Prediction Model For Acute Kidney Injury Research Articles

Machine learning-based model to predict severe acute kidney injury after total aortic arch replacement for acute type A aortic dissection

BackgroundSevere acute kidney injury (AKI) after total aortic arch replacement (TAAR) is related to adverse outcomes in patients with acute type A aortic dissection (ATAAD). However, the early prediction of severe AKI remains a challenge. This study aimed to develop a novel model to predict severe AKI after TAAR in ATAAD patients using machine learning algorithms. MethodsA total of 572 ATAAD patients undergoing TAAR were enrolled in this retrospective study, and randomly divided into a training set (70 %) and a validation set (30 %). Lasso regression, support vector machine-recursive feature elimination and random forest algorithms were used to screen indicators for severe AKI (defined as AKI stage III) in the training set, respectively. Then the intersection indicators were selected to construct models through artificial neural network (ANN) and logistic regression. The AUC-ROC curve was employed to ascertain the prediction efficacy of the ANN and logistic regression models. ResultsThe incidence of severe AKI after TAAR was 22.9 % among ATAAD patients. The intersection predictors identified by different machine learning algorithms were baseline serum creatinine and ICU admission variables, including serum cystatin C, procalcitonin, aspartate transaminase, platelet, lactic dehydrogenase, urine N-acetyl-β-d-glucosidase and Acute Physiology and Chronic Health Evaluation II score. The ANN model showed a higher AUC-ROC than logistic regression (0.938 vs 0.908, p < 0.05). Furthermore, the ANN model could predict 89.1 % of severe AKI cases beforehand. In the validation set, the superior performance of the ANN model was further confirmed in terms of discrimination ability (AUC = 0.916), calibration curve analysis and decision curve analysis. ConclusionThis study developed a novel and reliable clinical prediction model for severe AKI after TAAR in ATAAD patients using machine learning algorithms. Importantly, the ANN model showed a higher predictive ability for severe AKI than logistic regression.

Machine learning-based model to predict severe acute kidney injury after total aortic arch replacement for acute type A aortic dissection

BackgroundSevere acute kidney injury (AKI) after total aortic arch replacement (TAAR) is related to adverse outcomes in patients with acute type A aortic dissection (ATAAD). However, the early prediction of severe AKI remains a challenge. This study aimed to develop a novel model to predict severe AKI after TAAR in ATAAD patients using machine learning algorithms. MethodsA total of 572 ATAAD patients undergoing TAAR were enrolled in this retrospective study, and randomly divided into a training set (70%) and a validation set (30%). Lasso regression, support vector machine-recursive feature elimination and random forest algorithms were used to screen indicators for severe AKI (defined as AKI stage III) in the training set, respectively. Then the intersection indicators were selected to construct models through artificial neural network (ANN) and logistic regression. The AUC-ROC curve was employed to ascertain the prediction efficacy of the ANN and logistic regression models. ResultsThe incidence of severe AKI after TAAR was 22.9% among ATAAD patients. The intersection predictors identified by different machine learning algorithms were baseline serum creatinine and ICU admission variables, including serum cystatin C, procalcitonin, aspartate transaminase, platelet, lactic dehydrogenase, urine N-acetyl-β-D-glucosidase and Acute Physiology and Chronic Health Evaluation II score. The ANN model showed a higher AUC-ROC than logistic regression (0.938 vs 0.908, p <0.05). Furthermore, the ANN model could predict 89.1% of severe AKI cases beforehand. In the validation set, the superior performance of the ANN model was further confirmed in terms of discrimination ability (AUC=0.916), calibration curve analysis and decision curve analysis. ConclusionThis study developed a novel and reliable clinical prediction model for severe AKI after TAAR in ATAAD patients using machine learning algorithms. Importantly, the ANN model showed a higher predictive ability for severe AKI than logistic regression.

Validation of a Prediction Model for Acute Kidney Injury after Cardiac Surgery in a Retrospective Asian Cohort.

