The discovery cohort for our study consisted of patients selected from the electronic Intensive Care Unit Collaborative Research Database (eICU, version 2.0), which is a comprehensive database containing detailed information from multiple intensive care units (ICUs). The database includes data from over 200,000 ICU admissions that took place between 2014 and 2015 in 208 hospitals across the United States [12]. The validation cohort data were obtained from the Medical Information Mart for Intensive Care III (MIMIC-III, version 1.4), which is an extensive database containing the medical records of 60,000 intensive care unit (ICU) patients treated at the Beth Israel Deaconess Medical Center in Boston, Massachusetts, USA, from 2001 to 2012 [13]. In the context of routine clinical activities, all data are systematically acquired by computerized means, obviating the direct involvement of medical personnel in the data collection process. One author (LL) completed a web course titled “Protecting Human Research Subjects” offered by the National Institutes of Health, and obtained the corresponding certificate (ID: 35965741). Possessing the requisite qualifications, LL is duly authorized to initiate inquiries into the information repositories. It is noteworthy that the establishment of the databases received explicit approval from the Massachusetts Institute of Technology, granted under an exemption for informed consent.

We analyzed data from consecutive patients aged 18 years or older who were diagnosed with AMI. All patients received Guideline-Directed Medical Therapy. The definition of AMI was based on the clinical practice guideline [14]. AKI was diagnosed using the criteria proposed by Kidney Disease: Improving Global Outcomes. These criteria include an increase in serum creatinine (SCr) by $\ge$0.3 mg/dL (or $\ge$26.5 µmol/L) within 48 hours, an increase in SCr to 1.5 times the baseline level within 7 days, and a urine output (UO) of $\le$0.5 mL/kg/h for 6 hours [15]. 963 (11.1%) patients were excluded due to lack of data.

Patients’ data, including demographics (age, gender, race, weight), vital signs (temperature, blood pressure, heart rate, respiratory rate, UO), common comorbidities (congestive heart failure, hypertension, diabetes, chronic pulmonary obstructive disease, sepsis, renal disfunction, hepatic disfunction, stroke, cancer), and laboratory tests results (white blood cell count, red blood cell count, HCT, hemoglobin, platelet count, serum sodium, serum potassium, serum calcium, serum chloride, serum glucose, serum creatine kinase, serum aniongap, serum bicarbonate, SCr, blood urine nitrogen) were included in the initial analysis. In the two public databases, for patients with multiple ICU admissions, only the data from the first ICU admission were selected for analysis.

The principal determinant examined in this analysis was the initial HCT level upon hospitalization. The categorization of HCT was performed in accordance with the World Health Organization’s definition of anemia [16], stratifying it into five groups: (1) normal range (HCT $\ge$36% in females and $\ge$39% in males), (2) mild reduction (33% $\le$ HCT 36% in females or 33% $\le$ HCT 39% in males), (3) 30% $\le$ HCT 33%, (4) 27% $\le$ HCT 30%, and finally, (5) HCT 27%. Previous research has posited that male patients exhibit a heightened susceptibility to AKI in comparison to their female counterparts when confronted with a decline in HCT levels [9]. In order to mitigate potential sex stratification effects, a subgroup analysis stratified by gender was performed.

We aimed to identify the relationship between HCT and renal dysfunction in patients with AMI. The main outcome of interest was AKI during hospitalization. The secondary outcomes included RRT and hospital mortality.

The assessment of data normal distribution was conducted utilizing the Kolmogorov-Smirnov test. Continuous variables were presented as mean $\pm$ standard deviation (SD) and subjected to comparison through the t-test. Levene’s test was used to assess the assumption of homogeneity of variance. When the assumption of homoscedasticity was violated, intergroup comparisons were conducted using the Welch’s t-test. Proportions were used to represent categorical data, and the chi-squared test was employed for their analysis. Multiple imputation techniques were employed to address missing variables [17].

Multivariable logistic regression analysis employing stepwise backward selection methodology was employed for covariate adjustment, incorporating variables identified to exhibit marginal association (p 0.10) on univariate analysis with AKI. Adjusted risk estimates were obtained by constructing odds ratios (OR) along with 95% confidence intervals (CI). The marginal effects of HCT in predicting AKI were investigated. In addition, an evaluation of the variance inflation factor (VIF) among the covariates in a logistic model was used to assess potential multicollinearity. Variables demonstrating a VIF $>$4.0, indicative of multicollinearity, were excluded in the multivariate logistic regression analysis. Statistical significance was defined as p-values less than 0.05 (two-tailed). All statistical analyses were conducted using Stata version 15.0 (StataCorp, College Station, TX, USA) and R software version 4.0.4 (R Foundation for Statistical Computing, Vienna, Austria).

