Acute Kidney Injury Risk Prediction Score for Critically-ill Surgical Patients

Konlawij Trongtrakul; Jayanton Patumanond; Suneerat Kongsayreepong; Sunthiti Morakul; Tanyong Pipanmekaporn; Osaree Akaraborworn; Sujaree Poopipatpab


BMC Anesthesiol. 2020;20(140) 

In This Article


Overview of AKI

A series of 4652 cases in THAI-SICU study were assessed for their eligibility. Patients aged below 18 (n = 28), medical reasons for surgical ICU admission (n = 998), and admitted to ICU less 24 h (n = 152) were excluded from the analysis. Study flow is provided in Additional file 1: Figure S1. Finally, 3474 cases were eligible for developing an AKI prediction score. Of these, 333 (9.6%) cases experienced AKI within 7 days of ICU admission.

Figure S1.

Study flow

In general, the AKI group were older, had more males, and a higher illness severity than the non-AKI group. No differences in patients' pre-existing comorbidities were found between groups. More multiple abnormalities were identified on basic investigations in the AKI than in the non-AKI patients; including more anemia, less albuminemia, higher serum creatinine, lower PiO2/FiO2 ratio, abnormal chest imaging, and abnormal ECG. Regarding surgical intervention, AKI had higher class of ASA classification, more frequent emergency surgery, and had undergone more abdomino-colorectal surgery than non-AKI. A shorter duration of operative time, with a greater blood loss, and a lesser urine output were also found in AKI than in non-AKI. As for the outcomes, there was significantly greater risk of ICU and day-28 mortality in AKI than non-AKI, together with longer ICU length of stay and hospital length of stay in the AKI group (Table 1). Univariable logistic regression analysis relating each predictor to AKI is shown in Additional file 1: Table S1.

Predictors That Determined AKI Within 7 Days of ICU Admission

Table 2 shows the best AKI predictors using multivariable logistic regression analysis. The final selected predictors included age of patient, SOFA non-renal score, sepsis, emergency surgery, perioperative blood loss, and perioperative urine output. After arranging into a scoring system, the AKI prediction score was ranged between 0 to 16.5. Figure 1 illustrates the number of cases distributed according to each score level comparing AKI and non-AKI. The AKI prediction score had a good discriminative ability with AuROC = 0.839; 95%CI, 0.825–0.852 (Figure 2), and fitted to the original dataset by HL-GOF, p value = 0.302. The sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) of the score was 72.3, 80.6, 28.8, and 96.4%, respectively. The model remained a good discriminative ability, AuROC = 0.821; 95%CI, 0.797–0.845, after internal validating by bootstrapping method (1000 replications).

Figure 1.

Percentage distribution of each AKI prediction score categorized by AKI and non-AKI

Figure 2.

The discriminative ability of acute kidney injury prediction score in critically-ill surgical patients reported by Area under the Receiver Operating Characteristic Curve (AuROC)

The higher the score the greater the risk of AKI, and the predicted risk of the score was closely correlated with the reality (observed risk), as shown graphically (Figure 3). We then classified the score into fourth probability risk of AKI; as low, moderate, high, and very high risk. The LH+ of AKI was at 0.117, 0.927, 5.190, and 9.982, respectively (Table 3).

Figure 3.

Observed risk (circle) vs predicted risk (solid line) of AKI, the size of circle represents frequency of patients in each score level