Background: The incidence of postoperative acute kidney injury (AKI) is relatively high in some Asian regions. The objective of this study was to examine the performance of an AKI prediction model developed based on data from a White-dominant population in a retrospective Asian cohort of patients undergoing cardiovascular surgery. Methods: We retrospectively identified 549 patients who underwent elective major cardiovascular surgery (coronary artery bypass graft, valve surgery, and aorta surgery), and excluded those who underwent a percutaneous cardiovascular procedure. Patients with a baseline estimated glomerular filtration rate (eGFR) < 60 mL/min/1.73 m2 were also excluded. AKI was defined according to the Kidney Disease: Improving Global Outcomes (KDIGO) definition. Performance of the prediction model for AKI was expressed as area under the receiver operating characteristic curve (AUC). Results: The prediction model had a good predictive accuracy for postoperative AKI (all AUC > 0.92). The AUC of the prediction model in subgroups of age (<65 years and ≥65 years), sex (male and female), hypertension, and diabetes were all >0.85 (all p values < 0.001). Conclusions: The model could be used to predict postoperative AKI in Asian patients undergoing cardiovascular surgery with a baseline eGFR ≥ 60 mL/min/1.73 m2.

Development of interpretable machine learning models for prediction of acute kidney injury after noncardiac surgery: a retrospective cohort study.

Early identification of patients at high-risk of postoperative acute kidney injury (AKI) can facilitate the development of preventive approaches. This study aimed to develop prediction models for postoperative AKI in noncardiac surgery using machine learning algorithms. The authors also evaluated the predictive performance of models that included only preoperative variables or only important predictors. Adult patients undergoing noncardiac surgery were retrospectively included in the study (76457 patients in the discovery cohort and 11910 patients in the validation cohort). AKI was determined using the KDIGO criteria. The prediction model was developed using 87 variables (56 preoperative variables and 31 intraoperative variables). A variety of machine learning algorithms were employed to develop the model, including logistic regression, random forest, extreme gradient boosting, and gradient boosting decision trees. The performance of different models was compared using the area under the receiver operating characteristic curve (AUROC). Shapley Additive Explanations (SHAP) analysis was employed for model interpretation. The patients in the discovery cohort had a median age of 52 years (IQR: 42-61 years), and 1179 patients (1.5%) developed AKI after surgery. The gradient boosting decision trees algorithm showed the best predictive performance using all available variables, or only preoperative variables. The AUROCs were 0.849 (95% CI: 0.835-0.863) and 0.828 (95% CI: 0.813-0.843), respectively. The SHAP analysis showed that age, surgical duration, preoperative serum creatinine, and gamma-glutamyltransferase, as well as American Society of Anesthesiologists physical status III were the most important five features. When gradually reducing the features, the AUROCs decreased from 0.852 (including the top 40 features) to 0.839 (including the top 10 features). In the validation cohort, the authors observed a similar pattern regarding the models' predictive performance. The machine learning models the authors developed had satisfactory predictive performance for identifying high-risk postoperative AKI patients. Furthermore, the authors found that model performance was only slightly affected when only preoperative variables or only the most important predictive features were included.

Interpretable machine learning-based individual analysis of acute kidney injury in immune checkpoint inhibitor therapy.

Acute kidney injury (AKI) is a critical complication of immune checkpoint inhibitor therapy. Since the etiology of AKI in patients undergoing cancer therapy varies, clarifying underlying causes in individual cases is critical for optimal cancer treatment. Although it is essential to individually analyze immune checkpoint inhibitor-treated patients for underlying pathologies for each AKI episode, these analyses have not been realized. Herein, we aimed to individually clarify the underlying causes of AKI in immune checkpoint inhibitor-treated patients using a new clustering approach with Shapley Additive exPlanations (SHAP). We developed a gradient-boosting decision tree-based machine learning model continuously predicting AKI within 7 days, using the medical records of 616 immune checkpoint inhibitor-treated patients. The temporal changes in individual predictive reasoning in AKI prediction models represented the key features contributing to each AKI prediction and clustered AKI patients based on the features with high predictive contribution quantified in time series by SHAP. We searched for common clinical backgrounds of AKI patients in each cluster, compared with annotation by three nephrologists. One hundred and twelve patients (18.2%) had at least one AKI episode. They were clustered per the key feature, and their SHAP value patterns, and the nephrologists assessed the clusters' clinical relevance. Receiver operating characteristic analysis revealed that the area under the curve was 0.880. Patients with AKI were categorized into four clusters with significant prognostic differences (p = 0.010). The leading causes of AKI for each cluster, such as hypovolemia, drug-related, and cancer cachexia, were all clinically interpretable, which conventional approaches cannot obtain. Our results suggest that the clustering method of individual predictive reasoning in machine learning models can be applied to infer clinically critical factors for developing each episode of AKI among patients with multiple AKI risk factors, such as immune checkpoint inhibitor-treated patients.

