any component: all-cause death or initiation of renal replacement therapy (RRT) was recorded at its actual date, whereas persistent renal dysfunction was evaluated at day 90 and, when present, was assigned an event time of 90 days; all patients had complete 90-day follow-up. The proportional hazards assumption was tested; time-varying Cox models were used if violated. Multicollinearity was assessed in multivariable models to minimize the risk of collinearity among predictor variables. Restricted cubic splines with four knots placed at the 5th, 35th, 65th, and 95th percentiles of their respective distributions were used to explore non-linear associations between MAPonset (as a continuous variable) and MAKE-90. Machine learning models: To complement Cox regression and assess incremental predictive value, random forest models were constructed. A baseline model was trained, followed by models sequentially incorporating MAPonset, MAP change rate, or MAP recovery patterns. The dataset was randomly split into a training set (70%) and a testing set (30%). Feature selection was performed using the least absolute shrinkage and selection operator LASSO regression within the training set. The test set was held out entirely and used solely for independent validation, with no involvement in feature screening, model fitting, or hyperparameter optimization. The hyperparameters of the random forest were optimized using random grid search with 5-fold cross-validation within the training set, yielding final values of ntree = 1000, mtry = 2, and node size = 5. The test set was held out exclusively for independent validation and was not involved in feature selection, model training, or hyperparameter tuning. Discrimination was evaluated on the test set using the area under the receiver operating characteristic curve (AUC), with 95% confidence intervals estimated from 1,000 bootstrap resamples. Calibration was assessed using decile-based calibration plots and Brier scores. Feature importance was visualized using SHapley Additive exPlanations (SHAP) values. Given the single-center design and the absence of an external cohort, all models were internally evaluated only. Sensitivity Analyses employed stricter thresholds, defining decline as MAPonset≤ 80% of MAPbaseline and recovery as >80% of MAPbaseline for severely declining patients in the integrated phenotype (detailed in Supplementary Table S1). In addition, three complementary analyses were performed to assess the robustness of the primary findings. First, restricted mean survival time (RMST) analysis at τ = 90 days was performed as an assumption-free complement to the Cox models, which does not rely on the proportional-hazards assumption. Second, the composite endpoint was decomposed, and each component was analyzed with the model best suited to its nature: the composite of death or RRT was analyzed using Cox regression, and persistent renal dysfunction among 90-day survivors was analyzed using logistic regression. Third, MAKE-90 was re-analyzed as a binary outcome using logistic regression with identical covariate adjustment. Results Patient Characteristics and Clinical Outcomes A total of 1248 AKI patients were enrolled in this study, with a median age of 90 years. According to the KDIGO criteria, 837 patients (67.1%) were classified as AKI Stage 1, 251 (20.1%) as Stage 2, and 160 (12.8%) as Stage 3. The median baseline eGFR was 70.67 mL/min/1.73m2, and the median MAPbaseline was 87.33mmHg. Comorbidities and laboratory parameters at the onset of AKI are detailed in Table 2.Table 2Baseline Characteristics of the ParticipantsCharacteristicTotalNon-MAKE-90MAKE-90P Value1248617631Age [median (P25, P75)]90 (87, 94)90 (86, 94)91 (87, 94)0.03Gender1248 (100%)617 (100%)631 (100%)1BMI [median (P25, P75)] (kg/m2)22.98 (21.11, 25.35)23.15 (21.26, 25,39)22.86 (21.09, 25.06)0.15Smoke (%)547 (43.8)257 (41.7)290 (46.0)0.14ComorbidityHypertension (%)958 (76.8)489 (79.3)469 (74.3)0.046Diabetes (%)668 (53.5)337 (54.6)331 (52.5)0.478Hyperlipidemia (%)556 (44.6)283 (45.9)273 (43.3)0.385Hyperuricemia (%)751 (60.2)354 (57.4)397 (62.9)0.052Coronary heart disease (%)1022 (81.9)505 (81.8)517 (81.9)1Chronic heart failure (%)912 (73.1)390 (63.2)522 (82.7)<0.001CKD (%)552 (44.2)243 (39.4)309 (49.0)0.001Cerebral infarction (%)620 (49.7)293 (47.5)327 (51.8)0.14COPD (%)444 (35.6)210 (34.0)234 (37.1)0.287Breathing machine (%)496 (39.7)168 (27.2)328 (52.0)<0.001Tumor (%)465 (37.3)180 (29.2)285 (45.2)<0.001Liver injury332 (26.6)119 (19.3)213 (33.8)<0.001Laboratory parametersProteinuria (%)663 (53.1)278 (45.1)385 (61.0)<0.001Baseline eGFR [median (P25, P75)] (mL/min/1.73m2)70.67 [49.18, 84.04]73.72 [52.60, 85.39]66.22 [45.64, 81.91]<0.001HB [median (P25, P75)] (g/L)105 (90, 121)112 (98, 125)99 (85, 113)<0.001PLT [median (P25, P75)] (*10^9/L)156 (112, 208)166 (129, 220)141 (98, 191)<0.001CRP [median (P25, P75)] (mg/dL)4.40 (1.08, 11.10)2.38 (0.65, 8.37)6.44 (2.09, 13.30)<0.001FBG [median (P25, P75)] (mmol/L)7.37 (5.74, 10.21)7.00 (5.60, 9.60)7.75 (5.94, 10.83)<0.001Albumin [median (P25, P75)] (g/L)33.80 (30.3, 37.3)35,30 (32.10, 38.50)32.30 (29.10, 35.80)<0.001Urea [median (P25, P75)] (mmol/L)15.00 (10.00, 25.60)12.10 (8.50, 17.00)19.10 (12.76, 28.40)<0.001Uric Acid [median (P25, P75)] (mmol/L)376.00 (285.00, 479.07)357.00 (270.40, 443.10)402.00 (297.90, 512.00)<0.001Potassium [median (P25, P75)] (mmol/L)4.29 (3.90, 4.71)4.28 (3.89, 4.70)4.30 (3.90, 4.80)0.202Na [median (P25, P75)] (mmol/L)140.00 (136.00, 145.00)138.00 (135.00, 143.00)142.00 (137.00, 147.50)<0.001Ca [median (P25, P75)] (mmol/L)2.23 (2.11, 2.37)2.24 (2.13, 2.37)2.23 (2.08, 2.38)0.095Cl [median (P25, P75)] (mmol/L)103.00 (98.80, 107.60)102.00 (98.00, 106.00)104.00 (99.00, 110.00)<0.001P [median (P25, P75)] (mmol/L)1.10 (0.90, 1.38)1.10 (0.90, 1.30)1.17 (0.90, 1.44)0.001Bicarbonate [median (P25, P75)] (mmol/L)24.40 (21.70, 27.30)24.60 (22.00, 27.70)24.10 (21.00, 27.10)0.072TNI [median (P25, P75)] (μg/L)0.05 (0.02, 0.12)0.04 (0.01, 0.08)0.06 (0.03, 0.16)<0.001NT-proBNP [median (P25, P75)] (pg/mL)1991.50 (564.98, 6115.80)1140.40 (346.60, 3495.00)3105.00 (1079.75, 9015.75)<0.001AKI stage (%)<0.0011837 (67.1)494 (59.02)343 (40.98)2251 (20.1)103 (41.04)148 (58.96)3160 (12.8)20 (12.5)140 (87.5)MAP (mmHg)MAPbaseline [median (P25, P75)]87.33 (81.67, 93.33)88.00 (82.00, 94.67)87.00 (81.50, 92.83)0.058MAPonset [median (P25, P75)]80.33 (71.67, 89.33)82.67 (74.33, 90.67)77.67 (70.00, 87.67)<0.001MAP change Rate [median (P25, P75)]−7.97 (−17.94, 3.08)−6.25 (−15.83, 4.42)−10.20 (−19.69, 1.15)0.001MAP 24h after AKI [median (P25, P75)]84.00 (76.67, 91.38)85.00 (78.67, 91.67)82.33 (74.25, 91.08)<0.001MAP 48h after AKI [median (P25, P75)]82.00 (75.00, 89.33)83.33 (77.33, 90.00)80.33 (73.33, 88.83)<0.001MAP 72h after AKI [median (P25, P75)]83.33 (76.33, 90.33)85.00 (79.00, 91.33)81.50 (73.92, 89.08)<0.001Endpoints90-day mortality (%)457 (36.6)0 (0.0)457 (100)<0.00190-day RRT (%)9 (0.72)0 (0.0)9 (100)<0.00190-day persistent renal dysfunction (%)165 (13.2)0 (0.0)165 (100)<0.001Abbreviations: AKI, acute kidney injury; BMI, body mass index; BNP, B-type natriuretic peptide; CKD, chronic kidney disease; COPD, chronic obstructive pulmonary disease; CRP, C-reactive protein; eGFR, estimated glomerular filtration rate; FBG, fasting blood glucose; HB, hemoglobin; MAP, mean arterial pressure; MAPbaseline, baseline mean arterial pressure; MAPonset, mean arterial pressure at the onset of AKI; PLT, platelet; RRT, renal replacement therapy; TNI, troponin I. Table 2Baseline Characteristics of the ParticipantsCharacteristicTotalNon-MAKE-90MAKE-90P Value1248617631Age [median (P25, P75)]90 (87, 94)90 (86, 94)91 (87, 94)0.03Gender1248 (100%)617 (100%)631 (100%)1BMI [median (P25, P75)] (kg/m2)22.98 (21.11, 25.35)23.15 (21.26, 25,39)22.86 (21.09, 25.06)0.15Smoke (%)547 (43.8)257 (41.7)290 (46.0)0.14ComorbidityHypertension (%)958 (76.8)489 (79.3)469 (74.3)0.046Diabetes (%)668 (53.5)337 (54.6)331 (52.5)0.478Hyperlipidemia (%)556 (44.6)283 (45.9)273 (43.3)0.385Hyperuricemia (%)751 (60.2)354 (57.4)397 (62.9)0.052Coronary heart disease (%)1022 (81.9)505 (81.8)517 (81.9)1Chronic heart failure (%)912 (73.1)390 (63.2)522 (82.7)<0.001CKD (%)552 (44.2)243 (39.4)309 (49.0)0.001Cerebral infarction (%)620 (49.7)293 (47.5)327 (51.8)0.14COPD (%)444 (35.6)210 (34.0)234 (37.1)0.287Breathing machine (%)496 (39.7)168 (27.2)328 (52.0)<0.001Tumor (%)465 (37.3)180 (29.2)285 (45.2)<0.001Liver injury332 (26.6)119 (19.3)213 (33.8)<0.001Laboratory parametersProteinuria (%)663 (53.1)278 (45.1)385 (61.0)<0.001Baseline eGFR [median (P25, P75)] (mL/min/1.73m2)70.67 [49.18, 84.04]73.72 [52.60, 85.39]66.22 [45.64, 81.91]<0.001HB [median (P25, P75)] (g/L)105 (90, 121)112 (98, 125)99 (85, 113)<0.001PLT [median (P25, P75)] (*10^9/L)156 (112, 208)166 (129, 220)141 (98, 191)<0.001CRP [median (P25, P75)] (mg/dL)4.40 (1.08, 11.10)2.38 (0.65, 8.37)6.44 (2.09, 13.30)<0.001FBG [median (P25, P75)] (mmol/L)7.37 (5.74, 10.21)7.00 (5.60, 9.60)7.75 (5.94, 10.83)<0.001Albumin [median (P25, P75)] (g/L)33.80 (30.3, 37.3)35,30 (32.10, 38.50)32.30 (29.10, 35.80)<0.001Urea [median (P25, P75)] (mmol/L)15.00 (10.00, 25.60)12.10 (8.50, 17.00)19.10 (12.76, 28.40)<0.001Uric Acid [median (P25, P75)] (mmol/L)376.00 (285.00, 479.07)357.00 (270.40, 443.10)402.00 (297.90, 512.00)<0.001Potassium [median (P25, P75)] (mmol/L)4.29 (3.90, 4.71)4.28 (3.89, 4.70)4.30 (3.90, 4.80)0.202Na [median (P25, P75)] (mmol/L)140.00 (136.00, 145.00)138.00 (135.00, 143.00)142.00 (137.00, 147.50)<0.001Ca [median (P25, P75)] (mmol/L)2.23 (2.11, 2.37)2.24 (2.13, 2.37)2.23 (2.08, 2.38)0.095Cl [median (P25, P75)] (mmol/L)103.00 (98.80, 107.60)102.00 (98.00, 106.00)104.00 (99.00, 110.00)<0.001P [median (P25, P75)] (mmol/L)1.10 (0.90, 1.38)1.10 (0.90, 1.30)1.17 (0.90, 1.44)0.001Bicarbonate [median (P25, P75)] (mmol/L)24.40 (21.70, 27.30)24.60 (22.00, 27.70)24.10 (21.00, 27.10)0.072TNI [median (P25, P75)] (μg/L)0.05 (0.02, 0.12)0.04 (0.01, 0.08)0.06 (0.03, 0.16)<0.001NT-proBNP [median (P25, P75)] (pg/mL)1991.50 (564.98, 6115.80)1140.40 (346.60, 3495.00)3105.00 (1079.75, 9015.75)<0.001AKI stage (%)<0.0011837 (67.1)494 (59.02)343 (40.98)2251 (20.1)103 (41.04)148 (58.96)3160 (12.8)20 (12.5)140 (87.5)MAP (mmHg)MAPbaseline [median (P25, P75)]87.33 (81.67, 93.33)88.00 (82.00, 94.67)87.00 (81.50, 92.83)0.058MAPonset [median (P25, P75)]80.33 (71.67, 89.33)82.67 (74.33, 90.67)77.67 (70.00, 87.67)<0.001MAP change Rate [median (P25, P75)]−7.97 (−17.94, 3.08)−6.25 (−15.83, 4.42)−10.20 (−19.69, 1.15)0.001MAP 24h after AKI [median (P25, P75)]84.00 (76.67, 91.38)85.00 (78.67, 91.67)82.33 (74.25, 91.08)<0.001MAP 48h after AKI [median (P25, P75)]82.00 (75.00, 89.33)83.33 (77.33, 90.00)80.33 (73.33, 88.83)<0.001MAP 72h after AKI [median (P25, P75)]83.33 (76.33, 90.33)85.00 (79.00, 