The discovery cohort consisted of a total of 7712 AMI patients. Among them, 950 patients (12.3%) experienced AKI during hospitalization (Supplementary Table 1). In short, mean age was 65.0 $\pm$ 12.6 years, 4924 (63.8%) were men, and 3944 (51.1%) were diagnosed as ST-elevation myocardial infarction. AKI group compared to non-AKI group were older (68.4 $\pm$ 12.5 vs. 64.5 $\pm$ 12.6, p 0.001) and had significantly lower HCT (36.4 $\pm$ 7.8 vs. 39.5 $\pm$ 6.5, p 0.001). We identified 17 variables using stepwise backward approach to build logistic regression model in the discovery cohort, and only data of the 17 variables of the validation cohort were extracted for convenience. In the validation cohort, a total of 1980 patients with AMI were included, 279 patients (14.1%) were diagnosed as AKI and patients with AKI were found to be older compared to those without AKI (72.3 $\pm$ 12.9 vs. 67.1 $\pm$ 14.2, p 0.001). The AKI group exhibited lower HCT levels compared to the non-AKI group (31.7 $\pm$ 4.4 vs. 33.7 $\pm$ 4.8, p 0.001). Table 1 presents additional comparisons between the non-AKI and AKI groups in both the discovery and validation cohorts.

A multivariable logistic regression analysis was conducted to investigate the correlation between HCT and AKI. In the unadjusted model, low HCT was found to be significantly associated with an increased risk of AKI in AMI patients, with OR of 1.876 (95% CI: 1.572–2.238, p 0.001), 2.933 (95% CI: 2.346–3.665, p 0.001), 3.592 (95% CI: 2.770–4.657, p 0.001) and 4.399 (95% CI: 3.446–5.615, p 0.001) for HCT in mild reduction, 30–33%, 27–30% and 27% groups, respectively, with normal HCT as the reference. Furthermore, 17 variables were incorporated into the covariate adjustment of the univariate logistic analysis (Supplementary Table 2). HCT in 30–33%, 27–30% or 27% were identified as independent risk factors for AKI in the model adjusted by the 17 variables, with OR of 1.739 (95% CI: 1.178–2.566, p = 0.005), 1.802 (95% CI: 1.115–2.911, p = 0.016) and 2.502 (95% CI: 1.467–4.266, p 0.001), respectively. In the validation cohort, we also found that low HCT was an independent risk factor for AKI in AMI patients in the multivariate logistic model, with OR of 1.660 (95% CI: 1.051–2.618, p = 0.030) and 1.613 (95% CI: 1.022–2.552, p = 0.039) for HCT in 27–30% or 27% respectively (Table 2). To assess the presence of multicollinearity among the 17 variables, a VIF test was performed. All the VIF values were below 4.0, indicating that there was no significant multicollinearity. The average VIF values in the discovery and validation cohorts were 1.49 and 1.16, respectively.

The marginal effect analysis was employed to evaluate the association between HCT and AKI. Our findings revealed a clear negative linear correlation between HCT and AKI (Fig. 1A). With an increase in the initial HCT upon ICU admission, there was a linear decline in the probability of AKI. In the validation cohort, the linear relationship between HCT and AKI was also found in the marginal effect analysis (Fig. 1B).

Fig. 1.

Relationship between initial HCT and risk of AKI evaluated by marginal effect analysis in patients with AMI. (A) Marginal effect in the discovery cohort. (B) Marginal effect in the validation cohort. HCT, hematocrit; AKI, acute kidney injury; AMI, acute myocardial infarction.

As shown in Supplementary Fig. 1, after adjusting for the 17 variables in the logistic model, low HCT was found to be significantly associated with hospital mortality. The OR for HCT between 27–30% and below 27% were 1.557 (95% CI: 1.007–2.407, p = 0.047) and 1.755 (95% CI: 1.169–2.634, p = 0.007), respectively. Furthermore, in the multivariate logistic model, there was a significant association between low HCT and the need for RRT during hospitalization. The OR for HCT levels in mild reduction, 30–33%, 27–30%, and below 27% were 2.353 (95% CI: 1.510–3.670, p 0.001), 3.159 (95% CI: 1.878–5.312, p 0.001), 3.575 (95% CI: 1.991–26.418, p 0.001), and 3.800 (95% CI: 2.182–6.618, p 0.001), respectively (Supplementary Fig. 2).

In order to reduce a possible sex stratification effect, we performed a subgroup analysis separated by sex. After adjusting for other variables in the multivariate logistic model, HCT below 27% was independently associated with an increased risk of AKI in both male and female patients. The OR in male patients was 2.214 (95% CI: 1.416–3.463, p 0.001), while in female patients, the OR was 2.714 (95% CI: 1.686–4.369, p 0.001) (Fig. 2). The predictive marginal effect analysis showed a linear relationship between HCT and AKI in both male and female patients (Supplementary Fig. 3).

Fig. 2.

Subgroup analysis of sex on the relationship between HCT and AKI assessed by the multivariate logistic regression. HCT, hematocrit; AKI, acute kidney injury.