Acute kidney injury-attributable mortality in critically ill patients with sepsis

Participants

Critically ill adult patients were enrolled from two distinct retrospective ICU cohorts: (1) Medical Information Mart for Intensive Care IV (MIMIC IV) cohort and (2) Wenzhou cohort study (Zhou et al., 2021). The MIMIC IV cohort was enrolled from a relational database containing comprehensive information on over 250,000 patients hospitalized between 2008 and 2019 at Beth Israel Deaconess Medical Center in Boston, MA, USA. The Wenzhou cohort included critically ill adult patients from ICUs at the Second Affiliated Hospital of Wenzhou Medical University in Wenzhou, Zhejiang, China. The MIMIC IV public database was approved by the institutional review board (IRB). Wenzhou cohort was approved by the Second Affiliated Hospital of Wenzhou Medical University IRB. Informed consent was waived due to retrospective nature of the study.

Patients

Inclusion criteria for adult patients followed the Third International Consensus Definitions for Sepsis and Septic Shock (Sepsis-3) i.e., a known or suspected infection plus acute increase Sequential Organ Failure Assessment (SOFA) ≥ 2 points for organ dysfunction (Shankar-Hari et al., 2016; Singer et al., 2016) from the MIMIC IV and Wenzhou cohorts. We excluded patients with a history of chronic kidney disease (glomerulonephritis, diabetic nephropathy, hypertensive nephropathy, hereditary nephritis, and chronic kidney failure caused by a variety of other diseases), multiple hospitalizations (only the first hospitalization was considered), and ICU length of stay (LOS) less than 24 h. In both cohorts, the SOFA score (Vincent et al., 1996), Acute Physiology Score (APS) III (Knaus et al., 1991), Logistic Organ Dysfunction Score (LODS) (Le Gall et al., 1996), and Oxford Acute Severity of Illness Score (OASIS) (Johnson, Kramer & Clifford, 2013) were employed to evaluate the severity of illness. Calculations for the modified SOFA score, modified APS III and modified LODS were obtained through the exclusion of points associated with renal function. The shock and respiratory failure variables were both collected. We defined shock as the need for vasopressor within the first 48 h of hospital admission, while respiratory failure was defined as the need for invasive mechanical ventilation.

Outcomes

AKI is the primary outcome. Patients were defined as having AKI if they met the Kidney Disease: Improving Global Outcomes (KDIGO) serum creatinine diagnostic criteria for AKI (Supplemental Information) (Kellum et al., 2012). The presence of AKI and its severity were defined according to KDIGO criteria. The secondary outcomes included hospital and ICU mortality, hospital LOS and ICU LOS.

Statistical methods

Wilcoxon rank-sum test, Student’s t-test, and Chi-squared test were employed to compare the baseline characteristic variables. Before data analysis, the potential confounders and mediating variables between AKI and death were depicted in the directed acyclic graph (DAG) (Fig. 1) (Lederer et al., 2019). Potential confounders are variables related to both the exposure and outcome of interest, and therefore should be controlled for in analyses. Variables that may be on the indirect causal path between the exposure and outcome partially mediate the association between exposure and outcome. It would be incorrect to control for these factors, as that would partially close the causal path, attenuating the observed association between the exposure and outcome (Lederer et al., 2019).

Guided by the conceptual model illustrated in the DAG, multivariate logistic, and linear binomial regression models were developed for primary and secondary outcomes. Specifically, we assessed each of the variables identified on the DAG and listed in Table 1 for inclusion in the final multivariable model along with several pre-specified variables considered clinically significant pre-hoc. Pre-specified variables included illness severity score (modified APS III, SOFA score, modified LODS and OASIS), age, gender, race, and shock. Chronic co-morbidity variables for the final model were selected after assessing how each potential confounder affected the overall odds ratio (OR) when added to a multivariable model with the pre-specified variables. In generating final models, we retained only variables that altered the OR by greater than 10%, so as to avoid over-fitting.

We performed a sensitivity analysis of hospital and ICU LOS for all patients (both survivors and non-survivors). Sepsis-associated AKI was re-defined according to the urine output diagnostic criteria of KDIGO for AKI (sensitivity analysis with urine output criteria, considered only urine output criteria, not the worse between serum creatinine and urine output criteria) (Supplemental Information). Furthermore, sensitivity analyses of hospital and ICU mortality were conducted based on the urine output diagnostic criteria.