Acute kidney injury prediction model utility in premature myocardial infarction

The incidence of premature myocardial infarction (PMI) has been rising and acute kidney injury (AKI) occurring in PMI patients severely impacts prognosis. This study aimed to develop and validate a prediction model for AKI specific to PMI patients. The MIMIC-Ⅲ-CV and MIMIC-Ⅳ databases were utilized for model derivation of PMI patients. Single-center data served for external validation. There were 571 and 182 AKI patients in the training set (n= 937) and external validation set (n= 292) cohorts, respectively. Finally, a 7-variable model consisting of: Sequential Organ Failure Assessment (SOFA) score, coronary artery bypass grafting (CABG), ICU stay time, loop diuretics, estimated glomerular filtration rate (eGFR) HCO3- and Albumin was developed, achieving an AUC of 0.85 (95% CI: 0.83-0.88) in the training set. External validation also confirmed model robustness. This model may assist clinicians in the early identification of patients at elevated risk for PMI. Further validation is warranted before clinical application.

Prognostic significance of preoperative nutritional status for postoperative acute kidney injury in older patients undergoing major abdominal surgery: a retrospective cohort study.

The association between malnutrition and postoperative acute kidney injury (AKI) has not been well studied. In this study, the authors examined the association between preoperative nutritional status and postoperative AKI in older patients who underwent major abdominal surgery, as well as the predictive value of malnutrition for AKI. The authors retrospectively included patients aged 65 or older who underwent major elective abdominal surgery. The nutritional status of the patient was evaluated using three objective nutritional indices, such as the geriatric nutritional risk index (GNRI), the prognostic nutritional index (PNI), and the controlling nutritional status (CONUT). AKI was determined using the KDIGO criteria. The authors performed logistic regression analysis to investigate the association between preoperative nutritional status and postoperative AKI, as well as the predictive value of nutritional scores for postoperative AKI. A total of 2775 patients were included in the study, of which 707 (25.5%), 291 (10.5%), and 517 (18.6%) had moderate to severe malnutrition according to GNRI, PNI, and CONUT calculations. After surgery, 144 (5.2%) patients developed AKI, 86.1% at stage 1, 11.1% at stage 2, and 2.8% at stage 3 as determined by KDIGO criteria. After adjustment for traditional risk factors, worse nutritional scores were associated with a higher AKI risk. In addition to traditional risk factors, these nutritional indices improved the predictive ability of AKI prediction models, as demonstrated by significant improvements in integrated discrimination and net reclassification. Poor preoperative nutritional status, as assessed by GNRI, PNI, and CONUT scores, was associated with an increased risk of postoperative AKI. Incorporating these scores into AKI prediction models improved their performance. These findings emphasize the need for screening surgical patients for malnutrition risk. Further research is needed to determine whether preoperative malnutrition assessment and intervention can reduce postoperative AKI incidence.

Construction and validation of a risk assessment model for acute kidney injury in patients with acute pancreatitis in the intensive care unit

BackgroundTo construct and validate a risk assessment model for acute kidney injury (AKI) in patients with acute pancreatitis (AP) in the intensive care unit (ICU).MethodsA total of 963 patients diagnosed with acute pancreatitis (AP) from the Medical Information Mart for Intensive Care IV (MIMIC-IV) database was included. These patients were randomly divided into training set (N = 674) and validation set (N = 289) at a ratio of 7:3. Clinical characteristics were utilized to establish a nomogram for the prediction of AKI during ICU stay. These variables were selected by the least absolute shrinkage and selection operation (LASSO) regression and included in multivariate logistic regression analysis. Variables with P-values less than 0.05 were included in the final model. A nomogram was constructed based on the final model. The predicted accuracy of the nomogram was assessed by calculating the receiver operating characteristic curve (ROC) and the area under the curve (AUC). Moreover, calibration curves and Hosmer-Lemeshow goodness-of-fit test (HL test) were performed to evaluate model performance. Decision curve analysis (DCA) evaluated the clinical net benefit of the model.ResultsA multivariable model that included 6 variables: weight, SOFA score, white blood cell count, albumin, chronic heart failure, and sepsis. The C-index of the nomogram was 0.82, and the area under the receiver operating characteristic curve (AUC) of the training set and validation set were 0.82 (95% confidence interval:0.79–0.86) and 0.76 (95% confidence interval: 0.70–0.82), respectively. Calibration plots showed good consistency between predicted and observed outcomes in both the training and validation sets. DCA confirmed the clinical value of the model and its good impact on actual decision-making.ConclusionWe identified risk factors associated with the development of AKI in patients with AP. A risk prediction model for AKI in ICU patients with AP was constructed, and improving the treatment strategy of relevant factors in the model can reduce the risk of AKI in AP patients.

