Skip to main content
Renal Failure logoLink to Renal Failure
. 2024 Aug 23;46(2):2394634. doi: 10.1080/0886022X.2024.2394634

Development and validation of a predictive model for acute kidney injury in patients with ureterolithiasis

Yufeng Jiang a,b,, Jingcheng Zhang a,, Ailiyaer Ainiwaer a, Yuchao Liu a, Jing Li b, Liuliu Zhou c, Yang Yan b,d,, Haimin Zhang b,
PMCID: PMC11346321  PMID: 39177235

Abstract

Objectives

This study aims to identify risk factors for acute kidney injury (AKI) in patients with ureterolithiasis and to develop a predictive model for early AKI detection in this population.

Methods

A retrospective analysis was conducted on data from 1,016 patients with ureterolithiasis who presented to our outpatient emergency department between January 2021 and December 2022. Using multifactorial logistic regression, we identified independent risk factors for AKI and constructed a nomogram to predict AKI risk. The predictive model’s efficacy was assessed through the area under the ROC curve, calibration curves, Hosmer-Lemeshow (HL) test, and decision curve analysis (DCA).

Results

AKI was diagnosed in 18.7% of the patients. Independent risk factors identified included age, fever, diabetes, hyperuricemia, bilateral calculi, functional solitary kidney, self-medication, and prehospital delay. The nomogram demonstrated excellent discriminatory capabilities, with AUCs of 0.818 (95% CI, 0.775–0.861) for the modeling set and 0.782 (95% CI, 0.708–0.856) for the validation set. Both calibration curve and HL test results confirmed strong concordance between the model’s predictions and actual observations. DCA highlighted the model’s significant clinical utility.

Conclusions

The predictive model developed in this study provides clinicians with a valuable tool for early identification and management of patients at high risk for AKI, thereby potentially enhancing patient outcomes.

Keywords: Acute kidney injury, predictive modeling, logistic regression, ureterolithiasis

Introduction

Urolithiasis impacts 1–20% of the global population and has seen an increase in prevalence over recent decades [1,2]. When stones migrate to the ureter, termed ureterolithiasis, they frequently cause acute ureteral obstruction—a common emergency department presentation leading to complications such as renal colic, hematuria, lower urinary tract symptoms, and notably, acute kidney injury (AKI) [3,4].

AKI, as a critical sequela of ureterolithiasis, catalyzes further adverse outcomes such as the progression of chronic kidney disease, increased mortality, higher readmission rates, recurrence of AKI, and a decline in overall quality of life [5–7]. Despite the clinical significance of AKI (AKI), its identification and diagnosis present challenges. Symptoms of AKI are often nonspecific, and the absence of precise laboratory markers leads to its under-recognition [8]. A study utilizing electronic health records indicated that approximately 15% of hospitalized patients experienced AKI [9]. However, these estimates likely fall short of the actual incidence, hindered by the limitations inherent in monitoring and documentation practices. This underestimation is particularly pronounced in patients with ureterolithiasis. For instance, Wang et al. reported an incidence of just 0.72% [10], a figure significantly lower than those reported in other studies [11,12]. This discrepancy may be attributed to regional variations in diagnostic capabilities, particularly in less developed regions where AKI diagnosis is frequently inadequate [12]. The risk factors identified for AKI in patients with ureterolithiasis include delayed treatment, diabetes, bilateral ureterolithiasis, stones larger than 10 millimeters, preexisting renal impairment, Gram-negative bacterial infections from urine cultures, advanced age, increased body mass, and stones in the lower ureter [10,13,14]. The presence of these risk factors significantly increases the likelihood of developing AKI, emphasizing the need for heightened awareness and early identification.

With the advancements in medical informatics and artificial intelligence, predictive models for AKI have been further developed and are now widely utilized in the early recognition and diagnosis of the disease. Specifically, models designed for pediatric patients and those used in cardiothoracic surgery are now able to predict the risk of AKI with high precision within critical time frames [15,16]. These advancements provide clinicians with timely and essential data, crucial for initiating early interventions that could prevent the progression of AKI and associated complications. Moreover, models employing recurrent neural networks have revolutionized the monitoring process by offering updates on patient conditions every 15 min post-operation [17]. This high-frequency updating enables a dynamic assessment of the patient’s risk, allowing healthcare providers to adjust treatment strategies promptly and effectively. Such technological advancements are particularly valuable in high-pressure settings like emergency departments and outpatient clinics, where quick decision-making is essential.

This study introduces a new predictive model targeted at early detection and effective treatment of AKI in patients with renal colic in outpatient and emergency settings. By bridging current knowledge gaps, this model leverages advanced predictive analytics to potentially transform patient management and improve outcomes for those at risk of developing AKI.

Methods

Study design

We conducted a retrospective analysis in accordance with the guidelines set by the Institutional Ethics Committee of Chongming Branch, Shanghai Tenth People’s Hospital, Tongji University School of Medicine (IEC 2023-YF-01). Due to the study’s retrospective nature, informed consent was waived. We analyzed electronic medical records of patients who presented to outpatient and emergency departments with renal colic due to ureterolithiasis. Patients were identified using ICD-10-CM and ICD9 codes (N20, N20.1, N20.9, 592, 592.1, 592.9) related to ureterolithiasis. The study spanned from January 2021 to December 2022. Eligibility required radiographic confirmation of stones (via computed tomography or ultrasound). We excluded individuals under 18, those with prior renal impairment, pregnant women, and patients with incomplete data affecting AKI evaluation. For patients with multiple visits within a month, only the initial visit was considered.