91.33)81.50 (73.92, 89.08)<0.001Endpoints90-day mortality (%)457 (36.6)0 (0.0)457 (100)<0.00190-day RRT (%)9 (0.72)0 (0.0)9 (100)<0.00190-day persistent renal dysfunction (%)165 (13.2)0 (0.0)165 (100)<0.001Abbreviations: AKI, acute kidney injury; BMI, body mass index; BNP, B-type natriuretic peptide; CKD, chronic kidney disease; COPD, chronic obstructive pulmonary disease; CRP, C-reactive protein; eGFR, estimated glomerular filtration rate; FBG, fasting blood glucose; HB, hemoglobin; MAP, mean arterial pressure; MAPbaseline, baseline mean arterial pressure; MAPonset, mean arterial pressure at the onset of AKI; PLT, platelet; RRT, renal replacement therapy; TNI, troponin I. Baseline Characteristics of the Participants Abbreviations: AKI, acute kidney injury; BMI, body mass index; BNP, B-type natriuretic peptide; CKD, chronic kidney disease; COPD, chronic obstructive pulmonary disease; CRP, C-reactive protein; eGFR, estimated glomerular filtration rate; FBG, fasting blood glucose; HB, hemoglobin; MAP, mean arterial pressure; MAPbaseline, baseline mean arterial pressure; MAPonset, mean arterial pressure at the onset of AKI; PLT, platelet; RRT, renal replacement therapy; TNI, troponin I. The endpoint MAKE-90 occurred in 50.6% (631/1248) of patients. Of these, 36.6% (n = 457) died within 90 days, 0.72% (n = 9) required renal replacement therapy, and 13.2% (n = 165) had persistent renal impairment. The incidence of endpoint events significantly differed across AKI stages: 40.98% (343/837) in AKI Stage 1, 58.96% (148/251) in AKI Stage 2, and 87.5% (140/160) in AKI Stage 3 (P<0.001). No significant difference in MAPbaseline was observed between patients with and without MAKE-90. However, patients who reached the primary endpoint exhibited significantly lower MAPonset and maintained persistently lower levels within the first three days following AKI onset (Table 2). Association Between MAP at AKI Onset (MAPonset) with MAKE-90 Patients were first categorized into three groups based on MAPonset: MAPonset≤65mmHg group, 65–80mmHg group, and >80 mmHg group. The incidence of MAKE-90 differed significantly across groups (P < 0.001, Table 1). Kaplan-Meier analysis demonstrated progressively worsening prognosis with lower MAPonset (P<0.001, Figure 2). Figure 2Kaplan-Meier curves for the risk of MAKE-90 stratified by different MAP parameters. (A) Grouped by MAPonset: Group 1, MAPonset ≤ 65 mmHg; Group 2, 65 < MAPonset ≤ 80 mmHg; Group 3, MAPonset > 80 mmHg. (B) Grouped by MAP change rate: Group 1, ≤ –20%; Group 2, –20% to –10%; Group 3, –10% to 10%; Group 4, 10% to ≤ 20%; and Group 5, >20%. (C) Grouped by MAP recovery pattern: Group 1, No decline; Group 2, Recovery within 24h; Group 3, Recovery within 48h; Group 4, Recovery within 72h; and Group 5, No recovery by 72h. (D) Grouped by integrated phenotype: Group 1, No decline; Group 2, Mild decline with recovery; Group 3, Mild decline without recovery; Group 4, severe decline with recovery; and Group 5, severe decline without recovery.Four Kaplan-Meier survival graphs illustrate survival probability over the follow-up period. Graph A compares survival by mean arterial pressure (MAP) onset: less than or equal to 65 mmHg, 65-80 mmHg and greater than 80 mmHg, with the greater than 80 mmHg group showing higher survival at the end of follow-up. Graph B examines MAP change rates: less than or equal to -20 percent, negative 20 percent to negative 10 percent, negative 10 percent to 10 percent, 10 percent to 20 percent and greater than 20 percent, with the 10 percent to 20 percent group having better survival. Graph C looks at MAP recovery patterns: No decline, Recovery within 24h, 48h, 72h and No recovery by 72h, with No decline showing the highest survival. Graph D compares integrated phenotypes: No decline, Mild decline with/without recovery, Severe decline with/without recovery, with No decline again showing the highest survival. Each graph includes a table of numbers at risk over time. Groups with better MAP or recovery patterns have the highest survival probabilities. All graphs show significant differences (P less than 0.001), with colors distinguishing groups.Four Kaplan-Meier graphs comparing survival probability by mean arterial pressure categories and recovery patterns.Abbreviations: AKI, acute kidney injury; MAKE-90, major adverse kidney events at 90 days; MAP, mean arterial pressure; MAPbaseline, baseline mean arterial pressure; MAPonset, mean arterial pressure at the onset of AKI. Kaplan-Meier curves for the risk of MAKE-90 stratified by different MAP parameters. (A) Grouped by MAPonset: Group 1, MAPonset ≤ 65 mmHg; Group 2, 65 < MAPonset ≤ 80 mmHg; Group 3, MAPonset > 80 mmHg. (B) Grouped by MAP change rate: Group 1, ≤ –20%; Group 2, –20% to –10%; Group 3, –10% to 10%; Group 4, 10% to ≤ 20%; and Group 5, >20%. (C) Grouped by MAP recovery pattern: Group 1, No decline; Group 2, Recovery within 24h; Group 3, Recovery within 48h; Group 4, Recovery within 72h; and Group 5, No recovery by 72h. (D) Grouped by integrated phenotype: Group 1, No decline; Group 2, Mild decline with recovery; Group 3, Mild decline without recovery; Group 4, severe decline with recovery; and Group 5, severe decline without recovery. In the univariate analysis, taking the MPonset 65–80 mmHg group as the reference group, a lower MAPonset value predicted a worse prognosis (HR 1.30, 95% CI 1.03–1.65). Multivariate