Federated machine learning for predicting acute kidney injury in critically ill patients: a multicenter study in Taiwan

PurposeTo address the contentious data sharing across hospitals, this study adopted a novel approach, federated learning (FL), to establish an aggregate model for acute kidney injury (AKI) prediction in critically ill patients in Taiwan.MethodsThis study used data from the Critical Care Database of Taichung Veterans General Hospital (TCVGH) from 2015 to 2020 and electrical medical records of the intensive care units (ICUs) between 2018 and 2020 of four referral centers in different areas across Taiwan. AKI prediction models were trained and validated thereupon. An FL-based prediction model across hospitals was then established.ResultsThe study included 16,732 ICU admissions from the TCVGH and 38,424 ICU admissions from the other four hospitals. The complete model with 60 features and the parsimonious model with 21 features demonstrated comparable accuracies using extreme gradient boosting, neural network (NN), and random forest, with an area under the receiver-operating characteristic (AUROC) curve of approximately 0.90. The Shapley Additive Explanations plot demonstrated that the selected features were the key clinical components of AKI for critically ill patients. The AUROC curve of the established parsimonious model for external validation at the four hospitals ranged from 0.760 to 0.865. NN-based FL slightly improved the model performance at the four centers.ConclusionA reliable prediction model for AKI in ICU patients was developed with a lead time of 24 h, and it performed better when the novel FL platform across hospitals was implemented.

Machine learning clinical prediction models for acute kidney injury: the impact of baseline creatinine on prediction efficacy

BackgroundThere are many Machine Learning (ML) models which predict acute kidney injury (AKI) for hospitalised patients. While a primary goal of these models is to support clinical decision-making, the adoption of inconsistent methods of estimating baseline serum creatinine (sCr) may result in a poor understanding of these models’ effectiveness in clinical practice. Until now, the performance of such models with different baselines has not been compared on a single dataset. Additionally, AKI prediction models are known to have a high rate of false positive (FP) events regardless of baseline methods. This warrants further exploration of FP events to provide insight into potential underlying reasons.ObjectiveThe first aim of this study was to assess the variance in performance of ML models using three methods of baseline sCr on a retrospective dataset. The second aim was to conduct an error analysis to gain insight into the underlying factors contributing to FP events.Materials and methodsThe Intensive Care Unit (ICU) patients of the Medical Information Mart for Intensive Care (MIMIC)-IV dataset was used with the KDIGO (Kidney Disease Improving Global Outcome) definition to identify AKI episodes. Three different methods of estimating baseline sCr were defined as (1) the minimum sCr, (2) the Modification of Diet in Renal Disease (MDRD) equation and the minimum sCr and (3) the MDRD equation and the mean of preadmission sCr. For the first aim of this study, a suite of ML models was developed for each baseline and the performance of the models was assessed. An analysis of variance was performed to assess the significant difference between eXtreme Gradient Boosting (XGB) models across all baselines. To address the second aim, Explainable AI (XAI) methods were used to analyse the XGB errors with Baseline 3.ResultsRegarding the first aim, we observed variances in discriminative metrics and calibration errors of ML models when different baseline methods were adopted. Using Baseline 1 resulted in a 14% reduction in the f1 score for both Baseline 2 and Baseline 3. There was no significant difference observed in the results between Baseline 2 and Baseline 3. For the second aim, the FP cohort was analysed using the XAI methods which led to relabelling data with the mean of sCr in 180 to 0 days pre-ICU as the preferred sCr baseline method. The XGB model using this relabelled data achieved an AUC of 0.85, recall of 0.63, precision of 0.54 and f1 score of 0.58. The cohort size was 31,586 admissions, of which 5,473 (17.32%) had AKI.ConclusionIn the absence of a widely accepted method of baseline sCr, AKI prediction studies need to consider the impact of different baseline methods on the effectiveness of ML models and their potential implications in real-world implementations. The utilisation of XAI methods can be effective in providing insight into the occurrence of prediction errors. This can potentially augment the success rate of ML implementation in routine care.