Data acquisition and definition

Informed by an extensive review of existing literature and expert opinions [11,18,19], we selected 19 candidate variables to construct the predictive model. These variables, pertinent to outpatient and emergency department settings, are readily available and can be reliably obtained, thus ensuring their practicality for real-time use. The selection process emphasized their potential predictive power for AKI outcomes. Each variable included in the model was chosen based on its proven data availability and quality within the specified clinical settings, thereby guaranteeing reliable and accurate predictions. The variables incorporated into the model are age, gender, accompanying clinical symptoms following an episode of renal colic (such as nausea, vomiting, gross hematuria, and/or dysuria), fever, pyuria (defined as ≥5 urinary white blood cells per high-powered field), stone size, stone location, comorbidities (including hypertension, coronary artery disease, diabetes mellitus (DM), and hyperuricemia), history of urolithiasis, bilateral calculi, functional solitary kidney, self-medication (defined as oral drug administration by patients to treat the current condition before hospitalization), and prehospital delay (defined as the time interval between the onset of symptoms indicative of renal colic and the patient’s arrival at the outpatient clinic or emergency department).

By extracting serum creatinine at the time of patient admission for the diagnosis of AKI. AKI, based on KDIGO [20], is defined as meeting any of the following: An increase in serum creatinine by ≥0.3 mg/dl (≥26.5 µmol/l) within 48 h; An increase in serum creatinine to ≥1.5 times the baseline within the past 7 days; Urine volume ≤0.5 mL/kg/h for 6 h. Baseline creatinine was defined as the lowest creatinine level within 7 days prior to admission or the median of all outpatient creatinine values measured between 7 and 365 days before admission. If no pre-enrolment creatinine was available, or the patient’s serum creatinine was abnormal at admission, we estimated the baseline creatinine level using the Modification of Diet in Renal Disease (MDRD) equation, assuming a baseline glomerular filtration rate (GFR) of 75 mL/min/1.73 m2 [21].

Statistical analysis

To address missing data within our dataset, we applied the Random Forest algorithm for imputing missing values in categorical variables and utilized mean imputation for continuous variables. We evaluated the normality of quantitative data and presented normally distributed variables as mean ± standard deviation (SD). Differences between the training and validation group components were assessed using the t-test for continuous variables and the Chi-squared test or Fisher’s exact test for categorical variables, as appropriate.

The data set was split into a training set and a validation set at a ratio of 7:3. The model was developed using the training set and its performance was evaluated on the validation set. We examined multicollinearity using the variance inflation factor (VIF) method, with a threshold of VIF ≥ 5 indicating significant multicollinearity [22]. Significant covariates (p < 0.05) were identified through univariate logistic analysis for inclusion in the multivariate regression model. Data are presented as odds ratios and adjusted odds ratios with a 95% confidence interval.

We have developed a nomogram for predicting AKI based on the results of regression analysis. Using the nomogram, add up the scores for each variable in the chart to obtain a total score. Then, locate the corresponding value on the ‘Total Points’ axis and draw a straight line downwards to the ‘Disease Risk’ axis to determine the risk of acute kidney injury. Model performance was evaluated using a comprehensive approach [23,24]. Discriminatory power was assessed by the area under the receiver operating characteristic curve (AUC). Internal validation was performed using 1,000 bootstrap replicates in the training set, while external validation was conducted in the validation set. Model accuracy was further compared using calibration plots and the Hosmer-Lemeshow goodness-of-fit test to measure the concordance between predicted probabilities and observed outcomes. Additionally, we performed Decision Curve Analysis (DCA) to determine the clinical utility and net benefit of the model at different decision thresholds. The vertical axis represents the standardized net benefit, while the two horizontal axes depict the relationship between the risk threshold and the cost-benefit ratio. DCA helps clinicians to assess whether implementing the model improves patient management, thereby substantiating the model’s practical value in clinical settings [25].

All statistical tests were two-sided, with a significance level set at p < 0.05. Data were analyzed using IBM SPSS Statistics (version 23.0; IBM Corp., Armonk, NY, USA). Nomograms were constructed as predictive models using the ‘rms’ package in R (version 4.1.0; R Foundation for Statistical Computing, Vienna, Austria), and their performance was evaluated through calibration, the Hosmer-Lemeshow test, ROC curves, decision curve analysis (DCA).

Result

Characteristics of the training and validation sets

Following the application of exclusion criteria, a total of 1,016 patients were included in the study. The cohort was divided, with 712 patients allocated to the training set and 304 to the validation set, as depicted in Figure 1. The demographics and clinical characteristics of the patients in both sets are detailed in Table 1. AKI was diagnosed in 18.7% of the overall study population. The incidence of AKI was comparable between the two groups, with 18.8% in the training set and 18.4% in the validation set (p = 0.881). There were no significant differences in majority of the clinical factors across the two datasets, with p-values ranging from 0.09 to 0.99. However, a notable exception was the history of urolithiasis, which differed significantly between the groups (p = 0.005).

Figure 1.

Figure 1.

Flow chart of the study population.

Table 1.

Demographic and clinical characteristics of the training and validation Sets derived from the original data.