Cox regression analysis revealed that MAPonset ≤65 mmHg was an independent risk factor for MAKE-90 (HR 1.33, 95% CI 1.05–1.69 in Model 2; HR 1.08, 95% CI 0.85–1.39 in Model 3). Conversely, MAPonset >80 mmHg displayed considerable protective prognostic value (HR 0.67, 95% CI 0.57–0.80 in Model 2; HR 0.83, 95% CI 0.70–0.99 in Model 3) (Table 3).Table 3Prognostic Value of MAP Dimensions for MAKE-90Model 1Model 2Model 3HR (95% CI)P ValueHR (95% CI)P ValueHR (95% CI)P ValueMAPonset≤ 65 mmHg1.30 (1.03–1.65)0.031.33 (1.05–1.69)0.021.08 (0.85–1.39)0.5465–80 mmHgRefRefRef> 80 mmHg0.64 (0.54–0.76)<0.010.67 (0.57–0.80)<0.010.83 (0.70–0.99)0.04MAP change rate≤-20%1.50 (1.22–1.84)<0.011.46 (1.19–1.79)<0.0011.10 (0.89–1.36)0.36−20% to −10%1.29 (1.05–1.57)0.011.26 (1.03–1.54)0.021.00 (0.81–1.22)0.92−10% to 10%RefRefRef10% to 20%0.80 (0.59–1.10)0.170.76 (0.56–1.04)0.090.70 (0.51–0.97)0.03>20%0.95 (0.65–1.40)0.810.93 (0.63–1.37)0.700.74 (0.50–1.10)0.13MAP recovery patternNo-declineRefRefRefRecovery within 24 h1.25 (1.01–1.54)0.041.25 (1.01–1.55)0.041.12 (0.90–1.39)0.32Recovery within 48 h1.37 (0.99–1.89)0.051.41 (1.02–1.95)0.041.09 (0.78–1.53)0.60Recovery within 72 h1.27 (0.89–1.82)0.181.22 (0.86–1.74)0.270.90 (0.62–1.30)0.57No-recovery by 72 h1.73 (1.38–2.17)<0.011.71 (1.37–2.15)<0.011.32 (1.05–1.67)0.02Integrated PhenotypeNo-declineRefRefRefMild decline with recovery1.16 (0.90–1.49)0.261.14 (0.88–1.47)0.330.96 (0.74–1.25)0.77Mild decline without recovery1.57 (1.22–2.01)<0.011.59 (1.24–2.03)<0.011.25 (0.97–1.62)0.09Severe decline with recovery1.21 (0.90–1.63)0.211.18 (0.87–1.59)0.281.03 (0.75–1.40)0.09Severe decline without recovery1.68 (1.33–2.12)<0.011.72 (1.36–2.17)<0.011.33 (1.04–1.69)0.02Notes: Model 1, univariable model. Model 2 adjusted Model 1 for proteinuria, age, body mass index, smoking, hypertension, diabetes, hyperlipidemia, hyperuricemia, and history of chronic kidney disease. Model 3, based on Model 2 and additionally adjusted for AKI stage, coronary heart disease, heart failure, stroke, chronic obstructive pulmonary disease, mechanical ventilation, malignancy, hemoglobin, platelet, C-reactive protein, albumin, urea, and hepatic injury.Abbreviations: AKI, acute kidney injury; MAP, mean arterial pressure; MAP baseline, baseline MAP; MAPonset, MAP at the onset of acute kidney injury; REF, reference group. Table 3Prognostic Value of MAP Dimensions for MAKE-90Model 1Model 2Model 3HR (95% CI)P ValueHR (95% CI)P ValueHR (95% CI)P ValueMAPonset≤ 65 mmHg1.30 (1.03–1.65)0.031.33 (1.05–1.69)0.021.08 (0.85–1.39)0.5465–80 mmHgRefRefRef> 80 mmHg0.64 (0.54–0.76)<0.010.67 (0.57–0.80)<0.010.83 (0.70–0.99)0.04MAP change rate≤-20%1.50 (1.22–1.84)<0.011.46 (1.19–1.79)<0.0011.10 (0.89–1.36)0.36−20% to −10%1.29 (1.05–1.57)0.011.26 (1.03–1.54)0.021.00 (0.81–1.22)0.92−10% to 10%RefRefRef10% to 20%0.80 (0.59–1.10)0.170.76 (0.56–1.04)0.090.70 (0.51–0.97)0.03>20%0.95 (0.65–1.40)0.810.93 (0.63–1.37)0.700.74 (0.50–1.10)0.13MAP recovery patternNo-declineRefRefRefRecovery within 24 h1.25 (1.01–1.54)0.041.25 (1.01–1.55)0.041.12 (0.90–1.39)0.32Recovery within 48 h1.37 (0.99–1.89)0.051.41 (1.02–1.95)0.041.09 (0.78–1.53)0.60Recovery within 72 h1.27 (0.89–1.82)0.181.22 (0.86–1.74)0.270.90 (0.62–1.30)0.57No-recovery by 72 h1.73 (1.38–2.17)<0.011.71 (1.37–2.15)<0.011.32 (1.05–1.67)0.02Integrated PhenotypeNo-declineRefRefRefMild decline with recovery1.16 (0.90–1.49)0.261.14 (0.88–1.47)0.330.96 (0.74–1.25)0.77Mild decline without recovery1.57 (1.22–2.01)<0.011.59 (1.24–2.03)<0.011.25 (0.97–1.62)0.09Severe decline with recovery1.21 (0.90–1.63)0.211.18 (0.87–1.59)0.281.03 (0.75–1.40)0.09Severe decline without recovery1.68 (1.33–2.12)<0.011.72 (1.36–2.17)<0.011.33 (1.04–1.69)0.02Notes: Model 1, univariable model. Model 2 adjusted Model 1 for proteinuria, age, body mass index, smoking, hypertension, diabetes, hyperlipidemia, hyperuricemia, and history of chronic kidney disease. Model 3, based on Model 2 and additionally adjusted for AKI stage, coronary heart disease, heart failure, stroke, chronic obstructive pulmonary disease, mechanical ventilation, malignancy, hemoglobin, platelet, C-reactive protein, albumin, urea, and hepatic injury.Abbreviations: AKI, acute kidney injury; MAP, mean arterial pressure; MAP baseline, baseline MAP; MAPonset, MAP at the onset of acute kidney injury; REF, reference group. Prognostic Value of MAP Dimensions for MAKE-90 Notes: Model 1, univariable model. Model 2 adjusted Model 1 for proteinuria, age, body mass index, smoking, hypertension, diabetes, hyperlipidemia, hyperuricemia, and history of chronic kidney disease. Model 3, based on Model 2 and additionally adjusted for AKI stage, coronary heart disease, heart failure, stroke, chronic obstructive pulmonary disease, mechanical ventilation, malignancy, hemoglobin, platelet, C-reactive protein, albumin, urea, and hepatic injury. Abbreviations: AKI, acute kidney injury; MAP, mean arterial pressure; MAP baseline, baseline MAP; MAPonset, MAP at the onset of acute kidney injury; REF, reference group. Associations of MAP Change Rate with MAKE-90 Given the significant inter-individual variability in baseline blood pressure among the older population, we analyzed the relative MAP change rate, categorizing patients into five groups (≤−20%, −20% to −10%, −10% to 10%, 10% to 20%, >20%) (Table 1). Kaplan-Meier analysis showed a graded relationship between the