Prediction models combining zonulin, LPS, and LBP predict acute kidney injury and hepatorenal syndrome–acute kidney injury in cirrhotic patients

The development of acute kidney injury (AKI) and hepatorenal syndrome–acute kidney injury (HRS–AKI) in cirrhosis has been associated with intestinal barrier dysfunction and gut-kidney crosstalk. We use the related markers such as zonulin, lipopolysaccharides (LPS), and lipopolysaccharide-binding protein (LBP) to predict AKI and HRS–AKI in cirrhotic patients and evaluate their in vitro effects on intestinal (Caco-2) cells and renal tubular (HK-2) cells. From 2013 to 2020, we enrolled 70 cirrhotic patients and developed prediction models for AKI and HRS–AKI over a six-month period. There were 13 (18.6%) and 8 (11.4%) cirrhotic patients developed AKI and HRS–AKI. The prediction models incorporated zonulin, LPS, LBP, C-reactive protein, age, and history of hepatitis B for AKI, and zonulin, LPS, LBP, total bilirubin, and Child–Pugh score for HRS–AKI. The area under curve (AUC) for the prediction of AKI and HRS–AKI was 0.94 and 0.95, respectively. Furthermore, the conditioned medium of LPS+hrLBP pre-treated Caco-2 cells induced apoptosis, necrosis, and zonulin release in HK-2 cells, demonstrating the communication between them. This study found that zonulin, LPS, and LBP are potential practical markers for predicting AKI and HRS–AKI in cirrhotic patients, which may serve as potential targets for renal outcomes in cirrhotic patients.

Characterization of Risk Prediction Models for Acute Kidney Injury

Despite the expansion of published prediction models for acute kidney injury (AKI), there is little evidence of uptake of these models beyond their local derivation nor data on their association with patient outcomes. To systematically review published AKI prediction models across all clinical subsettings. MEDLINE via PubMed (January 1946 to April 2021) and Embase (January 1947 to April 2021) were searched using medical subject headings and text words related to AKI and prediction models. All studies that developed a prediction model for AKI, defined as a statistical model with at least 2 predictive variables to estimate future occurrence of AKI, were eligible for inclusion. There was no limitation on study populations or methodological designs. Two authors independently searched the literature, screened the studies, and extracted and analyzed the data following the Preferred Reporting Items for Systematic Review and Meta-analyses guideline. The data were pooled using a random-effects model, with subgroups defined by 4 clinical settings. Between-study heterogeneity was explored using multiple methods, and funnel plot analysis was used to identify publication bias. C statistic was used to measure the discrimination of prediction models. Of the 6955 studies initially identified through literature searching, 150 studies, with 14.4 million participants, met the inclusion criteria. The study characteristics differed widely in design, population, AKI definition, and model performance assessments. The overall pooled C statistic was 0.80 (95% CI, 0.79-0.81), with pooled C statistics in different clinical subsettings ranging from 0.78 (95% CI, 0.75-0.80) to 0.82 (95% CI, 0.78-0.86). Between-study heterogeneity was high overall and in the different clinical settings (eg, contrast medium-associated AKI: I2 = 99.9%; P < .001), and multiple methods did not identify any clear sources. A high proportion of models had a high risk of bias (126 [84.4%]) according to the Prediction Model Risk Of Bias Assessment Tool. In this study, the discrimination of the published AKI prediction models was good, reflected by high C statistics; however, the wide variation in the clinical settings, populations, and predictive variables likely drives the highly heterogenous findings that limit clinical utility. Standardized procedures for development and validation of prediction models are urgently needed.

Screening of Influencing Factors for AKI Based on Lasso Regression

Background: Acute kidney injury (AKI) is one of the common critical diseases in hospitalized patients. There are many factors affecting acute kidney injury, difficult to obtain disease data, and short treatment time window. Early detection of AKI is critical to reducing mortality. Methods: MIMIC-IV data were used to establish the prediction model of AKI in critical patients after operation. 1585 intensive care unit (ICU) patients were randomly divided into a training set (n=1109) and a prediction set (n=476) on a 7:3 basis. LASSO regression analysis was performed on the data, area under the operating characteristic curve (AUROC) was used to evaluate the model, and Hosmer-Lemeshow (HL) was used to test the degree of fit. Results: Spinal injury, implantation of temporary cardiac pacemaker, esophageal disease, contrast media, diuretic, epinephrine, aminoglycoside antibiotics, tumor chemotherapy drugs, age, gender and red blood cells were determined to be independent risk factors for acute kidney injury. The AUROC of the model in the train set was 0.901 (95% confidence interval: 0.856~0.946). In the validation set, AUROC was 0.851 (95% confidence interval: 0.808~0.894). AUROC of the above two groups indicated that the model had a good degree of differentiation, which was greater than 70% in both groups. Hosmer-Lemeshow (HL) test showed P&gt;0.05, indicating a good fit of the model. Conclusion: This prediction model can be used for the identification of post-operative AKI patients and provide a basis for clinicians or inpatients to more quickly self-measure the risk.