The attributable fraction (AF) of mortality from AKI (A F_AKI) and the population AF of mortality from AKI (population A F_AKI) were calculated as reported previously (detail for this calculation is provided in Supplemental Information) (Auriemma et al., 2020; Van Vught et al., 2016). The A F_AKI denoted the proportion of deaths attributable to AKI in septic patients with AKI. Population A F_AKI denoted the proportion of all deaths in the sepsis population attributable to AKI. Estimated value was generated by indirect standardization, performed within strata (additional details for this calculation are provided in Supplemental Information). All statistical analyses were conducted in R (version 3.6.1) used in our previous study (Weng et al., 2021; Xu et al., 2021); p-value < 0.05 denoted statistical significance.

Baseline characteristics and outcomes

Figure 2 illustrates patient selection flow chart, whereas Table 2 outlines the baseline patient characteristics. The Wenzhou cohort tended to be older with higher vasopressor use probability compared to those of the MIMIC IV cohort. The baseline modified SOFA score, modified APS III, modified LODS and OASIS were similar between the two cohorts. The proportion of patients requiring continuous renal replacement therapy (CRRT) were similar, however, more patients acquired AKI in the Wenzhou cohort compared to the MIMIC IV cohort. We reported more cases of stage 1 AKI in the Wenzhou population compared to the MIMIC IV population. While hospital and ICU LOS were longer in the Wenzhou cohort, hospital and ICU mortalities were higher in the MIMIC IV cohort.

Table 2 shows participant characteristics stratified by AKI status. in both cohorts, patients with AKI demonstrated a greater need for mechanical ventilation, vasopressor use, CRRT, and higher illness severity scores than patients without AKI; they also were characterized by higher mortality and longer LOS.

Comparison of clinical outcomes adjusted for severity of illness

MIMIC IV

We reported overall hospital mortality of 13.5%; briefly, 2,118 of 15,610 patients died before discharge (Table 3). Nearly 66% of non-survivors developed AKI, whereas 28% of survivors developed AKI (p < 0.001). More patient characteristics were shown in Table 3.

The unadjusted hospital mortality of sepsis with AKI was 27%, while that for sepsis without AKI was 7% (Table 4; OR = 4.82; 95% CI [4.37–5.31]; p < 0.001). In constructing the adjusted model, no other variables except the prespecified variables (illness severity score, age, gender, race, and shock) met the set criteria. In the multivariable regression model, the OR values for hospital mortality of patients with AKI were attenuated but remained statistically significant after adjustment for illness severity score (modified APS III, SOFA score, modified LODS, and OASIS), age, gender, race, and shock. Sensitivity analyses based on the urine output diagnostic criteria of KDIGO for AKI yielded similar results (Table S1). Septic patients who developed AKI, experienced longer hospital and ICU LOS than patients without AKI, whether among survivors or across all patients (Tables S2 and S3).

Furthermore, we conducted stratified analyses based on the severity of AKI. And found that stages 1, 2, and 3 AKI were all independently associated with hospital and ICU mortality in adjusted and unadjusted models (Fig. 3, Table S4). In four adjusted models, stage 3 AKI exhibited the most significant association with increased risk of hospital and ICU mortality. In the MIMIC IV cohort, the AF_AKI was 58.6% (CI [46.8%–70.3%]), whereas the population AF_AKI was 30.2% (95% CI [22.7%–37.8%]).

Wenzhou

Among 1,341 patients, 155 patients died before discharge, with overall hospital mortality of 10.3% (Table 3). Nearly 67% of non-survivors developed AKI, whereas 33% of survivors developed AKI (p < 0.001). More patient characteristics were shown in Table 3.

The unadjusted hospital mortality of sepsis with AKI was 21% while that for sepsis without AKI was 6% (Table 4; OR = 4.16; 95% CI [2.93–5.98]; p < 0.001). Similar to findings in the MIMIC IV cohort, development of AKI in the Wenzhou population was significantly associated with increased risk of hospital and ICU mortality in multivariate logistic regression after we adjusted for illness severity score (modified APS III, SOFA score, modified LODS and OASIS), age and shock. Similarly, in the sensitivity analyses, AKI was associated with ICU mortality, when we applied the urine output diagnostic criteria of KDIGO for AKI (Table S1).

As in the MIMIC IV cohort, patients with AKI had prolonged hospital and ICU (Tables S2 and S3). Moreover, the correlation of AKI with mortality was stratified according to the severity of AKI. In the Wenzhou cohort, stages 1, 2, and 3 AKI were independently associated with hospital and ICU mortality (Fig. 3, Table S4). In the Wenzhou cohort, the A F_AKI was 44.6% (95% CI [12.7%–76.4%]), whereas the population A F_AKI was 26.0% (95% CI [0%–56.8%]).