magnitude of MAP change rate and the cumulative incidence of MAKE-90 (P < 0.001), except for the group with a >20% increase (Figure 2). Univariate regression survival analysis showed that a decrease of even 10% decline from MAPbaseline was associated with an increased risk of MAKE-90 (HR 1.29, 95% CI 1.05–1.57), whereas a 20% decrease was associated with significantly higher risk of MAKE-90 (HR 1.50, 95% CI 1.22–1.84). After adjustment for potential confounders, the MAP change rate remained an independent risk factor for MAKE-90 (HR 1.26, 95% CI 1.03–1.54 for a 10% decrease, HR 1.46, 95% CI 1.19–1.79 for a 20% decrease). On the contrary, a mild elevation of MAP (within 10%–20% increase of MAPbaseline) appeared protective, but the benefits diminished as the increase exceeded this range (Table 3). Secondary endpoints, including 90-day all-cause mortality and persistent renal dysfunction, also had consistent trends (Supplementary Tables S2–S5). Based on the above findings, we proceeded to use the restricted cubic splines analysis to look for a potential nonlinear relation between MAPonset and the MAKE-90 event. The analysis confirmed a marked nonlinear U-shaped association between the two variables (P=0.0396). The nonlinear pattern remained statistically significant after controlling for multiple confounding factors in two multivariable-adjusted models (Model 2, P = 0.0428; Model 3, P = 0.0449) (Figure 3). Figure 3U-shaped association between MAP at the onset of AKI and MAKE-90. (A) Unadjusted model. (B) Model adjusted for proteinuria, age, body mass index, smoking, hypertension, diabetes, hyperlipidemia, hyperuricemia, and chronic kidney disease. (C) Model B further adjusted for AKI stage, coronary heart disease, heart failure, stroke, chronic obstructive pulmonary disease, mechanical ventilation, malignancy, and baseline levels of hemoglobin, platelets, C-reactive protein, albumin, urea, and hepatic injury.Three restricted cubic spline plots labeled MAKE90 show HR (95 percent CI) against MAPonset . Each graph has a dashed horizontal line at HR equals 1.0 and a smooth curve with a 95 percent confidence band. Image A: The overall P less than 0.001 and P for nonlinearity equals 0.0396. The curve lies above HR equals 1.0 near MAPonset of 60 mmHg, falls below 1.0 between 80 and 100 mmHg, and then rises modestly at the upper end. Image B: The overall P less than 0.001 and P for nonlinearity equals 0.0428. The curve is above 1.0 at MAPonset equals 60 mmHg, dips below 1.0 between 85 and 95 mmHg, and then rises slightly toward 100 mmHg.Image C: Overall P less than 0.001, nonlinear P equals 0.0449. The curve is above 1.0 from 60 to 70 mmHg, crosses near 1.0 around 80 mmHg, stays below 1.0 from 85 to 100 mmHg, and trends upward at the highest values.Three spline plots with 95% confidence bands illustrating the nonlinear association between MAKE90 hazard ratio and MAPonset across three models.Abbreviations: AKI, acute kidney injury; MAP, mean arterial pressure; MAKE-90, major adverse kidney events at 90 days. U-shaped association between MAP at the onset of AKI and MAKE-90. (A) Unadjusted model. (B) Model adjusted for proteinuria, age, body mass index, smoking, hypertension, diabetes, hyperlipidemia, hyperuricemia, and chronic kidney disease. (C) Model B further adjusted for AKI stage, coronary heart disease, heart failure, stroke, chronic obstructive pulmonary disease, mechanical ventilation, malignancy, and baseline levels of hemoglobin, platelets, C-reactive protein, albumin, urea, and hepatic injury. Association of MAP Recovery Patterns with MAKE-90 Among the 1157 patients with complete serial measurements, 542 (46.8%) experienced an initial hemodynamic insult (MAP change rate≤ −10%). Based on the time to recovery, patients were classified into five distinct patterns: “no decline”, and among those with an initial insult, “recovery within 24h”, “recovery within 48h”, “recovery within 72h”, and “no recovery by 72h” (Table 1). Kaplan-Meier analysis revealed a gradient of risk, with the worst prognosis in patients who did not recover by 72 hours (P < 0.001; Figure 2). This was confirmed by multivariable Cox regression. Using the “no decline” group as reference, “no recovery by 72h” was associated with the highest risk of MAKE-90 (HR 1.71, 95% CI 1.37–2.15 in Model 2; HR 1.32, 95% CI 1.05–1.67 in Model 3). Earlier recovery (within 72h) was associated with incrementally lower risks (Table 3). The association followed a consistent pattern for the secondary endpoints of 90-day mortality and persistent renal dysfunction (Supplementary Tables S6 and S7). Prognostic Impact of the Integrated High-Risk Hemodynamic Phenotype To identify the subgroup with the poorest prognosis, we integrated the severity of the initial insult with the 72-hour recovery status and grouped patients into five groups: “no decline”, “mild decline with recovery”, “mild decline without recovery”, “severe decline with recovery”, and “severe decline without decline” (Table 1). Kaplan-Meier analysis showed significant differences in outcomes among phenotypes (P < 0.001; Figure 2). Patients with the “severe decline without recovery” phenotype had the worst prognosis. Multiple Cox regression analyses revealed that a decline of more than 10% from MAPbaseline without recovery within 72 hours significantly increased the risk of MAKE-90 (HR 1.59, 95% CI 1.24–2.03 