Machine learning for acute kidney injury: Changing the traditional disease prediction mode.

Acute kidney injury (AKI) is a serious clinical comorbidity with clear short-term and long-term prognostic implications for inpatients. The diversity of risk factors for AKI has been recognized in previous studies, and a series of predictive models have been developed using traditional statistical methods in conjunction with its preventability, but they have failed to meet the expectations in limited clinical applications, the rapid spread of electronic health records and artificial intelligence machine learning technology has brought new hope for the construction of AKI prediction models. In this article, we systematically review the definition and classification of machine learning methods, modeling ideas and evaluation methods, and the characteristics and current status of modeling studies. According to the modeling objectives, we subdivided them into critical care medical setting models, all medical environment models, special surgery models, special disease models, and special nephrotoxin exposure models. As the first review article to comprehensively summarize and analyze machine learning prediction models for AKI, we aim to objectively describe the advantages and disadvantages of machine learning approaches to modeling, and help other researchers more quickly and intuitively understand the current status of modeling research, inspire ideas and learn from experience, so as to guide and stimulate more research and more in-depth exploration in the future, which will ultimately provide greater help to improve the overall status of AKI diagnosis and treatment.

Value of Central Venous Pressure Monitoring in the Patients with Sepsis-Associated Acute Kidney Injury.

Although the measurement of central venous pressure (CVP) is a common clinical tool, the role of CVP monitoring in the outcome of sepsis is controversial because threshold values of CVP are uncertain, and there are only limited data on short-term survival of patients with septic acute kidney injury (AKI). This retrospective cohort study was based on the Medical Information Mart for Intensive Care IV (MIMIC-IV) database (source of the training dataset). Multivariate regression analysis was performed to clarify the relation between CVP measurement and clinical outcomes, and a univariate regression model after propensity score matching was utilized to validate our findings. A mortality prediction model for septic AKI and a risk stratification scoring approach were developed, and the emergency intensive care unit (eICU) database was used for external validation. Of the 9170 patients in the training set, 2446 (26.7%) underwent CVP measurement. No significant association was found between CVP monitoring and 28-day mortality among patients with septic AKI (odds ratio = 0.479; 95% confidence interval 0.213-1.076, P = 0.075), even after adjustments (propensity score matching; P = 0.178). Length of ICU stay and hospital stay was markedly reduced in patients undergoing CVP measurement within 3 hours (median 6.2 and 10.9 days, respectively, P < 0.001). The addition of the mean perfusion pressure initial, CVP, and the magnitude of the CVP change within 48 hours to the model significantly increased model discrimination (area under the receiver operating characteristic curve: 0.867 and 0.780, respectively, P < 0.001). These findings suggest that CVP measurement alone has little effect on the outcome of septic AKI. Nonetheless, initial CVP levels and the dynamic changes in CVP within the first 48 hours after ICU admission and the mean perfusion pressure initial can improve the accuracy of outcome prediction models.

Generalizability of an acute kidney injury prediction model across health systems.

Delays in the identification of acute kidney injury (AKI) in hospitalized patients are a major barrier to the development of effective interventions to treat AKI. A recent study by Tomasev and colleagues at DeepMind described a model that achieved a state-of-the-art performance in predicting AKI up to 48 hours in advance.1 Because this model was trained in a population of US Veterans that was 94% male, questions have arisen about its reproducibility and generalizability. In this study, we aimed to reproduce key aspects of this model, trained and evaluated it in a similar population of US Veterans, and evaluated its generalizability in a large academic hospital setting. We found that the model performed worse in predicting AKI in females in both populations, with miscalibration in lower stages of AKI and worse discrimination (a lower area under the curve) in higher stages of AKI. We demonstrate that while this discrepancy in performance can be largely corrected in non-Veterans by updating the original model using data from a sex-balanced academic hospital cohort, the worse model performance persists in Veterans. Our study sheds light on the importance of reproducing artificial intelligence studies, and on the complexity of discrepancies in model performance in subgroups that cannot be explained simply on the basis of sample size.