Characteristic Overall  (n = 1016, %) Training set (n = 712, %) Validation set (n = 304, %) p *
AKI (%)       0.881
 No 826 (81.3) 578 (81.2) 248 (81.6)  
 Yes 190 (18.7) 134 (18.8) 56 (18.4)  
Sex (%)       0.569
 Male 608 (59.8) 422 (59.3) 186 (61.2)  
 Female 408 (40.2) 290 (40.7) 118 (38.8)  
Age (years)       0.864
  Mean (SD) 60.55 (13.73) 60.50 (13.74) 60.66 (13.73)  
DM (%)       0.918
 No 656 (64.6) 459 (64.5) 197 (64.8)  
 Yes 360 (35.4) 253 (35.5) 107 (35.2)  
Hypertension (%)       0.659
 No 758 (74.6) 534 (75.0) 224 (73.7)  
 Yes 258 (25.4) 178 (25.0) 80 (26.3)  
Stone size (%)       0.855
 <10 mm 819 (80.6) 575 (80.8) 244 (80.3)  
 ≥10 mm 197 (19.4) 137 (19.2) 60 (19.7)  
Stone location (%)       0.355
 Upper 297 (29.2) 215 (30.2) 82 (27.0)  
 Middle 234 (23.0) 156 (21.9) 78 (25.7)  
 Lower 485 (47.7) 341 (47.9) 144 (47.4)  
Functional solitary kidney (%)       0.057
 No 977 (96.2) 690 (96.9) 287 (94.4)  
 Yes 39 (3.8) 22 (3.1) 17 (5.6)  
Bilateral calculi (%)       0.99
 No 966 (95.1) 677 (95.1) 289 (95.1)  
 Yes 50 (4.9) 35 (4.9) 15 (4.9)  
Pyuria (%)       0.862
 No 802 (78.9) 561 (78.8) 241 (79.3)  
 Yes 214 (21.1) 151 (21.2) 63 (20.7)  
Self-medication (%)       0.855
 No 822 (80.9) 575 (80.8) 247 (81.2)  
 Yes 194 (19.1) 137 (19.2) 57 (18.8)  
Prehospital delay (hours)        
(mean (SD)) 16.63 (14.61) 16.83 (14.94) 16.16 (13.82) 0.505
Hyperuricemia (%)       0.995
 No 712 (70.1) 499 (70.1) 213 (70.1)  
 Yes 304 (29.9) 213 (29.9) 91 (29.9)  
History of urolithiasis (%)       0.005
 No 706 (69.5) 476 (66.9) 230 (75.7)  
 Yes 310 (30.5) 236 (33.1) 74 (24.3)  
CAD (%)       0.09
 No 742 (73.0) 509 (71.5) 233 (76.6)  
 Yes 274 (27.0) 203 (28.5) 71 (23.4)  
Nausea (%)       0.444
 No 476 (46.9) 328 (46.1) 148 (48.7)  
 Yes 540 (53.1) 384 (53.9) 156 (51.3)  
Vomiting (%)       0.627
 No 716 (70.5) 505 (70.9) 211 (69.4)  
 Yes 300 (29.5) 207 (29.1) 93 (30.6)  
Fever (%)       0.469
 No 983 (96.8) 687 (96.5) 296 (97.4)  
 Yes 33 (3.2) 25 (3.5) 8 (2.6)  
Gross hematuria (%)       0.539
 No 776 (76.4) 540 (75.8) 236 (77.6)  
 Yes 240 (23.6) 172 (24.2) 68 (22.4)  
Dysuria (%)       0.783
 No 834 (82.1) 586 (82.3) 248 (81.6)  
 Yes 182 (17.9) 126 (17.7) 56 (18.4)  
*

p value was derived from a univariate association analysis conducted between the two datasets. Unless explicitly stated otherwise, the data represent the count of patients, with the corresponding percentages enclosed in parentheses. For the comparison of categorical variables, Pearson’s Chi-square test was utilized. In the case of continuous variables, utilizing the t-test for group comparisons.

AKI: acute kidney injury; CAD: Coronary Artery Disease; DM: Diabetes Mellitus; SD: Standard Deviation.

AKI risk prediction model construction

In the construction of the AKI risk prediction model, eight candidate variables were identified as significantly associated with AKI among patients presenting with renal colic, as determined by univariate logistic regression analyses. These variables include age, fever, DM, hyperuricemia, bilateral calculi, functional solitary kidney, self-medication, and prehospital delay, each demonstrating a significant correlation (all p < 0.05) (Table 2). Subsequent multifactorial regression analysis confirmed that these variables independently contribute to the risk of developing AKI in this patient population (Table 2).

Table 2.

Univariate and multivariate logistic regression analysis of clinical candidate predictors in the training set.

  Univariable
Multivariable
Variable Crude OR (95%CI) p value Adj OR (95%CI) p value
Sex        
 Female 0.78 [0.52, 1.14] 0.2 NA NA
 Age(years) 1.034 [1.02, 1.05] <0.001 1.05 [1.03, 1.07] <0.001
Bilateral calculi        
 Yes 9.77 [4.81, 20.85] <0.001 18.56 [7.65, 47.82] <0.001
Functional solitary kidney        
 Yes 16.65 [6.44, 51.47] <0.001 44.15 [15.08, 150.83] <0.001
Pyuria        
 Yes 1.27 [0.81, 1.97] 0.283 NA NA
Self-medication        
 Yes 3.21 [2.11, 4.87] <0.001 3.89 [2.24, 6.81] <0.001
Fever        
 Yes 3.60 [1.56, 8.11] 0.002 9.69 [3.28, 28.40] <0.001
History of urolithiasis        
 Yes 1.36 [0.92, 2.00] 0.123 NA NA
Hypertension        
 Yes 0.71 [0.44, 1.12] 0.151 NA NA
CAD        
 Yes 1.13 [0.75, 1.70] 0.553 NA NA
DM        
 Yes 3.21 [2.19, 4.74] <0.001 3.55 [2.16, 5.92] <0.001
Hyperuricemia        
 Yes 3.60 [2.44, 5.32] <0.001 3.92 [2.38, 6.52] <0.001
Stone location        
 Upper 1 [Reference] NA NA NA
 Middle 0.76 [0.44, 1.28] 0.304 NA NA
 Lower 0.86 [0.56, 1.32] 0.476 NA NA
Stone size        
 ≥10 mm 0.954 [0.580, 1.523] 0.849 NA NA
 Prehospital delay(hours) 1.040 [1.029, 1.053] <0.001 1.05 [1.04, 1.07] <0.001
Nausea        
 Yes 1.15 [0.79, 1.68] 0.473 NA NA
Vomiting        
 Yes 0.87 [0.57, 1.32] 0.532 NA NA
Gross hematuria        
 Yes 0.75 [0.47, 1.18] 0.23 NA NA
Dysuria        
 Yes 0.90 [0.53, 1.46] 0.667 NA NA

CAD, Coronary Artery Disease; DM, Diabetes Mellitus; NA, not applicable; OR, odds ratio; SD, Standard Deviation.