in Model 2; HR 1.25, 95% CI 0.97–1.62 in Model 3). After adjusting for all potential confounders, patients with the “severe decline without recovery” exhibited a further increased risk (HR 1.72, 95% CI 1.36–2.17 in Model 2; HR 1.33, 95% CI 1.04–1.69 in Model 3) (Table 3). The increased risk remained robust and consistent across all adjusted models when evaluating the secondary endpoint (Supplementary Tables S8 and S9). Sensitivity and Robustness Analyses All primary findings remained consistent in sensitivity analyses that employed: 1) changing a hypotension threshold (MAPonset ≤80% of MAPbaseline) and 2) alternative classifications of the integrated phenotype (detailed grouping in Supplementary Table S1, Supplementary Tables S10 and S11). In addition, three complementary analyses corroborated the primary findings. First, restricted mean survival time (RMST) analysis at τ = 90 days, which does not rely on the proportional-hazards assumption, yielded consistent results: compared with the “no-decline” group, patients with “no recovery by 72h” lost 18.2 MAKE-free days (95% CI 3.0–33.5; P = 0.019), and those with the “severe decline without recovery” phenotype lost 17.8 MAKE-free days (95% CI 2.4–33.2; P = 0.023). Second, component-wise analyses showed that the associations were strongest for the composite of death or RRT: in fully adjusted Cox models, both “no recovery by 72h” (HR 1.62, 95% CI 1.25–2.09) and the “severe decline without recovery” phenotype (HR 1.56, 95% CI 1.19–2.05) were associated with death or RRT (Supplementary Table S12); estimates for persistent renal dysfunction among 90-day survivors were imprecise owing to the limited number of events (148 events among 723 survivors) and competing mortality (Supplementary Table S13). Third, when MAKE-90 was re-analyzed as a binary outcome using logistic regression with identical covariate adjustment, the directions of all associations were consistent with the Cox models (eg, “no recovery by 72h” vs “no decline”: OR 1.86, 95% CI 1.31–2.67 in Model 2), with attenuation toward the null after full adjustment, as expected because the binary model discards event timing (Supplementary Table S14). Incremental Predictive Value of Dynamic Mean Arterial Pressure Assessment To objectively evaluate the added value of dynamic MAP assessment beyond conventional risk factors, we employed a machine learning approach. First, LASSO regression identified the seven most predictive clinical variables from a large set of all available traditional risk factors to construct a robust baseline model (Model 1): including proteinuria, AKI stage, requirement for mechanical ventilation, history of malignant tumors, chronic heart failure, liver injury, and hypertension. On the held-out test set, the AUCs (95% CI) were 0.71 (0.66–0.77), 0.73 (0.67–0.78), 0.75 (0.69–0.80), 0.72 (0.67–0.77), and 0.76 (0.71–0.81) for Models 1–5, respectively, with Brier scores ranging from 0.206 to 0.238 and acceptable calibration across deciles (Figure 4A and Supplementary Figure S1). Because the 95% confidence intervals overlapped across models, the incremental value of the MAP parameters is interpreted as numerically higher discrimination rather than statistically significant improvement. Figure 4Assessment of the incremental predictive value of dynamic mean arterial pressure assessment for MAKE-90. (A) Receiver operating characteristic (ROC) curves demonstrating the incremental predictive value of successive models for MAKE-90. Model 1: The seven most predictive clinical variables identified by LASSO regression, including proteinuria, AKI stage, requirement for mechanical ventilation, history of malignancy, heart failure, diagnosis of liver injury, and history of hypertension. Model 2: Model 1 + the absolute value of MAPonset. Model 3: Model 1 + MAP change rate. Model 4: Model 1 + MAP recovery pattern. Model 5: Model 1 + MAP integrated phenotype. B-E. Shapley additive explanations (SHAP) summary plots illustrating the directional influence of each MAP-related parameter on the prediction of MAKE-90 in the respective model. (B) Influence of MAPonset. (C) Influence of the MAP change rate. (D) Influence of MAP recovery pattern. (E) Influence of MAP integrated phenotype.Image A displays a receiver operating characteristic graph with Specificity on the x-axis and Sensitivity on the y-axis. Five models are compared: Model 1 (AUC 0.71), Model 2 (AUC 0.73), Model 3 (AUC 0.75), Model 4 (AUC 0.72) and Model 5 (AUC 0.76). All curves start near sensitivity 0 at specificity 1.00 and rise towards sensitivity 1.00 as specificity decreases. Image B shows a SHAP summary plot with SHAP values on the x-axis. Features include Breathing machine (0.137), AKI stage (0.115) and others, with Breathing machine and AKI stage having the widest spreads. Image C′s SHAP plot lists Breathing machine (0.118) and AKI stage (0.116) as top features, with distributions extending rightward. Image D′s SHAP plot highlights Breathing machine (0.164) with the broadest spread. Image E′s SHAP plot shows AKI stage (0.138) with the widest spread, while Hypertension clusters near 0.00.Five plots Ashowing ROC curves and SHAP distributions for successive MAKE-90 prediction models.Abbreviations: AKI, acute kidney injury; MAKE-90, major adverse kidney events at 90 days; MAP, mean arterial pressure; MAPonset, MAP at the onset of AKI. Assessment