With the help of the ADE-ASAS-2, we retrospectively monitored a total of 51,772 patients in our hospital. 1270 patients triggered alarm signals and 332 of them were diagnosed with D-AKI. This study found that the total incidence of D-AKI in our adult inpatient population was about 0.64%. At present, there is still a lack of accurate incidence of D-AKI in all hospitalized populations. Previous reports in the literature have shown that the incidence of AKI in inpatients ranges from 0.7–77% depending on the definition and study population [22, 23], with a drug-related rate of 25% [24]. Therefore, although the incidence of D-AKI that we obtained is lower than that in general studies, it is still within a reasonable range.
Currently, AKI remains a common complication associated with high mortality and prolonged hospital stay[3, 24]. Previous studies have demonstrated that AKI can be prevented by early intervention in individual[10]. However, most of the current studies on AKI prediction models are based on specific patient populations[14, 15, 21], especially to predict the risk of contrast-related AKI during percutaneous coronary intervention (PCI)[13, 25]. Currently, there is no nomogram for predicting D-AKI risk based on the general Chinese patient population to accurately assess the risk of D-AKI in Chinese hospitalized patients in a multi-drug environment. Therefore, this research was dedicated to developing a simple and easy-to-use D-AKI prediction nomogram for adult inpatients to help clinicians to more accurately identify patients with potential D-AKI risk and take measures to prevent the occurrence of D-AKI.
The D-AKI risk prediction nomogram was developed by the development group and verified in the validation group. Previous research experience has shown that an AUC value greater than 0.7 indicates good predictive performance of the model [21, 26]. In our model, the AUC values for the development and validation groups were 0.787 (95%CI: 0.752–0.823) and 0.788 (95%CI: 0.736–0.840), respectively. The calibration of the model was verified by the GiViTI calibration belt. The P-values were greater than 0.05 in both the development and validation groups. The above shows that our model has a good ability to distinguish D-AKI patients from non-D-AKI patients without overestimating or underestimating the risk of occurrence.
The nomogram model predicts D-AKI in hospitalized patients based on six predictors, including baseline RBC count, neutrophil count ≥ 7×109/L, use of NSAIDs, use of diuretics, alcohol abuse, and previous CKD. Again, these six selected predictors have been shown to be strongly associated with kidney damage in several studies[3, 15, 21, 27, 28]. In addition, we found that in the initial population of this study, D-AKI patients were older than non-D-AKI patients (median 65 vs. 56 years; P < 0.001), and advanced age has become a major risk factor for AKI due to changes in renal structure and function in older adults, which has been confirmed and incorporated into many AKI risk models[13, 15, 26]. In order to find more targeted D-AKI predictors, we balanced demographic data including age; length of hospitalization and duration of suspected drug exposure by propensity score matching. The predictors we finally found were routinely available in the HIS.
Our study concluded that the reduction in erythrocyte count and neutrophil count ≥ 7×109/L has predictive significance for D-AKI. At present, there are few studies on the relationship between erythrocyte count and AKI, and the mechanism remains unknown. Both inflammation and oxidative stress may play a key role in the progression of AKI[29, 30]. When inflammation occurs, iron metabolism and bone marrow function are affected and the proliferation and maturation of erythrocytes are inhibited[29], resulting in a decrease in erythrocytes counts. In addition, neutrophil count as a marker of inflammation has been shown to provide additional information on the prognosis of AKI[31, 32]. Therefore, the role of neutrophil count in the prediction of D-AKI should be valued clinically.
In this nomogram model, we found that the combination of non-steroidal anti-inflammatory drugs and diuretics contributed approximately 32 and 45 points to the predicted total score, respectively. Previous studies have shown that the two aforementioned drugs together with angiotensin-converting enzyme constitute a "triple whammy " theory[31, 33, 34]. The pathological mechanism by which NSAIDs precipitate hemodynamically mediated kidney injury is the inhibition of renal prostaglandins, causing renal vasoconstriction to occur preferentially in the afferent arteries [27, 35]. In addition, use of diuretics was found to be a risk factor for D-AKI in many studies[17, 31, 36]. The mechanism is to stimulate the sympathetic nervous system and the renin-angiotensin system (RAS), which leads to hemodynamic changes and eventually renal perfusion deficit leading to AKI [37]. CKD is another important predictor of D-AKI in inpatients. The relationship between CKD and AKI has been mentioned in many studies[4, 21, 27]. Recent studies haves shown that both are mutual risk factors and risk factors for cardiovascular disease [4]. Similarly, eGFR, the main diagnostic indicator of CKD, has been used as an independent predictor by multiple AKI prediction models[26, 38]. We also found a significant correlation between alcohol abuse and D-AKI. The latest research has pointed out that frequent and occasional binge drinking are associated with a 2.2-fold and 2.0-fold higher risk of CKD progression, respectively, compared with no alcohol consumption [39]. Hence, clinicians should focus on the renal function of hospitalized patients with a history of chronic kidney disease or alcohol abuse.
The main strength of our study is the first analysis of D-AKI episodes in Chinese hospitalized adult patients by the self-developed ADE-ASAS-2 in a multi-drug environment. We also constructed a D-AKI risk prediction nomogram, which was well identified and calibrated. Compared with the predictors involved in other studies[15, 27], our model was based on 6 variables that are widely used and easily accessible in clinical practice, and thus be applied in various medical environments. In addition, we found that both amphotericin B and its liposomes had a high incidence of D-AKI and that the incidence differed between dosage forms, which provides a direction for the next comparative study with a large sample.
However, there are several limitations of our study. First, this was a single-center retrospective study and sample selection bias was inevitable. For the same reason, we did not perform external verification, and the obtained results need to be verified by future prospective multi-center studies. Besides, the construction of this model was based on a case-control study with the risk of observation bias and confounding bias, which was minimized by applying the propensity score matching protocol. Furthermore, some variables that affect the predictive performance of D-AKI, such as cystatin C and interleukin 18[40, 41], were ignored due to the high percentage of missing values. If these variables are combined, the predictive value may be improved. However, the sample size of this study was insufficient to adequately analyze additional variables, which means that prospective studies with more detailed data and studies with larger sample sizes are needed to verify or update.