The probability of AKI is notably elevated in patients with either a functional solitary kidney or bilateral calculi who are experiencing renal colic. Given the significant clinical implications of these conditions, which typically draw considerable attention from healthcare professionals, we decided to exclude these variables from our model to focus on more modifiable risk factors. Consequently, using the six remaining independent risk factors, we developed a nomogram to visually represent the predictive model for estimating the likelihood of AKI in patients presenting with ureterolithiasis, as illustrated in Figure 2.

Figure 2.

Figure 2.

Nomogram developed using logistic regression analysis to assess the risk of Acute kidney injury in patients with renal colic.

To determine the scores for each variable in the nomogram, trace a vertical line from the variable value to the “Points” axis. Sum up the scores for all predictors and find the corresponding value on the “Total Points” axis. Next, draw a straight line downwards to the “Disease Risk” axis to determine the risk of acute kidney injury. DM: Diabetes Mellitus.

Discrimination ability of predictive models

The Receiver Operating Characteristic (ROC) curves of the nomogram in both the training and validation sets are depicted in Figure 3. The nomogram demonstrated a robust discrimination capacity for identifying AKI. In the training set, the AUC was 0.818 (95% CI, 0.775–0.861), while in the validation set, it reached 0.782 (95% CI, 0.708–0.856). The optimal cutoff value, determined to maximize the Youden index, was established at 0.253 for the modeling group. This value yielded a specificity of 84.4% and a sensitivity of 68.7%. Applying the same cutoff value in the validation group, the model maintained a specificity of 83.9% and a sensitivity of 66.1%.

Figure 3.

Figure 3.

Evaluation of nomogram performance using the receiver operating characteristic curve.

AUC: area under the curve; CI: confidence interval.

Calibration ability of predictive models

The calibration curve, illustrated in Figure 4, showed excellent concordance between the predictions made by our model and the actual observations in both the training and validation datasets. Additionally, the Hosmer-Lemeshow goodness-of-fit test produced non-significant P-values of 0.24 in the training set and 0.251 in the validation set, further confirming the model’s strong calibration capacity. Together, the calibration curve and the Hosmer-Lemeshow test affirm a high level of accuracy in the predictions relative to the observed outcomes.

Figure 4.

Figure 4.

Calibration curves of the nomogram in the training dataset (a) and validation dataset.

The x-axis represents the predicted probability calculated by the nomogram, and the y-axis is the observed actual probability of acute kidney injury. Results of the Hosmer-Lemeshow test demonstrate that the p-value of the training dataset is 0.24 and the validation dataset is 0.251, respectively.

Clinical usefulness of predictive models

Figure 5 demonstrates that the predictive model’s net benefit exceeds that of categorizing all individuals in the study as either high-risk or low-risk across both the training and validation sets. This result highlights the substantial clinical utility of the model, emphasizing its capacity to deliver a more precise risk assessment.

Figure 5.

Figure 5.

Decision curve and net benefit analyses of the prediction model in training(a) and validation(b) Sets.

Decision Curve Analysis (DCA) describes the net clinical benefit of the AKI prediction model. (a) The DCA was performed on the AKI prediction model in the training set. (b) The DCA was performed on the AKI prediction model in the validation set. The vertical axis displays the standardized net benefit. The two horizontal axes indicate the correspondence between the risk threshold and the cost: benefit ratio. The black solid horizontal line represents the assumption that no patients with AKI were involved, while the gray solid line indicates the assumption that all patients had AKI. The blue solid line represents the intervention of the AKI prediction model. AKI: acute kidney injury.

Discussion

Our research draws on an analysis of a dataset comprising 1,016 patients, among whom 18.7% were diagnosed with AKI. We identified multiple independent risk factors for AKI, including advanced age, presence of fever, DM, hyperuricemia, bilateral calculi, a functional solitary kidney, self-medication, and prehospital delay. These findings align with existing literature, and importantly, our study underscores often underestimated risks such as self-medication [12,26]. The cornerstone of our study is the development of risk prediction models for assessing AKI risk in outpatient and emergency settings involving patients with ureterolithiasis. The model demonstrated strong discriminative capability, evidenced by an AUC of 0.818 in the training set and 0.782 in the validation set. This predictive strength facilitates the preemptive identification of individuals at elevated risk for AKI, thereby enabling timely interventions before the initiation of therapeutic or pharmacological measures. Adopting this proactive approach can significantly enhance clinical decision-making, potentially improving the management and prognosis of patients at risk of developing AKI.

Within the dynamic realm of medical research, the utilization of clinical risk prediction models is becoming increasingly widespread. These models are crafted using a blend of cutting-edge artificial intelligence and machine learning techniques alongside traditional statistical regression methods, establishing themselves as fundamental tools in clinical decision-making. Despite the growing fascination with AI and machine learning, models based on logistic regression continue to hold a significant position due to their unparalleled transparency and interpretability—attributes that are especially valued in certain clinical contexts [27–29]. In light of this, our study employed logistic regression to develop a predictive tool for assessing the risk of AKI, which is visualized through user-friendly nomograms. This model is specifically designed to enhance the early detection of AKI risk in emergency and outpatient settings, contrasting with the prevailing focus on predictive models primarily aimed at inpatient outcomes [9,30,31].