of the incremental predictive value of dynamic mean arterial pressure assessment for MAKE-90. (A) Receiver operating characteristic (ROC) curves demonstrating the incremental predictive value of successive models for MAKE-90. Model 1: The seven most predictive clinical variables identified by LASSO regression, including proteinuria, AKI stage, requirement for mechanical ventilation, history of malignancy, heart failure, diagnosis of liver injury, and history of hypertension. Model 2: Model 1 + the absolute value of MAPonset. Model 3: Model 1 + MAP change rate. Model 4: Model 1 + MAP recovery pattern. Model 5: Model 1 + MAP integrated phenotype. B-E. Shapley additive explanations (SHAP) summary plots illustrating the directional influence of each MAP-related parameter on the prediction of MAKE-90 in the respective model. (B) Influence of MAPonset. (C) Influence of the MAP change rate. (D) Influence of MAP recovery pattern. (E) Influence of MAP integrated phenotype. Then we developed and compared a series of random forest models to assess the incremental predictive value of various MAP parameters separately. The baseline model (Model 1) achieved an AUC of 0.71 for predicting MAKE-90 (Figure 4A). Sequentially incorporating the dynamic MAP parameters improved predictive performance: adding MAPonset (Model 2) increased the AUC to 0.73; adding MAP change rate (Model 3) increased it to 0.75; adding MAP recovery pattern (Model 4) achieved the AUC of 0.72 and adding the integrated phenotype (Model 5) achieved the highest AUC of 0.76 (Figure 4A). It is noteworthy that the contribution of each MAP characteristic (quantified by its mean absolute SHAP values) varies with model complexity. Absolute MAPonset, MAP relative change, and MAP recovery patterns were all identified as influencing factors (Figure 4B–4E). Discussion This study introduces and validates a novel, three-dimensional framework for the hemodynamic assessment of older inpatients with AKI, moving beyond static thresholds to a dynamic, individualized paradigm. We demonstrate that integrating (1) the absolute severity at AKI onset, (2) the relative change from baseline, and (3) the recovery trajectory within a critical 72-hour window provides superior prognostic stratification. Most importantly, we identify a distinct high-risk phenotype characterized by severe initial hypotension with delayed recovery, which identifies patients at the highest risk for adverse outcomes and who may benefit most from closer monitoring. Consistent with previous literature and clinical practice, we found that MAP ≤ 65 mmHg at the onset of AKI predicted worse outcomes in older AKI patients.6,22 This threshold is well-established, with a systematic review identifying MAP below 65 mmHg as the most consistently cited value associated with increased postoperative AKI risk.8 It is also observed in studies that attempted to maintain renal perfusion in sepsis-associated AKI.23,24 Similarly, a recent study in patients with NSTEMI undergoing PCI found that lower admission MAP was strongly associated with the development of contrast-associated nephropathy within 72 hours.25 But this threshold of MAP has several limitations, as previous studies showed that 46% of patients who had a MAP greater than 65 mmHg still developed AKI.26,27 Similarly, the results of a subgroup analysis of 776 patients with septic shock showed that maintenance of MAP at a higher level (80–85 mmHg) in patients with prior hypertension could decrease the rate of severe AKI and renal replacement therapy.28 This evidence highlights the limitation of uniform threshold standards of MAP. Recent reviews also confirmed that uniform blood pressure targets are inadequate for complex scenarios, advocating for personal pressure goals based on patients’ baseline status, particularly for those with chronic hypertension.15 Previous studies have found that the incidence of AKI could be reduced by perioperative individualized blood pressure target management.29 However, a randomized controlled trial on the effects of individualized blood pressure management during major abdominal surgery showed no significant improvement in major postoperative complications, including AKI.30 The clinical efficacy thus remains limited and inconclusive. More importantly, there has been a substantial gap in knowledge regarding the role of individualized blood pressure management in older AKI patients. To fill this void, we looked at the MAP relative change rate from baseline in our cohort. We found the MAP change rate as an independent predictor for poor outcomes, which indicated that normal-looking absolute pressure values may conceal relative hypotension risk. This issue is of great importance in older patients since they often have chronic hypertension or frailty-related hypotension at baseline. In patients with chronic hypertension, even blood pressure readings within the conventional “normal range” may represent a significant relative decrease from baseline levels. Since the kidneys of these patients have adapted to a higher perfusion pressure,24 this reduction in blood pressure might compromise renal blood flow, thereby establishing a state of relative hypotension to induce kidney injury. So, using individualized blood pressure assessment in older patients with AKI would provide another layer of information on top of the absolute threshold