Analysis of a machine learning-based risk stratification scheme for acute kidney injury in vancomycin.

Vancomycin-associated acute kidney injury (AKI) continues to pose a major challenge to both patients and healthcare providers. The purpose of this study is to construct a machine learning framework for stratified predicting and interpreting vancomycin-associated AKI. Our study is a retrospective analysis of medical records of 724 patients who have received vancomycin therapy from 1 January 2015 through 30 September 2020. The basic clinical information, vancomycin dosage and days, comorbidities and medication, laboratory indicators of the patients were recorded. Machine learning algorithm of XGBoost was used to construct a series risk prediction model for vancomycin-associated AKI in different underlying diseases. The vast majority of sub-model performed best on the corresponding sub-dataset. Additionally, the aim of this study was to explain each model and to explore the influence of clinical variables on prediction. As the results of the analysis showed that in addition to the common indicators (serum creatinine and creatinine clearance rate), some other underappreciated indicators such as serum cystatin and cumulative days of vancomycin administration, weight and age, neutrophils and hemoglobin were the risk factors for cancer, diabetes mellitus, heptic insufficiency respectively. Stratified analysis of the comorbidities in patients with vancomycin-associated AKI further confirmed the necessity for different patient populations to be studied.

Development and validation of a nomogram for the early prediction of acute kidney injury in hospitalized COVID-19 patients.

Acute kidney injury (AKI) is a prevalent complication of coronavirus disease 2019 (COVID-19) and is closely linked with a poorer prognosis. The aim of this study was to develop and validate an easy-to-use and accurate early prediction model for AKI in hospitalized COVID-19 patients. Data from 480 COVID-19-positive patients (336 in the training set and 144 in the validation set) were obtained from the public database of the Cancer Imaging Archive (TCIA). The least absolute shrinkage and selection operator (LASSO) regression method and multivariate logistic regression were used to screen potential predictive factors to construct the prediction nomogram. Receiver operating curves (ROC), calibration curves, as well as decision curve analysis (DCA) were adopted to assess the effectiveness of the nomogram. The prognostic value of the nomogram was also examined. A predictive nomogram for AKI was developed based on arterial oxygen saturation, procalcitonin, C-reactive protein, glomerular filtration rate, and the history of coronary artery disease. In the training set, the nomogram produced an AUC of 0.831 (95% confidence interval [CI]: 0.774-0.889) with a sensitivity of 85.2% and a specificity of 69.9%. In the validation set, the nomogram produced an AUC of 0.810 (95% CI: 0.737-0.871) with a sensitivity of 77.4% and a specificity of 78.8%. The calibration curve shows that the nomogram exhibited excellent calibration and fit in both the training and validation sets. DCA suggested that the nomogram has promising clinical effectiveness. In addition, the median length of stay (m-LS) for patients in the high-risk group for AKI (risk score ≥ 0.122) was 14.0 days (95% CI: 11.3-16.7 days), which was significantly longer than 8.0 days (95% CI: 7.1-8.9 days) for patients in the low-risk group (risk score <0.122) (hazard ratio (HR): 1.98, 95% CI: 1.55-2.53, p < 0.001). Moreover, the mortality rate was also significantly higher in the high-risk group than that in the low-risk group (20.6 vs. 2.9%, odd ratio (OR):8.61, 95%CI: 3.45-21.52). The newly constructed nomogram model could accurately identify potential COVID-19 patients who may experience AKI during hospitalization at the very beginning of their admission and may be useful for informing clinical prognosis.

An Acute Kidney Injury Prediction Model for 24-hour Ultramarathon Runners.

Acute kidney injury (AKI) is frequently seen in ultrarunners, and in this study, an AKI prediction model for 24-hour ultrarunners was built based on the runner's prerace blood, urine, and body composition data. Twenty-two ultrarunners participated in the study. The risk of acquiring AKI was evaluated by a support vector machine (SVM) model, which is a statistical model commonly used for classification tasks. The inputs of the SVM model were the data collected 1 hour before the race, and the output of the SVM model was the decision of acquiring AKI. Our best AKI prediction model achieved accuracy of 96% in training and 90% in cross-validation tests. In addition, the sensitivity and specificity of the model were 90% and 100%, respectively. In accordance with the AKI prediction model components, ultra-runners are suggested to have high muscle mass and undergo regular ultra-endurance sports training to reduce the risk of acquiring AKI after participating in a 24-hour ultramarathon.

