Results

During the study period, there were 113, 723 adult admissions, of which 15,114 (13.3%) were not eligible for this analysis and a further 28,372 (25.9% of the total) were excluded because they had no data on SCr during hospitalisation (Figure 1). After a random selection of one admission per patient, the final study data set included 48,835 patients. Ambulatory preadmission SCr was ascertained in 20.7% of admissions; for the remaining the baseline SCr was imputed.

AKI incidence

AKI was identified in 7,427 patients giving an overall incidence of 15.2%. In 2,847 patients, criteria of AKI were fulfilled within the first 24 hours after admission, accounting for 5.8% of all hospital admissions and 38.3% of all AKI episodes. In 4,580 patients, AKI was not present within the first day, but developed later during their stay, with the earliest SCr keeping with the definition, on average, about five days after admission, irrespective the HA-AKI stage (p=0.05) (Figure 2).

Patients’ characteristics

In comparison to No AKI patients, those with AKI were older, more likely to be admitted via the emergency room, had more comorbidities and more often required critical care (Table 1).
Patients with CA- and HA-AKI had a very similar distribution of comorbidities, except for CKD, which we found more prevalent in those with AKI at admission. The type of AKI also related to the principal diagnosis, with genitourinary system and infectious diseases being most frequent in CA-AKI and circulatory system disease in HA-AKI patients. Serum creatinine at admission was, on average, twice as high as the baseline value in patients with AKI at admission. These patients were also more likely to endure more severe AKI (23.2% stage 2 and 30.9% stage 3 in CA-AKI vs 21.4% and 14.5%, in HA-AKI, respectively).

Outcomes

Greater severity of AKI when apparent at admission translated into a higher number of patients requiring in-patient RRT and slightly higher in-hospital mortality (8.3% and 23.3%, respectively, vs 3.9% and 20.6% in HA-AKI) (Table 1). Patients with CA-AKI had a median length of hospital stay of 8.1 days (P25, P75: 4.8, 14.4) which was shorter, on average, almost seven days than a length of stay in HA-AKI group. In the latter, significant differences in the total length of stay between severity of AKI (p<0.001) arose from the number of days after, not the number of days before AKI detection (Figure 2). Among patients who survived hospitalisation, 6.4 % of CA-AKI and 5.3% of HA-AKI died up to 6th-month after discharge.
Generally, the outcomes of our interest worsened according to staging, whether AKI was present at admission or acquired during hospital stay (Figure 3). One should notice that an initially low in-hospital mortality in HA-AKI stage 2 and 3 started steadily increasing and in the post-discharge period exceeded the mortality rate observed in the group of most severe CA-AKI.
After multivariable adjustment in the Cox model, the gradual relationship between the severity of AKI and the risk of the adverse outcome remained significant (Table 2). Patients with HA-AKI stage 1 were least likely to die during hospitalisation among all AKI patients, but still, over twice more likely than No AKI patients (hazard ratio (HR) 2.28, 95% confidence interval (CI): 2.03–2.56). Patients in stage 3 of both CA- and HA-AKI were in a much higher risk of death, either during hospital (HR 5.65, 95% CI: 4.81–6.63 and HR 6.70, 95% CI: 5.84–7.68) or within six months of discharge (HR 2.50, 95% CI: 1.79–3.49 and HR 2.18, 95% CI: 1.53-3.11), than patients in other stages.

Hazard function

The survival regression with the Weibull distribution of time to AKI occurrence coincided better with the function determined by Kaplan-Meier curve and showed a better fit to our data in comparison to the model with an exponential distribution of time (AIC 47437 vs 47588) (Figure 4 A). The probability of AKI occurring in subsequent days decreased from 1.68 in the first day to 1.27 in the 30th day of hospitalisation (Figure 4 B).