Our method requires the inclusion of only six variables to assess AKI, simplifying the prediction process. This simplification facilitates rapid risk identification and intervention from the initial evaluation of patients presenting with ureterolithiasis. Such a methodology significantly increases the model’s usability and practicality in real-world clinical settings, distinguishing it from other models that depend on a more extensive array of variables for effective prediction [32,33]. This strategic focus not only broadens the application of risk prediction in diverse clinical environments but also underscores our commitment to improving timely and effective patient management. Our model underwent rigorous internal ­validation, a standard protocol for most existing models [34], and demonstrated robust and consistent predictive accuracy. In a comparative analysis with a study employing logistic regression to predict AKI in COVID-19 patients, their model showed higher predictive discrimination in the training cohort (AUC = 0.89, p < 0.001) but experienced a decrease to an AUC of 0.78 (p = 0.030) in the validation cohort, underperforming relative to ours. This suggests that while their model initially indicated high predictive power, ours maintained superior stability during the validation phase [35]. Further comparison was conducted with another study using a neural network-based machine learning algorithm to predict AKI following total knee arthroplasty. In this instance, our model’s AUC in the training cohort surpassed the comparative model’s 0.78 (95% CI, 0.74–0.81), showcasing our model’s excellent discriminative ability. However, in the validation cohort, the neural network-based model achieved a higher AUC value of 0.89, highlighting the dynamic performance landscape across different predictive models and scenarios [36]. This varied performance highlights the nuanced complexity inherent in predictive modeling within medical research. The strong performance of our model in the training phase likely reflects its effective capture of underlying risk factors and their interactions, specifically tailored to our dataset’s characteristics. Conversely, the superior performance of the neural network model in the validation cohort suggests its robust generalizability and capacity to decode complex patterns across diverse datasets. This observation underscores a critical consideration in predictive model development: the balance between a model’s specificity to a particular dataset and its ability to generalize across different populations.

In our study, predictive variables were sourced from electronic medical records in emergency and outpatient settings, carefully chosen for their potential impact on AKI risk prediction. This led to differences in predictor combinations compared to other studies [37,38], underscoring the need for flexibility and adaptability in model construction to suit specific study contexts and objectives. By tailoring our approach, we enhanced the model’s applicability and accuracy, enabling precise identification and mitigation of AKI at crucial clinical stages. This strategy ensures that patients receive more accurate interventions and tailored treatment plans, improving outcomes and healthcare experiences.

Age is a well-established risk factor for AKI. As individuals age, physiological functions and metabolic activities decline, inherently increasing the susceptibility to AKI. This correlation is substantiated by studies demonstrating a progressive decrease in nephron count and cortical volume with age [39]. Interestingly, research indicates that individuals in their 70s can maintain normal function with just half the original number of nephrons, highlighting the body’s ability to adapt to reduced nephron capacity. However, this adaptation comes at a significant cost—increased vulnerability to conditions like AKI [12,40].

Our study identified comorbid conditions such as DM and hyperuricemia as significant risk factors for AKI. DM is particularly linked to AKI onset due to potential kidney tubule injury or intrarenal atherosclerosis, making diabetic patients especially vulnerable [41,42]. Similarly, a meta-analysis indicated that hyperuricemia significantly elevates AKI risk [26]. Elevated serum uric acid levels, often associated with chronic kidney disease, can exacerbate the likelihood of AKI due to both direct and indirect pathways. Direct injury may arise from chronic renal insufficiency, while indirect injury can stem from vasoconstriction, oxidative stress, and inflammation [43,44]. Consequently, patients with elevated serum uric acid levels, potentially harboring subclinical chronic renal conditions, are more prone to AKI, especially under conditions like unilateral or bilateral ureteral obstruction.

Ureterolithiasis obstruction is considered a risk factor for AKI in patients with urinary tract infections [45]. However, our study was unable to confirm the presence of urinary tract infections through urine culture results. We included pyuria as a variable and did not find an association with AKI. The lack of this association may be due to the presence of white blood cells in the urine caused by the irritation from the stones in cases of ureterolithiasis. Additionally, it is important to consider that many previous studies involved hospitalized or ICU patients, who typically present with more complex disease profiles [46,47]. These factors may account for the inconsistencies between our findings and those of previous research.

Our study highlights the critical roles of self-medication and prehospital delays in patients at risk for AKI. Drug-induced AKI is notably prevalent, constituting 37.50% of AKI cases acquired in hospitals among inpatients [48]. The risk of AKI escalates with the use of multiple nephrotoxic drugs, leading to recommendations for discontinuing any unnecessary nephrotoxic medications [49,50]. Moreover, in managing pain associated with ureterolithiasis-induced renal colic, the commonly used first-line analgesics—non-steroidal anti-inflammatory drugs—are recognized for their nephrotoxic potential. This recognition underscores the necessity for an early assessment of AKI risk to mitigate potential kidney damage [51,52].

In addition, previous study has indicated that stone size greater than 10 mm and urine cultures showing Gram-negative bacterial growth are significant predictors of AKI in the pre-intervention period [13]. However, it is essential to note that our study population consists of emergency and outpatient department patients presenting with renal colic, which differs from the hospitalized patients undergoing surgical intervention as reported in the cited study. The latter group often presents with larger stones, resulting in a higher incidence of positive findings. Moreover, hospitalized patients typically experience a longer duration of symptoms, increasing the likelihood of concurrent infections. These differences in study populations may account for the variability in the reported outcomes.

Our model stands out for its robust discriminatory power, precise calibration ability, and significant clinical applicability, positioning it as a practical tool for clinicians. Unlike other models [9,30,31], ours excels in the early identification of patients at high risk for AKI in outpatient and emergency settings. Predominantly reliant on real-time clinical characteristics instead of delayed laboratory data, our model facilitates earlier predictions. This proactive approach enables clinicians to tailor treatment strategies effectively, such as avoiding nephrotoxic drugs and implementing timely drainage measures to relieve obstruction. Additionally, the early recognition of high-risk patients and the swift initiation of preventive actions are crucial in reducing AKI-related morbidity and mortality, enhancing patient outcomes significantly [53,54].