Comparison of machine learning and Logistic regression model in predicting acute kidney injury after cardiac surgery: data analysis based on MIMIC-III database

To establish an acute kidney injury (AKI) prediction model in patients after cardiac surgery by extreme gradient boosting (XGBoost) machine learning model, and to explore the risk and protective factors for AKI in patients after cardiac surgery. All patients who underwent cardiac surgery in Medical Information Mart for Intensive Care-III (MIMIC-III) database were enrolled, and they were divided into AKI group and non-AKI group according to whether AKI developed within 14 days after cardiac surgery. Their clinical characteristics were compared. Based on five-fold cross-validation, XGBoost and Logistic regression were used to establish the prediction model of AKI after cardiac surgery. And the area under the receiver operator characteristic curve (AUC) of the models was compared. The output model of XGBoost was interpreted by Shapley additive explanations (SHAP). A total of 6 912 patients were included, of which 5 681 (82.2%) developed AKI within 14 days after the operation, and 1 231 (17.8%) did not. Compared with the non-AKI group, the main characteristics of AKI group included older age [years: 68.0 (59.0, 76.0) vs. 62.0 (52.0, 71.0)], higher incidence of emergency admission and complicated with obesity and diabetes (52.4% vs. 47.8%, 9.0% vs. 4.0%, 32.0% vs. 22.2%), lower respiratory rate [RR; bpm: times/min: 17.0 (14.0, 20.0) vs. 19.0 (15.0, 22.0)], lower heart rate [HR; bpm: 80.0 (67.0, 89.0) vs. 82.0 (71.5, 93.0)], higher blood pressure [mmHg (1 mmHg ≈ 0.133 kPa): 80.0 (70.7, 90.0) vs. 78.0 (70.0, 88.0)], higher hemoglobin (Hb), blood glucose, blood K+ level and serum creatinine [SCr; Hb (g/L): 122.0 (109.0, 136.0) vs. 120.0 (106.0, 135.0), blood glucose (mmol/L): 7.3 (6.1, 8.9) vs. 6.8 (5.7, 8.5), blood K+ level (mmol/L): 4.2 (3.9, 4.7) vs. 4.2 (3.8, 4.6), SCr (μmol/L): 88.4 (70.7, 106.1) vs. 79.6 (70.7, 97.2)], lower albumin (ALB) and triacylglycerol [TG; ALB (g/L): 38.0 (35.0, 41.0) vs. 39.0 (37.0, 42.0), TG (mmol/L): 1.4 (1.0, 2.0) vs. 1.5 (1.0, 2.2)] as well as higher incidence of multiple organ dysfunction syndrome (MODS) and sepsis (30.6% vs. 16.2%, 3.3% vs. 1.9%), with significant differences (all P < 0.05). In the output model of Logistic regression, important predictors were lactic acid [Lac; odds ratio (OR) = 1.062, 95% confidence interval (95%CI) was 1.030-1.100, P = 0.005], obesity (OR = 2.234, 95%CI was 1.900-2.640, P < 0.001), male (OR = 0.858, 95%CI was 0.794-0.928, P = 0.049), diabetes (OR = 1.820, 95%CI was 1.680-1.980, P < 0.001) and emergency admission (OR = 1.278, 95%CI was 1.190-1.380, P < 0.001). Receiver operator characteristic curve (ROC curve) analysis showed that the AUC of the Logistic regression model for predicting AKI after cardiac surgery was 0.62 (95%CI was 0.61-0.67). After optimizing the XGBoost model parameters by grid search combined with five-fold cross-validation, the model was trained well with no overfitting or overfitting. ROC analysis showed that the AUC of XGBoost model for predicting AKI after cardiac surgery was 0.77 (95%CI was 0.75-0.80), which was significantly higher than that of Logistic regression model (P < 0.01). After SHAP treatment, in the output model of XGBoost, age and ALB were the most important predictors of the final outcome, where age was the risk factor (average |SHAP value| was 0.434), and ALB was the protective factor (average |SHAP value| was 0.221). Age is an important risk factor for AKI after cardiac surgery, and ALB is a protective factor. The performance of machine learning in predicting cardiac and vascular surgery-associated AKI is better than the traditional Logistic regression. XGBoost can analyze the more complex relationship between variables and outcomes, and can predict the risk of postoperative AKI more accurately and individually.