assessment, which is crucial from a clinical perspective for this group of patients. Interestingly, we observed that the effect of MAPonset on outcomes might be a non-linear and U-shaped relationship. Moderately elevated MAPonset confers a protective effect, but this benefit diminishes when MAPonset exceeds 20% above baseline values. Inversely, a lower MAPonset level was associated with a higher risk of adverse outcomes. This aligns with previous investigations where U-shaped hemodynamic relationships have been observed in non-kidney tissues.31,32 A particularly relevant study showed that in non-critically ill patients with AKI, systolic blood pressure presented a similar U-shaped association with AKI, and the lowest incidence of events occurred when systolic blood pressure was in the 110–129 mmHg range.33 Collectively, these findings suggest that there is an individualized optimal hemodynamic range for older AKI patients. Deviations beyond this range, whether hypotensive or hypertensive, can be detrimental. A key advance of our study is the focus on post-insult recovery dynamics, which was a previously overlooked dimension. We found that an early return of MAP to baseline level after AKI might lower the risk of major adverse events. This observation might help us to better understand the pathophysiological mechanisms underlying AKI in older patients. Multiple risk factors render older patients more susceptible to AKI, including inadequate renal perfusion, sepsis, cardiovascular events, use of nephrotoxic medications, and a history of chronic kidney disease. Notably, dehydration and volume depletion represent the most prevalent risk factors in this population.34 Rapid blood pressure recovery likely reflects preserved neurovascular compensatory capacity and rehabilitated renal autoregulation in these patients, enabling effective restoration of renal perfusion.35 Conversely, delayed recovery from hypotension suggests exhaustion of these adaptive responses, potentially reflecting microvascular dysfunction or persistent inflammatory-mediated vasodilation, although residual confounding from underlying illness severity cannot be entirely excluded. The duration and severity of hypotension are equally important since they jointly determine renal outcomes. Combining the degree of hypotension at the onset of AKI with the recovery patterns into the prognostic model strongly confirms their synergistic effect. Patients with profound initial hypotension and delayed restoration constituted a high-risk phenotype exhibiting the most unfavorable clinical prognosis. This underscores that the duration of hemodynamic compromise is as critical as its initial severity. The potential prognostic value of this dynamic framework is supported by our predictive modeling. The sequential addition of MAPonset, MAP change rate, MAP recovery pattern, and integrated phenotype to a baseline clinical model significantly improved the prediction of MAKE-90 (AUC increased from 0.71 to 0.76). Notably, integrated phenotype emerged as a top contributor in the machine learning model, validating it as a cornerstone of prognostication. Furthermore, the interpretation of our findings should consider the composite nature of MAKE-90, which includes mortality, renal replacement therapy, and persistent renal dysfunction. Given the high mortality rate in older AKI patients, a substantial proportion of the observed adverse events are driven by death. Therefore, the associations identified by our dynamic MAP framework likely reflect a combination of kidney-specific injury and overall patient vulnerability. Future studies with larger sample sizes should aim to delineate the differential impact of hemodynamic phenotypes on renal-specific versus overall adverse outcomes. This study has several limitations. First, even though multivariate adjustments for potential confounders were used in the study, it is an observational design and thus does not establish causality. Second, as a single-center retrospective study involving an older male cohort, its findings have limited generalizability to other populations. Thirdly, the relatively low rate of renal replacement therapy may reflect age-related conservative management preferences. Fourth, the observation duration was relatively short, and there were still potential residual confounders, which call for prospective, multicenter studies with extended follow-up durations in the future. Fifth, baseline blood pressure was measured seated while inpatient measurements were supine. This postural difference may introduce systematic bias in MAP comparisons, particularly in older adults with reduced baroreflex sensitivity. Future studies should adopt a unified measurement protocol to minimize this source of variability. In conclusion, this study demonstrates that delayed MAP recovery and the integrated hemodynamic phenotype are robust prognostic markers for adverse outcomes in older inpatients with AKI. A composite assessment of the initial hemodynamic insult and the subsequent recovery kinetics within a critical 72-hour window may enhance risk stratification beyond static thresholds. The modest improvement in model discrimination (AUC from 0.71 to 0.76) supports the potential prognostic value of this dynamic framework for identifying high-risk patients, though prospective validation is required before clinical implementation.