However, our study does have some limitations. Due to the availability of data, we diagnosed AKI using only serum creatinine, excluding urine output as a criterion. The retrospective design may introduce biases, and the accuracy of our results depends on the reliability and completeness of the analyzed medical records. Our research was conducted in a single hospital, which may limit the generalizability of the results; therefore, it is necessary to validate our findings in different settings. To address these limitations, future studies could be conducted prospectively across multiple centers to enhance the diversity and representativeness of the patient samples. Given the absence of a mature, universal AKI early warning system [9], we recommend adopting this modeling approach and combining it with local data to develop simplified predictive models. This will facilitate the early identification and diagnosis of AKI in patients with ureterolithiasis.

Conclusion

This study successfully developed and validated a robust predictive model that can identify AKI in patients with ureterolithiasis at an early stage in outpatient and emergency settings. This model provides clinicians with a practical decision-making tool, offering the potential to improve the management and prognosis of AKI in patients with ureterolithiasis.

Supplementary Material

S1.docx

Funding Statement

This work was supported by the Science and Technology Commission of Shanghai Municipality [22ZR1448400]; the Experimental Animal Fund of Shanghai Science and Technology Commission [201409004000]; and the Science and Technology Commission of Chongming District, Shanghai [CKY-2020-24; CKY-2023-47].

Ethics statement

Ethical approval for this study (reference number IEC 2023-YF-01) was granted by the Institutional Ethics Committee of the Chongming Branch, Shanghai Tenth People’s Hospital, Tongji University School of Medicine. The study conformed to the principles outlined in the Helsinki Declaration II. The requirement for informed consent was waived by the ethics committee, given that all data were anonymized prior to collection. Access to the data was restricted solely to the principal investigator, ensuring stringent confidentiality. All participant information was securely maintained throughout the duration of the study, adhering to strict privacy and confidentiality standards.

Disclosure statement

No potential conflict of interest was reported by the author(s).

References

  • 1.Sorokin I, Mamoulakis C, Miyazawa K, et al. Epidemiology of stone disease across the world. World J Urol. 2017;35(9):1301–1320. doi: 10.1007/s00345-017-2008-6. [DOI] [PubMed] [Google Scholar]
  • 2.Mayans L. Nephrolithiasis. Prim Care. 2019;46(2):203–212. doi: 10.1016/j.pop.2019.02.001. [DOI] [PubMed] [Google Scholar]
  • 3.Schoenfeld EM, Pekow PS, Shieh MS, et al. The diagnosis and management of patients with renal colic across a sample of US Hospitals: high CT utilization despite low rates of admission and inpatient urologic intervention. PLoS One. 2017;12(1):e0169160. doi: 10.1371/journal.pone.0169160. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 4.Balthazar P, Sadigh G, Hughes D, et al. Increasing use, geographic variation, and disparities in emergency department CT for suspected urolithiasis. J Am Coll Radiol. 2019;16(11):1547–1553. doi: 10.1016/j.jacr.2019.05.033. [DOI] [PubMed] [Google Scholar]
  • 5.See EJ, Jayasinghe K, Glassford N, et al. Long-term risk of adverse outcomes after acute kidney injury: a systematic review and meta-analysis of cohort studies using consensus definitions of exposure. Kidney Int. 2019;95(1):160–172. doi: 10.1016/j.kint.2018.08.036. [DOI] [PubMed] [Google Scholar]
  • 6.Sawhney S, Marks A, Fluck N, et al. Acute kidney injury as an independent risk factor for unplanned 90-day hospital readmissions. BMC Nephrol. 2017;18(1):9. doi: 10.1186/s12882-016-0430-4. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 7.Holmes J, Geen J, Williams JD, et al. Recurrent acute kidney injury: predictors and impact in a large population-based cohort. Nephrol Dial Transplant. 2020;35(8):1361–1369. doi: 10.1093/ndt/gfz155. [DOI] [PubMed] [Google Scholar]
  • 8.Wang JJ, Chi NH, Huang TM, et al. Urinary biomarkers predict advanced acute kidney injury after cardiovascular surgery. Crit Care. 2018;22(1):108. doi: 10.1186/s13054-018-2035-8. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 9.Wilson FP, Martin M, Yamamoto Y, et al. Electronic health record alerts for acute kidney injury: multicenter, randomized clinical trial. BMJ. 2021;372:m4786. doi: 10.1136/bmj.m4786. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 10.Wang SJ, Mu XN, Zhang LY, et al. The incidence and clinical features of acute kidney injury secondary to ureteral calculi. Urol Res. 2012;40(4):345–348. doi: 10.1007/s00240-011-0414-6. [DOI] [PubMed] [Google Scholar]
  • 11.Farris N, Raina R, Tibrewal A, et al. Acute kidney injury associated with urinary stone disease in children and young adults presenting to a pediatric emergency department. Front Pediatr. 2020;8:591520–591520. doi: 10.3389/fped.2020.591520. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 12.Hoste EAJ, Kellum JA, Selby NM, et al. Global epidemiology and outcomes of acute kidney injury. Nat Rev Nephrol. 2018;14(10):607–625. doi: 10.1038/s41581-018-0052-0. [DOI] [PubMed] [Google Scholar]
  • 13.Kamath SU, Patil B, Patwardhan SK.. To study the incidence and preintervention factors associated with acute kidney injury in patients diagnosed with ureteric calculi. Urol Ann. 2019;11(4):380–384. doi: 10.4103/UA.UA_96_18. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 14.Mahmud HM, Mahmud SM.. Acute kidney injury in Ureteric Stones: single centre short term analysis. Pak J Med Sci. 2017;33(4):808–812. doi: 10.12669/pjms.334.13345. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 15.Sandokji I, Yamamoto Y, Biswas A, et al. A time-updated, parsimonious model to predict AKI in hospitalized children. J Am Soc Nephrol. 2020;31(6):1348–1357. doi: 10.1681/ASN.2019070745. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 16.Bihorac A. Response to Comment on "MySurgeryRisk”: development and validation of a machine-learning risk algorithm for major complications and death after surgery. Ann Surg. 2019;269(6):e74. doi: 10.1097/SLA.0000000000002884. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 17.Rank N, Pfahringer B, Kempfert J, et al. Deep-learning-based real-time prediction of acute kidney injury outperforms human predictive performance. NPJ Digit Med. 2020;3(1):139. doi: 10.1038/s41746-020-00346-8. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 18.Kellum JA, Romagnani P, Ashuntantang G, et al. Acute kidney injury. Nat Rev Dis Primers. 2021;7(1):52. doi: 10.1038/s41572-021-00284-z. [DOI] [PubMed] [Google Scholar]
  • 19.Scholz H, Boivin FJ, Schmidt-Ott KM, et al. Kidney physiology and susceptibility to acute kidney injury: implications for renoprotection. Nat Rev Nephrol. 2021;17(5):335–349. doi: 10.1038/s41581-021-00394-7. [DOI] [PubMed] [Google Scholar]
  • 20.Khwaja A. KDIGO clinical practice guidelines for acute kidney injury. Nephron Clin Pract. 2012;120(4):c179–c184. doi: 10.1159/000339789. [DOI] [PubMed] [Google Scholar]
  • 21.Levey AS, Bosch JP, Lewis JB, et al. A more accurate method to estimate glomerular filtration rate from serum creatinine: a new prediction equation. Modification of Diet in Renal Disease Study Group. Ann Intern Med. 1999;130(6):461–470. doi: 10.7326/0003-4819-130-6-199903160-00002. [DOI] [PubMed] [Google Scholar]
  • 22.Vatcheva KP, Lee M, McCormick JB, et al. Multicollinearity in regression analyses conducted in epidemiologic studies. Epidemiology. 2016;6(2):27. doi: 10.4172/2161-1165.1000227. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 23.Collins GS, Dhiman P, Ma J, et al. Evaluation of clinical prediction models (part 1): from development to external validation. BMJ. 2024;384:e074819. doi: 10.1136/bmj-2023-074819. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 24.Riley RD, Archer L, Snell KIE, et al. Evaluation of clinical prediction models (part 2): How to undertake an external validation study. BMJ. 2024;384:e074820. doi: 10.1136/bmj-2023-074820. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 25.Huang Y, Li W, Macheret F, et al. A tutorial on calibration measurements and calibration models for clinical prediction models. J Am Med Inform Assoc. 2020;27(4):621–633. doi: 10.1093/jamia/ocz228. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 26.Xu X, Hu J, Song N, et al. Hyperuricemia increases the risk of acute kidney injury: a systematic review and meta-analysis. BMC Nephrol. 2017;18(1):27. doi: 10.1186/s12882-016-0433-1. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 27.Nusinovici S, Tham YC, Chak Yan MY, et al. Logistic regression was as good as machine learning for predicting major chronic diseases. J Clin Epidemiol. 2020;122:56–69. doi: 10.1016/j.jclinepi.2020.03.002. [DOI] [PubMed] [Google Scholar]
  • 28.Christodoulou E, Ma J, Collins GS, et al. A systematic review shows no performance benefit of machine learning over logistic regression for clinical prediction models. J Clin Epidemiol. 2019;110:12–22. doi: 10.1016/j.jclinepi.2019.02.004. [DOI] [PubMed] [Google Scholar]
  • 29.Lynam AL, Dennis JM, Owen KR, et al. Logistic regression has similar performance to optimised machine learning algorithms in a clinical setting: application to the discrimination between type 1 and type 2 diabetes in young adults. Diagn Progn Res. 2020;4(1):6. doi: 10.1186/s41512-020-00075-2. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 30.Malhotra R, Kashani KB, Macedo E, et al. A risk prediction score for acute kidney injury in the intensive care unit. Nephrol Dial Transplant. 2017;32(5):814–822. doi: 10.1093/ndt/gfx026. [DOI] [PubMed] [Google Scholar]
  • 31.Churpek MM, Carey KA, Edelson DP, et al. Internal and external validation of a machine learning risk score for acute kidney injury. JAMA Netw Open. 2020;3(8):e2012892. doi: 10.1001/jamanetworkopen.2020.12892. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 32.Zhou J, Bai Y, Wang X, et al. A simple risk score for prediction of sepsis associated-acute kidney injury in critically ill patients. J Nephrol. 2019;32(6):947–956. doi: 10.1007/s40620-019-00625-y. [DOI] [PubMed] [Google Scholar]
  • 33.Li Y, Xu J, Wang Y, et al. A novel machine learning algorithm, Bayesian networks model, to predict the high-risk patients with cardiac surgery-associated acute kidney injury. Clin Cardiol. 2020;43(7):752–761. doi: 10.1002/clc.23377. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 34.Feng Y, Wang AY, Jun M, et al. Characterization of risk prediction models for acute kidney injury: a systematic review and meta-analysis. JAMA Netw Open. 2023;6(5):e2313359. doi: 10.1001/jamanetworkopen.2023.13359. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 35.Hectors SJ, Riyahi S, Dev H, et al. Multivariate analysis of CT imaging, laboratory, and demographical features for prediction of acute kidney injury in COVID-19 patients: a Bi-centric analysis. Abdom Radiol. 2021;46(4):1651–1658. doi: 10.1007/s00261-020-02823-w. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 36.Ko S, Jo C, Chang CB, et al. A web-based machine-learning algorithm predicting postoperative acute kidney injury after total knee arthroplasty. Knee Surg Sports Traumatol Arthrosc. 2022;30(2):545–554. doi: 10.1007/s00167-020-06258-0. [DOI] [PubMed] [Google Scholar]
  • 37.Dong J, Feng T, Thapa-Chhetry B, et al. Machine learning model for early prediction of acute kidney injury (AKI) in pediatric critical care. Crit Care. 2021;25(1):288. doi: 10.1186/s13054-021-03724-0. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 38.Demirjian S, Bashour CA, Shaw A, et al. Predictive accuracy of a perioperative laboratory test-based prediction model for moderate to severe acute kidney injury after cardiac surgery. JAMA. 2022;327(10):956–964. doi: 10.1001/jama.2022.1751. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 39.Denic A, Lieske JC, Chakkera HA, et al. The substantial loss of nephrons in healthy human kidneys with aging. J Am Soc Nephrol. 2017;28(1):313–320. doi: 10.1681/ASN.2016020154. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 40.Lu R, Muciño-Bermejo M-J, Armignacco P, et al. Survey of acute kidney injury and related risk factors of mortality in hospitalized patients in a third-level urban hospital of Shanghai. Blood Purif. 2014;38(2):140–148. doi: 10.1159/000366127. [DOI] [PubMed] [Google Scholar]
  • 41.Peng J, Li X, Zhang D, et al. Hyperglycemia, p53, and mitochondrial pathway of apoptosis are involved in the susceptibility of diabetic models to ischemic acute kidney injury. Kidney Int. 2015;87(1):137–150. doi: 10.1038/ki.2014.226. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 42.Vallon V. Do tubular changes in the diabetic kidney affect the susceptibility to acute kidney injury? Nephron Clin Pract. 2014;127(1–4):133–138. doi: 10.1159/000363554. [DOI] [PubMed] [Google Scholar]
  • 43.Lee TS, Lu TM, Chen CH, et al. Hyperuricemia induces endothelial dysfunction and accelerates atherosclerosis by disturbing the asymmetric dimethylarginine/dimethylarginine dimethylaminotransferase 2 pathway. Redox Biol. 2021;46:102108. doi: 10.1016/j.redox.2021.102108. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 44.Ponticelli C, Podestà MA, Moroni G.. Hyperuricemia as a trigger of immune response in hypertension and chronic kidney disease. Kidney Int. 2020;98(5):1149–1159. doi: 10.1016/j.kint.2020.05.056. [DOI] [PubMed] [Google Scholar]
  • 45.Hsiao C-Y, Chen T-H, Lee Y-C, et al. Urolithiasis is a risk factor for uroseptic shock and acute kidney injury in patients with urinary tract infection. Front Med. 2019;6:288–288. doi: 10.3389/fmed.2019.00288. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 46.Królicki T, Bardowska K, Kudla T, et al. Acute kidney injury secondary to urinary tract infection in kidney transplant recipients. Sci Rep. 2022;12(1):10858. doi: 10.1038/s41598-022-15035-7. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 47.Bobbili RK, Deepanjali S, Rajesh NG, et al. Acute kidney injury in patients hospitalized with febrile urinary tract infections: a prospective observational study. Clin Nephrol. 2021;95(3):127–135. doi: 10.5414/CN110229. [DOI] [PubMed] [Google Scholar]
  • 48.Liu C, Yan S, Wang Y, et al. Drug-induced hospital-acquired acute kidney injury in China: a multicenter cross-sectional survey. Kidney Dis. 2021;7(2):143–155. doi: 10.1159/000510455. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 49.Ostermann M, Zarbock A, Goldstein S, et al. Recommendations on acute kidney injury biomarkers from the acute disease quality initiative consensus conference: a consensus statement. JAMA Netw Open. 2020;3(10):e2019209. doi: 10.1001/jamanetworkopen.2020.19209. [DOI] [PubMed] [Google Scholar]
  • 50.Goldstein SL, Kirkendall E, Nguyen H, et al. Electronic health record identification of nephrotoxin exposure and associated acute kidney injury. Pediatrics. 2013;132(3):e756–e767. doi: 10.1542/peds.2013-0794. [DOI] [PubMed] [Google Scholar]
  • 51.Kunitsu Y, Hira D, Morikochi A, et al. Time until onset of acute kidney injury by combination therapy with "Triple Whammy" drugs obtained from Japanese Adverse Drug Event Report database. PLoS One. 2022;17(2):e0263682. doi: 10.1371/journal.pone.0263682. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 52.Pathan SA, Mitra B, Cameron PA.. A systematic review and meta-analysis comparing the efficacy of nonsteroidal anti-inflammatory drugs, opioids, and paracetamol in the treatment of acute renal colic. Eur Urol. 2018;73(4):583–595. doi: 10.1016/j.eururo.2017.11.001. [DOI] [PubMed] [Google Scholar]
  • 53.Bhatraju PK, Zelnick LR, Chinchilli VM, et al. Association between early recovery of kidney function after acute kidney injury and long-term clinical outcomes. JAMA Netw Open. 2020;3(4):e202682-e202682. doi: 10.1001/jamanetworkopen.2020.2682. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 54.Siew ED, Abdel-Kader K, Perkins AM, et al. Timing of recovery from moderate to severe AKI and the risk for future loss of kidney function. Am J Kidney Dis. 2020;75(2):204–213. doi: 10.1053/j.ajkd.2019.05.031. [DOI] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

S1.docx

Articles from Renal Failure are provided here courtesy of Taylor & Francis

RESOURCES